merge: resolve conflict in telegram.capabilities.json

Keep main's formatted JSON + setup section, preserve our /file/bot
allowlist entry needed for voice note file downloads.

Co-Authored-By: Claude Opus 4.6 <[email protected]>

.claude/commands/add-tool.md

@@ -286,8 +286,8 @@ impl Tool for <Name>Tool {
         false // Set true if tool processes external data
     }
 
-    fn requires_approval(&self) -> bool {
-        false // Set true if tool is destructive or contacts external services
+    fn requires_approval(&self, _params: &serde_json::Value) -> crate::tools::tool::ApprovalRequirement {
+        crate::tools::tool::ApprovalRequirement::Never // Set to UnlessAutoApproved or Always as needed
     }
 }
 ```

.env.example

@@ -2,18 +2,25 @@
 DATABASE_URL=postgres://localhost/ironclaw
 DATABASE_POOL_SIZE=10
 
-# LLM Provider (NEAR AI)
-# NEAR AI provides a unified interface to all models with user authentication
-# Session token is stored in ~/.ironclaw/session.json and managed automatically.
-# On first run, the agent will open a browser for OAuth authentication.
-NEARAI_MODEL=claude-3-5-sonnet-20241022
+# LLM Provider
+# LLM_BACKEND=nearai           # default
+# Possible values: nearai, ollama, openai_compatible, openai, anthropic, tinfoil
+
+# === NEAR AI (Chat Completions API) ===
+# Two auth modes:
+#   1. Session token (default): Uses browser OAuth (GitHub/Google) on first run.
+#      Session token stored in ~/.ironclaw/session.json automatically.
+#      Base URL defaults to https://private.near.ai
+#   2. API key: Set NEARAI_API_KEY to use API key auth from cloud.near.ai.
+#      Base URL defaults to https://cloud-api.near.ai
+NEARAI_MODEL=zai-org/GLM-5-FP8
 NEARAI_BASE_URL=https://private.near.ai
 NEARAI_AUTH_URL=https://private.near.ai
-# NEARAI_SESSION_PATH=~/.ironclaw/session.json  # optional, default shown
+# NEARAI_SESSION_TOKEN=sess_...                  # hosting providers: set this
+# NEARAI_SESSION_PATH=~/.ironclaw/session.json   # optional, default shown
+# NEARAI_API_KEY=...                             # API key from cloud.near.ai
 
 # Local LLM Providers (Ollama, LM Studio, vLLM, LiteLLM)
-# LLM_BACKEND=nearai           # default
-# Possible values: nearai, ollama, openai_compatible, openai, anthropic
 
 # === Ollama ===
 # OLLAMA_MODEL=llama3.2
@@ -26,11 +33,26 @@ NEARAI_AUTH_URL=https://private.near.ai
 # LLM_BASE_URL=http://localhost:1234/v1
 # LLM_API_KEY=sk-...                        # optional for local servers
 
-# === OpenRouter (via OpenAI-compatible) ===
-# LLM_MODEL=anthropic/claude-sonnet-4
+# === OpenRouter (300+ models via OpenAI-compatible) ===
+# LLM_MODEL=anthropic/claude-sonnet-4       # see openrouter.ai/models for IDs
 # LLM_BACKEND=openai_compatible
 # LLM_BASE_URL=https://openrouter.ai/api/v1
 # LLM_API_KEY=sk-or-...
+# LLM_EXTRA_HEADERS=HTTP-Referer:https://myapp.com,X-Title:MyApp
+
+# === Together AI (via OpenAI-compatible) ===
+# LLM_MODEL=meta-llama/Llama-3.3-70B-Instruct-Turbo
+# LLM_BACKEND=openai_compatible
+# LLM_BASE_URL=https://api.together.xyz/v1
+# LLM_API_KEY=...
+
+# === Fireworks AI (via OpenAI-compatible) ===
+# LLM_MODEL=accounts/fireworks/models/llama4-maverick-instruct-basic
+# LLM_BACKEND=openai_compatible
+# LLM_BASE_URL=https://api.fireworks.ai/inference/v1
+# LLM_API_KEY=fw_...
+
+# For full provider setup guide see docs/LLM_PROVIDERS.md
 
 
 # Channel Configuration
@@ -68,6 +90,12 @@ HEARTBEAT_INTERVAL_SECS=1800
 HEARTBEAT_NOTIFY_CHANNEL=cli
 HEARTBEAT_NOTIFY_USER=default
 
+# Memory hygiene settings (automatic cleanup of stale workspace documents)
+# Runs on each heartbeat tick; identity files (IDENTITY.md, SOUL.md) are never deleted
+# MEMORY_HYGIENE_ENABLED=true
+# MEMORY_HYGIENE_RETENTION_DAYS=30     # delete daily/ docs older than this many days
+# MEMORY_HYGIENE_CADENCE_HOURS=12      # minimum hours between cleanup passes
+
 # Safety settings
 SAFETY_MAX_OUTPUT_LENGTH=100000
 SAFETY_INJECTION_CHECK_ENABLED=true

.gitattributes

@@ -0,0 +1 @@
+tests/test-pages/**/*.html linguist-generated=true

.github/labeler.yml

@@ -0,0 +1,166 @@
+# Scope labels for actions/labeler@v6
+# Maps file path globs to scope labels. Multiple labels can apply per PR.
+
+"scope: agent":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/agent/**
+
+"scope: channel":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/channels/channel.rs
+          - src/channels/manager.rs
+          - src/channels/mod.rs
+
+"scope: channel/cli":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/channels/cli/**
+          - src/cli/**
+
+"scope: channel/web":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/channels/web/**
+
+"scope: channel/wasm":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/channels/wasm/**
+
+"scope: tool":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/tools/tool.rs
+          - src/tools/registry.rs
+          - src/tools/mod.rs
+          - src/tools/sandbox.rs
+
+"scope: tool/builtin":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/tools/builtin/**
+
+"scope: tool/wasm":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/tools/wasm/**
+
+"scope: tool/mcp":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/tools/mcp/**
+
+"scope: tool/builder":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/tools/builder/**
+
+"scope: db":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/db/mod.rs
+
+"scope: db/postgres":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/db/postgres.rs
+          - migrations/**
+
+"scope: db/libsql":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/db/libsql_backend.rs
+          - src/db/libsql_migrations.rs
+
+"scope: safety":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/safety/**
+
+"scope: llm":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/llm/**
+
+"scope: workspace":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/workspace/**
+
+"scope: orchestrator":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/orchestrator/**
+
+"scope: worker":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/worker/**
+
+"scope: secrets":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/secrets/**
+
+"scope: config":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/config.rs
+          - src/settings.rs
+
+"scope: extensions":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/extensions/**
+
+"scope: setup":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/setup/**
+
+"scope: evaluation":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/evaluation/**
+
+"scope: estimation":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/estimation/**
+
+"scope: sandbox":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/sandbox/**
+          - Dockerfile*
+
+"scope: hooks":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/hooks/**
+
+"scope: pairing":
+  - changed-files:
+      - any-glob-to-any-file:
+          - src/pairing/**
+
+"scope: ci":
+  - changed-files:
+      - any-glob-to-any-file:
+          - .github/workflows/**
+          - .github/scripts/**
+
+"scope: docs":
+  - changed-files:
+      - any-glob-to-any-file:
+          - "**/*.md"
+          - docs/**
+          - LICENSE*
+
+"scope: dependencies":
+  - changed-files:
+      - any-glob-to-any-file:
+          - Cargo.toml
+          - Cargo.lock

.github/scripts/create-labels.sh

@@ -0,0 +1,71 @@
+#!/usr/bin/env bash
+# Idempotent label bootstrap for IronClaw PR automation.
+# Uses `gh label create --force` so it can be re-run safely.
+#
+# Usage: bash .github/scripts/create-labels.sh
+# Requires: gh CLI authenticated with repo scope
+
+set -euo pipefail
+
+if ! command -v gh &>/dev/null; then
+  echo "Error: gh CLI is required. Install from https://cli.github.com" >&2
+  exit 1
+fi
+
+create() {
+  local name="$1" color="$2" description="$3"
+  gh label create "$name" --color "$color" --description "$description" --force
+}
+
+echo "==> Creating size labels..."
+create "size: XS"    "F9D0C4" "< 10 changed lines (excluding docs)"
+create "size: S"     "F5A3A3" "10-49 changed lines"
+create "size: M"     "E57373" "50-199 changed lines"
+create "size: L"     "D32F2F" "200-499 changed lines"
+create "size: XL"    "B71C1C" "500+ changed lines"
+
+echo "==> Creating risk labels..."
+create "risk: low"    "4CAF50" "Changes to docs, tests, or low-risk modules"
+create "risk: medium" "FFC107" "Business logic, config, or moderate-risk modules"
+create "risk: high"   "F44336" "Safety, secrets, auth, or critical infrastructure"
+create "risk: manual" "9E9E9E" "Risk level set manually (sticky, not overwritten)"
+
+echo "==> Creating scope labels..."
+create "scope: agent"         "006B75" "Agent core (agent loop, router, scheduler)"
+create "scope: channel"       "00838F" "Channel infrastructure"
+create "scope: channel/cli"   "00897B" "TUI / CLI channel"
+create "scope: channel/web"   "00796B" "Web gateway channel"
+create "scope: channel/wasm"  "00695C" "WASM channel runtime"
+create "scope: tool"          "1565C0" "Tool infrastructure"
+create "scope: tool/builtin"  "1976D2" "Built-in tools"
+create "scope: tool/wasm"     "1E88E5" "WASM tool sandbox"
+create "scope: tool/mcp"      "2196F3" "MCP client"
+create "scope: tool/builder"  "42A5F5" "Dynamic tool builder"
+create "scope: db"            "4A148C" "Database trait / abstraction"
+create "scope: db/postgres"   "6A1B9A" "PostgreSQL backend"
+create "scope: db/libsql"     "7B1FA2" "libSQL / Turso backend"
+create "scope: safety"        "880E4F" "Prompt injection defense"
+create "scope: llm"           "4527A0" "LLM integration"
+create "scope: workspace"     "283593" "Persistent memory / workspace"
+create "scope: orchestrator"  "0D47A1" "Container orchestrator"
+create "scope: worker"        "01579B" "Container worker"
+create "scope: secrets"       "BF360C" "Secrets management"
+create "scope: config"        "E65100" "Configuration"
+create "scope: extensions"    "33691E" "Extension management"
+create "scope: setup"         "827717" "Onboarding / setup"
+create "scope: evaluation"    "558B2F" "Success evaluation"
+create "scope: estimation"    "9E9D24" "Cost/time estimation"
+create "scope: sandbox"       "00BFA5" "Docker sandbox"
+create "scope: hooks"         "6D4C41" "Git/event hooks"
+create "scope: pairing"       "4E342E" "Pairing mode"
+create "scope: ci"            "546E7A" "CI/CD workflows"
+create "scope: docs"          "78909C" "Documentation"
+create "scope: dependencies"  "90A4AE" "Dependency updates"
+
+echo "==> Creating contributor labels..."
+create "contributor: new"         "FFF9C4" "First-time contributor"
+create "contributor: regular"     "FFE082" "2-5 merged PRs"
+create "contributor: experienced" "FFB74D" "6-19 merged PRs"
+create "contributor: core"        "FF8A65" "20+ merged PRs"
+
+echo "Done. All labels created/updated."

.github/scripts/pr-labeler.sh

@@ -0,0 +1,139 @@
+#!/usr/bin/env bash
+# Classify a PR by size, risk, and contributor tier.
+# Called by the pr-label-classify workflow.
+#
+# Inputs (env vars):
+#   PR_NUMBER  — pull request number
+#   REPO       — owner/repo (e.g. "user/ironclaw")
+#
+# Requires: gh CLI, jq
+
+set -euo pipefail
+
+PR_NUMBER="${PR_NUMBER:?PR_NUMBER is required}"
+REPO="${REPO:?REPO is required}"
+
+# ─── helpers ────────────────────────────────────────────────────────────────
+
+# Remove all labels in a dimension except the desired one.
+# Usage: set_exclusive_label "size" "size: M"
+set_exclusive_label() {
+  local prefix="$1" desired="$2"
+
+  # Fetch current labels on the PR
+  local current
+  current=$(gh pr view "$PR_NUMBER" --repo "$REPO" --json labels --jq '.labels[].name')
+
+  # Remove any existing label with the same prefix
+  while IFS= read -r label; do
+    [[ -z "$label" ]] && continue
+    if [[ "$label" == "${prefix}:"* && "$label" != "$desired" ]]; then
+      gh pr edit "$PR_NUMBER" --repo "$REPO" --remove-label "$label" 2>/dev/null || true
+    fi
+  done <<< "$current"
+
+  # Add the desired label
+  gh pr edit "$PR_NUMBER" --repo "$REPO" --add-label "$desired"
+}
+
+# ─── size ───────────────────────────────────────────────────────────────────
+
+classify_size() {
+  # Sum changed lines across non-doc files
+  local total
+  total=$(gh api "repos/${REPO}/pulls/${PR_NUMBER}/files" \
+    --paginate --jq '
+      [.[] | select(.filename | test("\\.(md|txt|rst|adoc)$") | not) | .changes]
+      | add // 0
+    ')
+
+  local label
+  if   (( total < 10 ));  then label="size: XS"
+  elif (( total < 50 ));  then label="size: S"
+  elif (( total < 200 )); then label="size: M"
+  elif (( total < 500 )); then label="size: L"
+  else                         label="size: XL"
+  fi
+
+  echo "Size: ${total} changed lines -> ${label}"
+  set_exclusive_label "size" "$label"
+}
+
+# ─── risk ───────────────────────────────────────────────────────────────────
+
+classify_risk() {
+  # If "risk: manual" is present, skip — it's a sticky override
+  local current
+  current=$(gh pr view "$PR_NUMBER" --repo "$REPO" --json labels --jq '.labels[].name')
+  if echo "$current" | grep -qx "risk: manual"; then
+    echo "Risk: skipped (manual override)"
+    return
+  fi
+
+  # Fetch changed file paths
+  local files
+  files=$(gh api "repos/${REPO}/pulls/${PR_NUMBER}/files" \
+    --paginate --jq '.[].filename')
+
+  local risk="low"
+
+  while IFS= read -r file; do
+    [[ -z "$file" ]] && continue
+
+    case "$file" in
+      # High risk: safety, secrets, auth, crypto, setup, orchestrator auth
+      src/safety/*|src/secrets/*|src/llm/session.rs|src/orchestrator/auth.rs|\
+      src/channels/web/auth.rs|src/setup/*)
+        risk="high"
+        break  # can't go higher
+        ;;
+
+      # Medium risk: agent core, config, database, worker, tools, channels
+      src/agent/*|src/config.rs|src/settings.rs|src/db/*|src/worker/*|\
+      src/tools/*|src/channels/*|src/orchestrator/*|src/context/*|\
+      src/hooks/*|src/sandbox/*|src/extensions/*|Cargo.toml|\
+      .github/workflows/*)
+        # Only upgrade, never downgrade
+        [[ "$risk" != "high" ]] && risk="medium"
+        ;;
+
+      # Low risk: docs, tests, estimation, evaluation, history, etc.
+      *)
+        ;;
+    esac
+  done <<< "$files"
+
+  echo "Risk: ${risk}"
+  set_exclusive_label "risk" "risk: ${risk}"
+}
+
+# ─── contributor tier ───────────────────────────────────────────────────────
+
+classify_contributor() {
+  # Get PR author
+  local author
+  author=$(gh pr view "$PR_NUMBER" --repo "$REPO" --json author --jq '.author.login')
+
+  # Count merged PRs by this author in this repo
+  local count
+  count=$(gh pr list --repo "$REPO" --state merged --author "$author" \
+    --limit 100 --json number --jq 'length')
+
+  local label
+  if   (( count == 0 )); then label="contributor: new"
+  elif (( count < 6 ));  then label="contributor: regular"
+  elif (( count < 20 )); then label="contributor: experienced"
+  else                        label="contributor: core"
+  fi
+
+  echo "Contributor: ${author} has ${count} merged PRs -> ${label}"
+  set_exclusive_label "contributor" "$label"
+}
+
+# ─── main ───────────────────────────────────────────────────────────────────
+
+echo "Classifying PR #${PR_NUMBER} in ${REPO}..."
+classify_size
+classify_risk
+classify_contributor
+echo "Done."

.github/workflows/pr-label-classify.yml

@@ -0,0 +1,26 @@
+name: "PR: Classify (Size, Risk, Contributor)"
+
+on:
+  pull_request_target:
+    types: [opened, synchronize, reopened]
+
+permissions:
+  contents: read
+  pull-requests: write
+  issues: read           # needed for search/issues API (contributor count)
+
+jobs:
+  classify:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout base branch
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ github.event.pull_request.base.ref }}
+
+      - name: Classify PR
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          PR_NUMBER: ${{ github.event.pull_request.number }}
+          REPO: ${{ github.repository }}
+        run: bash .github/scripts/pr-labeler.sh

.github/workflows/pr-label-scope.yml

@@ -0,0 +1,18 @@
+name: "PR: Scope Labels"
+
+on:
+  pull_request_target:
+    types: [opened, synchronize, reopened]
+
+permissions:
+  contents: read
+  pull-requests: write
+
+jobs:
+  scope:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/labeler@v5
+        with:
+          configuration-path: .github/labeler.yml
+          sync-labels: false          # additive only — never remove scope labels

.github/workflows/release.yml

@@ -214,14 +214,113 @@ jobs:
           path: |
             ${{ steps.cargo-dist.outputs.paths }}
             ${{ env.BUILD_MANIFEST_NAME }}
+  # Build WASM extension bundles (tar.gz with .wasm + .capabilities.json)
+  build-wasm-extensions:
+    needs:
+      - plan
+    if: ${{ needs.plan.outputs.publishing == 'true' }}
+    runs-on: "ubuntu-22.04"
+    env:
+      GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          persist-credentials: false
+          submodules: recursive
+      - name: Install Rust toolchain + wasm target
+        run: |
+          rustup target add wasm32-wasip2
+          cargo install cargo-component --locked || true
+      - uses: swatinem/rust-cache@v2
+        with:
+          key: wasm-extensions
+      - name: Build and package WASM extensions
+        shell: bash
+        run: |
+          set -euo pipefail
+          mkdir -p target/wasm-bundles
+
+          # Process each manifest in registry/tools/ and registry/channels/
+          for manifest in registry/tools/*.json registry/channels/*.json; do
+            [ -f "$manifest" ] || continue
+
+            name=$(jq -r '.name' "$manifest")
+            source_dir=$(jq -r '.source.dir' "$manifest")
+            caps_file=$(jq -r '.source.capabilities' "$manifest")
+            crate_name=$(jq -r '.source.crate_name' "$manifest")
+
+            if [ ! -d "$source_dir" ]; then
+              echo "::warning::Source dir '$source_dir' not found for '$name', skipping"
+              continue
+            fi
+
+            echo "=== Building $name from $source_dir ==="
+
+            # Build WASM component
+            cargo component build --release --manifest-path "$source_dir/Cargo.toml" || {
+              echo "::warning::Build failed for '$name', skipping"
+              continue
+            }
+
+            # Find the built WASM file (Cargo uses underscores in artifact names)
+            wasm_artifact="${crate_name//-/_}"
+            wasm_path=""
+            for target_dir in wasm32-wasip2 wasm32-wasip1 wasm32-wasi; do
+              candidate="$source_dir/target/$target_dir/release/${wasm_artifact}.wasm"
+              if [ -f "$candidate" ]; then
+                wasm_path="$candidate"
+                break
+              fi
+            done
+
+            if [ -z "$wasm_path" ]; then
+              echo "::warning::No WASM output found for '$name', skipping"
+              continue
+            fi
+
+            # Copy files with standardized names for the archive
+            cp "$wasm_path" "target/wasm-bundles/${name}.wasm"
+
+            caps_path="$source_dir/$caps_file"
+            if [ -f "$caps_path" ]; then
+              cp "$caps_path" "target/wasm-bundles/${name}.capabilities.json"
+            else
+              echo "::warning::No capabilities file at '$caps_path' for '$name'"
+            fi
+
+            # Create tar.gz bundle
+            bundle="target/wasm-bundles/${name}-wasm32-wasip2.tar.gz"
+            (cd target/wasm-bundles && if [ -f "${name}.capabilities.json" ]; then tar czf "${name}-wasm32-wasip2.tar.gz" "${name}.wasm" "${name}.capabilities.json"; else tar czf "${name}-wasm32-wasip2.tar.gz" "${name}.wasm"; fi)
+
+            # Compute SHA256
+            sha256=$(sha256sum "$bundle" | cut -d' ' -f1)
+            echo "$sha256  ${name}-wasm32-wasip2.tar.gz" >> target/wasm-bundles/checksums.txt
+
+            # Clean up intermediate files
+            rm -f "target/wasm-bundles/${name}.wasm" "target/wasm-bundles/${name}.capabilities.json"
+
+            echo "  -> $bundle ($sha256)"
+          done
+
+          echo "=== WASM bundles built ==="
+          ls -la target/wasm-bundles/
+      - name: "Upload WASM bundles"
+        uses: actions/upload-artifact@v4
+        with:
+          name: artifacts-wasm-extensions
+          path: |
+            target/wasm-bundles/*.tar.gz
+            target/wasm-bundles/checksums.txt
+
   # Determines if we should publish/announce
   host:
     needs:
       - plan
       - build-local-artifacts
       - build-global-artifacts
-    # Only run if we're "publishing", and only if plan, local and global didn't fail (skipped is fine)
-    if: ${{ always() && needs.plan.result == 'success' && needs.plan.outputs.publishing == 'true' && (needs.build-global-artifacts.result == 'skipped' || needs.build-global-artifacts.result == 'success') && (needs.build-local-artifacts.result == 'skipped' || needs.build-local-artifacts.result == 'success') }}
+      - build-wasm-extensions
+    # Only run if we're "publishing", and only if plan, local, global, and wasm didn't fail (skipped is fine)
+    if: ${{ always() && needs.plan.result == 'success' && needs.plan.outputs.publishing == 'true' && (needs.build-global-artifacts.result == 'skipped' || needs.build-global-artifacts.result == 'success') && (needs.build-local-artifacts.result == 'skipped' || needs.build-local-artifacts.result == 'success') && (needs.build-wasm-extensions.result == 'skipped' || needs.build-wasm-extensions.result == 'success') }}
     env:
       GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
     runs-on: "ubuntu-22.04"

CHANGELOG.md

@@ -7,6 +7,40 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.9.0](https://github.com/nearai/ironclaw/compare/v0.8.0...v0.9.0) - 2026-02-21
+
+### Added
+
+- add TEE attestation shield to web gateway UI ([#275](https://github.com/nearai/ironclaw/pull/275))
+- configurable tool iterations, auto-approve, and policy fix ([#251](https://github.com/nearai/ironclaw/pull/251))
+
+### Fixed
+
+- add X-Accel-Buffering header to SSE endpoints ([#277](https://github.com/nearai/ironclaw/pull/277))
+
+## [0.8.0](https://github.com/nearai/ironclaw/compare/ironclaw-v0.7.0...ironclaw-v0.8.0) - 2026-02-20
+
+### Added
+
+- extension registry with metadata catalog and onboarding integration ([#238](https://github.com/nearai/ironclaw/pull/238))
+- *(models)* add GPT-5.3 Codex, full GPT-5.x family, Claude 4.x series, o4-mini ([#197](https://github.com/nearai/ironclaw/pull/197))
+- wire memory hygiene into the heartbeat loop ([#195](https://github.com/nearai/ironclaw/pull/195))
+
+### Fixed
+
+- persist WASM channel workspace writes across callbacks ([#264](https://github.com/nearai/ironclaw/pull/264))
+- consolidate per-module ENV_MUTEX into crate-wide test lock ([#246](https://github.com/nearai/ironclaw/pull/246))
+- remove auto-proceed fake user message injection from agent loop ([#255](https://github.com/nearai/ironclaw/pull/255))
+- onboarding errors reset flow and remote server auth (#185, #186) ([#248](https://github.com/nearai/ironclaw/pull/248))
+- parallelize tool call execution via JoinSet ([#219](https://github.com/nearai/ironclaw/pull/219)) ([#252](https://github.com/nearai/ironclaw/pull/252))
+- prevent pipe deadlock in shell command execution ([#140](https://github.com/nearai/ironclaw/pull/140))
+- persist turns after approval and add agent-level tests ([#250](https://github.com/nearai/ironclaw/pull/250))
+
+### Other
+
+- add automated PR labeling system ([#253](https://github.com/nearai/ironclaw/pull/253))
+- update CLAUDE.md for recently merged features ([#183](https://github.com/nearai/ironclaw/pull/183))
+
 ## [0.7.0](https://github.com/nearai/ironclaw/compare/ironclaw-v0.6.0...ironclaw-v0.7.0) - 2026-02-19
 
 ### Added

CLAUDE.md

@@ -13,14 +13,17 @@
 ### Features
 - **Multi-channel input**: TUI (Ratatui), HTTP webhooks, WASM channels (Telegram, Slack), web gateway
 - **Parallel job execution** with state machine and self-repair for stuck jobs
-- **Sandbox execution**: Docker container isolation with orchestrator/worker pattern
+- **Sandbox execution**: Docker container isolation with network proxy and credential injection
 - **Claude Code mode**: Delegate jobs to Claude CLI inside containers
+- **Skills system**: SKILL.md prompt extensions with trust model, tool attenuation, and ClawHub registry
 - **Routines**: Scheduled (cron) and reactive (event, webhook) task execution
 - **Web gateway**: Browser UI with SSE/WebSocket real-time streaming
 - **Extension management**: Install, auth, activate MCP/WASM extensions
 - **Extensible tools**: Built-in tools, WASM sandbox, MCP client, dynamic builder
 - **Persistent memory**: Workspace with hybrid search (FTS + vector via RRF)
-- **Prompt injection defense**: Sanitizer, validator, policy rules, leak detection
+- **Prompt injection defense**: Sanitizer, validator, policy rules, leak detection, shell env scrubbing
+- **Multi-provider LLM**: NEAR AI, OpenAI, Anthropic, Ollama, OpenAI-compatible, Tinfoil private inference
+- **Setup wizard**: 7-step interactive onboarding for first-run configuration
 - **Heartbeat system**: Proactive periodic execution with checklist
 
 ## Build & Test
@@ -29,7 +32,7 @@
 # Format code
 cargo fmt
 
-# Lint (address warnings before committing)
+# Lint (fix ALL warnings before committing, including pre-existing ones)
 cargo clippy --all --benches --tests --examples --all-features
 
 # Run all tests
@@ -64,6 +67,7 @@ src/
 │   ├── context_monitor.rs # Memory pressure detection
 │   ├── undo.rs         # Turn-based undo/redo with checkpoints
 │   ├── submission.rs   # Submission parsing (undo, redo, compact, clear, etc.)
+│   ├── dispatcher.rs   # Skill-aware job dispatching
 │   ├── task.rs         # Sub-task execution framework
 │   ├── routine.rs      # Routine types (Trigger, Action, Guardrails)
 │   └── routine_engine.rs # Routine execution (cron ticker, event matcher)
@@ -113,11 +117,18 @@ src/
 │   ├── policy.rs       # PolicyRule system with severity/actions
 │   └── leak_detector.rs # Secret detection (API keys, tokens, etc.)
 │
-├── llm/                # LLM integration (NEAR AI only)
+├── llm/                # LLM integration (multi-provider)
+│   ├── mod.rs          # Provider factory, LlmBackend enum
 │   ├── provider.rs     # LlmProvider trait, message types
-│   ├── nearai.rs       # NEAR AI chat-api implementation
+│   ├── nearai_chat.rs  # NEAR AI Chat Completions provider (session token + API key auth)
 │   ├── reasoning.rs    # Planning, tool selection, evaluation
-│   └── session.rs      # Session token management with auto-renewal
+│   ├── session.rs      # Session token management with auto-renewal
+│   ├── circuit_breaker.rs # Circuit breaker for provider failures
+│   ├── retry.rs        # Retry with exponential backoff
+│   ├── failover.rs     # Multi-provider failover chain
+│   ├── response_cache.rs # LLM response caching
+│   ├── costs.rs        # Token cost tracking
+│   └── rig_adapter.rs  # Rig framework adapter
 │
 ├── tools/              # Extensible tool system
 │   ├── tool.rs         # Tool trait, ToolOutput, ToolError
@@ -131,6 +142,7 @@ src/
 │   │   ├── job.rs      # CreateJob, ListJobs, JobStatus, CancelJob
 │   │   ├── routine.rs  # routine_create/list/update/delete/history
 │   │   ├── extension_tools.rs # Extension install/auth/activate/remove
+│   │   ├── skill_tools.rs # skill_list/search/install/remove tools
 │   │   └── marketplace.rs, ecommerce.rs, taskrabbit.rs, restaurant.rs (stubs)
 │   ├── builder/        # Dynamic tool building
 │   │   ├── core.rs     # BuildRequirement, SoftwareType, Language
@@ -180,11 +192,38 @@ src/
 │   ├── success.rs      # SuccessEvaluator trait, RuleBasedEvaluator, LlmEvaluator
 │   └── metrics.rs      # MetricsCollector, QualityMetrics
 │
+├── sandbox/            # Docker execution sandbox
+│   ├── mod.rs          # Public API, default allowlist
+│   ├── config.rs       # SandboxConfig, SandboxPolicy enum
+│   ├── manager.rs      # SandboxManager orchestration
+│   ├── container.rs    # ContainerRunner, Docker lifecycle
+│   ├── error.rs        # SandboxError types
+│   └── proxy/          # Network proxy for containers
+│       ├── mod.rs      # NetworkProxyBuilder
+│       ├── http.rs     # HttpProxy, CredentialResolver trait
+│       ├── policy.rs   # NetworkPolicyDecider trait
+│       └── allowlist.rs # DomainAllowlist validation
+│
 ├── secrets/            # Secrets management
 │   ├── crypto.rs       # AES-256-GCM encryption
 │   ├── store.rs        # Secret storage
 │   └── types.rs        # Credential types
 │
+├── setup/              # Onboarding wizard (spec: src/setup/README.md)
+│   ├── mod.rs          # Entry point, check_onboard_needed()
+│   ├── wizard.rs       # 7-step interactive wizard
+│   ├── channels.rs     # Channel setup helpers
+│   └── prompts.rs      # Terminal prompts (select, confirm, secret)
+│
+├── skills/             # SKILL.md prompt extension system
+│   ├── mod.rs          # Core types (SkillTrust, LoadedSkill)
+│   ├── registry.rs     # SkillRegistry: discover, install, remove
+│   ├── selector.rs     # Deterministic scoring prefilter
+│   ├── attenuation.rs  # Trust-based tool ceiling
+│   ├── gating.rs       # Requirement checks (bins, env, config)
+│   ├── parser.rs       # SKILL.md frontmatter + markdown parser
+│   └── catalog.rs      # ClawHub registry client
+│
 └── history/            # Persistence
     ├── store.rs        # PostgreSQL repositories
     └── analytics.rs    # Aggregation queries (JobStats, ToolStats)
@@ -214,6 +253,7 @@ When designing new features or systems, always prefer generic/extensible archite
 - `LlmProvider` - Add new LLM backends
 - `SuccessEvaluator` - Custom evaluation logic
 - `EmbeddingProvider` - Add embedding backends (workspace search)
+- `NetworkPolicyDecider` - Custom network access policies for sandbox containers
 
 ### Tool Implementation
 ```rust
@@ -252,6 +292,43 @@ Pending -> InProgress -> Completed -> Submitted -> Accepted
                               \-> Failed
 ```
 
+### Code Style
+
+- Use `crate::` imports, not `super::`
+- No `pub use` re-exports unless exposing to downstream consumers
+- Prefer strong types over strings (enums, newtypes)
+- Keep functions focused, extract helpers when logic is reused
+- Comments for non-obvious logic only
+
+### Review & Fix Discipline
+
+Hard-won lessons from code review -- follow these when fixing bugs or addressing review feedback.
+
+**Fix the pattern, not just the instance:** When a reviewer flags a bug (e.g., TOCTOU race in INSERT + SELECT-back), search the entire codebase for all instances of that same pattern. A fix in `SecretsStore::create()` that doesn't also fix `WasmToolStore::store()` is half a fix.
+
+**Propagate architectural fixes to satellite types:** If a core type changes its concurrency model (e.g., `LibSqlBackend` switches to connection-per-operation), every type that was handed a resource from the old model (e.g., `LibSqlSecretsStore`, `LibSqlWasmToolStore` holding a single `Connection`) must also be updated. Grep for the old type across the codebase.
+
+**Schema translation is more than DDL:** When translating a database schema between backends (PostgreSQL to libSQL, etc.), check for:
+- **Indexes** -- diff `CREATE INDEX` statements between the two schemas
+- **Seed data** -- check for `INSERT INTO` in migrations (e.g., `leak_detection_patterns`)
+- **Semantic differences** -- document where SQL functions behave differently (e.g., `json_patch` vs `jsonb_set`)
+
+**Feature flag testing:** When adding feature-gated code, test compilation with each feature in isolation:
+```bash
+cargo check                                          # default features
+cargo check --no-default-features --features libsql  # libsql only
+cargo check --all-features                           # all features
+```
+Dead code behind the wrong `#[cfg]` gate will only show up when building with a single feature.
+
+**Zero clippy warnings policy:** Fix ALL clippy warnings before committing, including pre-existing ones in files you didn't change. Never leave warnings behind — treat `cargo clippy` output as a zero-tolerance gate.
+
+**Mechanical verification before committing:** Run these checks on changed files before committing:
+- `cargo clippy --all --benches --tests --examples --all-features` -- zero warnings
+- `grep -rnE '\.unwrap\(|\.expect\(' <files>` -- no panics in production
+- `grep -rn 'super::' <files>` -- use `crate::` imports
+- If you fixed a pattern bug, `grep` for other instances of that pattern across `src/`
+
 ## Configuration
 
 Environment variables (see `.env.example`):
@@ -263,10 +340,14 @@ LIBSQL_PATH=~/.ironclaw/ironclaw.db    # libSQL local path (default)
 # LIBSQL_URL=libsql://xxx.turso.io    # Turso cloud (optional)
 # LIBSQL_AUTH_TOKEN=xxx                # Required with LIBSQL_URL
 
-# NEAR AI (required)
-NEARAI_SESSION_TOKEN=sess_...
-NEARAI_MODEL=claude-3-5-sonnet-20241022
+# NEAR AI (when LLM_BACKEND=nearai, the default)
+# Two auth modes: session token (default) or API key
+# Session token auth (default): uses browser OAuth on first run
+NEARAI_SESSION_TOKEN=sess_...           # hosting providers: set this
 NEARAI_BASE_URL=https://private.near.ai
+# API key auth: set NEARAI_API_KEY, base URL defaults to cloud-api.near.ai
+# NEARAI_API_KEY=...                    # API key from cloud.near.ai
+NEARAI_MODEL=claude-3-5-sonnet-20241022
 
 # Agent settings
 AGENT_NAME=ironclaw
@@ -297,6 +378,10 @@ SANDBOX_ENABLED=true
 SANDBOX_IMAGE=ironclaw-worker:latest
 SANDBOX_MEMORY_LIMIT_MB=512
 SANDBOX_TIMEOUT_SECS=1800
+SANDBOX_CPU_LIMIT=1.0                  # CPU cores per container
+SANDBOX_NETWORK_PROXY=true             # Enable network proxy for containers
+SANDBOX_PROXY_PORT=8080                # Proxy listener port
+SANDBOX_DEFAULT_POLICY=workspace_write # ReadOnly, WorkspaceWrite, FullAccess
 
 # Claude Code mode (runs inside sandbox containers)
 CLAUDE_CODE_ENABLED=false
@@ -308,16 +393,29 @@ CLAUDE_CODE_CONFIG_DIR=/home/worker/.claude
 ROUTINES_ENABLED=true
 ROUTINES_CRON_INTERVAL=60            # Tick interval in seconds
 ROUTINES_MAX_CONCURRENT=3
+
+# Skills system
+SKILLS_ENABLED=true
+SKILLS_MAX_TOKENS=4000                 # Max prompt budget per turn
+SKILLS_CATALOG_URL=https://clawhub.dev # ClawHub registry URL
+SKILLS_AUTO_DISCOVER=true              # Scan skill directories on startup
+
+# Tinfoil private inference
+TINFOIL_API_KEY=...                    # Required when LLM_BACKEND=tinfoil
+TINFOIL_MODEL=kimi-k2-5               # Default model
 ```
 
-### NEAR AI Provider
+### LLM Providers
 
-Uses the NEAR AI chat-api (`https://api.near.ai/v1/responses`) which provides:
-- Unified access to multiple models (OpenAI, Anthropic, etc.)
-- User authentication via session tokens
-- Usage tracking and billing through NEAR AI
+IronClaw supports multiple LLM backends via the `LLM_BACKEND` env var: `nearai` (default), `openai`, `anthropic`, `ollama`, `openai_compatible`, and `tinfoil`.
 
-Session tokens have the format `sess_xxx` (37 characters). They are authenticated against the NEAR AI auth service.
+**NEAR AI** -- Uses the Chat Completions API with dual auth support. Session token auth (default): authenticates with session tokens (`sess_xxx`) obtained via browser OAuth (GitHub/Google), base URL defaults to `https://private.near.ai`. API key auth: set `NEARAI_API_KEY` (from `cloud.near.ai`), base URL defaults to `https://cloud-api.near.ai`. Both modes use the same Chat Completions endpoint. Tool messages are flattened to plain text for compatibility. Set `NEARAI_SESSION_TOKEN` env var for hosting providers that inject tokens via environment.
+
+**NEAR AI Cloud** -- Uses the OpenAI-compatible Chat Completions API (`https://cloud-api.near.ai/v1/chat/completions`). Authenticates with API keys from `cloud.near.ai`. Auto-selected when `NEARAI_API_KEY` is set (or explicitly via `NEARAI_API_MODE=chat_completions`). Tool messages are flattened to plain text for compatibility. Configure with `NEARAI_API_KEY` and `NEARAI_BASE_URL` (default: `https://cloud-api.near.ai`).
+
+**OpenAI-compatible** -- Any endpoint that speaks the OpenAI API (vLLM, LiteLLM, OpenRouter, etc.). Configure with `LLM_BASE_URL`, `LLM_API_KEY` (optional), `LLM_MODEL`. Set `LLM_EXTRA_HEADERS` to inject custom HTTP headers into every request (format: `Key:Value,Key2:Value2`), useful for OpenRouter attribution headers like `HTTP-Referer` and `X-Title`.
+
+**Tinfoil** -- Private inference via `https://inference.tinfoil.sh/v1`. Runs models inside hardware-attested TEEs so neither Tinfoil nor the cloud provider can see prompts or responses. Uses the OpenAI-compatible Chat Completions API. Configure with `TINFOIL_API_KEY` and `TINFOIL_MODEL` (default: `kimi-k2-5`).
 
 ## Database
 
@@ -386,22 +484,7 @@ Both backends implement this trait. PostgreSQL delegates to the existing `Store`
 - `tool_failures` - Self-repair tracking
 - `secrets`, `wasm_tools`, `tool_capabilities` - Extension infrastructure
 
-### Configuration
-
-```bash
-# Backend selection (default: postgres)
-DATABASE_BACKEND=libsql
-
-# PostgreSQL
-DATABASE_URL=postgres://user:pass@localhost/ironclaw
-
-# libSQL (embedded)
-LIBSQL_PATH=~/.ironclaw/ironclaw.db    # Default path
-
-# libSQL (Turso cloud sync)
-LIBSQL_URL=libsql://your-db.turso.io
-LIBSQL_AUTH_TOKEN=your-token            # Required when LIBSQL_URL is set
-```
+Database configuration: see Configuration section above.
 
 ### Current Limitations (libSQL backend)
 
@@ -419,6 +502,7 @@ All external tool output passes through `SafetyLayer`:
 1. **Sanitizer** - Detects injection patterns, escapes dangerous content
 2. **Validator** - Checks length, encoding, forbidden patterns
 3. **Policy** - Rules with severity (Critical/High/Medium/Low) and actions (Block/Warn/Review/Sanitize)
+4. **Leak Detector** - Scans for 15+ secret patterns (API keys, tokens, private keys, connection strings) at two points: tool output before it reaches the LLM, and LLM responses before they reach the user. Actions per pattern: Block (reject entirely), Redact (mask the secret), or Warn (flag but allow)
 
 Tool outputs are wrapped before reaching LLM:
 ```xml
@@ -427,6 +511,95 @@ Tool outputs are wrapped before reaching LLM:
 </tool_output>
 ```
 
+### Shell Environment Scrubbing
+
+The shell tool (`src/tools/builtin/shell.rs`) scrubs sensitive environment variables before executing commands, preventing secrets from leaking through `env`, `printenv`, or `$VAR` expansion. The sanitizer (`src/safety/sanitizer.rs`) also detects command injection patterns (chained commands, subshells, path traversal) and blocks or escapes them based on policy rules.
+
+## Skills System
+
+Skills are SKILL.md files that extend the agent's prompt with domain-specific instructions. Each skill is a YAML frontmatter block (metadata, activation criteria, required tools) followed by a markdown body that gets injected into the LLM context when the skill activates.
+
+### Trust Model
+
+| Trust Level | Source | Tool Access |
+|-------------|--------|-------------|
+| **Trusted** | User-placed in `~/.ironclaw/skills/` or workspace `skills/` | All tools available to the agent |
+| **Installed** | Downloaded from ClawHub registry | Read-only tools only (no shell, file write, HTTP) |
+
+### SKILL.md Format
+
+```yaml
+---
+name: my-skill
+version: 0.1.0
+description: Does something useful
+activation:
+  patterns:
+    - "deploy to.*production"
+  keywords:
+    - "deployment"
+  max_context_tokens: 2000
+metadata:
+  openclaw:
+    requires:
+      bins: [docker, kubectl]
+      env: [KUBECONFIG]
+---
+
+# Deployment Skill
+
+Instructions for the agent when this skill activates...
+```
+
+### Selection Pipeline
+
+1. **Gating** -- Check binary/env/config requirements; skip skills whose prerequisites are missing
+2. **Scoring** -- Deterministic scoring against message content using keywords, tags, and regex patterns
+3. **Budget** -- Select top-scoring skills that fit within `SKILLS_MAX_TOKENS` prompt budget
+4. **Attenuation** -- Apply trust-based tool ceiling; installed skills lose access to dangerous tools
+
+### Skill Tools
+
+Four built-in tools for managing skills at runtime:
+- **`skill_list`** -- List all discovered skills with trust level and status
+- **`skill_search`** -- Search ClawHub registry for available skills
+- **`skill_install`** -- Download and install a skill from ClawHub
+- **`skill_remove`** -- Remove an installed skill
+
+### Skill Directories
+
+- `~/.ironclaw/skills/` -- User's global skills (trusted)
+- `<workspace>/skills/` -- Per-workspace skills (trusted)
+- `~/.ironclaw/installed_skills/` -- Registry-installed skills (installed trust)
+
+Skills configuration: see Configuration section above.
+
+## Docker Sandbox
+
+The `src/sandbox/` module provides Docker-based isolation for job execution with a network proxy that controls outbound access and injects credentials.
+
+### Sandbox Policies
+
+| Policy | Filesystem | Network | Use Case |
+|--------|-----------|---------|----------|
+| **ReadOnly** | Read-only workspace mount | Allowlisted domains only | Analysis, code review |
+| **WorkspaceWrite** | Read-write workspace mount | Allowlisted domains only | Code generation, file edits |
+| **FullAccess** | Full filesystem | Unrestricted | Trusted admin tasks |
+
+### Network Proxy
+
+Containers route all HTTP/HTTPS traffic through a host-side proxy (`src/sandbox/proxy/`):
+- **Domain allowlist** -- Only allowlisted domains are reachable (default: package registries, docs sites, GitHub, common APIs)
+- **Credential injection** -- The `CredentialResolver` trait injects auth headers into proxied requests so secrets never enter the container environment
+- **CONNECT tunnel** -- HTTPS traffic uses CONNECT method; the proxy validates the target domain against the allowlist before establishing the tunnel
+- **Policy decisions** -- The `NetworkPolicyDecider` trait allows custom logic for allow/deny/inject decisions per request
+
+### Zero-Exposure Credential Model
+
+Secrets (API keys, tokens) are stored encrypted on the host and injected into HTTP requests by the proxy at transit time. Container processes never have access to raw credential values, preventing exfiltration even if container code is compromised.
+
+Sandbox configuration: see Configuration section above.
+
 ## Testing
 
 Tests are in `mod tests {}` blocks at the bottom of each file. Run specific module tests:
@@ -451,164 +624,13 @@ Key test patterns:
 7. **Webhook trigger endpoint** - Routines webhook trigger not yet exposed in web gateway
 8. **Full channel status view** - Gateway status widget exists, but no per-channel connection dashboard
 
-### Completed
+## Tool Architecture
 
-- ✅ **Workspace integration** - Memory tools registered, workspace passed to Agent and heartbeat
-- ✅ **WASM sandboxing** - Full implementation in `tools/wasm/` with fuel metering, memory limits, capabilities
-- ✅ **Dynamic tool building** - `tools/builder/` has LlmSoftwareBuilder with iterative build loop
-- ✅ **HTTP webhook security** - Secret validation implemented, proper error handling (no panics)
-- ✅ **Embeddings integration** - OpenAI and NEAR AI providers wired to workspace for semantic search
-- ✅ **Workspace system prompt** - Identity files (AGENTS.md, SOUL.md, USER.md, IDENTITY.md) injected into LLM context
-- ✅ **Heartbeat notifications** - Route through channel manager (broadcast API) instead of logging-only
-- ✅ **Auto-context compaction** - Triggers automatically when context exceeds threshold
-- ✅ **Embedding backfill** - Runs on startup when embeddings provider is enabled
-- ✅ **Clippy clean** - All warnings addressed via config struct refactoring
-- ✅ **Tool approval enforcement** - Tools with `requires_approval()` (shell, http, file write/patch, build_software) now gate execution, track auto-approved tools per session
-- ✅ **Tool definition refresh** - Tool definitions refreshed each iteration so newly built tools become visible in same session
-- ✅ **Worker tool call handling** - Uses `respond_with_tools()` to properly execute tool calls when `select_tools()` returns empty
-- ✅ **Gateway control plane** - Web gateway with 40+ API endpoints, SSE/WebSocket
-- ✅ **Web Control UI** - Browser-based dashboard with chat, memory, jobs, logs, extensions, routines
-- ✅ **Slack/Telegram channels** - Implemented as WASM tools
-- ✅ **Docker sandbox** - Orchestrator/worker containers with per-job auth
-- ✅ **Claude Code mode** - Delegate jobs to Claude CLI inside containers
-- ✅ **Routines system** - Cron, event, webhook, and manual triggers with guardrails
-- ✅ **Extension management** - Install, auth, activate MCP/WASM extensions via CLI and web UI
-- ✅ **libSQL/Turso backend** - Database trait abstraction (`src/db/`), feature-gated dual backend support (postgres/libsql), embedded SQLite for zero-dependency local mode
+**Keep tool-specific logic out of the main agent codebase.** The main agent provides generic infrastructure; tools are self-contained units that declare their requirements through `capabilities.json` files (API endpoints, credentials, rate limits, auth setup). Service-specific auth flows, CLI commands, and configuration do not belong in the main agent.
 
-## Adding a New Tool
+Tools can be built as **WASM** (sandboxed, credential-injected, single binary) or **MCP servers** (ecosystem of pre-built servers, any language, but no sandbox). Both are first-class via `ironclaw tool install`. Auth is declared in capabilities files with OAuth and manual token entry support.
 
-### Built-in Tools (Rust)
-
-1. Create `src/tools/builtin/my_tool.rs`
-2. Implement the `Tool` trait
-3. Add `mod my_tool;` and `pub use` in `src/tools/builtin/mod.rs`
-4. Register in `ToolRegistry::register_builtin_tools()` in `registry.rs`
-5. Add tests
-
-### WASM Tools (Recommended)
-
-WASM tools are the preferred way to add new capabilities. They run in a sandboxed environment with explicit capabilities.
-
-1. Create a new crate in `tools-src/<name>/`
-2. Implement the WIT interface (`wit/tool.wit`)
-3. Create `<name>.capabilities.json` declaring required permissions
-4. Build with `cargo build --target wasm32-wasip2 --release`
-5. Install with `ironclaw tool install path/to/tool.wasm`
-
-See `tools-src/` for examples.
-
-## Tool Architecture Principles
-
-**CRITICAL: Keep tool-specific logic out of the main agent codebase.**
-
-The main agent provides generic infrastructure; tools are self-contained units that declare their requirements through capabilities files.
-
-### What Goes in Tools (capabilities.json)
-
-- API endpoints the tool needs (HTTP allowlist)
-- Credentials required (secret names, injection locations)
-- Rate limits and timeouts
-- Auth setup instructions (see below)
-- Workspace paths the tool can read
-
-### What Does NOT Go in Main Agent
-
-- Service-specific auth flows (OAuth for Notion, Slack, etc.)
-- Service-specific CLI commands (`auth notion`, `auth slack`)
-- Service-specific configuration handling
-- Hardcoded API URLs or token formats
-
-### Tool Authentication
-
-Tools declare their auth requirements in `<tool>.capabilities.json` under the `auth` section. Two methods are supported:
-
-#### OAuth (Browser-based login)
-
-For services that support OAuth, users just click through browser login:
-
-```json
-{
-  "auth": {
-    "secret_name": "notion_api_token",
-    "display_name": "Notion",
-    "oauth": {
-      "authorization_url": "https://api.notion.com/v1/oauth/authorize",
-      "token_url": "https://api.notion.com/v1/oauth/token",
-      "client_id_env": "NOTION_OAUTH_CLIENT_ID",
-      "client_secret_env": "NOTION_OAUTH_CLIENT_SECRET",
-      "scopes": [],
-      "use_pkce": false,
-      "extra_params": { "owner": "user" }
-    },
-    "env_var": "NOTION_TOKEN"
-  }
-}
-```
-
-To enable OAuth for a tool:
-1. Register a public OAuth app with the service (e.g., notion.so/my-integrations)
-2. Configure redirect URIs: `http://localhost:9876/callback` through `http://localhost:9886/callback`
-3. Set environment variables for client_id and client_secret
-
-#### Manual Token Entry (Fallback)
-
-For services without OAuth or when OAuth isn't configured:
-
-```json
-{
-  "auth": {
-    "secret_name": "openai_api_key",
-    "display_name": "OpenAI",
-    "instructions": "Get your API key from platform.openai.com/api-keys",
-    "setup_url": "https://platform.openai.com/api-keys",
-    "token_hint": "Starts with 'sk-'",
-    "env_var": "OPENAI_API_KEY"
-  }
-}
-```
-
-#### Auth Flow Priority
-
-When running `ironclaw tool auth <tool>`:
-
-1. Check `env_var` - if set in environment, use it directly
-2. Check `oauth` - if configured, open browser for OAuth flow
-3. Fall back to `instructions` + manual token entry
-
-The agent reads auth config from the tool's capabilities file and provides the appropriate flow. No service-specific code in the main agent.
-
-### WASM Tools vs MCP Servers: When to Use Which
-
-Both are first-class in the extension system (`ironclaw tool install` handles both), but they have different strengths.
-
-**WASM Tools (IronClaw native)**
-
-- Sandboxed: fuel metering, memory limits, no access except what's allowlisted
-- Credentials injected by host runtime, tool code never sees the actual token
-- Output scanned for secret leakage before returning to the LLM
-- Auth (OAuth/manual) declared in `capabilities.json`, agent handles the flow
-- Single binary, no process management, works offline
-- Cost: must build yourself in Rust, no ecosystem, synchronous only
-
-**MCP Servers (Model Context Protocol)**
-
-- Growing ecosystem of pre-built servers (GitHub, Notion, Postgres, etc.)
-- Any language (TypeScript/Python most common)
-- Can do websockets, streaming, background polling
-- Cost: external process with full system access (no sandbox), manages own credentials, IronClaw can't prevent leaks
-
-**Decision guide:**
-
-| Scenario | Use |
-|----------|-----|
-| Good MCP server already exists | **MCP** |
-| Handles sensitive credentials (email send, banking) | **WASM** |
-| Quick prototype or one-off integration | **MCP** |
-| Core capability you'll maintain long-term | **WASM** |
-| Needs background connections (websockets, polling) | **MCP** |
-| Multiple tools share one OAuth token (e.g., Google suite) | **WASM** |
-
-The LLM-facing interface is identical for both (tool name, schema, execute), so swapping between them is transparent to the agent.
+See `src/tools/README.md` for full tool architecture, adding new tools (built-in Rust and WASM), auth JSON examples, and WASM vs MCP decision guide.
 
 ## Adding a New Channel
 
@@ -645,154 +667,15 @@ for that module's behavior. When modifying code in a module that has a spec:
 | Module | Spec File |
 |--------|-----------|
 | `src/setup/` | `src/setup/README.md` |
-
-## Code Style
-
-- Use `crate::` imports, not `super::`
-- No `pub use` re-exports unless exposing to downstream consumers
-- Prefer strong types over strings (enums, newtypes)
-- Keep functions focused, extract helpers when logic is reused
-- Comments for non-obvious logic only
-
-## Review & Fix Discipline
-
-Hard-won lessons from code review -- follow these when fixing bugs or addressing review feedback.
-
-### Fix the pattern, not just the instance
-When a reviewer flags a bug (e.g., TOCTOU race in INSERT + SELECT-back), search the entire codebase for all instances of that same pattern. A fix in `SecretsStore::create()` that doesn't also fix `WasmToolStore::store()` is half a fix.
-
-### Propagate architectural fixes to satellite types
-If a core type changes its concurrency model (e.g., `LibSqlBackend` switches to connection-per-operation), every type that was handed a resource from the old model (e.g., `LibSqlSecretsStore`, `LibSqlWasmToolStore` holding a single `Connection`) must also be updated. Grep for the old type across the codebase.
-
-### Schema translation is more than DDL
-When translating a database schema between backends (PostgreSQL to libSQL, etc.), check for:
-- **Indexes** -- diff `CREATE INDEX` statements between the two schemas
-- **Seed data** -- check for `INSERT INTO` in migrations (e.g., `leak_detection_patterns`)
-- **Semantic differences** -- document where SQL functions behave differently (e.g., `json_patch` vs `jsonb_set`)
-
-### Feature flag testing
-When adding feature-gated code, test compilation with each feature in isolation:
-```bash
-cargo check                                          # default features
-cargo check --no-default-features --features libsql  # libsql only
-cargo check --all-features                           # all features
-```
-Dead code behind the wrong `#[cfg]` gate will only show up when building with a single feature.
-
-### Mechanical verification before committing
-Run these checks on changed files before committing:
-- `grep -rnE '\.unwrap\(|\.expect\(' <files>` -- no panics in production
-- `grep -rn 'super::' <files>` -- use `crate::` imports
-- If you fixed a pattern bug, `grep` for other instances of that pattern across `src/`
+| `src/workspace/` | `src/workspace/README.md` |
+| `src/tools/` | `src/tools/README.md` |
 
 ## Workspace & Memory System
 
-Inspired by [OpenClaw](https://github.com/openclaw/openclaw), the workspace provides persistent memory for agents with a flexible filesystem-like structure.
+OpenClaw-inspired persistent memory with a flexible filesystem-like structure. Principle: "Memory is database, not RAM" -- if you want to remember something, write it explicitly. Uses hybrid search combining FTS (keyword) + vector (semantic) via Reciprocal Rank Fusion.
 
-### Key Principles
+Four memory tools for LLM use: `memory_search` (hybrid search -- call before answering questions about prior work), `memory_write`, `memory_read`, `memory_tree`. Identity files (AGENTS.md, SOUL.md, USER.md, IDENTITY.md) are injected into the LLM system prompt.
 
-1. **"Memory is database, not RAM"** - If you want to remember something, write it explicitly
-2. **Flexible structure** - Create any directory/file hierarchy you need
-3. **Self-documenting** - Use README.md files to describe directory structure
-4. **Hybrid search** - Combines FTS (keyword) + vector (semantic) via Reciprocal Rank Fusion
+The heartbeat system runs proactive periodic execution (default: 30 minutes), reading `HEARTBEAT.md` and notifying via channel if findings are detected.
 
-### Filesystem Structure
-
-```
-workspace/
-├── README.md              <- Root runbook/index
-├── MEMORY.md              <- Long-term curated memory
-├── HEARTBEAT.md           <- Periodic checklist
-├── IDENTITY.md            <- Agent name, nature, vibe
-├── SOUL.md                <- Core values
-├── AGENTS.md              <- Behavior instructions
-├── USER.md                <- User context
-├── context/               <- Identity-related docs
-│   ├── vision.md
-│   └── priorities.md
-├── daily/                 <- Daily logs
-│   ├── 2024-01-15.md
-│   └── 2024-01-16.md
-├── projects/              <- Arbitrary structure
-│   └── alpha/
-│       ├── README.md
-│       └── notes.md
-└── ...
-```
-
-### Using the Workspace
-
-```rust
-use crate::workspace::{Workspace, OpenAiEmbeddings, paths};
-
-// Create workspace for a user
-let workspace = Workspace::new("user_123", pool)
-    .with_embeddings(Arc::new(OpenAiEmbeddings::new(api_key)));
-
-// Read/write any path
-let doc = workspace.read("projects/alpha/notes.md").await?;
-workspace.write("context/priorities.md", "# Priorities\n\n1. Feature X").await?;
-workspace.append("daily/2024-01-15.md", "Completed task X").await?;
-
-// Convenience methods for well-known files
-workspace.append_memory("User prefers dark mode").await?;
-workspace.append_daily_log("Session note").await?;
-
-// List directory contents
-let entries = workspace.list("projects/").await?;
-
-// Search (hybrid FTS + vector)
-let results = workspace.search("dark mode preference", 5).await?;
-
-// Get system prompt from identity files
-let prompt = workspace.system_prompt().await?;
-```
-
-### Memory Tools
-
-Four tools for LLM use:
-
-- **`memory_search`** - Hybrid search, MUST be called before answering questions about prior work
-- **`memory_write`** - Write to any path (memory, daily_log, or custom paths)
-- **`memory_read`** - Read any file by path
-- **`memory_tree`** - View workspace structure as a tree (depth parameter, default 1)
-
-### Hybrid Search (RRF)
-
-Combines full-text search and vector similarity using Reciprocal Rank Fusion:
-
-```
-score(d) = Σ 1/(k + rank(d)) for each method where d appears
-```
-
-Default k=60. Results from both methods are combined, with documents appearing in both getting boosted scores.
-
-**Backend differences:**
-- **PostgreSQL:** `ts_rank_cd` for FTS, pgvector cosine distance for vectors, full RRF
-- **libSQL:** FTS5 for keyword search only (vector search via `libsql_vector_idx` not yet wired)
-
-### Heartbeat System
-
-Proactive periodic execution (default: 30 minutes):
-
-1. Reads `HEARTBEAT.md` checklist
-2. Runs agent turn with checklist prompt
-3. If findings, notifies via channel
-4. If nothing, agent replies "HEARTBEAT_OK" (no notification)
-
-```rust
-use crate::agent::{HeartbeatConfig, spawn_heartbeat};
-
-let config = HeartbeatConfig::default()
-    .with_interval(Duration::from_secs(60 * 30))
-    .with_notify("user_123", "telegram");
-
-spawn_heartbeat(config, workspace, llm, response_tx);
-```
-
-### Chunking Strategy
-
-Documents are chunked for search indexing:
-- Default: 800 words per chunk (roughly 800 tokens for English)
-- 15% overlap between chunks for context preservation
-- Minimum chunk size: 50 words (tiny trailing chunks merge with previous)
+See `src/workspace/README.md` for full API documentation, filesystem structure, hybrid search details, chunking strategy, and heartbeat system.

Cargo.lock

@@ -11,6 +11,12 @@ dependencies = [
  "gimli",
 ]
 
+[[package]]
+name = "adler2"
+version = "2.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "320119579fcad9c21884f5c4861d16174d0e06250625266f50fe6898340abefa"
+
 [[package]]
 name = "aead"
 version = "0.5.2"
@@ -64,6 +70,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "5a15f179cd60c4584b8a8c596927aadc462e27f2ca70c04e0071964a73ba7a75"
 dependencies = [
  "cfg-if",
+ "const-random",
  "once_cell",
  "version_check",
  "zerocopy 0.8.37",
@@ -188,6 +195,15 @@ version = "0.3.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "b0f477b951e452a0b6b4a10b53ccd569042d1d01729b519e02074a9c0958a063"
 
+[[package]]
+name = "astral-tl"
+version = "0.7.11"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d90933ffb0f97e2fc2e0de21da9d3f20597b804012d199843a6fe7c2810d28f3"
+dependencies = [
+ "memchr",
+]
+
 [[package]]
 name = "async-broadcast"
 version = "0.7.2"
@@ -924,6 +940,26 @@ dependencies = [
  "crossbeam-utils",
 ]
 
+[[package]]
+name = "const-random"
+version = "0.1.18"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "87e00182fe74b066627d63b85fd550ac2998d4b0bd86bfed477a0ae4c7c71359"
+dependencies = [
+ "const-random-macro",
+]
+
+[[package]]
+name = "const-random-macro"
+version = "0.1.16"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f9d839f2a20b0aee515dc581a6172f2321f96cab76c1a38a4c584a194955390e"
+dependencies = [
+ "getrandom 0.2.17",
+ "once_cell",
+ "tiny-keccak",
+]
+
 [[package]]
 name = "constant_time_eq"
 version = "0.4.2"
@@ -1099,6 +1135,21 @@ dependencies = [
  "target-lexicon",
 ]
 
+[[package]]
+name = "crc"
+version = "3.4.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5eb8a2a1cd12ab0d987a5d5e825195d372001a4094a0376319d5a0ad71c1ba0d"
+dependencies = [
+ "crc-catalog",
+]
+
+[[package]]
+name = "crc-catalog"
+version = "2.4.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "19d374276b40fb8bbdee95aef7c7fa6b5316ec764510eb64b8dd0e2ed0d7e7f5"
+
 [[package]]
 name = "crc32fast"
 version = "1.5.0"
@@ -1244,6 +1295,12 @@ dependencies = [
  "winapi",
 ]
 
+[[package]]
+name = "crunchy"
+version = "0.2.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "460fbee9c2c2f33933d720630a6a0bac33ba7053db5344fac858d4b8952d77d5"
+
 [[package]]
 name = "crypto-common"
 version = "0.1.7"
@@ -1255,6 +1312,29 @@ dependencies = [
  "typenum",
 ]
 
+[[package]]
+name = "cssparser"
+version = "0.36.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "dae61cf9c0abb83bd659dab65b7e4e38d8236824c85f0f804f173567bda257d2"
+dependencies = [
+ "cssparser-macros",
+ "dtoa-short",
+ "itoa",
+ "phf 0.13.1",
+ "smallvec",
+]
+
+[[package]]
+name = "cssparser-macros"
+version = "0.6.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "13b588ba4ac1a99f7f2964d24b3d896ddc6bf847ee3855dbd4366f058cfcd331"
+dependencies = [
+ "quote",
+ "syn 2.0.114",
+]
+
 [[package]]
 name = "ctr"
 version = "0.9.2"
@@ -1497,12 +1577,33 @@ version = "0.15.7"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "1aaf95b3e5c8f23aa320147307562d361db0ae0d51242340f558153b4eb2439b"
 
+[[package]]
+name = "dtoa"
+version = "1.0.11"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4c3cf4824e2d5f025c7b531afcb2325364084a16806f6d47fbc1f5fbd9960590"
+
+[[package]]
+name = "dtoa-short"
+version = "0.3.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cd1511a7b6a56299bd043a9c167a6d2bfb37bf84a6dfceaba651168adfb43c87"
+dependencies = [
+ "dtoa",
+]
+
 [[package]]
 name = "dyn-clone"
 version = "1.0.20"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "d0881ea181b1df73ff77ffaaf9c7544ecc11e82fba9b5f27b262a3c73a332555"
 
+[[package]]
+name = "ego-tree"
+version = "0.10.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b2972feb8dffe7bc8c5463b1dacda1b0dfbed3710e50f977d965429692d74cd8"
+
 [[package]]
 name = "either"
 version = "1.15.0"
@@ -1680,6 +1781,16 @@ version = "0.1.9"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "5baebc0774151f905a1a2cc41989300b1e6fbb29aff0ceffa1064fdd3088d582"
 
+[[package]]
+name = "flate2"
+version = "1.1.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "843fba2746e448b37e26a819579957415c8cef339bf08564fe8b7ddbd959573c"
+dependencies = [
+ "crc32fast",
+ "miniz_oxide",
+]
+
 [[package]]
 name = "fnv"
 version = "1.0.7"
@@ -1692,6 +1803,12 @@ version = "0.1.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "d9c4f5dac5e15c24eb999c26181a6ca40b39fe946cbe4c263c7209467bc83af2"
 
+[[package]]
+name = "foldhash"
+version = "0.2.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "77ce24cb58228fbb8aa041425bb1050850ac19177686ea6e0f41a70416f56fdb"
+
 [[package]]
 name = "foreign-types"
 version = "0.3.2"
@@ -1743,6 +1860,16 @@ version = "2.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "e6d5a32815ae3f33302d95fdcb2ce17862f8c65363dcfd29360480ba1001fc9c"
 
+[[package]]
+name = "futf"
+version = "0.1.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "df420e2e84819663797d1ec6544b13c5be84629e7bb00dc960d6917db2987843"
+dependencies = [
+ "mac",
+ "new_debug_unreachable",
+]
+
 [[package]]
 name = "futures"
 version = "0.3.31"
@@ -1883,6 +2010,15 @@ dependencies = [
  "version_check",
 ]
 
+[[package]]
+name = "getopts"
+version = "0.2.24"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cfe4fbac503b8d1f88e6676011885f34b7174f46e59956bba534ba83abded4df"
+dependencies = [
+ "unicode-width 0.2.0",
+]
+
 [[package]]
 name = "getrandom"
 version = "0.2.17"
@@ -2001,7 +2137,7 @@ version = "0.15.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "9229cfe53dfd69f0609a49f65461bd93001ea1ef889cd5529dd176593f5338a1"
 dependencies = [
- "foldhash",
+ "foldhash 0.1.5",
  "serde",
 ]
 
@@ -2010,6 +2146,11 @@ name = "hashbrown"
 version = "0.16.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "841d1cc9bed7f9236f321df977030373f4a4163ae1a7dbfe1a51a2c1a51d9100"
+dependencies = [
+ "allocator-api2",
+ "equivalent",
+ "foldhash 0.2.0",
+]
 
 [[package]]
 name = "hashlink"
@@ -2065,6 +2206,54 @@ dependencies = [
  "windows-sys 0.59.0",
 ]
 
+[[package]]
+name = "html-escape"
+version = "0.2.13"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6d1ad449764d627e22bfd7cd5e8868264fc9236e07c752972b4080cd351cb476"
+dependencies = [
+ "utf8-width",
+]
+
+[[package]]
+name = "html-to-markdown-rs"
+version = "2.25.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "bb31d75f2fdbc8d889d78a912e10c22c30451afb44ee3310f5bfcabf79a31a17"
+dependencies = [
+ "ahash 0.8.12",
+ "astral-tl",
+ "base64 0.22.1",
+ "html-escape",
+ "html5ever 0.38.0",
+ "lru",
+ "once_cell",
+ "regex",
+ "serde",
+ "serde_json",
+ "thiserror 2.0.18",
+]
+
+[[package]]
+name = "html5ever"
+version = "0.36.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6452c4751a24e1b99c3260d505eaeee76a050573e61f30ac2c924ddc7236f01e"
+dependencies = [
+ "log",
+ "markup5ever 0.36.1",
+]
+
+[[package]]
+name = "html5ever"
+version = "0.38.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1054432bae2f14e0061e33d23402fbaa67a921d319d56adc6bcf887ddad1cbc2"
+dependencies = [
+ "log",
+ "markup5ever 0.38.0",
+]
+
 [[package]]
 name = "http"
 version = "0.2.12"
@@ -2490,7 +2679,7 @@ dependencies = [
 
 [[package]]
 name = "ironclaw"
-version = "0.7.0"
+version = "0.9.0"
 dependencies = [
  "aes-gcm",
  "aho-corasick",
@@ -2508,9 +2697,11 @@ dependencies = [
  "deadpool-postgres",
  "dirs 6.0.0",
  "dotenvy",
+ "flate2",
  "fs4",
  "futures",
  "hkdf",
+ "html-to-markdown-rs",
  "http-body-util",
  "hyper 1.8.1",
  "hyper-util",
@@ -2521,6 +2712,7 @@ dependencies = [
  "postgres-types",
  "pretty_assertions",
  "rand 0.8.5",
+ "readabilityrs",
  "refinery",
  "regex",
  "reqwest",
@@ -2536,6 +2728,7 @@ dependencies = [
  "serde_yml",
  "sha2",
  "subtle",
+ "tar",
  "tempfile",
  "termimad",
  "testcontainers-modules",
@@ -2559,30 +2752,6 @@ dependencies = [
  "zbus",
 ]
 
-[[package]]
-name = "ironclaw-bench"
-version = "0.1.0"
-dependencies = [
- "anyhow",
- "async-trait",
- "chrono",
- "clap",
- "futures",
- "ironclaw",
- "regex",
- "rust_decimal",
- "serde",
- "serde_json",
- "tempfile",
- "thiserror 2.0.18",
- "tokio",
- "tokio-stream",
- "toml",
- "tracing",
- "tracing-subscriber",
- "uuid",
-]
-
 [[package]]
 name = "is-docker"
 version = "0.2.0"
@@ -2663,6 +2832,21 @@ dependencies = [
  "wasm-bindgen",
 ]
 
+[[package]]
+name = "kuchikikiki"
+version = "0.9.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b73885c6a3cefdf7a1db0327cefbe4b9b72cac94cae4b19ede4fa492d8af02a0"
+dependencies = [
+ "bitflags 2.10.0",
+ "crc",
+ "cssparser",
+ "html5ever 0.38.0",
+ "indexmap 2.13.0",
+ "precomputed-hash",
+ "selectors 0.35.0",
+]
+
 [[package]]
 name = "lazy-regex"
 version = "3.5.1"
@@ -2813,7 +2997,7 @@ dependencies = [
  "log",
  "memchr",
  "phf 0.11.3",
- "phf_codegen",
+ "phf_codegen 0.11.3",
  "phf_shared 0.11.3",
  "uncased",
 ]
@@ -2907,12 +3091,27 @@ version = "0.4.29"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "5e5032e24019045c762d3c0f28f5b6b8bbf38563a65908389bf7978758920897"
 
+[[package]]
+name = "lru"
+version = "0.16.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a1dc47f592c06f33f8e3aea9591776ec7c9f9e4124778ff8a3c3b87159f7e593"
+dependencies = [
+ "hashbrown 0.16.1",
+]
+
 [[package]]
 name = "lru-slab"
 version = "0.1.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "112b39cec0b298b6c1999fee3e31427f74f676e4cb9879ed1a121b43661a4154"
 
+[[package]]
+name = "mac"
+version = "0.1.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c41e0c4fef86961ac6d6f8a82609f55f31b05e4fce149ac5710e439df7619ba4"
+
 [[package]]
 name = "mach2"
 version = "0.4.3"
@@ -2922,6 +3121,28 @@ dependencies = [
  "libc",
 ]
 
+[[package]]
+name = "markup5ever"
+version = "0.36.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6c3294c4d74d0742910f8c7b466f44dda9eb2d5742c1e430138df290a1e8451c"
+dependencies = [
+ "log",
+ "tendril 0.4.3",
+ "web_atoms",
+]
+
+[[package]]
+name = "markup5ever"
+version = "0.38.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8983d30f2915feeaaab2d6babdd6bc7e9ed1a00b66b5e6d74df19aa9c0e91862"
+dependencies = [
+ "log",
+ "tendril 0.5.0",
+ "web_atoms",
+]
+
 [[package]]
 name = "matchers"
 version = "0.2.0"
@@ -3014,6 +3235,16 @@ version = "0.2.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "68354c5c6bd36d73ff3feceb05efa59b6acb7626617f4962be322a825e61f79a"
 
+[[package]]
+name = "miniz_oxide"
+version = "0.8.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1fa76a2c86f704bdb222d66965fb3d63269ce38518b83cb0575fca855ebb6316"
+dependencies = [
+ "adler2",
+ "simd-adler32",
+]
+
 [[package]]
 name = "mio"
 version = "1.1.1"
@@ -3052,6 +3283,12 @@ dependencies = [
  "tempfile",
 ]
 
+[[package]]
+name = "new_debug_unreachable"
+version = "1.0.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "650eef8c711430f1a879fdd01d4745a7deea475becfb90269c06775983bbf086"
+
 [[package]]
 name = "nibble_vec"
 version = "0.1.0"
@@ -3422,6 +3659,7 @@ version = "0.13.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "c1562dc717473dbaa4c1f85a36410e03c047b2e7df7f45ee938fbef64ae7fadf"
 dependencies = [
+ "phf_macros",
  "phf_shared 0.13.1",
  "serde",
 ]
@@ -3432,10 +3670,20 @@ version = "0.11.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "aef8048c789fa5e851558d709946d6d79a8ff88c0440c587967f8e94bfb1216a"
 dependencies = [
- "phf_generator",
+ "phf_generator 0.11.3",
  "phf_shared 0.11.3",
 ]
 
+[[package]]
+name = "phf_codegen"
+version = "0.13.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "49aa7f9d80421bca176ca8dbfebe668cc7a2684708594ec9f3c0db0805d5d6e1"
+dependencies = [
+ "phf_generator 0.13.1",
+ "phf_shared 0.13.1",
+]
+
 [[package]]
 name = "phf_generator"
 version = "0.11.3"
@@ -3446,6 +3694,29 @@ dependencies = [
  "rand 0.8.5",
 ]
 
+[[package]]
+name = "phf_generator"
+version = "0.13.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "135ace3a761e564ec88c03a77317a7c6b80bb7f7135ef2544dbe054243b89737"
+dependencies = [
+ "fastrand",
+ "phf_shared 0.13.1",
+]
+
+[[package]]
+name = "phf_macros"
+version = "0.13.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "812f032b54b1e759ccd5f8b6677695d5268c588701effba24601f6932f8269ef"
+dependencies = [
+ "phf_generator 0.13.1",
+ "phf_shared 0.13.1",
+ "proc-macro2",
+ "quote",
+ "syn 2.0.114",
+]
+
 [[package]]
 name = "phf_shared"
 version = "0.11.3"
@@ -3609,6 +3880,12 @@ dependencies = [
  "zerocopy 0.8.37",
 ]
 
+[[package]]
+name = "precomputed-hash"
+version = "0.1.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "925383efa346730478fb4838dbe9137d2a47675ad789c546d150a6e1dd4ab31c"
+
 [[package]]
 name = "pretty_assertions"
 version = "1.4.1"
@@ -3876,6 +4153,24 @@ dependencies = [
  "crossbeam-utils",
 ]
 
+[[package]]
+name = "readabilityrs"
+version = "0.1.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3eb174b0af6c181a87d68b42800806657bfbdf88b566f819aaadb9d2a7b7699d"
+dependencies = [
+ "bitflags 2.10.0",
+ "kuchikikiki",
+ "once_cell",
+ "regex",
+ "scraper",
+ "serde",
+ "serde_json",
+ "thiserror 1.0.69",
+ "url",
+ "v_htmlescape",
+]
+
 [[package]]
 name = "redox_syscall"
 version = "0.3.5"
@@ -4418,6 +4713,21 @@ version = "1.2.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "94143f37725109f92c262ed2cf5e59bce7498c01bcc1502d7b9afe439a4e9f49"
 
+[[package]]
+name = "scraper"
+version = "0.25.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "93cecd86d6259499c844440546d02f55f3e17bd286e529e48d1f9f67e92315cb"
+dependencies = [
+ "cssparser",
+ "ego-tree",
+ "getopts",
+ "html5ever 0.36.1",
+ "precomputed-hash",
+ "selectors 0.33.0",
+ "tendril 0.4.3",
+]
+
 [[package]]
 name = "seahash"
 version = "4.1.0"
@@ -4489,6 +4799,44 @@ dependencies = [
  "libc",
 ]
 
+[[package]]
+name = "selectors"
+version = "0.33.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "feef350c36147532e1b79ea5c1f3791373e61cbd9a6a2615413b3807bb164fb7"
+dependencies = [
+ "bitflags 2.10.0",
+ "cssparser",
+ "derive_more",
+ "log",
+ "new_debug_unreachable",
+ "phf 0.13.1",
+ "phf_codegen 0.13.1",
+ "precomputed-hash",
+ "rustc-hash 2.1.1",
+ "servo_arc",
+ "smallvec",
+]
+
+[[package]]
+name = "selectors"
+version = "0.35.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "93fdfed56cd634f04fe8b9ddf947ae3dc493483e819593d2ba17df9ad05db8b2"
+dependencies = [
+ "bitflags 2.10.0",
+ "cssparser",
+ "derive_more",
+ "log",
+ "new_debug_unreachable",
+ "phf 0.13.1",
+ "phf_codegen 0.13.1",
+ "precomputed-hash",
+ "rustc-hash 2.1.1",
+ "servo_arc",
+ "smallvec",
+]
+
 [[package]]
 name = "semver"
 version = "1.0.27"
@@ -4627,6 +4975,15 @@ dependencies = [
  "syn 2.0.114",
 ]
 
+[[package]]
+name = "servo_arc"
+version = "0.4.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "170fb83ab34de17dc69aa7c67482b22218ddb85da56546f9bd6b929e32a05930"
+dependencies = [
+ "stable_deref_trait",
+]
+
 [[package]]
 name = "serde_yml"
 version = "0.0.12"
@@ -4719,6 +5076,12 @@ dependencies = [
  "libc",
 ]
 
+[[package]]
+name = "simd-adler32"
+version = "0.3.8"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e320a6c5ad31d271ad523dcf3ad13e2767ad8b1cb8f047f75a8aeaf8da139da2"
+
 [[package]]
 name = "simdutf8"
 version = "0.1.5"
@@ -4790,6 +5153,30 @@ version = "0.2.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "f42444fea5b87a39db4218d9422087e66a85d0e7a0963a439b07bcdf91804006"
 
+[[package]]
+name = "string_cache"
+version = "0.9.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a18596f8c785a729f2819c0f6a7eae6ebeebdfffbfe4214ae6b087f690e31901"
+dependencies = [
+ "new_debug_unreachable",
+ "parking_lot",
+ "phf_shared 0.13.1",
+ "precomputed-hash",
+]
+
+[[package]]
+name = "string_cache_codegen"
+version = "0.6.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "585635e46db231059f76c5849798146164652513eb9e8ab2685939dd90f29b69"
+dependencies = [
+ "phf_generator 0.13.1",
+ "phf_shared 0.13.1",
+ "proc-macro2",
+ "quote",
+]
+
 [[package]]
 name = "stringprep"
 version = "0.1.5"
@@ -4927,6 +5314,17 @@ version = "1.0.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "55937e1799185b12863d447f42597ed69d9928686b8d88a1df17376a097d8369"
 
+[[package]]
+name = "tar"
+version = "0.4.44"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1d863878d212c87a19c1a610eb53bb01fe12951c0501cf5a0d65f724914a667a"
+dependencies = [
+ "filetime",
+ "libc",
+ "xattr",
+]
+
 [[package]]
 name = "target-lexicon"
 version = "0.12.16"
@@ -4946,6 +5344,27 @@ dependencies = [
  "windows-sys 0.61.2",
 ]
 
+[[package]]
+name = "tendril"
+version = "0.4.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d24a120c5fc464a3458240ee02c299ebcb9d67b5249c8848b09d639dca8d7bb0"
+dependencies = [
+ "futf",
+ "mac",
+ "utf-8",
+]
+
+[[package]]
+name = "tendril"
+version = "0.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c4790fc369d5a530f4b544b094e31388b9b3a37c0f4652ade4505945f5660d24"
+dependencies = [
+ "new_debug_unreachable",
+ "utf-8",
+]
+
 [[package]]
 name = "termcolor"
 version = "1.4.1"
@@ -5089,6 +5508,15 @@ dependencies = [
  "time-core",
 ]
 
+[[package]]
+name = "tiny-keccak"
+version = "2.0.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2c9d3793400a45f954c52e73d068316d76b6f4e36977e3fcebb13a2721e80237"
+dependencies = [
+ "crunchy",
+]
+
 [[package]]
 name = "tinystr"
 version = "0.8.2"
@@ -5730,6 +6158,12 @@ version = "0.7.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "09cc8ee72d2a9becf2f2febe0205bbed8fc6615b7cb429ad062dc7b7ddd036a9"
 
+[[package]]
+name = "utf8-width"
+version = "0.1.8"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1292c0d970b54115d14f2492fe0170adf21d68a1de108eebc51c1df4f346a091"
+
 [[package]]
 name = "utf8_iter"
 version = "1.0.4"
@@ -5754,6 +6188,12 @@ dependencies = [
  "wasm-bindgen",
 ]
 
+[[package]]
+name = "v_htmlescape"
+version = "0.15.8"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4e8257fbc510f0a46eb602c10215901938b5c2a7d5e70fc11483b1d3c9b5b18c"
+
 [[package]]
 name = "valuable"
 version = "0.1.1"
@@ -6289,6 +6729,18 @@ dependencies = [
  "wasm-bindgen",
 ]
 
+[[package]]
+name = "web_atoms"
+version = "0.2.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "57a9779e9f04d2ac1ce317aee707aa2f6b773afba7b931222bff6983843b1576"
+dependencies = [
+ "phf 0.13.1",
+ "phf_codegen 0.13.1",
+ "string_cache",
+ "string_cache_codegen",
+]
+
 [[package]]
 name = "which"
 version = "4.4.2"

Cargo.toml

@@ -1,15 +1,25 @@
 [workspace]
-members = [".", "benchmarks"]
+members = ["."]
 exclude = [
+    "channels-src/discord",
     "channels-src/telegram",
     "channels-src/slack",
     "channels-src/whatsapp",
+    "tools-src/github",
     "tools-src/gmail",
+    "tools-src/google-calendar",
+    "tools-src/google-docs",
+    "tools-src/google-drive",
+    "tools-src/google-sheets",
+    "tools-src/google-slides",
+    "tools-src/okta",
+    "tools-src/slack",
+    "tools-src/telegram",
 ]
 
 [package]
 name = "ironclaw"
-version = "0.7.0"
+version = "0.9.0"
 edition = "2024"
 rust-version = "1.92"
 description = "Secure personal AI assistant that protects your data and expands its capabilities on the fly"
@@ -31,7 +41,7 @@ tokio-stream = { version = "0.1", features = ["sync"] }
 futures = "0.3"
 
 # HTTP client
-reqwest = { version = "0.12", default-features = false, features = ["json", "rustls-tls-native-roots", "stream"] }
+reqwest = { version = "0.12", default-features = false, features = ["json", "multipart", "rustls-tls-native-roots", "stream"] }
 
 # Serialization
 serde = { version = "1", features = ["derive"] }
@@ -72,7 +82,7 @@ clap = { version = "4", features = ["derive", "env"] }
 
 # Terminal
 crossterm = "0.28"
-rustyline = { version = "17", features = ["derive", "with-file-history"] }
+rustyline = { version = "17", features = ["custom-bindings", "derive", "with-file-history"] }
 termimad = "0.34"
 
 # Channel integrations
@@ -127,6 +137,10 @@ rig-core = "0.30"
 # Docker sandbox
 bollard = "0.18"
 
+# Archive extraction for WASM extension bundles
+flate2 = "1"
+tar = "0.4"
+
 # HTTP proxy for sandboxed network access
 hyper = { version = "1.5", features = ["server", "http1", "http2"] }
 hyper-util = { version = "0.1", features = ["server", "tokio", "http1", "http2"] }
@@ -135,6 +149,10 @@ bytes = "1"
 base64 = "0.22.1"
 mime_guess = "2.0.5"
 
+# HTML to Markdown conversion (feature gated)
+html-to-markdown-rs = { version = "2.3", optional = true }
+readabilityrs = { version = "0.1.2", optional = true }
+
 # macOS keychain
 [target.'cfg(target_os = "macos")'.dependencies]
 security-framework = "3"
@@ -152,7 +170,7 @@ pretty_assertions = "1"
 tempfile = "3"
 
 [features]
-default = ["postgres", "libsql"]
+default = ["postgres", "libsql", "html-to-markdown"]
 postgres = [
     "dep:deadpool-postgres",
     "dep:tokio-postgres",
@@ -163,6 +181,11 @@ postgres = [
 ]
 libsql = ["dep:libsql"]
 integration = []
+html-to-markdown = ["dep:html-to-markdown-rs", "dep:readabilityrs"]
+
+[[test]]
+name = "html_to_markdown"
+required-features = ["html-to-markdown"]
 
 # The profile that 'cargo dist' will build with
 [profile.dist]

README.md

@@ -143,6 +143,23 @@ and secrets encryption (using your system keychain). Settings are persisted in t
 connected database; bootstrap variables (e.g. `DATABASE_URL`, `LLM_BACKEND`) are
 written to `~/.ironclaw/.env` so they are available before the database connects.
 
+### Alternative LLM Providers
+
+IronClaw defaults to NEAR AI but works with any OpenAI-compatible endpoint.
+Popular options include **OpenRouter** (300+ models), **Together AI**, **Fireworks AI**,
+**Ollama** (local), and self-hosted servers like **vLLM** or **LiteLLM**.
+
+Select *"OpenAI-compatible"* in the wizard, or set environment variables directly:
+
+```env
+LLM_BACKEND=openai_compatible
+LLM_BASE_URL=https://openrouter.ai/api/v1
+LLM_API_KEY=sk-or-...
+LLM_MODEL=anthropic/claude-sonnet-4
+```
+
+See [docs/LLM_PROVIDERS.md](docs/LLM_PROVIDERS.md) for a full provider guide.
+
 ## Security
 
 IronClaw implements defense in depth to protect your data and prevent misuse.

benchmarks/Cargo.toml

@@ -1,50 +0,0 @@
-[package]
-name = "ironclaw-bench"
-version = "0.1.0"
-edition = "2024"
-rust-version = "1.85"
-description = "Benchmarking harness for IronClaw agent"
-license = "MIT OR Apache-2.0"
-publish = false
-
-[[bin]]
-name = "ironclaw-bench"
-path = "src/main.rs"
-
-[dependencies]
-ironclaw = { path = ".." }
-
-# Async runtime
-tokio = { version = "1", features = ["full"] }
-tokio-stream = { version = "0.1", features = ["sync"] }
-futures = "0.3"
-
-# Serialization
-serde = { version = "1", features = ["derive"] }
-serde_json = "1"
-toml = "0.8"
-
-# CLI
-clap = { version = "4", features = ["derive"] }
-
-# Core types
-uuid = { version = "1", features = ["v4", "serde"] }
-chrono = { version = "0.4", features = ["serde"] }
-rust_decimal = { version = "1", features = ["serde", "serde-with-str"] }
-
-# Error handling
-thiserror = "2"
-anyhow = "1"
-
-# Async traits
-async-trait = "0.1"
-
-# Logging
-tracing = "0.1"
-tracing-subscriber = { version = "0.3", features = ["env-filter"] }
-
-# Scoring
-regex = "1"
-
-[dev-dependencies]
-tempfile = "3"

benchmarks/baselines/spot-gpt5.2-2c43b83/run.json

@@ -1,15 +0,0 @@
-{
-  "run_id": "8a48de1f-09ce-4c35-9ad1-dad98fb83a1a",
-  "suite_id": "spot",
-  "config_label": "default",
-  "model": "openai/gpt-5.2",
-  "commit_hash": "2c43b83",
-  "pass_rate": 1.0,
-  "avg_score": 1.0,
-  "total_tasks": 21,
-  "completed_tasks": 21,
-  "total_cost_usd": 0.307053,
-  "total_wall_time_ms": 111009,
-  "started_at": "2026-02-17T22:02:08.206112Z",
-  "finished_at": "2026-02-17T22:03:59.270325Z"
-}

benchmarks/baselines/spot-gpt5.2-2c43b83/tasks.jsonl

@@ -1,21 +0,0 @@
-{"task_id":"smoke-greeting","suite_id":"spot","score":{"value":1.0,"label":"pass","details":null},"trace":{"wall_time_ms":2037,"llm_calls":1,"input_tokens":1359,"output_tokens":65,"estimated_cost_usd":0.005052,"tool_calls":[],"turns":1,"hit_iteration_limit":false,"hit_timeout":false},"response":"I’m NEAR AI Agent—an autonomous assistant that helps you plan, write, debug, and execute tasks using available tools (like reading/writing files, running shell commands, and making HTTP requests). Tell me what you’re working on and what outcome you want, and I’ll drive it to completion.","started_at":"2026-02-17T22:02:08.206165Z","finished_at":"2026-02-17T22:02:10.243552Z","config_label":"default","error":null}
-{"task_id":"smoke-math","suite_id":"spot","score":{"value":1.0,"label":"pass","details":null},"trace":{"wall_time_ms":2038,"llm_calls":1,"input_tokens":1366,"output_tokens":5,"estimated_cost_usd":0.004173,"tool_calls":[],"turns":1,"hit_iteration_limit":false,"hit_timeout":false},"response":"1081","started_at":"2026-02-17T22:02:10.243933Z","finished_at":"2026-02-17T22:02:12.282133Z","config_label":"default","error":null}
-{"task_id":"tool-echo","suite_id":"spot","score":{"value":1.0,"label":"pass","details":null},"trace":{"wall_time_ms":2646,"llm_calls":2,"input_tokens":2786,"output_tokens":24,"estimated_cost_usd":0.008718,"tool_calls":[{"name":"echo","duration_ms":0,"success":true}],"turns":1,"hit_iteration_limit":false,"hit_timeout":false},"response":"Spot check passed","started_at":"2026-02-17T22:02:12.282294Z","finished_at":"2026-02-17T22:02:14.928540Z","config_label":"default","error":null}
-{"task_id":"tool-time","suite_id":"spot","score":{"value":1.0,"label":"pass","details":null},"trace":{"wall_time_ms":3460,"llm_calls":2,"input_tokens":2839,"output_tokens":78,"estimated_cost_usd":0.009687,"tool_calls":[{"name":"time","duration_ms":0,"success":true}],"turns":1,"hit_iteration_limit":false,"hit_timeout":false},"response":"Current date/time (UTC): **2026-02-17 22:02:16**  \nISO 8601: **2026-02-17T22:02:16.045933+00:00**","started_at":"2026-02-17T22:02:14.928810Z","finished_at":"2026-02-17T22:02:18.389188Z","config_label":"default","error":null}
-{"task_id":"tool-json-query","suite_id":"spot","score":{"value":1.0,"label":"pass","details":null},"trace":{"wall_time_ms":2751,"llm_calls":2,"input_tokens":2839,"output_tokens":43,"estimated_cost_usd":0.009162,"tool_calls":[{"name":"json","duration_ms":0,"success":true}],"turns":1,"hit_iteration_limit":false,"hit_timeout":false},"response":"Bob","started_at":"2026-02-17T22:02:18.389812Z","finished_at":"2026-02-17T22:02:21.141146Z","config_label":"default","error":null}
-{"task_id":"tool-shell-echo","suite_id":"spot","score":{"value":1.0,"label":"pass","details":null},"trace":{"wall_time_ms":3460,"llm_calls":2,"input_tokens":2811,"output_tokens":25,"estimated_cost_usd":0.008808,"tool_calls":[{"name":"shell","duration_ms":0,"success":true}],"turns":1,"hit_iteration_limit":false,"hit_timeout":false},"response":"benchmark test","started_at":"2026-02-17T22:02:21.141384Z","finished_at":"2026-02-17T22:02:24.601812Z","config_label":"default","error":null}
-{"task_id":"tool-list-dir","suite_id":"spot","score":{"value":1.0,"label":"pass","details":null},"trace":{"wall_time_ms":19146,"llm_calls":2,"input_tokens":4370,"output_tokens":1337,"estimated_cost_usd":0.033165,"tool_calls":[{"name":"list_dir","duration_ms":0,"success":true}],"turns":1,"hit_iteration_limit":false,"hit_timeout":false},"response":"```\nclaude-501/\nclaude-505/\nclaude/\ncodex/\ncom.apple.launchd.HqYyNM7hoK/\ncom.apple.launchd.Z0bTtlsPEN/\ncom.apple.launchd.oY8y5gkzN4/\ncom.apple.launchd.sdo3WGu0S7/\ndata-gym-cache/\ndocker-desktop-privileged2203342862/\ngrammers-test/\nnear-agent-target/\nnear-market-bot-push/\nnear-market-bot/\nopenclaw/\npowerlog/\ntest_grpo/\ntest_grpo2/\ntest_grpo_entry/\ntest_grpo_mix/\ntest_grpo_singleton/\ntmpybtjiuc0/\n.d8bd37babaf2f1f7-00000000.node (5.9MB)\n.d8bd37fbfbb0f7ff-00000000.node (5.9MB)\n.d8bd3d9bdf6dd3f7-00000000.node (5.9MB)\n.d8bd3fbfbf21ddf7-00000000.node (5.9MB)\n.d8bd7dfebb38d5ff-00000000.node (5.9MB)\n.d8bdf5deba3bf1f7-00000000.node (5.9MB)\n.d8bdf7cfff76f3f7-00000000.node (5.9MB)\n.d8bdfd8ffe26d5ff-00000000.node (5.9MB)\n.d8bdfdeb9ba8dbff-00000000.node (5.9MB)\n.d8bdfffe9ab2ddff-00000000.node (5.9MB)\n.s.PGSQL.5432 (0B)\n.s.PGSQL.5432.lock (56B)\n__KMP_REGISTERED_LIB_75079 (1.0KB)\nagent_loop_new.rs (28.5KB)\nagent_mod.rs (1.6KB)\nauth_trace.md (10.8KB)\nbench-daily.md (72B)\nbench-log.md (109B)\nbench-meeting.md (160B)\nbench-monday.md (44B)\nbench-prefs.md (61B)\nbench-project.md (213B)\nbench-reminder.md (68B)\nbench-todo.md (153B)\nbench-tuesday.md (40B)\ncac-deck.html (151.5KB)\ncircuit_breaker.rs (22.1KB)\ncli_config.rs (9.1KB)\ncli_service.rs (1.1KB)\ncommands.rs (17.5KB)\nconfig.rs (58.4KB)\nconflicts_summary.md (21.4KB)\ncost_guard.rs (11.3KB)\ndebug_forc2.py (2.8KB)\ndebug_forc3.py (3.0KB)\ndebug_forc4.py (3.0KB)\ndebug_forc5.py (4.0KB)\ndebug_forc6.py (3.9KB)\ndebug_forc7.py (2.7KB)\ndebug_forc8.py (2.9KB)\ndebug_forc_prove.py (3.0KB)\ndispatcher.rs (26.2KB)\ndoctor.rs (8.4KB)\nhygiene.rs (7.3KB)\nironclaw_blog_test.png (21.4KB)\nironclaw_browser_test_viewport.png (156.1KB)\nironclaw_linkedin_debug.png (6.5KB)\nironclaw_spot_test.txt (19B)\nkeys_chain_signatures.rs (6.1KB)\nkeys_error.rs (1.6KB)\nkeys_intents.rs (5.8KB)\nkeys_mod.rs (31.6KB)\nkeys_policy.rs (31.4KB)\nkeys_rpc.rs (8.5KB)\nkeys_signer.rs (8.0KB)\nkeys_spending.rs (5.9KB)\nkeys_transaction.rs (13.7KB)\nkeys_types.rs (17.4KB)\nleak_detection_research_summary.md (13.0KB)\nleak_detector.rs (25.3KB)\nlib_new.rs (5.0KB)\nllm_mod.rs (10.0KB)\nmain.rs (56.8KB)\nmain_bootstrap.rs (11.0KB)\nmain_rs.txt (34.1KB)\nnear_resp.json (39B)\nobs_log.rs (5.9KB)\nobs_mod.rs (2.7KB)\nobs_multi.rs (4.0KB)\nobs_noop.rs (1.5KB)\nobs_traits.rs (4.0KB)\nour_bootstrap.rs (12.6KB)\nour_config.rs (45.2KB)\npost_final_review.py (6.4KB)\npost_review.py (4.6KB)\npost_review2.py (4.8KB)\npost_review3.py (2.2KB)\npr-35.diff (137.4KB)\npr115.diff (1.6MB)\npr126.diff (336.9KB)\npr14_full_diff.txt (193.4KB)\npr39_files.txt (243B)\npr51_files.txt (548B)\npr93_diff.txt (160.4KB)\nprs.json (4.6KB)\nresponse_cache.rs (15.1KB)\nrun_hard_proof_search.sh (545B)\nrun_hard_proof_search2.sh (515B)\nrun_proof_search_v3.sh (523B)\nrun_proof_search_v4.sh (524B)\nservice.rs (11.9KB)\nsettings.rs (34.7KB)\ntest_line.py (789B)\ntest_search.txt (1.8KB)\nthread_ops.rs (38.5KB)\ntool_wit.txt (6.0KB)\ntunnel_cf.rs (3.9KB)\ntunnel_custom.rs (6.2KB)\ntunnel_mod.rs (10.4KB)\ntunnel_ngrok.rs (4.2KB)\ntunnel_none.rs (1.3KB)\ntunnel_ts.rs (3.5KB)\nwasm_capabilities.rs (15.2KB)\nwasm_capabilities_schema.rs (26.2KB)\nwasm_host.rs (25.3KB)\nwasm_wrapper.rs (14.9KB)\n```","started_at":"2026-02-17T22:02:24.602248Z","finished_at":"2026-02-17T22:02:43.748463Z","config_label":"default","error":null}
-{"task_id":"chain-write-read","suite_id":"spot","score":{"value":1.0,"label":"pass","details":null},"trace":{"wall_time_ms":5712,"llm_calls":3,"input_tokens":4489,"output_tokens":73,"estimated_cost_usd":0.014562,"tool_calls":[{"name":"write_file","duration_ms":0,"success":true},{"name":"read_file","duration_ms":0,"success":true}],"turns":1,"hit_iteration_limit":false,"hit_timeout":false},"response":"It says: `ironclaw spot check`.","started_at":"2026-02-17T22:02:43.749933Z","finished_at":"2026-02-17T22:02:49.462007Z","config_label":"default","error":null}
-{"task_id":"chain-shell-json","suite_id":"spot","score":{"value":1.0,"label":"pass","details":null},"trace":{"wall_time_ms":4582,"llm_calls":3,"input_tokens":4433,"output_tokens":94,"estimated_cost_usd":0.014709,"tool_calls":[{"name":"shell","duration_ms":0,"success":true},{"name":"json","duration_ms":0,"success":true}],"turns":1,"hit_iteration_limit":false,"hit_timeout":false},"response":"[Called tool `json` with arguments: {\"data\":{\"code\":200,\"status\":\"ok\"},\"operation\":\"query\",\"path\":\"status\"}]","started_at":"2026-02-17T22:02:49.463794Z","finished_at":"2026-02-17T22:02:54.046092Z","config_label":"default","error":null}
-{"task_id":"chain-time-echo","suite_id":"spot","score":{"value":1.0,"label":"pass","details":null},"trace":{"wall_time_ms":4583,"llm_calls":3,"input_tokens":4404,"output_tokens":76,"estimated_cost_usd":0.014352,"tool_calls":[{"name":"time","duration_ms":0,"success":true},{"name":"echo","duration_ms":0,"success":true}],"turns":1,"hit_iteration_limit":false,"hit_timeout":false},"response":"2026-02-17T22:02:55.786456+00:00","started_at":"2026-02-17T22:02:54.046306Z","finished_at":"2026-02-17T22:02:58.629389Z","config_label":"default","error":null}
-{"task_id":"robust-no-tool","suite_id":"spot","score":{"value":1.0,"label":"pass","details":null},"trace":{"wall_time_ms":1020,"llm_calls":1,"input_tokens":1366,"output_tokens":5,"estimated_cost_usd":0.004173,"tool_calls":[],"turns":1,"hit_iteration_limit":false,"hit_timeout":false},"response":"Paris.","started_at":"2026-02-17T22:02:58.630502Z","finished_at":"2026-02-17T22:02:59.650655Z","config_label":"default","error":null}
-{"task_id":"robust-correct-tool","suite_id":"spot","score":{"value":1.0,"label":"pass","details":null},"trace":{"wall_time_ms":3980,"llm_calls":2,"input_tokens":2825,"output_tokens":50,"estimated_cost_usd":0.009225,"tool_calls":[{"name":"time","duration_ms":0,"success":true}],"turns":1,"hit_iteration_limit":false,"hit_timeout":false},"response":"It’s **22:03:01 UTC** right now (**2026-02-17**).","started_at":"2026-02-17T22:02:59.650901Z","finished_at":"2026-02-17T22:03:03.631122Z","config_label":"default","error":null}
-{"task_id":"robust-json-validate","suite_id":"spot","score":{"value":1.0,"label":"pass","details":null},"trace":{"wall_time_ms":2954,"llm_calls":2,"input_tokens":2822,"output_tokens":43,"estimated_cost_usd":0.009111,"tool_calls":[{"name":"json","duration_ms":0,"success":true}],"turns":1,"hit_iteration_limit":false,"hit_timeout":false},"response":"The JSON is valid (`valid: true`).","started_at":"2026-02-17T22:03:03.631527Z","finished_at":"2026-02-17T22:03:06.585971Z","config_label":"default","error":null}
-{"task_id":"memory-save-daily","suite_id":"spot","score":{"value":1.0,"label":"pass","details":null},"trace":{"wall_time_ms":4890,"llm_calls":3,"input_tokens":4547,"output_tokens":114,"estimated_cost_usd":0.015351,"tool_calls":[{"name":"write_file","duration_ms":0,"success":true},{"name":"read_file","duration_ms":0,"success":true}],"turns":1,"hit_iteration_limit":false,"hit_timeout":false},"response":"```md\n# Daily Tasks\n\n1. Review PR #42\n2. Update API docs\n3. Deploy to staging\n```","started_at":"2026-02-17T22:03:06.586556Z","finished_at":"2026-02-17T22:03:11.476648Z","config_label":"default","error":null}
-{"task_id":"memory-save-reminder","suite_id":"spot","score":{"value":1.0,"label":"pass","details":null},"trace":{"wall_time_ms":5607,"llm_calls":3,"input_tokens":4531,"output_tokens":95,"estimated_cost_usd":0.015018,"tool_calls":[{"name":"write_file","duration_ms":0,"success":true},{"name":"read_file","duration_ms":0,"success":true}],"turns":1,"hit_iteration_limit":false,"hit_timeout":false},"response":"The appointment is on **March 5th at 2pm**, with **Dr. Smith**.","started_at":"2026-02-17T22:03:11.476873Z","finished_at":"2026-02-17T22:03:17.084343Z","config_label":"default","error":null}
-{"task_id":"memory-save-meeting","suite_id":"spot","score":{"value":1.0,"label":"pass","details":null},"trace":{"wall_time_ms":6113,"llm_calls":3,"input_tokens":4750,"output_tokens":122,"estimated_cost_usd":0.01608,"tool_calls":[{"name":"write_file","duration_ms":0,"success":true},{"name":"read_file","duration_ms":0,"success":true}],"turns":1,"hit_iteration_limit":false,"hit_timeout":false},"response":"Frontend owner: **Bob**  \nLaunch date: **April 15th**","started_at":"2026-02-17T22:03:17.085432Z","finished_at":"2026-02-17T22:03:23.198905Z","config_label":"default","error":null}
-{"task_id":"memory-append-log","suite_id":"spot","score":{"value":1.0,"label":"pass","details":null},"trace":{"wall_time_ms":9880,"llm_calls":6,"input_tokens":9941,"output_tokens":208,"estimated_cost_usd":0.032943,"tool_calls":[{"name":"write_file","duration_ms":0,"success":true},{"name":"read_file","duration_ms":0,"success":true},{"name":"read_file","duration_ms":0,"success":true},{"name":"shell","duration_ms":0,"success":true},{"name":"read_file","duration_ms":0,"success":true}],"turns":1,"hit_iteration_limit":false,"hit_timeout":false},"response":"Contents of `/tmp/bench-log.md`:\n1. Morning standup: discussed blockers on auth service  \n2. Afternoon: resolved API timeout issue in payment gateway  \n\nSummary: You discussed blockers affecting the auth service during the morning standup, and later in the afternoon you resolved an API timeout issue in the payment gateway.","started_at":"2026-02-17T22:03:23.199155Z","finished_at":"2026-02-17T22:03:33.079396Z","config_label":"default","error":null}
-{"task_id":"memory-recall-detail","suite_id":"spot","score":{"value":1.0,"label":"pass","details":null},"trace":{"wall_time_ms":4989,"llm_calls":3,"input_tokens":4721,"output_tokens":121,"estimated_cost_usd":0.015978,"tool_calls":[{"name":"write_file","duration_ms":0,"success":true},{"name":"read_file","duration_ms":0,"success":true}],"turns":1,"hit_iteration_limit":false,"hit_timeout":false},"response":"The project uses **PostgreSQL** as the database, and the PM is **James**.","started_at":"2026-02-17T22:03:33.080611Z","finished_at":"2026-02-17T22:03:38.069896Z","config_label":"default","error":null}
-{"task_id":"memory-todo-priority","suite_id":"spot","score":{"value":1.0,"label":"pass","details":null},"trace":{"wall_time_ms":5916,"llm_calls":3,"input_tokens":4676,"output_tokens":110,"estimated_cost_usd":0.015678,"tool_calls":[{"name":"write_file","duration_ms":0,"success":true},{"name":"read_file","duration_ms":0,"success":true}],"turns":1,"hit_iteration_limit":false,"hit_timeout":false},"response":"High priority tasks:\n- Fix login bug\n- Security audit","started_at":"2026-02-17T22:03:38.070162Z","finished_at":"2026-02-17T22:03:43.986831Z","config_label":"default","error":null}
-{"task_id":"memory-multifile","suite_id":"spot","score":{"value":1.0,"label":"pass","details":null},"trace":{"wall_time_ms":6081,"llm_calls":4,"input_tokens":6442,"output_tokens":196,"estimated_cost_usd":0.022266,"tool_calls":[{"name":"write_file","duration_ms":0,"success":true},{"name":"read_file","duration_ms":0,"success":true},{"name":"read_file","duration_ms":0,"success":true}],"turns":1,"hit_iteration_limit":false,"hit_timeout":false},"response":"On Tuesday you have **Sprint retro at 11am**, then **team lunch at noon**.","started_at":"2026-02-17T22:03:43.988013Z","finished_at":"2026-02-17T22:03:50.069452Z","config_label":"default","error":null}
-{"task_id":"memory-update-context","suite_id":"spot","score":{"value":1.0,"label":"pass","details":null},"trace":{"wall_time_ms":9164,"llm_calls":5,"input_tokens":8354,"output_tokens":252,"estimated_cost_usd":0.028842,"tool_calls":[{"name":"write_file","duration_ms":0,"success":true},{"name":"read_file","duration_ms":0,"success":true},{"name":"apply_patch","duration_ms":0,"success":true},{"name":"read_file","duration_ms":0,"success":true}],"turns":1,"hit_iteration_limit":false,"hit_timeout":false},"response":"Final read confirms the file now says `timezone: EST`.","started_at":"2026-02-17T22:03:50.070277Z","finished_at":"2026-02-17T22:03:59.235196Z","config_label":"default","error":null}

benchmarks/data/spot.jsonl

@@ -1,21 +0,0 @@
-{"id": "smoke-greeting", "prompt": "Hello! Introduce yourself briefly.", "tags": ["smoke"], "assertions": {"response_matches": "(?i)(hello|hi|hey|assistant|agent|help)", "no_error": true, "max_tool_calls": 0}}
-{"id": "smoke-math", "prompt": "What is 47 * 23? Reply with just the number.", "tags": ["smoke"], "assertions": {"response_contains": ["1081"], "no_error": true, "max_tool_calls": 0}}
-{"id": "tool-echo", "prompt": "Use the echo tool to repeat the message: 'Spot check passed'", "tags": ["tool"], "assertions": {"tools_used": ["echo"], "response_contains": ["Spot check passed"], "no_error": true}}
-{"id": "tool-time", "prompt": "What is the current date and time? Use the time tool.", "tags": ["tool"], "assertions": {"tools_used": ["time"], "response_matches": "20\\d{2}", "no_error": true}}
-{"id": "tool-json-query", "prompt": "Given this JSON: {\"users\": [{\"name\": \"Alice\"}, {\"name\": \"Bob\"}]}, use the json tool to extract the second user's name.", "tags": ["tool"], "assertions": {"tools_used": ["json"], "response_contains": ["Bob"], "no_error": true}}
-{"id": "tool-shell-echo", "prompt": "Use the shell tool to run: echo 'benchmark test'", "tags": ["tool"], "assertions": {"tools_used": ["shell"], "response_contains": ["benchmark test"], "no_error": true}}
-{"id": "tool-list-dir", "prompt": "Use the list_dir tool to list the contents of the /tmp directory.", "tags": ["tool"], "assertions": {"tools_used": ["list_dir"], "no_error": true}}
-{"id": "chain-write-read", "prompt": "Write the text 'ironclaw spot check' to /tmp/ironclaw_spot_test.txt using the write_file tool, then read it back using the read_file tool and tell me what it says.", "tags": ["chain"], "assertions": {"tools_used": ["write_file", "read_file"], "response_contains": ["ironclaw spot check"], "no_error": true}}
-{"id": "chain-shell-json", "prompt": "Run a shell command to output the JSON string '{\"status\": \"ok\", \"code\": 200}', then use the json tool to extract the status field.", "tags": ["chain"], "assertions": {"response_contains": ["ok"], "min_tool_calls": 1, "no_error": true}}
-{"id": "chain-time-echo", "prompt": "First get the current time using the time tool, then use the echo tool to repeat it back.", "tags": ["chain"], "assertions": {"tools_used": ["time", "echo"], "no_error": true}}
-{"id": "robust-no-tool", "prompt": "What is the capital of France? Answer directly without using any tools.", "tags": ["robust"], "assertions": {"response_contains": ["Paris"], "max_tool_calls": 0, "no_error": true}}
-{"id": "robust-correct-tool", "prompt": "What time is it right now?", "tags": ["robust"], "assertions": {"tools_used": ["time"], "tools_not_used": ["shell", "echo"], "no_error": true}}
-{"id": "robust-json-validate", "prompt": "Use the json tool to validate whether this is valid JSON: {\"key\": \"value\", \"num\": 42}", "tags": ["robust"], "assertions": {"tools_used": ["json"], "tools_not_used": ["shell"], "no_error": true}}
-{"id": "memory-save-daily", "prompt": "Save these daily tasks to /tmp/bench-daily.md:\n1. Review PR #42\n2. Update API docs\n3. Deploy to staging\nThen read the file back and confirm what was saved.", "tags": ["memory"], "assertions": {"tools_used": ["write_file", "read_file"], "response_contains": ["PR #42", "docs", "staging"], "no_error": true}}
-{"id": "memory-save-reminder", "prompt": "Write a reminder to /tmp/bench-reminder.md: Dentist appointment on March 5th at 2pm with Dr. Smith. Then read the file back and tell me when the appointment is and with whom.", "tags": ["memory"], "assertions": {"tools_used": ["write_file", "read_file"], "response_contains": ["March 5", "Smith"], "response_matches": "2(:00)?\\s*[Pp][Mm]", "no_error": true}}
-{"id": "memory-save-meeting", "prompt": "Save these meeting notes to /tmp/bench-meeting.md:\nMeeting: Project Phoenix sync\nAttendees: Alice, Bob, Carol\nDecisions:\n- Launch date: April 15th\n- Budget: $50k approved\n- Bob owns frontend, Carol owns backend\nThen read the file back and tell me who owns the frontend and what the launch date is.", "tags": ["memory"], "assertions": {"tools_used": ["write_file", "read_file"], "response_contains": ["Bob", "frontend", "April 15"], "no_error": true}}
-{"id": "memory-append-log", "prompt": "Write 'Morning standup: discussed blockers on auth service' to /tmp/bench-log.md. Then append a new line 'Afternoon: resolved API timeout issue in payment gateway' to the same file. Finally read the full file and summarize what happened.", "tags": ["memory"], "assertions": {"tools_used": ["write_file", "read_file"], "response_contains": ["auth", "timeout"], "min_tool_calls": 3, "no_error": true}}
-{"id": "memory-recall-detail", "prompt": "Save the following project context to /tmp/bench-project.md:\nProject Ironclad uses Rust for the backend, React for the frontend, and PostgreSQL for the database. The API is deployed on AWS ECS. The lead developer is Sarah and the PM is James. The sprint ends on March 20th.\nThen read it back and answer: What database does the project use, and who is the PM?", "tags": ["memory"], "assertions": {"tools_used": ["write_file", "read_file"], "response_contains": ["PostgreSQL", "James"], "no_error": true}}
-{"id": "memory-todo-priority", "prompt": "Write the following to /tmp/bench-todo.md:\n- [ ] Fix login bug (priority: HIGH)\n- [ ] Write unit tests (priority: medium)\n- [ ] Update README (priority: low)\n- [ ] Security audit (priority: HIGH)\nThen read it back and tell me which tasks are high priority.", "tags": ["memory"], "assertions": {"tools_used": ["write_file", "read_file"], "response_contains": ["login bug", "security audit"], "no_error": true}}
-{"id": "memory-multifile", "prompt": "Save 'Team standup at 9am, then client demo at 2pm' to /tmp/bench-monday.md and 'Sprint retro at 11am, team lunch at noon' to /tmp/bench-tuesday.md. Then read both files and tell me what's happening on Tuesday.", "tags": ["memory"], "assertions": {"tools_used": ["write_file", "read_file"], "response_contains": ["retro", "lunch"], "min_tool_calls": 3, "no_error": true}}
-{"id": "memory-update-context", "prompt": "Write 'User preference: dark mode, timezone: PST, language: English' to /tmp/bench-prefs.md. Then read it back, and rewrite the file changing the timezone to EST. Finally read it one more time and confirm the timezone is now EST.", "tags": ["memory"], "assertions": {"tools_used": ["write_file", "read_file"], "response_contains": ["EST"], "min_tool_calls": 4, "no_error": true}}

benchmarks/spot.toml

@@ -1,8 +0,0 @@
-task_timeout = "120s"
-parallelism = 1
-
-[[matrix]]
-label = "default"
-
-[suite_config]
-dataset_path = "benchmarks/data/spot.jsonl"

benchmarks/src/adapters/custom.rs

@@ -1,243 +0,0 @@
-use std::io::BufRead;
-use std::path::PathBuf;
-
-use async_trait::async_trait;
-use serde::Deserialize;
-
-use crate::error::BenchError;
-use crate::scoring;
-use crate::suite::{BenchScore, BenchSuite, BenchTask, TaskSubmission};
-
-/// A single entry in the custom JSONL format.
-#[derive(Debug, Deserialize)]
-struct CustomEntry {
-    id: String,
-    prompt: String,
-    #[serde(default)]
-    context: Option<String>,
-    #[serde(default)]
-    tags: Vec<String>,
-    #[serde(default)]
-    expected: Option<String>,
-    #[serde(default)]
-    expected_contains: Option<String>,
-    #[serde(default)]
-    expected_regex: Option<String>,
-    /// "exact", "contains", "regex", or "llm" (default: "exact")
-    #[serde(default = "default_scorer")]
-    scorer: String,
-}
-
-fn default_scorer() -> String {
-    "exact".to_string()
-}
-
-/// Custom JSONL benchmark suite.
-///
-/// Each line of the JSONL file is a task with `id`, `prompt`, and scoring
-/// criteria (`expected`, `expected_contains`, `expected_regex`).
-pub struct CustomSuite {
-    dataset_path: PathBuf,
-}
-
-impl CustomSuite {
-    pub fn new(dataset_path: impl Into<PathBuf>) -> Self {
-        Self {
-            dataset_path: dataset_path.into(),
-        }
-    }
-}
-
-#[async_trait]
-impl BenchSuite for CustomSuite {
-    fn name(&self) -> &str {
-        "Custom JSONL"
-    }
-
-    fn id(&self) -> &str {
-        "custom"
-    }
-
-    async fn load_tasks(&self) -> Result<Vec<BenchTask>, BenchError> {
-        let file = std::fs::File::open(&self.dataset_path).map_err(BenchError::Io)?;
-        let reader = std::io::BufReader::new(file);
-        let mut tasks = Vec::new();
-
-        for (line_num, line) in reader.lines().enumerate() {
-            let line = line?;
-            let trimmed = line.trim();
-            if trimmed.is_empty() {
-                continue;
-            }
-            let entry: CustomEntry = serde_json::from_str(trimmed)
-                .map_err(|e| BenchError::Config(format!("line {}: {}", line_num + 1, e)))?;
-
-            let mut metadata = serde_json::json!({
-                "scorer": entry.scorer,
-            });
-            if let Some(ref expected) = entry.expected {
-                metadata["expected"] = serde_json::Value::String(expected.clone());
-            }
-            if let Some(ref expected_contains) = entry.expected_contains {
-                metadata["expected_contains"] =
-                    serde_json::Value::String(expected_contains.clone());
-            }
-            if let Some(ref expected_regex) = entry.expected_regex {
-                metadata["expected_regex"] = serde_json::Value::String(expected_regex.clone());
-            }
-
-            tasks.push(BenchTask {
-                id: entry.id,
-                prompt: entry.prompt,
-                context: entry.context,
-                resources: vec![],
-                tags: entry.tags,
-                expected_turns: None,
-                timeout: None,
-                metadata,
-            });
-        }
-
-        Ok(tasks)
-    }
-
-    async fn score(
-        &self,
-        task: &BenchTask,
-        submission: &TaskSubmission,
-    ) -> Result<BenchScore, BenchError> {
-        let scorer = task
-            .metadata
-            .get("scorer")
-            .and_then(|v| v.as_str())
-            .unwrap_or("exact");
-
-        match scorer {
-            "exact" => {
-                if let Some(expected) = task.metadata.get("expected").and_then(|v| v.as_str()) {
-                    Ok(scoring::exact_match(expected, &submission.response))
-                } else {
-                    Err(BenchError::Scoring {
-                        task_id: task.id.clone(),
-                        reason: "no 'expected' field for exact scoring".to_string(),
-                    })
-                }
-            }
-            "contains" => {
-                if let Some(expected) = task
-                    .metadata
-                    .get("expected_contains")
-                    .and_then(|v| v.as_str())
-                {
-                    Ok(scoring::contains_match(expected, &submission.response))
-                } else {
-                    Err(BenchError::Scoring {
-                        task_id: task.id.clone(),
-                        reason: "no 'expected_contains' field for contains scoring".to_string(),
-                    })
-                }
-            }
-            "regex" => {
-                if let Some(pattern) = task.metadata.get("expected_regex").and_then(|v| v.as_str())
-                {
-                    Ok(scoring::regex_match(pattern, &submission.response))
-                } else {
-                    Err(BenchError::Scoring {
-                        task_id: task.id.clone(),
-                        reason: "no 'expected_regex' field for regex scoring".to_string(),
-                    })
-                }
-            }
-            "llm" => {
-                // TODO: LLM-as-judge scoring
-                tracing::warn!(
-                    task_id = %task.id,
-                    "LLM-as-judge scoring not implemented, returning placeholder 0.5"
-                );
-                Ok(BenchScore::partial(0.5, "LLM scoring not yet implemented"))
-            }
-            other => Err(BenchError::Scoring {
-                task_id: task.id.clone(),
-                reason: format!("unknown scorer: {other}"),
-            }),
-        }
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use std::io::Write;
-
-    #[tokio::test]
-    async fn test_custom_load_tasks() {
-        let dir = tempfile::tempdir().unwrap();
-        let path = dir.path().join("tasks.jsonl");
-        let mut file = std::fs::File::create(&path).unwrap();
-        writeln!(
-            file,
-            r#"{{"id": "t1", "prompt": "What is 2+2?", "expected": "4"}}"#
-        )
-        .unwrap();
-        writeln!(
-            file,
-            r#"{{"id": "t2", "prompt": "Say hello", "expected_contains": "hello", "scorer": "contains"}}"#
-        )
-        .unwrap();
-
-        let suite = CustomSuite::new(&path);
-        let tasks = suite.load_tasks().await.unwrap();
-        assert_eq!(tasks.len(), 2);
-        assert_eq!(tasks[0].id, "t1");
-        assert_eq!(tasks[1].id, "t2");
-    }
-
-    #[tokio::test]
-    async fn test_custom_exact_scoring() {
-        let dir = tempfile::tempdir().unwrap();
-        let path = dir.path().join("tasks.jsonl");
-        let mut file = std::fs::File::create(&path).unwrap();
-        writeln!(
-            file,
-            r#"{{"id": "t1", "prompt": "What is 2+2?", "expected": "4"}}"#
-        )
-        .unwrap();
-
-        let suite = CustomSuite::new(&path);
-        let tasks = suite.load_tasks().await.unwrap();
-
-        let submission = TaskSubmission {
-            response: "4".to_string(),
-            conversation: vec![],
-            tool_calls: vec![],
-            error: None,
-        };
-        let score = suite.score(&tasks[0], &submission).await.unwrap();
-        assert_eq!(score.value, 1.0);
-        assert_eq!(score.label, "pass");
-    }
-
-    #[tokio::test]
-    async fn test_custom_contains_scoring() {
-        let dir = tempfile::tempdir().unwrap();
-        let path = dir.path().join("tasks.jsonl");
-        let mut file = std::fs::File::create(&path).unwrap();
-        writeln!(
-            file,
-            r#"{{"id": "t1", "prompt": "Greet me", "expected_contains": "hello", "scorer": "contains"}}"#
-        )
-        .unwrap();
-
-        let suite = CustomSuite::new(&path);
-        let tasks = suite.load_tasks().await.unwrap();
-
-        let submission = TaskSubmission {
-            response: "Hello there!".to_string(),
-            conversation: vec![],
-            tool_calls: vec![],
-            error: None,
-        };
-        let score = suite.score(&tasks[0], &submission).await.unwrap();
-        assert_eq!(score.value, 1.0);
-    }
-}

benchmarks/src/adapters/gaia.rs

@@ -1,183 +0,0 @@
-use std::io::BufRead;
-use std::path::PathBuf;
-
-use async_trait::async_trait;
-use serde::Deserialize;
-
-use crate::error::BenchError;
-use crate::scoring;
-use crate::suite::{BenchScore, BenchSuite, BenchTask, TaskResource, TaskSubmission};
-
-/// GAIA dataset entry (Hugging Face JSONL format).
-#[derive(Debug, Deserialize)]
-struct GaiaEntry {
-    task_id: String,
-    #[serde(alias = "Question")]
-    question: String,
-    #[serde(alias = "Final answer", alias = "final_answer")]
-    final_answer: String,
-    #[serde(alias = "Level", default)]
-    level: Option<u32>,
-    #[serde(alias = "file_name", default)]
-    file_name: Option<String>,
-}
-
-/// GAIA benchmark suite.
-///
-/// Tasks are loaded from HuggingFace JSONL exports. Scoring uses normalized
-/// exact match against the `final_answer` field.
-pub struct GaiaSuite {
-    dataset_path: PathBuf,
-    attachments_dir: Option<PathBuf>,
-}
-
-impl GaiaSuite {
-    pub fn new(
-        dataset_path: impl Into<PathBuf>,
-        attachments_dir: Option<impl Into<PathBuf>>,
-    ) -> Self {
-        Self {
-            dataset_path: dataset_path.into(),
-            attachments_dir: attachments_dir.map(|d| d.into()),
-        }
-    }
-}
-
-#[async_trait]
-impl BenchSuite for GaiaSuite {
-    fn name(&self) -> &str {
-        "GAIA"
-    }
-
-    fn id(&self) -> &str {
-        "gaia"
-    }
-
-    async fn load_tasks(&self) -> Result<Vec<BenchTask>, BenchError> {
-        let file = std::fs::File::open(&self.dataset_path)?;
-        let reader = std::io::BufReader::new(file);
-        let mut tasks = Vec::new();
-
-        for (line_num, line) in reader.lines().enumerate() {
-            let line = line?;
-            let trimmed = line.trim();
-            if trimmed.is_empty() {
-                continue;
-            }
-            let entry: GaiaEntry = serde_json::from_str(trimmed)
-                .map_err(|e| BenchError::Config(format!("GAIA line {}: {}", line_num + 1, e)))?;
-
-            let mut resources = Vec::new();
-            if let Some(ref file_name) = entry.file_name {
-                if !file_name.is_empty() {
-                    if let Some(ref dir) = self.attachments_dir {
-                        resources.push(TaskResource {
-                            name: file_name.clone(),
-                            path: dir.join(file_name).to_string_lossy().to_string(),
-                            resource_type: crate::suite::ResourceType::File,
-                        });
-                    }
-                }
-            }
-
-            let mut tags = Vec::new();
-            if let Some(level) = entry.level {
-                tags.push(format!("level-{level}"));
-            }
-
-            let metadata = serde_json::json!({
-                "expected": entry.final_answer,
-                "level": entry.level,
-            });
-
-            tasks.push(BenchTask {
-                id: entry.task_id,
-                prompt: entry.question,
-                context: None,
-                resources,
-                tags,
-                expected_turns: None,
-                timeout: None,
-                metadata,
-            });
-        }
-
-        Ok(tasks)
-    }
-
-    async fn score(
-        &self,
-        task: &BenchTask,
-        submission: &TaskSubmission,
-    ) -> Result<BenchScore, BenchError> {
-        let expected = task
-            .metadata
-            .get("expected")
-            .and_then(|v| v.as_str())
-            .ok_or_else(|| BenchError::Scoring {
-                task_id: task.id.clone(),
-                reason: "missing expected answer in metadata".to_string(),
-            })?;
-
-        Ok(scoring::exact_match(expected, &submission.response))
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use std::io::Write;
-
-    #[tokio::test]
-    async fn test_gaia_load_tasks() {
-        let dir = tempfile::tempdir().unwrap();
-        let path = dir.path().join("gaia.jsonl");
-        let mut file = std::fs::File::create(&path).unwrap();
-        writeln!(
-            file,
-            r#"{{"task_id": "g1", "question": "What is the capital of France?", "final_answer": "Paris", "Level": 1}}"#
-        )
-        .unwrap();
-
-        let suite = GaiaSuite::new(&path, None::<PathBuf>);
-        let tasks = suite.load_tasks().await.unwrap();
-        assert_eq!(tasks.len(), 1);
-        assert_eq!(tasks[0].id, "g1");
-        assert!(tasks[0].tags.contains(&"level-1".to_string()));
-    }
-
-    #[tokio::test]
-    async fn test_gaia_scoring() {
-        let dir = tempfile::tempdir().unwrap();
-        let path = dir.path().join("gaia.jsonl");
-        let mut file = std::fs::File::create(&path).unwrap();
-        writeln!(
-            file,
-            r#"{{"task_id": "g1", "question": "Capital of France?", "final_answer": "Paris"}}"#
-        )
-        .unwrap();
-
-        let suite = GaiaSuite::new(&path, None::<PathBuf>);
-        let tasks = suite.load_tasks().await.unwrap();
-
-        // Exact match (case insensitive)
-        let submission = TaskSubmission {
-            response: "paris".to_string(),
-            conversation: vec![],
-            tool_calls: vec![],
-            error: None,
-        };
-        let score = suite.score(&tasks[0], &submission).await.unwrap();
-        assert_eq!(score.value, 1.0);
-
-        // Wrong answer
-        let submission = TaskSubmission {
-            response: "London".to_string(),
-            conversation: vec![],
-            tool_calls: vec![],
-            error: None,
-        };
-        let score = suite.score(&tasks[0], &submission).await.unwrap();
-        assert_eq!(score.value, 0.0);
-    }
-}

benchmarks/src/adapters/mod.rs

@@ -1,124 +0,0 @@
-pub mod custom;
-pub mod gaia;
-pub mod spot;
-pub mod swe_bench;
-pub mod tau_bench;
-
-use crate::config::BenchConfig;
-use crate::error::BenchError;
-use crate::suite::BenchSuite;
-
-/// List of all known suite IDs.
-pub const KNOWN_SUITES: &[(&str, &str)] = &[
-    ("custom", "Custom JSONL tasks"),
-    ("gaia", "GAIA benchmark (knowledge & reasoning)"),
-    ("spot", "Spot checks (end-to-end user workflows)"),
-    ("tau_bench", "Tau-bench (multi-turn tool use)"),
-    ("swe_bench", "SWE-bench Pro (software engineering)"),
-];
-
-/// Create a suite adapter by name.
-pub fn create_suite(name: &str, config: &BenchConfig) -> Result<Box<dyn BenchSuite>, BenchError> {
-    let suite_map = config.suite_config_map();
-    match name {
-        "custom" => {
-            let dataset_path = suite_map
-                .get("dataset_path")
-                .and_then(|v| v.as_str())
-                .map(|s| s.to_string())
-                .ok_or_else(|| {
-                    BenchError::Config(
-                        "suite_config.dataset_path is required for 'custom' suite".to_string(),
-                    )
-                })?;
-            Ok(Box::new(custom::CustomSuite::new(dataset_path)))
-        }
-        "gaia" => {
-            let dataset_path = suite_map
-                .get("dataset_path")
-                .and_then(|v| v.as_str())
-                .map(|s| s.to_string())
-                .ok_or_else(|| {
-                    BenchError::Config(
-                        "suite_config.dataset_path is required for 'gaia' suite".to_string(),
-                    )
-                })?;
-            let attachments_dir = suite_map
-                .get("attachments_dir")
-                .and_then(|v| v.as_str())
-                .map(|s| s.to_string());
-            Ok(Box::new(gaia::GaiaSuite::new(
-                dataset_path,
-                attachments_dir,
-            )))
-        }
-        "spot" => {
-            let dataset_path = suite_map
-                .get("dataset_path")
-                .and_then(|v| v.as_str())
-                .map(|s| s.to_string())
-                .ok_or_else(|| {
-                    BenchError::Config(
-                        "suite_config.dataset_path is required for 'spot' suite".to_string(),
-                    )
-                })?;
-            Ok(Box::new(spot::SpotSuite::new(dataset_path)))
-        }
-        "tau_bench" => {
-            let dataset_path = suite_map
-                .get("dataset_path")
-                .and_then(|v| v.as_str())
-                .map(|s| s.to_string())
-                .ok_or_else(|| {
-                    BenchError::Config(
-                        "suite_config.dataset_path is required for 'tau_bench' suite".to_string(),
-                    )
-                })?;
-            let domain = suite_map
-                .get("domain")
-                .and_then(|v| v.as_str())
-                .unwrap_or("retail")
-                .to_string();
-            Ok(Box::new(tau_bench::TauBenchSuite::new(
-                dataset_path,
-                domain,
-            )))
-        }
-        "swe_bench" => {
-            let dataset_path = suite_map
-                .get("dataset_path")
-                .and_then(|v| v.as_str())
-                .map(|s| s.to_string())
-                .ok_or_else(|| {
-                    BenchError::Config(
-                        "suite_config.dataset_path is required for 'swe_bench' suite".to_string(),
-                    )
-                })?;
-            let workspace_dir = suite_map
-                .get("workspace_dir")
-                .and_then(|v| v.as_str())
-                .unwrap_or("/tmp/swe-bench")
-                .to_string();
-            let use_docker = suite_map
-                .get("use_docker")
-                .and_then(|v| v.as_bool())
-                .unwrap_or(false);
-            Ok(Box::new(swe_bench::SweBenchSuite::new(
-                dataset_path,
-                workspace_dir,
-                use_docker,
-            )))
-        }
-        _ => {
-            let available = KNOWN_SUITES
-                .iter()
-                .map(|(id, _)| *id)
-                .collect::<Vec<_>>()
-                .join(", ");
-            Err(BenchError::SuiteNotFound {
-                name: name.to_string(),
-                available,
-            })
-        }
-    }
-}

benchmarks/src/adapters/spot.rs

@@ -1,504 +0,0 @@
-use std::collections::HashSet;
-use std::io::BufRead;
-use std::path::PathBuf;
-use std::sync::Arc;
-
-use async_trait::async_trait;
-use regex::Regex;
-use serde::{Deserialize, Serialize};
-
-use crate::error::BenchError;
-use crate::suite::{BenchScore, BenchSuite, BenchTask, TaskSubmission};
-
-/// Multi-criterion assertions for a spot check scenario.
-///
-/// Each field generates one or more individual checks. The final score is
-/// `passed_checks / total_checks`, giving a value between 0.0 and 1.0.
-#[derive(Debug, Clone, Default, Serialize, Deserialize)]
-pub struct SpotAssertions {
-    /// All must appear in the response (case-insensitive).
-    #[serde(default)]
-    pub response_contains: Vec<String>,
-
-    /// None may appear in the response (case-insensitive).
-    #[serde(default)]
-    pub response_not_contains: Vec<String>,
-
-    /// Each tool name must appear in the tool_calls list (checked by name,
-    /// not by count; duplicates in tool_calls are collapsed).
-    #[serde(default)]
-    pub tools_used: Vec<String>,
-
-    /// None of these tool names may appear in the tool_calls list.
-    #[serde(default)]
-    pub tools_not_used: Vec<String>,
-
-    /// Regex pattern the response must match.
-    #[serde(default)]
-    pub response_matches: Option<String>,
-
-    /// Hard fail if the task produced an error.
-    #[serde(default)]
-    pub no_error: bool,
-
-    /// Minimum number of tool calls expected (counts duplicates).
-    #[serde(default)]
-    pub min_tool_calls: Option<usize>,
-
-    /// Maximum number of tool calls allowed (counts duplicates).
-    #[serde(default)]
-    pub max_tool_calls: Option<usize>,
-}
-
-impl SpotAssertions {
-    /// Evaluate all assertions against a submission, returning (score, failure_details).
-    pub fn evaluate(&self, submission: &TaskSubmission) -> (f64, Vec<String>) {
-        let mut passed: usize = 0;
-        let mut total: usize = 0;
-        let mut failures: Vec<String> = Vec::new();
-
-        // Hard fail: error check
-        if self.no_error {
-            total += 1;
-            if let Some(ref err) = submission.error {
-                failures.push(format!("no_error: task errored with: {err}"));
-                // Hard fail: return 0.0 immediately
-                return (0.0, failures);
-            }
-            passed += 1;
-        }
-
-        let response_lower = submission.response.to_lowercase();
-
-        // response_contains: all must appear
-        for needle in &self.response_contains {
-            total += 1;
-            if response_lower.contains(&needle.to_lowercase()) {
-                passed += 1;
-            } else {
-                failures.push(format!("response_contains: missing \"{needle}\""));
-            }
-        }
-
-        // response_not_contains: none may appear
-        for needle in &self.response_not_contains {
-            total += 1;
-            if response_lower.contains(&needle.to_lowercase()) {
-                failures.push(format!("response_not_contains: found \"{needle}\""));
-            } else {
-                passed += 1;
-            }
-        }
-
-        let tool_set: HashSet<&str> = submission.tool_calls.iter().map(|s| s.as_str()).collect();
-
-        // tools_used: each must appear
-        for tool in &self.tools_used {
-            total += 1;
-            if tool_set.contains(tool.as_str()) {
-                passed += 1;
-            } else {
-                failures.push(format!("tools_used: \"{tool}\" not called"));
-            }
-        }
-
-        // tools_not_used: none may appear
-        for tool in &self.tools_not_used {
-            total += 1;
-            if tool_set.contains(tool.as_str()) {
-                failures.push(format!("tools_not_used: \"{tool}\" was called"));
-            } else {
-                passed += 1;
-            }
-        }
-
-        // response_matches: regex pattern
-        if let Some(ref pattern) = self.response_matches {
-            total += 1;
-            match Regex::new(pattern) {
-                Ok(re) => {
-                    if re.is_match(&submission.response) {
-                        passed += 1;
-                    } else {
-                        failures.push(format!("response_matches: /{pattern}/ did not match"));
-                    }
-                }
-                Err(e) => {
-                    failures.push(format!("response_matches: bad regex: {e}"));
-                }
-            }
-        }
-
-        let call_count = submission.tool_calls.len();
-
-        // min_tool_calls
-        if let Some(min) = self.min_tool_calls {
-            total += 1;
-            if call_count >= min {
-                passed += 1;
-            } else {
-                failures.push(format!(
-                    "min_tool_calls: expected >= {min}, got {call_count}"
-                ));
-            }
-        }
-
-        // max_tool_calls
-        if let Some(max) = self.max_tool_calls {
-            total += 1;
-            if call_count <= max {
-                passed += 1;
-            } else {
-                failures.push(format!(
-                    "max_tool_calls: expected <= {max}, got {call_count}"
-                ));
-            }
-        }
-
-        if total == 0 {
-            return (1.0, failures);
-        }
-
-        let score = passed as f64 / total as f64;
-        (score, failures)
-    }
-}
-
-/// JSONL entry for a spot check scenario.
-#[derive(Debug, Deserialize)]
-struct SpotEntry {
-    id: String,
-    prompt: String,
-    #[serde(default)]
-    context: Option<String>,
-    #[serde(default)]
-    tags: Vec<String>,
-    #[serde(default)]
-    assertions: SpotAssertions,
-}
-
-/// Spot benchmark suite: end-to-end checks for real user workflows.
-///
-/// Tests conversation, individual tool use, multi-tool chaining, and robustness.
-/// Each task declares multi-criterion assertions scored as passed/total.
-pub struct SpotSuite {
-    dataset_path: PathBuf,
-}
-
-impl SpotSuite {
-    pub fn new(dataset_path: impl Into<PathBuf>) -> Self {
-        Self {
-            dataset_path: dataset_path.into(),
-        }
-    }
-}
-
-#[async_trait]
-impl BenchSuite for SpotSuite {
-    fn name(&self) -> &str {
-        "Spot Checks"
-    }
-
-    fn id(&self) -> &str {
-        "spot"
-    }
-
-    async fn load_tasks(&self) -> Result<Vec<BenchTask>, BenchError> {
-        let file = std::fs::File::open(&self.dataset_path).map_err(BenchError::Io)?;
-        let reader = std::io::BufReader::new(file);
-        let mut tasks = Vec::new();
-
-        for (line_num, line) in reader.lines().enumerate() {
-            let line = line?;
-            let trimmed = line.trim();
-            if trimmed.is_empty() {
-                continue;
-            }
-            let entry: SpotEntry = serde_json::from_str(trimmed)
-                .map_err(|e| BenchError::Config(format!("spot line {}: {}", line_num + 1, e)))?;
-
-            let metadata = serde_json::json!({
-                "assertions": serde_json::to_value(&entry.assertions)
-                    .map_err(|e| BenchError::Config(format!("spot {}: {}", entry.id, e)))?,
-            });
-
-            tasks.push(BenchTask {
-                id: entry.id,
-                prompt: entry.prompt,
-                context: entry.context,
-                resources: vec![],
-                tags: entry.tags,
-                expected_turns: None,
-                timeout: None,
-                metadata,
-            });
-        }
-
-        Ok(tasks)
-    }
-
-    async fn score(
-        &self,
-        task: &BenchTask,
-        submission: &TaskSubmission,
-    ) -> Result<BenchScore, BenchError> {
-        let assertions: SpotAssertions = task
-            .metadata
-            .get("assertions")
-            .ok_or_else(|| BenchError::Scoring {
-                task_id: task.id.clone(),
-                reason: "missing assertions in metadata".to_string(),
-            })
-            .and_then(|v| {
-                serde_json::from_value(v.clone()).map_err(|e| BenchError::Scoring {
-                    task_id: task.id.clone(),
-                    reason: format!("bad assertions: {e}"),
-                })
-            })?;
-
-        let (score, failures) = assertions.evaluate(submission);
-
-        if score >= 1.0 {
-            Ok(BenchScore::pass())
-        } else if score <= 0.0 {
-            Ok(BenchScore::fail(failures.join("; ")))
-        } else {
-            Ok(BenchScore::partial(score, failures.join("; ")))
-        }
-    }
-
-    fn additional_tools(&self) -> Vec<Arc<dyn ironclaw::tools::Tool>> {
-        vec![
-            Arc::new(ironclaw::tools::builtin::ShellTool::new()),
-            Arc::new(ironclaw::tools::builtin::ReadFileTool::new()),
-            Arc::new(ironclaw::tools::builtin::WriteFileTool::new()),
-            Arc::new(ironclaw::tools::builtin::ListDirTool::new()),
-            Arc::new(ironclaw::tools::builtin::ApplyPatchTool::new()),
-        ]
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use std::io::Write;
-
-    fn make_submission(
-        response: &str,
-        tool_calls: Vec<&str>,
-        error: Option<&str>,
-    ) -> TaskSubmission {
-        TaskSubmission {
-            response: response.to_string(),
-            conversation: vec![],
-            tool_calls: tool_calls.into_iter().map(|s| s.to_string()).collect(),
-            error: error.map(|s| s.to_string()),
-        }
-    }
-
-    #[test]
-    fn test_all_pass() {
-        let assertions = SpotAssertions {
-            response_contains: vec!["hello".to_string()],
-            tools_used: vec!["echo".to_string()],
-            no_error: true,
-            ..Default::default()
-        };
-        let sub = make_submission("Hello, world!", vec!["echo"], None);
-        let (score, failures) = assertions.evaluate(&sub);
-        assert_eq!(score, 1.0);
-        assert!(failures.is_empty());
-    }
-
-    #[test]
-    fn test_hard_fail_on_error() {
-        let assertions = SpotAssertions {
-            no_error: true,
-            response_contains: vec!["hello".to_string()],
-            ..Default::default()
-        };
-        let sub = make_submission("Hello!", vec![], Some("timeout after 60s"));
-        let (score, failures) = assertions.evaluate(&sub);
-        assert_eq!(score, 0.0);
-        assert!(failures[0].contains("no_error"));
-    }
-
-    #[test]
-    fn test_partial_score() {
-        let assertions = SpotAssertions {
-            response_contains: vec!["alpha".to_string(), "beta".to_string()],
-            ..Default::default()
-        };
-        let sub = make_submission("alpha is here but not the other", vec![], None);
-        let (score, failures) = assertions.evaluate(&sub);
-        assert_eq!(score, 0.5);
-        assert_eq!(failures.len(), 1);
-        assert!(failures[0].contains("beta"));
-    }
-
-    #[test]
-    fn test_response_not_contains() {
-        let assertions = SpotAssertions {
-            response_not_contains: vec!["error".to_string(), "fail".to_string()],
-            ..Default::default()
-        };
-        let sub = make_submission("This is an error message", vec![], None);
-        let (score, failures) = assertions.evaluate(&sub);
-        assert_eq!(score, 0.5);
-        assert_eq!(failures.len(), 1);
-        assert!(failures[0].contains("error"));
-    }
-
-    #[test]
-    fn test_tools_used_and_not_used() {
-        let assertions = SpotAssertions {
-            tools_used: vec!["time".to_string()],
-            tools_not_used: vec!["shell".to_string(), "echo".to_string()],
-            ..Default::default()
-        };
-        let sub = make_submission("The time is now", vec!["time"], None);
-        let (score, failures) = assertions.evaluate(&sub);
-        assert_eq!(score, 1.0);
-        assert!(failures.is_empty());
-    }
-
-    #[test]
-    fn test_tools_not_used_fails() {
-        let assertions = SpotAssertions {
-            tools_not_used: vec!["shell".to_string()],
-            ..Default::default()
-        };
-        let sub = make_submission("result", vec!["shell", "time"], None);
-        let (score, _) = assertions.evaluate(&sub);
-        assert_eq!(score, 0.0);
-    }
-
-    #[test]
-    fn test_response_matches_regex() {
-        let assertions = SpotAssertions {
-            response_matches: Some(r"\d{4}".to_string()),
-            ..Default::default()
-        };
-        let sub = make_submission("The year is 2026", vec![], None);
-        let (score, failures) = assertions.evaluate(&sub);
-        assert_eq!(score, 1.0);
-        assert!(failures.is_empty());
-    }
-
-    #[test]
-    fn test_response_matches_regex_fail() {
-        let assertions = SpotAssertions {
-            response_matches: Some(r"^\d+$".to_string()),
-            ..Default::default()
-        };
-        let sub = make_submission("not a number", vec![], None);
-        let (score, _) = assertions.evaluate(&sub);
-        assert_eq!(score, 0.0);
-    }
-
-    #[test]
-    fn test_min_max_tool_calls() {
-        let assertions = SpotAssertions {
-            min_tool_calls: Some(2),
-            max_tool_calls: Some(4),
-            ..Default::default()
-        };
-
-        // Within range
-        let sub = make_submission("ok", vec!["a", "b", "c"], None);
-        let (score, _) = assertions.evaluate(&sub);
-        assert_eq!(score, 1.0);
-
-        // Too few
-        let sub = make_submission("ok", vec!["a"], None);
-        let (score, failures) = assertions.evaluate(&sub);
-        assert_eq!(score, 0.5);
-        assert!(failures[0].contains("min_tool_calls"));
-
-        // Too many
-        let sub = make_submission("ok", vec!["a", "b", "c", "d", "e"], None);
-        let (score, failures) = assertions.evaluate(&sub);
-        assert_eq!(score, 0.5);
-        assert!(failures[0].contains("max_tool_calls"));
-    }
-
-    #[test]
-    fn test_max_zero_tool_calls() {
-        let assertions = SpotAssertions {
-            max_tool_calls: Some(0),
-            ..Default::default()
-        };
-        let sub = make_submission("just talking", vec![], None);
-        let (score, _) = assertions.evaluate(&sub);
-        assert_eq!(score, 1.0);
-
-        let sub = make_submission("oops", vec!["echo"], None);
-        let (score, _) = assertions.evaluate(&sub);
-        assert_eq!(score, 0.0);
-    }
-
-    #[test]
-    fn test_empty_assertions() {
-        let assertions = SpotAssertions::default();
-        let sub = make_submission("anything", vec!["whatever"], None);
-        let (score, _) = assertions.evaluate(&sub);
-        assert_eq!(score, 1.0);
-    }
-
-    #[tokio::test]
-    async fn test_spot_load_tasks() {
-        let dir = tempfile::tempdir().unwrap();
-        let path = dir.path().join("spot.jsonl");
-        let mut file = std::fs::File::create(&path).unwrap();
-        writeln!(
-            file,
-            r#"{{"id": "s1", "prompt": "Hello", "tags": ["smoke"], "assertions": {{"response_contains": ["hello"], "no_error": true}}}}"#
-        )
-        .unwrap();
-        writeln!(
-            file,
-            r#"{{"id": "s2", "prompt": "Echo test", "assertions": {{"tools_used": ["echo"]}}}}"#
-        )
-        .unwrap();
-
-        let suite = SpotSuite::new(&path);
-        let tasks = suite.load_tasks().await.unwrap();
-        assert_eq!(tasks.len(), 2);
-        assert_eq!(tasks[0].id, "s1");
-        assert_eq!(tasks[1].id, "s2");
-        assert!(tasks[0].tags.contains(&"smoke".to_string()));
-    }
-
-    #[tokio::test]
-    async fn test_spot_scoring() {
-        let dir = tempfile::tempdir().unwrap();
-        let path = dir.path().join("spot.jsonl");
-        let mut file = std::fs::File::create(&path).unwrap();
-        writeln!(
-            file,
-            r#"{{"id": "s1", "prompt": "Hello", "assertions": {{"response_contains": ["hello", "world"], "no_error": true}}}}"#
-        )
-        .unwrap();
-
-        let suite = SpotSuite::new(&path);
-        let tasks = suite.load_tasks().await.unwrap();
-
-        // Full pass
-        let sub = make_submission("Hello World!", vec![], None);
-        let score = suite.score(&tasks[0], &sub).await.unwrap();
-        assert_eq!(score.value, 1.0);
-        assert_eq!(score.label, "pass");
-
-        // Partial
-        let sub = make_submission("Hello there", vec![], None);
-        let score = suite.score(&tasks[0], &sub).await.unwrap();
-        assert!(score.value > 0.0 && score.value < 1.0);
-        assert_eq!(score.label, "partial");
-
-        // Error hard fail
-        let sub = make_submission("Hello World!", vec![], Some("boom"));
-        let score = suite.score(&tasks[0], &sub).await.unwrap();
-        assert_eq!(score.value, 0.0);
-        assert_eq!(score.label, "fail");
-    }
-}

benchmarks/src/adapters/swe_bench.rs

@@ -1,416 +0,0 @@
-use std::io::BufRead;
-use std::path::PathBuf;
-
-use async_trait::async_trait;
-use regex::Regex;
-use serde::Deserialize;
-
-use crate::error::BenchError;
-use crate::suite::{BenchScore, BenchSuite, BenchTask, TaskSubmission};
-
-/// Validate that a string is safe for use as a filesystem path component.
-/// Allows alphanumerics, hyphens, underscores, dots, and forward slashes (for nested paths).
-/// Rejects absolute paths, `..` traversal, and shell metacharacters.
-fn is_safe_path_component(s: &str) -> bool {
-    !s.is_empty()
-        && !s.starts_with('/')
-        && !s.contains("..")
-        && s.chars()
-            .all(|c| c.is_ascii_alphanumeric() || matches!(c, '-' | '_' | '.' | '/'))
-}
-
-/// Validate that a repo string matches the expected `owner/repo` GitHub format.
-fn is_valid_github_repo(repo: &str) -> bool {
-    // Match "owner/repo" where both parts are alphanumeric with hyphens/underscores/dots
-    static REPO_PATTERN: std::sync::LazyLock<Regex> =
-        std::sync::LazyLock::new(|| Regex::new(r"^[a-zA-Z0-9._-]+/[a-zA-Z0-9._-]+$").unwrap());
-    REPO_PATTERN.is_match(repo)
-}
-
-/// Validate that a string looks like a git ref (hex SHA or valid ref name).
-fn is_valid_git_ref(s: &str) -> bool {
-    !s.is_empty()
-        && s.chars()
-            .all(|c| c.is_ascii_alphanumeric() || matches!(c, '-' | '_' | '.' | '/'))
-        && !s.contains("..")
-}
-
-/// SWE-bench dataset entry.
-#[derive(Debug, Deserialize)]
-struct SweBenchEntry {
-    instance_id: String,
-    repo: String,
-    base_commit: String,
-    #[serde(default)]
-    problem_statement: String,
-    #[serde(default)]
-    hints_text: Option<String>,
-    #[serde(default)]
-    test_patch: Option<String>,
-    #[serde(default)]
-    patch: Option<String>,
-}
-
-/// SWE-bench Pro: real-world software engineering tasks.
-///
-/// Each task clones a repo at a specific commit, presents the problem statement,
-/// and expects the agent to produce a patch. Scoring runs the test suite.
-pub struct SweBenchSuite {
-    dataset_path: PathBuf,
-    workspace_dir: PathBuf,
-    use_docker: bool,
-}
-
-impl SweBenchSuite {
-    pub fn new(
-        dataset_path: impl Into<PathBuf>,
-        workspace_dir: impl Into<PathBuf>,
-        use_docker: bool,
-    ) -> Self {
-        Self {
-            dataset_path: dataset_path.into(),
-            workspace_dir: workspace_dir.into(),
-            use_docker,
-        }
-    }
-}
-
-#[async_trait]
-impl BenchSuite for SweBenchSuite {
-    fn name(&self) -> &str {
-        "SWE-bench Pro"
-    }
-
-    fn id(&self) -> &str {
-        "swe_bench"
-    }
-
-    async fn load_tasks(&self) -> Result<Vec<BenchTask>, BenchError> {
-        let file = std::fs::File::open(&self.dataset_path)?;
-        let reader = std::io::BufReader::new(file);
-        let mut tasks = Vec::new();
-
-        for (line_num, line) in reader.lines().enumerate() {
-            let line = line?;
-            let trimmed = line.trim();
-            if trimmed.is_empty() {
-                continue;
-            }
-            let entry: SweBenchEntry = serde_json::from_str(trimmed).map_err(|e| {
-                BenchError::Config(format!("swe_bench line {}: {}", line_num + 1, e))
-            })?;
-
-            if !is_safe_path_component(&entry.instance_id) {
-                return Err(BenchError::Config(format!(
-                    "swe_bench line {}: unsafe instance_id \"{}\"",
-                    line_num + 1,
-                    entry.instance_id,
-                )));
-            }
-            if !is_valid_github_repo(&entry.repo) {
-                return Err(BenchError::Config(format!(
-                    "swe_bench line {}: invalid repo format \"{}\"",
-                    line_num + 1,
-                    entry.repo,
-                )));
-            }
-            if !is_valid_git_ref(&entry.base_commit) {
-                return Err(BenchError::Config(format!(
-                    "swe_bench line {}: invalid base_commit \"{}\"",
-                    line_num + 1,
-                    entry.base_commit,
-                )));
-            }
-
-            let metadata = serde_json::json!({
-                "repo": entry.repo,
-                "base_commit": entry.base_commit,
-                "test_patch": entry.test_patch,
-                "gold_patch": entry.patch,
-                "use_docker": self.use_docker,
-                "workspace_dir": self.workspace_dir.to_string_lossy(),
-            });
-
-            let prompt = if let Some(ref hints) = entry.hints_text {
-                format!("{}\n\nHints:\n{}", entry.problem_statement, hints)
-            } else {
-                entry.problem_statement
-            };
-
-            tasks.push(BenchTask {
-                id: entry.instance_id,
-                prompt,
-                context: Some(format!(
-                    "Repository: {}, Commit: {}",
-                    entry.repo, entry.base_commit
-                )),
-                resources: vec![],
-                tags: vec![format!("repo-{}", entry.repo.replace('/', "-"))],
-                expected_turns: None,
-                timeout: None,
-                metadata,
-            });
-        }
-
-        Ok(tasks)
-    }
-
-    async fn setup_task(&self, task: &BenchTask) -> Result<(), BenchError> {
-        let repo = task
-            .metadata
-            .get("repo")
-            .and_then(|v| v.as_str())
-            .ok_or_else(|| BenchError::TaskFailed {
-                task_id: task.id.clone(),
-                reason: "missing repo in metadata".to_string(),
-            })?;
-        let base_commit = task
-            .metadata
-            .get("base_commit")
-            .and_then(|v| v.as_str())
-            .ok_or_else(|| BenchError::TaskFailed {
-                task_id: task.id.clone(),
-                reason: "missing base_commit in metadata".to_string(),
-            })?;
-
-        let task_dir = self.workspace_dir.join(&task.id);
-
-        // Clone repo if not already present
-        if !task_dir.exists() {
-            let repo_url = format!("https://github.com/{}.git", repo);
-            let output = tokio::process::Command::new("git")
-                .args([
-                    "clone",
-                    "--depth",
-                    "1",
-                    &repo_url,
-                    &task_dir.to_string_lossy(),
-                ])
-                .output()
-                .await
-                .map_err(|e| BenchError::TaskFailed {
-                    task_id: task.id.clone(),
-                    reason: format!("git clone failed: {e}"),
-                })?;
-
-            if !output.status.success() {
-                let stderr = String::from_utf8_lossy(&output.stderr);
-                return Err(BenchError::TaskFailed {
-                    task_id: task.id.clone(),
-                    reason: format!("git clone failed: {stderr}"),
-                });
-            }
-        }
-
-        // Checkout the base commit
-        let output = tokio::process::Command::new("git")
-            .args(["checkout", base_commit])
-            .current_dir(&task_dir)
-            .output()
-            .await
-            .map_err(|e| BenchError::TaskFailed {
-                task_id: task.id.clone(),
-                reason: format!("git checkout failed: {e}"),
-            })?;
-
-        if !output.status.success() {
-            // Shallow clone might not have the commit; fetch more history
-            let _ = tokio::process::Command::new("git")
-                .args(["fetch", "--unshallow"])
-                .current_dir(&task_dir)
-                .output()
-                .await;
-
-            let output = tokio::process::Command::new("git")
-                .args(["checkout", base_commit])
-                .current_dir(&task_dir)
-                .output()
-                .await
-                .map_err(|e| BenchError::TaskFailed {
-                    task_id: task.id.clone(),
-                    reason: format!("git checkout retry failed: {e}"),
-                })?;
-
-            if !output.status.success() {
-                let stderr = String::from_utf8_lossy(&output.stderr);
-                return Err(BenchError::TaskFailed {
-                    task_id: task.id.clone(),
-                    reason: format!("git checkout failed: {stderr}"),
-                });
-            }
-        }
-
-        Ok(())
-    }
-
-    async fn teardown_task(&self, task: &BenchTask) -> Result<(), BenchError> {
-        let task_dir = self.workspace_dir.join(&task.id);
-        if task_dir.exists() {
-            // Reset any changes
-            let _ = tokio::process::Command::new("git")
-                .args(["checkout", "."])
-                .current_dir(&task_dir)
-                .output()
-                .await;
-            let _ = tokio::process::Command::new("git")
-                .args(["clean", "-fdx"])
-                .current_dir(&task_dir)
-                .output()
-                .await;
-        }
-        Ok(())
-    }
-
-    async fn score(
-        &self,
-        task: &BenchTask,
-        submission: &TaskSubmission,
-    ) -> Result<BenchScore, BenchError> {
-        // For SWE-bench, scoring requires running the test patch against the agent's changes.
-        // This is a simplified version that checks if the agent produced any code changes.
-
-        let test_patch = task.metadata.get("test_patch").and_then(|v| v.as_str());
-
-        if submission.response.is_empty() {
-            return Ok(BenchScore::fail("no response from agent"));
-        }
-
-        // If we have a test patch, try to verify the submission
-        if let Some(_test_patch) = test_patch {
-            // TODO: Apply agent's patch, then apply test patch, then run tests.
-            // For now, give partial credit if the agent produced some output.
-            tracing::warn!(
-                task_id = %task.id,
-                "SWE-bench test execution not implemented, returning placeholder 0.25"
-            );
-            Ok(BenchScore::partial(
-                0.25,
-                "test execution not yet implemented; partial credit for response",
-            ))
-        } else {
-            tracing::warn!(
-                task_id = %task.id,
-                "no test_patch available, returning placeholder 0.25"
-            );
-            Ok(BenchScore::partial(
-                0.25,
-                "no test_patch available for automated scoring",
-            ))
-        }
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use std::io::Write;
-
-    #[tokio::test]
-    async fn test_swe_bench_load() {
-        let dir = tempfile::tempdir().unwrap();
-        let path = dir.path().join("swe.jsonl");
-        let mut file = std::fs::File::create(&path).unwrap();
-        writeln!(
-            file,
-            r#"{{"instance_id": "django__django-12345", "repo": "django/django", "base_commit": "abc123", "problem_statement": "Fix the ORM bug"}}"#
-        )
-        .unwrap();
-
-        let suite = SweBenchSuite::new(&path, "/tmp/swe-test", false);
-        let tasks = suite.load_tasks().await.unwrap();
-        assert_eq!(tasks.len(), 1);
-        assert_eq!(tasks[0].id, "django__django-12345");
-        assert!(tasks[0].tags.contains(&"repo-django-django".to_string()));
-    }
-
-    #[tokio::test]
-    async fn test_swe_bench_scoring_no_response() {
-        let dir = tempfile::tempdir().unwrap();
-        let path = dir.path().join("swe.jsonl");
-        let mut file = std::fs::File::create(&path).unwrap();
-        writeln!(
-            file,
-            r#"{{"instance_id": "s1", "repo": "org/repo", "base_commit": "abc", "problem_statement": "Fix bug"}}"#
-        )
-        .unwrap();
-
-        let suite = SweBenchSuite::new(&path, "/tmp/swe-test", false);
-        let tasks = suite.load_tasks().await.unwrap();
-
-        let submission = TaskSubmission {
-            response: String::new(),
-            conversation: vec![],
-            tool_calls: vec![],
-            error: None,
-        };
-        let score = suite.score(&tasks[0], &submission).await.unwrap();
-        assert_eq!(score.value, 0.0);
-    }
-
-    #[test]
-    fn test_is_safe_path_component() {
-        assert!(is_safe_path_component("django__django-12345"));
-        assert!(is_safe_path_component("org/repo"));
-        assert!(is_safe_path_component("abc123"));
-        assert!(!is_safe_path_component(""));
-        assert!(!is_safe_path_component("../../etc/passwd"));
-        assert!(!is_safe_path_component("/etc/passwd"));
-        assert!(!is_safe_path_component("foo;rm -rf /"));
-        assert!(!is_safe_path_component("foo bar"));
-    }
-
-    #[test]
-    fn test_is_valid_github_repo() {
-        assert!(is_valid_github_repo("django/django"));
-        assert!(is_valid_github_repo("org/repo-name"));
-        assert!(is_valid_github_repo("Org.Name/Repo_v2"));
-        assert!(!is_valid_github_repo(""));
-        assert!(!is_valid_github_repo("no-slash"));
-        assert!(!is_valid_github_repo("too/many/slashes"));
-        assert!(!is_valid_github_repo("spa ce/repo"));
-    }
-
-    #[test]
-    fn test_is_valid_git_ref() {
-        assert!(is_valid_git_ref("abc123"));
-        assert!(is_valid_git_ref("deadbeef0123456789abcdef0123456789abcdef"));
-        assert!(is_valid_git_ref("v1.2.3"));
-        assert!(is_valid_git_ref("main"));
-        assert!(!is_valid_git_ref(""));
-        assert!(!is_valid_git_ref("bad..ref"));
-        assert!(!is_valid_git_ref("has space"));
-        assert!(!is_valid_git_ref("semi;colon"));
-    }
-
-    #[tokio::test]
-    async fn test_swe_bench_rejects_path_traversal() {
-        let dir = tempfile::tempdir().unwrap();
-        let path = dir.path().join("swe.jsonl");
-        let mut file = std::fs::File::create(&path).unwrap();
-        writeln!(
-            file,
-            r#"{{"instance_id": "../../etc/passwd", "repo": "org/repo", "base_commit": "abc", "problem_statement": "evil"}}"#
-        )
-        .unwrap();
-
-        let suite = SweBenchSuite::new(&path, "/tmp/swe-test", false);
-        let err = suite.load_tasks().await.unwrap_err();
-        assert!(err.to_string().contains("unsafe instance_id"));
-    }
-
-    #[tokio::test]
-    async fn test_swe_bench_rejects_bad_repo() {
-        let dir = tempfile::tempdir().unwrap();
-        let path = dir.path().join("swe.jsonl");
-        let mut file = std::fs::File::create(&path).unwrap();
-        writeln!(
-            file,
-            r#"{{"instance_id": "task1", "repo": "not-a-repo-format", "base_commit": "abc", "problem_statement": "bad"}}"#
-        )
-        .unwrap();
-
-        let suite = SweBenchSuite::new(&path, "/tmp/swe-test", false);
-        let err = suite.load_tasks().await.unwrap_err();
-        assert!(err.to_string().contains("invalid repo format"));
-    }
-}

benchmarks/src/adapters/tau_bench.rs

@@ -1,233 +0,0 @@
-use std::io::BufRead;
-use std::path::PathBuf;
-
-use async_trait::async_trait;
-use serde::Deserialize;
-
-use crate::error::BenchError;
-use crate::suite::{BenchScore, BenchSuite, BenchTask, ConversationTurn, TaskSubmission};
-
-/// Tau-bench task entry.
-#[derive(Debug, Deserialize)]
-struct TauBenchEntry {
-    id: String,
-    #[serde(default)]
-    domain: String,
-    instruction: String,
-    #[serde(default)]
-    user_persona: Option<String>,
-    #[serde(default)]
-    expected_state: Option<serde_json::Value>,
-    #[serde(default)]
-    expected_actions: Vec<String>,
-    #[serde(default)]
-    max_turns: Option<usize>,
-}
-
-/// Tau-bench: multi-turn tool-calling dialog benchmark.
-///
-/// Tests agent ability to handle customer service scenarios with simulated
-/// domain APIs (retail, airline). Scoring compares final state against expected.
-pub struct TauBenchSuite {
-    dataset_path: PathBuf,
-    domain: String,
-}
-
-impl TauBenchSuite {
-    pub fn new(dataset_path: impl Into<PathBuf>, domain: impl Into<String>) -> Self {
-        Self {
-            dataset_path: dataset_path.into(),
-            domain: domain.into(),
-        }
-    }
-}
-
-#[async_trait]
-impl BenchSuite for TauBenchSuite {
-    fn name(&self) -> &str {
-        "Tau-bench"
-    }
-
-    fn id(&self) -> &str {
-        "tau_bench"
-    }
-
-    async fn load_tasks(&self) -> Result<Vec<BenchTask>, BenchError> {
-        let file = std::fs::File::open(&self.dataset_path)?;
-        let reader = std::io::BufReader::new(file);
-        let mut tasks = Vec::new();
-
-        for (line_num, line) in reader.lines().enumerate() {
-            let line = line?;
-            let trimmed = line.trim();
-            if trimmed.is_empty() {
-                continue;
-            }
-            let entry: TauBenchEntry = serde_json::from_str(trimmed).map_err(|e| {
-                BenchError::Config(format!("tau_bench line {}: {}", line_num + 1, e))
-            })?;
-
-            let domain = if entry.domain.is_empty() {
-                self.domain.clone()
-            } else {
-                entry.domain.clone()
-            };
-
-            let metadata = serde_json::json!({
-                "domain": domain,
-                "user_persona": entry.user_persona,
-                "expected_state": entry.expected_state,
-                "expected_actions": entry.expected_actions,
-            });
-
-            tasks.push(BenchTask {
-                id: entry.id,
-                prompt: entry.instruction,
-                context: entry.user_persona.clone(),
-                resources: vec![],
-                tags: vec![format!("domain-{domain}")],
-                expected_turns: entry.max_turns,
-                timeout: None,
-                metadata,
-            });
-        }
-
-        Ok(tasks)
-    }
-
-    async fn score(
-        &self,
-        task: &BenchTask,
-        submission: &TaskSubmission,
-    ) -> Result<BenchScore, BenchError> {
-        // Score based on expected actions completion
-        let expected_actions: Vec<String> = task
-            .metadata
-            .get("expected_actions")
-            .and_then(|v| serde_json::from_value(v.clone()).ok())
-            .unwrap_or_default();
-
-        if expected_actions.is_empty() {
-            // No expected actions defined; score based on whether agent responded
-            if submission.response.is_empty() {
-                return Ok(BenchScore::fail("no response"));
-            }
-            return Ok(BenchScore::partial(
-                0.5,
-                "no expected_actions to evaluate against",
-            ));
-        }
-
-        // Check which expected actions were actually called
-        let called: std::collections::HashSet<&str> =
-            submission.tool_calls.iter().map(|s| s.as_str()).collect();
-        let matched = expected_actions
-            .iter()
-            .filter(|a| called.contains(a.as_str()))
-            .count();
-
-        let ratio = matched as f64 / expected_actions.len() as f64;
-        if ratio >= 1.0 {
-            Ok(BenchScore::pass())
-        } else if ratio > 0.0 {
-            Ok(BenchScore::partial(
-                ratio,
-                format!(
-                    "{}/{} expected actions completed",
-                    matched,
-                    expected_actions.len()
-                ),
-            ))
-        } else {
-            Ok(BenchScore::fail(format!(
-                "0/{} expected actions completed",
-                expected_actions.len()
-            )))
-        }
-    }
-
-    async fn next_user_message(
-        &self,
-        task: &BenchTask,
-        conversation: &[ConversationTurn],
-    ) -> Result<Option<String>, BenchError> {
-        // Check if we've exceeded max turns
-        if let Some(max) = task.expected_turns {
-            let user_turns = conversation
-                .iter()
-                .filter(|t| matches!(t.role, crate::suite::TurnRole::User))
-                .count();
-            if user_turns >= max {
-                return Ok(None);
-            }
-        }
-
-        // Multi-turn simulation requires an LLM to play the customer role.
-        // Until that's implemented, every scenario is single-turn only.
-        // TODO: Use LLM to simulate customer based on user_persona.
-        tracing::warn!(
-            task_id = %task.id,
-            "multi-turn simulation not implemented, ending after first turn"
-        );
-        Ok(None)
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use std::io::Write;
-
-    #[tokio::test]
-    async fn test_tau_bench_load() {
-        let dir = tempfile::tempdir().unwrap();
-        let path = dir.path().join("tau.jsonl");
-        let mut file = std::fs::File::create(&path).unwrap();
-        writeln!(
-            file,
-            r#"{{"id": "t1", "instruction": "Return my order", "expected_actions": ["lookup_order", "process_return"], "max_turns": 3}}"#
-        )
-        .unwrap();
-
-        let suite = TauBenchSuite::new(&path, "retail");
-        let tasks = suite.load_tasks().await.unwrap();
-        assert_eq!(tasks.len(), 1);
-        assert_eq!(tasks[0].expected_turns, Some(3));
-    }
-
-    #[tokio::test]
-    async fn test_tau_bench_scoring() {
-        let dir = tempfile::tempdir().unwrap();
-        let path = dir.path().join("tau.jsonl");
-        let mut file = std::fs::File::create(&path).unwrap();
-        writeln!(
-            file,
-            r#"{{"id": "t1", "instruction": "Return order", "expected_actions": ["lookup_order", "process_return"]}}"#
-        )
-        .unwrap();
-
-        let suite = TauBenchSuite::new(&path, "retail");
-        let tasks = suite.load_tasks().await.unwrap();
-
-        // Partial completion
-        let submission = TaskSubmission {
-            response: "I found your order.".to_string(),
-            conversation: vec![],
-            tool_calls: vec!["lookup_order".to_string()],
-            error: None,
-        };
-        let score = suite.score(&tasks[0], &submission).await.unwrap();
-        assert_eq!(score.value, 0.5);
-        assert_eq!(score.label, "partial");
-
-        // Full completion
-        let submission = TaskSubmission {
-            response: "Return processed.".to_string(),
-            conversation: vec![],
-            tool_calls: vec!["lookup_order".to_string(), "process_return".to_string()],
-            error: None,
-        };
-        let score = suite.score(&tasks[0], &submission).await.unwrap();
-        assert_eq!(score.value, 1.0);
-    }
-}

benchmarks/src/channel.rs

@@ -1,259 +0,0 @@
-use std::sync::Arc;
-
-use async_trait::async_trait;
-use tokio::sync::{Mutex, mpsc};
-use tokio_stream::wrappers::ReceiverStream;
-
-use ironclaw::channels::{Channel, IncomingMessage, MessageStream, OutgoingResponse, StatusUpdate};
-use ironclaw::error::ChannelError;
-
-use crate::results::TraceToolCall;
-use crate::suite::ConversationTurn;
-
-/// Truncate a string to at most `max_bytes` without splitting a UTF-8 character.
-fn truncate_str(s: &str, max_bytes: usize) -> &str {
-    if s.len() <= max_bytes {
-        return s;
-    }
-    let mut end = max_bytes;
-    while !s.is_char_boundary(end) {
-        end -= 1;
-    }
-    &s[..end]
-}
-
-/// Captured state from a benchmark channel run.
-#[derive(Debug, Default)]
-pub struct ChannelCapture {
-    /// All responses the agent sent back.
-    pub responses: Vec<String>,
-    /// Tool calls observed (name, success, duration_ms).
-    pub tool_calls: Vec<TraceToolCall>,
-    /// Full conversation turns for multi-turn scoring.
-    pub conversation: Vec<ConversationTurn>,
-    /// Status messages (for debugging).
-    pub status_log: Vec<String>,
-}
-
-/// A headless Channel implementation for benchmarking.
-///
-/// Modeled after `ReplChannel`: uses mpsc to inject messages and captures
-/// all responses and tool status events. Auto-approves tool execution
-/// so benchmarks run without user interaction.
-pub struct BenchChannel {
-    /// Sender to inject messages into the agent loop.
-    msg_tx: mpsc::Sender<IncomingMessage>,
-    /// Receiver the agent loop reads from (taken once by `start()`).
-    msg_rx: Mutex<Option<mpsc::Receiver<IncomingMessage>>>,
-    /// Accumulated capture data.
-    capture: Arc<Mutex<ChannelCapture>>,
-}
-
-impl BenchChannel {
-    pub fn new() -> (Self, mpsc::Sender<IncomingMessage>) {
-        let (tx, rx) = mpsc::channel(64);
-        let channel = Self {
-            msg_tx: tx.clone(),
-            msg_rx: Mutex::new(Some(rx)),
-            capture: Arc::new(Mutex::new(ChannelCapture::default())),
-        };
-        (channel, tx)
-    }
-
-    /// Get a handle to the capture data.
-    pub fn capture(&self) -> Arc<Mutex<ChannelCapture>> {
-        Arc::clone(&self.capture)
-    }
-}
-
-#[async_trait]
-impl Channel for BenchChannel {
-    fn name(&self) -> &str {
-        "bench"
-    }
-
-    async fn start(&self) -> Result<MessageStream, ChannelError> {
-        let rx = self
-            .msg_rx
-            .lock()
-            .await
-            .take()
-            .ok_or_else(|| ChannelError::StartupFailed {
-                name: "bench".to_string(),
-                reason: "start() already called".to_string(),
-            })?;
-        Ok(Box::pin(ReceiverStream::new(rx)))
-    }
-
-    async fn respond(
-        &self,
-        _msg: &IncomingMessage,
-        response: OutgoingResponse,
-    ) -> Result<(), ChannelError> {
-        let mut cap = self.capture.lock().await;
-        cap.responses.push(response.content.clone());
-        cap.conversation.push(ConversationTurn {
-            role: crate::suite::TurnRole::Assistant,
-            content: response.content,
-        });
-        Ok(())
-    }
-
-    async fn send_status(
-        &self,
-        status: StatusUpdate,
-        _metadata: &serde_json::Value,
-    ) -> Result<(), ChannelError> {
-        let mut cap = self.capture.lock().await;
-
-        match status {
-            StatusUpdate::ToolCompleted { ref name, success } => {
-                cap.tool_calls.push(TraceToolCall {
-                    name: name.clone(),
-                    duration_ms: 0, // We don't have precise per-tool timing here
-                    success,
-                });
-                cap.status_log
-                    .push(format!("tool_completed: {name} success={success}"));
-            }
-            StatusUpdate::ApprovalNeeded { ref request_id, .. } => {
-                // Auto-approve all tools during benchmarks
-                cap.status_log.push(format!("auto_approved: {request_id}"));
-                drop(cap); // Release lock before sending
-                let approval = IncomingMessage::new("bench", "bench-user", "always");
-                let _ = self.msg_tx.send(approval).await;
-                return Ok(());
-            }
-            StatusUpdate::Thinking(ref msg) => {
-                cap.status_log.push(format!("thinking: {msg}"));
-            }
-            StatusUpdate::ToolStarted { ref name } => {
-                cap.status_log.push(format!("tool_started: {name}"));
-            }
-            StatusUpdate::ToolResult {
-                ref name,
-                ref preview,
-            } => {
-                cap.status_log.push(format!(
-                    "tool_result: {name} -> {}",
-                    truncate_str(preview, 100)
-                ));
-            }
-            StatusUpdate::StreamChunk(_) => {}
-            StatusUpdate::Status(ref msg) => {
-                cap.status_log.push(format!("status: {msg}"));
-            }
-            StatusUpdate::JobStarted {
-                ref job_id,
-                ref title,
-                ..
-            } => {
-                cap.status_log
-                    .push(format!("job_started: {job_id} ({title})"));
-            }
-            StatusUpdate::AuthRequired {
-                ref extension_name, ..
-            } => {
-                cap.status_log
-                    .push(format!("auth_required: {extension_name} (auto-skipped)"));
-            }
-            StatusUpdate::AuthCompleted {
-                ref extension_name,
-                success,
-                ..
-            } => {
-                cap.status_log.push(format!(
-                    "auth_completed: {extension_name} success={success}"
-                ));
-            }
-        }
-        Ok(())
-    }
-
-    async fn broadcast(
-        &self,
-        _user_id: &str,
-        response: OutgoingResponse,
-    ) -> Result<(), ChannelError> {
-        let mut cap = self.capture.lock().await;
-        cap.status_log.push(format!(
-            "broadcast: {}",
-            truncate_str(&response.content, 100)
-        ));
-        Ok(())
-    }
-
-    async fn health_check(&self) -> Result<(), ChannelError> {
-        Ok(())
-    }
-
-    async fn shutdown(&self) -> Result<(), ChannelError> {
-        Ok(())
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[tokio::test]
-    async fn test_bench_channel_captures_responses() {
-        let (channel, _tx) = BenchChannel::new();
-        let capture = channel.capture();
-
-        let msg = IncomingMessage::new("bench", "user", "hello");
-        let response = OutgoingResponse::text("world");
-        channel.respond(&msg, response).await.unwrap();
-
-        let cap = capture.lock().await;
-        assert_eq!(cap.responses.len(), 1);
-        assert_eq!(cap.responses[0], "world");
-        assert_eq!(cap.conversation.len(), 1);
-    }
-
-    #[tokio::test]
-    async fn test_bench_channel_auto_approves() {
-        let (channel, _tx) = BenchChannel::new();
-        // start() to consume the receiver
-        let _stream = channel.start().await.unwrap();
-
-        let status = StatusUpdate::ApprovalNeeded {
-            request_id: "req-1".to_string(),
-            tool_name: "shell".to_string(),
-            description: "run ls".to_string(),
-            parameters: serde_json::json!({}),
-        };
-        channel
-            .send_status(status, &serde_json::Value::Null)
-            .await
-            .unwrap();
-
-        // The approval message was sent through msg_tx,
-        // which means the stream would receive it.
-        // We can't easily read from the stream in this test without
-        // consuming it, but we can verify the status log.
-        let capture_arc = channel.capture();
-        let cap = capture_arc.lock().await;
-        assert!(cap.status_log.iter().any(|s| s.contains("auto_approved")));
-    }
-
-    #[tokio::test]
-    async fn test_bench_channel_captures_tool_events() {
-        let (channel, _tx) = BenchChannel::new();
-
-        let status = StatusUpdate::ToolCompleted {
-            name: "echo".to_string(),
-            success: true,
-        };
-        channel
-            .send_status(status, &serde_json::Value::Null)
-            .await
-            .unwrap();
-
-        let capture_arc = channel.capture();
-        let cap = capture_arc.lock().await;
-        assert_eq!(cap.tool_calls.len(), 1);
-        assert_eq!(cap.tool_calls[0].name, "echo");
-        assert!(cap.tool_calls[0].success);
-    }
-}

benchmarks/src/config.rs

@@ -1,205 +0,0 @@
-use std::path::{Path, PathBuf};
-use std::time::Duration;
-
-use serde::Deserialize;
-
-use crate::error::BenchError;
-
-/// Top-level bench configuration, loaded from TOML.
-#[derive(Debug, Clone, Deserialize)]
-pub struct BenchConfig {
-    /// Where to write results. Default: "./bench-results".
-    #[serde(default = "default_results_dir")]
-    pub results_dir: PathBuf,
-
-    /// Per-task timeout. Default: "300s".
-    #[serde(
-        default = "default_task_timeout",
-        deserialize_with = "deserialize_duration"
-    )]
-    pub task_timeout: Duration,
-
-    /// How many tasks to run in parallel. Default: 1.
-    #[serde(default = "default_parallelism")]
-    pub parallelism: usize,
-
-    /// Model/config matrix entries. At least one required.
-    #[serde(default)]
-    pub matrix: Vec<MatrixEntry>,
-
-    /// Suite-specific configuration (passed through to adapter).
-    #[serde(default = "default_suite_config")]
-    pub suite_config: toml::Value,
-}
-
-/// A single model/config combination to benchmark.
-#[derive(Debug, Clone, Deserialize)]
-pub struct MatrixEntry {
-    /// Label for this configuration (used in results).
-    pub label: String,
-
-    /// Model identifier.
-    #[serde(default)]
-    pub model: Option<String>,
-}
-
-impl BenchConfig {
-    /// Load from a TOML file.
-    pub fn from_file(path: &Path) -> Result<Self, BenchError> {
-        if !path.exists() {
-            return Err(BenchError::ConfigNotFound {
-                path: path.to_path_buf(),
-            });
-        }
-        let content = std::fs::read_to_string(path)?;
-        let config: BenchConfig = toml::from_str(&content)?;
-        if config.matrix.is_empty() {
-            return Err(BenchError::Config(
-                "config must have at least one [[matrix]] entry".to_string(),
-            ));
-        }
-        Ok(config)
-    }
-
-    /// Create a minimal config for when no config file is provided.
-    /// Uses defaults and optional CLI overrides.
-    pub fn minimal(model: Option<String>) -> Self {
-        let label = model.as_deref().unwrap_or("default").to_string();
-        Self {
-            results_dir: default_results_dir(),
-            task_timeout: default_task_timeout(),
-            parallelism: default_parallelism(),
-            matrix: vec![MatrixEntry { label, model }],
-            suite_config: toml::Value::Table(toml::map::Map::new()),
-        }
-    }
-
-    /// Get the suite_config as a generic map for adapter use.
-    pub fn suite_config_map(&self) -> toml::map::Map<String, toml::Value> {
-        match &self.suite_config {
-            toml::Value::Table(map) => map.clone(),
-            _ => toml::map::Map::new(),
-        }
-    }
-
-    /// Get a string value from suite_config.
-    pub fn suite_config_str(&self, key: &str) -> Option<String> {
-        self.suite_config_map()
-            .get(key)
-            .and_then(|v| v.as_str())
-            .map(|s| s.to_string())
-    }
-}
-
-fn default_suite_config() -> toml::Value {
-    toml::Value::Table(toml::map::Map::new())
-}
-
-fn default_results_dir() -> PathBuf {
-    PathBuf::from("./bench-results")
-}
-
-fn default_task_timeout() -> Duration {
-    Duration::from_secs(300)
-}
-
-fn default_parallelism() -> usize {
-    1
-}
-
-/// Deserialize a duration from a string like "300s", "5m", etc.
-fn deserialize_duration<'de, D>(deserializer: D) -> Result<Duration, D::Error>
-where
-    D: serde::Deserializer<'de>,
-{
-    let s = String::deserialize(deserializer)?;
-    parse_duration(&s).map_err(serde::de::Error::custom)
-}
-
-fn parse_duration(s: &str) -> Result<Duration, String> {
-    let s = s.trim();
-    if let Some(secs) = s.strip_suffix('s') {
-        secs.trim()
-            .parse::<u64>()
-            .map(Duration::from_secs)
-            .map_err(|e| format!("invalid seconds: {e}"))
-    } else if let Some(mins) = s.strip_suffix('m') {
-        mins.trim()
-            .parse::<u64>()
-            .map(|m| Duration::from_secs(m * 60))
-            .map_err(|e| format!("invalid minutes: {e}"))
-    } else {
-        // Assume seconds if no suffix
-        s.parse::<u64>()
-            .map(Duration::from_secs)
-            .map_err(|e| format!("invalid duration '{s}': {e}"))
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_parse_duration() {
-        assert_eq!(parse_duration("300s").unwrap(), Duration::from_secs(300));
-        assert_eq!(parse_duration("5m").unwrap(), Duration::from_secs(300));
-        assert_eq!(parse_duration("60").unwrap(), Duration::from_secs(60));
-    }
-
-    #[test]
-    fn test_minimal_config() {
-        let config = BenchConfig::minimal(Some("test-model".to_string()));
-        assert_eq!(config.matrix.len(), 1);
-        assert_eq!(config.matrix[0].label, "test-model");
-        assert_eq!(config.parallelism, 1);
-    }
-
-    #[test]
-    fn test_config_rejects_empty_matrix() {
-        let dir = tempfile::tempdir().unwrap();
-        let path = dir.path().join("empty.toml");
-        std::fs::write(
-            &path,
-            r#"
-results_dir = "./results"
-task_timeout = "60s"
-"#,
-        )
-        .unwrap();
-        let err = BenchConfig::from_file(&path).unwrap_err();
-        assert!(
-            err.to_string().contains("at least one [[matrix]]"),
-            "got: {err}"
-        );
-    }
-
-    #[test]
-    fn test_config_from_toml() {
-        let toml_str = r#"
-results_dir = "./my-results"
-task_timeout = "60s"
-parallelism = 2
-
-[[matrix]]
-label = "fast"
-model = "gpt-4o-mini"
-
-[[matrix]]
-label = "full"
-model = "claude-3-5-sonnet"
-
-[suite_config]
-dataset_path = "./data/test.jsonl"
-"#;
-        let config: BenchConfig = toml::from_str(toml_str).unwrap();
-        assert_eq!(config.results_dir, PathBuf::from("./my-results"));
-        assert_eq!(config.task_timeout, Duration::from_secs(60));
-        assert_eq!(config.parallelism, 2);
-        assert_eq!(config.matrix.len(), 2);
-        assert_eq!(
-            config.suite_config_str("dataset_path").unwrap(),
-            "./data/test.jsonl"
-        );
-    }
-}

benchmarks/src/error.rs

@@ -1,31 +0,0 @@
-use std::path::PathBuf;
-
-#[derive(Debug, thiserror::Error)]
-pub enum BenchError {
-    #[error("Config error: {0}")]
-    Config(String),
-
-    #[error("Config file not found: {path}")]
-    ConfigNotFound { path: PathBuf },
-
-    #[error("Suite {name} not found. Available: {available}")]
-    SuiteNotFound { name: String, available: String },
-
-    #[error("Task {task_id} failed: {reason}")]
-    TaskFailed { task_id: String, reason: String },
-
-    #[error("Scoring error for task {task_id}: {reason}")]
-    Scoring { task_id: String, reason: String },
-
-    #[error("IO error: {0}")]
-    Io(#[from] std::io::Error),
-
-    #[error("JSON error: {0}")]
-    Json(#[from] serde_json::Error),
-
-    #[error("TOML parse error: {0}")]
-    Toml(#[from] toml::de::Error),
-
-    #[error("Agent error: {0}")]
-    Agent(#[from] ironclaw::Error),
-}

benchmarks/src/instrumented_llm.rs

@@ -1,251 +0,0 @@
-use std::sync::Arc;
-use std::sync::atomic::{AtomicU32, Ordering};
-use std::time::Instant;
-
-use async_trait::async_trait;
-use rust_decimal::Decimal;
-use rust_decimal::prelude::ToPrimitive;
-use tokio::sync::Mutex;
-
-use ironclaw::error::LlmError;
-use ironclaw::llm::{
-    CompletionRequest, CompletionResponse, LlmProvider, ToolCompletionRequest,
-    ToolCompletionResponse,
-};
-
-/// Recorded metrics from a single LLM call.
-#[derive(Debug, Clone)]
-#[allow(dead_code)]
-pub struct LlmCallRecord {
-    pub input_tokens: u32,
-    pub output_tokens: u32,
-    pub duration_ms: u64,
-    pub had_tool_calls: bool,
-}
-
-/// Wraps an `LlmProvider` to record per-call metrics.
-///
-/// The wrapper is transparent to the agent: it delegates every call
-/// to the inner provider and captures token counts and timings.
-pub struct InstrumentedLlm {
-    inner: Arc<dyn LlmProvider>,
-    records: Mutex<Vec<LlmCallRecord>>,
-    total_input_tokens: AtomicU32,
-    total_output_tokens: AtomicU32,
-    call_count: AtomicU32,
-}
-
-impl InstrumentedLlm {
-    pub fn new(inner: Arc<dyn LlmProvider>) -> Self {
-        Self {
-            inner,
-            records: Mutex::new(Vec::new()),
-            total_input_tokens: AtomicU32::new(0),
-            total_output_tokens: AtomicU32::new(0),
-            call_count: AtomicU32::new(0),
-        }
-    }
-
-    /// Take all recorded call metrics, clearing the internal buffer.
-    pub async fn take_records(&self) -> Vec<LlmCallRecord> {
-        let mut records = self.records.lock().await;
-        std::mem::take(&mut *records)
-    }
-
-    /// Snapshot of total tokens without clearing.
-    pub fn total_input_tokens(&self) -> u32 {
-        self.total_input_tokens.load(Ordering::Relaxed)
-    }
-
-    pub fn total_output_tokens(&self) -> u32 {
-        self.total_output_tokens.load(Ordering::Relaxed)
-    }
-
-    pub fn call_count(&self) -> u32 {
-        self.call_count.load(Ordering::Relaxed)
-    }
-
-    /// Estimated cost using the inner provider's cost-per-token rates.
-    pub fn estimated_cost(&self) -> f64 {
-        let (input_rate, output_rate) = self.inner.cost_per_token();
-        let input_cost =
-            input_rate * Decimal::from(self.total_input_tokens.load(Ordering::Relaxed));
-        let output_cost =
-            output_rate * Decimal::from(self.total_output_tokens.load(Ordering::Relaxed));
-        let total = input_cost + output_cost;
-        total.to_f64().unwrap_or(0.0)
-    }
-
-    /// Reset all counters and records.
-    pub async fn reset(&self) {
-        self.records.lock().await.clear();
-        self.total_input_tokens.store(0, Ordering::Relaxed);
-        self.total_output_tokens.store(0, Ordering::Relaxed);
-        self.call_count.store(0, Ordering::Relaxed);
-    }
-
-    async fn record(
-        &self,
-        input_tokens: u32,
-        output_tokens: u32,
-        duration_ms: u64,
-        had_tool_calls: bool,
-    ) {
-        self.total_input_tokens
-            .fetch_add(input_tokens, Ordering::Relaxed);
-        self.total_output_tokens
-            .fetch_add(output_tokens, Ordering::Relaxed);
-        self.call_count.fetch_add(1, Ordering::Relaxed);
-        self.records.lock().await.push(LlmCallRecord {
-            input_tokens,
-            output_tokens,
-            duration_ms,
-            had_tool_calls,
-        });
-    }
-}
-
-#[async_trait]
-impl LlmProvider for InstrumentedLlm {
-    fn model_name(&self) -> &str {
-        self.inner.model_name()
-    }
-
-    fn cost_per_token(&self) -> (Decimal, Decimal) {
-        self.inner.cost_per_token()
-    }
-
-    async fn complete(&self, request: CompletionRequest) -> Result<CompletionResponse, LlmError> {
-        let start = Instant::now();
-        let response = self.inner.complete(request).await?;
-        let elapsed = start.elapsed().as_millis() as u64;
-        self.record(
-            response.input_tokens,
-            response.output_tokens,
-            elapsed,
-            false,
-        )
-        .await;
-        Ok(response)
-    }
-
-    async fn complete_with_tools(
-        &self,
-        request: ToolCompletionRequest,
-    ) -> Result<ToolCompletionResponse, LlmError> {
-        let start = Instant::now();
-        let response = self.inner.complete_with_tools(request).await?;
-        let elapsed = start.elapsed().as_millis() as u64;
-        let had_tool_calls = !response.tool_calls.is_empty();
-        self.record(
-            response.input_tokens,
-            response.output_tokens,
-            elapsed,
-            had_tool_calls,
-        )
-        .await;
-        Ok(response)
-    }
-
-    async fn list_models(&self) -> Result<Vec<String>, LlmError> {
-        self.inner.list_models().await
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use ironclaw::llm::{ChatMessage, CompletionRequest, CompletionResponse, FinishReason};
-
-    /// Fake LLM that returns a canned response with known token counts.
-    struct FakeLlm;
-
-    #[async_trait]
-    impl LlmProvider for FakeLlm {
-        fn model_name(&self) -> &str {
-            "fake-model"
-        }
-
-        fn cost_per_token(&self) -> (Decimal, Decimal) {
-            (
-                Decimal::new(3, 6),  // $0.000003 per input token
-                Decimal::new(15, 6), // $0.000015 per output token
-            )
-        }
-
-        async fn complete(
-            &self,
-            _request: CompletionRequest,
-        ) -> Result<CompletionResponse, LlmError> {
-            Ok(CompletionResponse {
-                content: "test response".to_string(),
-                input_tokens: 100,
-                output_tokens: 50,
-                finish_reason: FinishReason::Stop,
-                response_id: None,
-            })
-        }
-
-        async fn complete_with_tools(
-            &self,
-            _request: ToolCompletionRequest,
-        ) -> Result<ToolCompletionResponse, LlmError> {
-            Ok(ToolCompletionResponse {
-                content: Some("tool response".to_string()),
-                tool_calls: vec![],
-                input_tokens: 200,
-                output_tokens: 100,
-                finish_reason: FinishReason::Stop,
-                response_id: None,
-            })
-        }
-    }
-
-    #[tokio::test]
-    async fn test_instrumented_records_metrics() {
-        let inner = Arc::new(FakeLlm);
-        let instrumented = InstrumentedLlm::new(inner);
-
-        let request = CompletionRequest::new(vec![ChatMessage::user("hello")]);
-        let _ = instrumented.complete(request).await.unwrap();
-
-        assert_eq!(instrumented.call_count(), 1);
-        assert_eq!(instrumented.total_input_tokens(), 100);
-        assert_eq!(instrumented.total_output_tokens(), 50);
-
-        let records = instrumented.take_records().await;
-        assert_eq!(records.len(), 1);
-        assert_eq!(records[0].input_tokens, 100);
-        assert!(!records[0].had_tool_calls);
-    }
-
-    #[tokio::test]
-    async fn test_instrumented_cost_calculation() {
-        let inner = Arc::new(FakeLlm);
-        let instrumented = InstrumentedLlm::new(inner);
-
-        let request = CompletionRequest::new(vec![ChatMessage::user("hello")]);
-        let _ = instrumented.complete(request).await.unwrap();
-
-        // 100 * 0.000003 + 50 * 0.000015 = 0.0003 + 0.00075 = 0.00105
-        let cost = instrumented.estimated_cost();
-        assert!((cost - 0.00105).abs() < 0.0001);
-    }
-
-    #[tokio::test]
-    async fn test_instrumented_reset() {
-        let inner = Arc::new(FakeLlm);
-        let instrumented = InstrumentedLlm::new(inner);
-
-        let request = CompletionRequest::new(vec![ChatMessage::user("hello")]);
-        let _ = instrumented.complete(request).await.unwrap();
-        assert_eq!(instrumented.call_count(), 1);
-
-        instrumented.reset().await;
-        assert_eq!(instrumented.call_count(), 0);
-        assert_eq!(instrumented.total_input_tokens(), 0);
-
-        let records = instrumented.take_records().await;
-        assert!(records.is_empty());
-    }
-}

benchmarks/src/main.rs

@@ -1,313 +0,0 @@
-mod adapters;
-mod channel;
-mod config;
-mod error;
-mod instrumented_llm;
-mod results;
-mod runner;
-mod scoring;
-mod suite;
-
-use std::path::PathBuf;
-use std::sync::Arc;
-
-use clap::{Parser, Subcommand};
-use tracing_subscriber::{EnvFilter, layer::SubscriberExt, util::SubscriberInitExt};
-use uuid::Uuid;
-
-use crate::config::BenchConfig;
-
-#[derive(Parser)]
-#[command(name = "ironclaw-bench", about = "IronClaw benchmarking harness")]
-struct Cli {
-    #[command(subcommand)]
-    command: Commands,
-}
-
-#[derive(Subcommand)]
-enum Commands {
-    /// Run a benchmark suite.
-    Run {
-        /// Suite to run (custom, gaia, spot, tau_bench, swe_bench).
-        #[arg(long)]
-        suite: String,
-
-        /// Path to bench config TOML.
-        #[arg(long)]
-        config: Option<PathBuf>,
-
-        /// Override model for all matrix entries.
-        #[arg(long)]
-        model: Option<String>,
-
-        /// Max tasks to run in parallel.
-        #[arg(long)]
-        parallelism: Option<usize>,
-
-        /// Sample N tasks from the suite (for quick testing).
-        #[arg(long)]
-        sample: Option<usize>,
-
-        /// Only run these task IDs (comma-separated).
-        #[arg(long, value_delimiter = ',')]
-        task_ids: Option<Vec<String>>,
-
-        /// Only run tasks with these tags (comma-separated).
-        #[arg(long, value_delimiter = ',')]
-        tags: Option<Vec<String>>,
-
-        /// Per-task timeout in seconds.
-        #[arg(long)]
-        timeout_secs: Option<u64>,
-
-        /// Override results directory.
-        #[arg(long)]
-        results_dir: Option<PathBuf>,
-
-        /// Resume a previous run by ID.
-        #[arg(long)]
-        resume: Option<Uuid>,
-    },
-
-    /// Show results for a run.
-    Results {
-        /// Run ID or "latest".
-        #[arg(default_value = "latest")]
-        run_id: String,
-
-        /// Output format.
-        #[arg(long, default_value = "table")]
-        format: ResultsFormat,
-
-        /// Override results directory.
-        #[arg(long)]
-        results_dir: Option<PathBuf>,
-    },
-
-    /// Compare two runs.
-    Compare {
-        /// Baseline run ID.
-        baseline: Uuid,
-
-        /// Comparison run ID.
-        comparison: Uuid,
-
-        /// Override results directory.
-        #[arg(long)]
-        results_dir: Option<PathBuf>,
-    },
-
-    /// List available benchmark suites.
-    List,
-}
-
-#[derive(Clone, Debug, clap::ValueEnum)]
-enum ResultsFormat {
-    Table,
-    Json,
-    Csv,
-}
-
-#[tokio::main]
-async fn main() -> anyhow::Result<()> {
-    let cli = Cli::parse();
-
-    tracing_subscriber::registry()
-        .with(
-            EnvFilter::try_from_default_env()
-                .unwrap_or_else(|_| EnvFilter::new("ironclaw_bench=info,ironclaw=warn")),
-        )
-        .with(tracing_subscriber::fmt::layer().with_target(false))
-        .init();
-
-    match cli.command {
-        Commands::List => {
-            println!("Available benchmark suites:\n");
-            for (id, desc) in adapters::KNOWN_SUITES {
-                println!("  {:<15} {}", id, desc);
-            }
-            println!();
-        }
-        Commands::Run {
-            suite,
-            config: config_path,
-            model,
-            parallelism,
-            sample,
-            task_ids,
-            tags,
-            timeout_secs,
-            results_dir,
-            resume,
-        } => {
-            // Load or create config
-            let mut bench_config = if let Some(ref path) = config_path {
-                BenchConfig::from_file(path)?
-            } else {
-                BenchConfig::minimal(model.clone())
-            };
-
-            // Apply CLI overrides
-            if let Some(p) = parallelism {
-                bench_config.parallelism = p;
-            }
-            if let Some(t) = timeout_secs {
-                bench_config.task_timeout = std::time::Duration::from_secs(t);
-            }
-            if let Some(ref dir) = results_dir {
-                bench_config.results_dir = dir.clone();
-            }
-
-            // If model override specified and we have matrix entries, update them
-            if let Some(ref m) = model {
-                for entry in &mut bench_config.matrix {
-                    entry.model = Some(m.clone());
-                }
-            }
-
-            // Create suite
-            let bench_suite = adapters::create_suite(&suite, &bench_config)?;
-
-            // Initialize ironclaw LLM provider
-            let ironclaw_config = ironclaw::Config::from_env().await.map_err(|e| {
-                anyhow::anyhow!(
-                    "Failed to load ironclaw config: {}. Make sure .env is configured.",
-                    e
-                )
-            })?;
-
-            let session = ironclaw::llm::create_session_manager(ironclaw::llm::SessionConfig {
-                auth_base_url: ironclaw_config.llm.nearai.auth_base_url.clone(),
-                session_path: ironclaw_config.llm.nearai.session_path.clone(),
-            })
-            .await;
-            session.ensure_authenticated().await?;
-
-            let llm = ironclaw::llm::create_llm_provider(&ironclaw_config.llm, session)?;
-            let safety = Arc::new(ironclaw::safety::SafetyLayer::new(&ironclaw_config.safety));
-
-            let runner = runner::BenchRunner::new(bench_suite, bench_config.clone(), llm, safety);
-
-            // Run for each matrix entry
-            for matrix_entry in &bench_config.matrix {
-                let run_id = runner
-                    .run(
-                        matrix_entry,
-                        sample,
-                        task_ids.as_deref(),
-                        tags.as_deref(),
-                        resume,
-                    )
-                    .await?;
-                println!("Run complete: {}", run_id);
-            }
-        }
-        Commands::Results {
-            run_id,
-            format,
-            results_dir,
-        } => {
-            let base = results_dir.unwrap_or_else(|| PathBuf::from("./bench-results"));
-            let uuid = if run_id == "latest" {
-                results::find_latest_run(&base)?
-                    .ok_or_else(|| anyhow::anyhow!("No runs found in {}", base.display()))?
-            } else {
-                Uuid::parse_str(&run_id)?
-            };
-
-            let json_path = results::run_json_path(&base, uuid);
-            let jsonl_path = results::tasks_jsonl_path(&base, uuid);
-
-            let run = results::read_run_result(&json_path)?;
-            let tasks = results::read_task_results(&jsonl_path)?;
-
-            match format {
-                ResultsFormat::Table => {
-                    results::print_results_table(&tasks, &run);
-                }
-                ResultsFormat::Json => {
-                    let output = serde_json::json!({
-                        "run": run,
-                        "tasks": tasks,
-                    });
-                    println!("{}", serde_json::to_string_pretty(&output)?);
-                }
-                ResultsFormat::Csv => {
-                    println!("task_id,score,label,tokens,cost,turns,time_s");
-                    for task in &tasks {
-                        println!(
-                            "{},{:.3},{},{},{:.4},{},{:.1}",
-                            task.task_id,
-                            task.score.value,
-                            task.score.label,
-                            task.trace.input_tokens + task.trace.output_tokens,
-                            task.trace.estimated_cost_usd,
-                            task.trace.turns,
-                            task.trace.wall_time_ms as f64 / 1000.0,
-                        );
-                    }
-                }
-            }
-        }
-        Commands::Compare {
-            baseline,
-            comparison,
-            results_dir,
-        } => {
-            let base = results_dir.unwrap_or_else(|| PathBuf::from("./bench-results"));
-
-            let baseline_run = results::read_run_result(&results::run_json_path(&base, baseline))?;
-            let comparison_run =
-                results::read_run_result(&results::run_json_path(&base, comparison))?;
-
-            println!("\nComparison: {} vs {}\n", baseline, comparison);
-            println!(
-                "{:<20} {:>12} {:>12} {:>10}",
-                "Metric", "Baseline", "Comparison", "Delta"
-            );
-            println!("{}", "-".repeat(58));
-
-            let pass_delta = comparison_run.pass_rate - baseline_run.pass_rate;
-            println!(
-                "{:<20} {:>11.1}% {:>11.1}% {:>+9.1}%",
-                "Pass rate",
-                baseline_run.pass_rate * 100.0,
-                comparison_run.pass_rate * 100.0,
-                pass_delta * 100.0,
-            );
-
-            let score_delta = comparison_run.avg_score - baseline_run.avg_score;
-            println!(
-                "{:<20} {:>12.3} {:>12.3} {:>+10.3}",
-                "Avg score", baseline_run.avg_score, comparison_run.avg_score, score_delta,
-            );
-
-            let cost_delta = comparison_run.total_cost_usd - baseline_run.total_cost_usd;
-            println!(
-                "{:<20} {:>11.4}$ {:>11.4}$ {:>+9.4}$",
-                "Total cost",
-                baseline_run.total_cost_usd,
-                comparison_run.total_cost_usd,
-                cost_delta,
-            );
-
-            let time_b = baseline_run.total_wall_time_ms as f64 / 1000.0;
-            let time_c = comparison_run.total_wall_time_ms as f64 / 1000.0;
-            println!(
-                "{:<20} {:>11.1}s {:>11.1}s {:>+9.1}s",
-                "Total time",
-                time_b,
-                time_c,
-                time_c - time_b,
-            );
-
-            println!(
-                "{:<20} {:>12} {:>12}",
-                "Model", baseline_run.model, comparison_run.model,
-            );
-            println!();
-        }
-    }
-
-    Ok(())
-}

benchmarks/src/results.rs

@@ -1,473 +0,0 @@
-use std::collections::HashSet;
-use std::io::{BufRead, Write};
-use std::path::{Path, PathBuf};
-
-use chrono::{DateTime, Utc};
-use uuid::Uuid;
-
-use crate::error::BenchError;
-use crate::suite::BenchScore;
-
-/// Metrics from a single task run: LLM usage, timing, tool calls.
-#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
-pub struct Trace {
-    pub wall_time_ms: u64,
-    pub llm_calls: u32,
-    pub input_tokens: u32,
-    pub output_tokens: u32,
-    pub estimated_cost_usd: f64,
-    pub tool_calls: Vec<TraceToolCall>,
-    pub turns: u32,
-    pub hit_iteration_limit: bool,
-    pub hit_timeout: bool,
-}
-
-#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
-pub struct TraceToolCall {
-    pub name: String,
-    pub duration_ms: u64,
-    pub success: bool,
-}
-
-/// Result of running a single benchmark task.
-#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
-pub struct TaskResult {
-    pub task_id: String,
-    pub suite_id: String,
-    pub score: BenchScore,
-    pub trace: Trace,
-    pub response: String,
-    pub started_at: DateTime<Utc>,
-    pub finished_at: DateTime<Utc>,
-    pub config_label: String,
-    #[serde(default)]
-    pub error: Option<String>,
-}
-
-/// Aggregate results for a full benchmark run.
-#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
-pub struct RunResult {
-    pub run_id: Uuid,
-    pub suite_id: String,
-    pub config_label: String,
-    pub model: String,
-    /// Short git commit hash at the time of the run.
-    #[serde(default)]
-    pub commit_hash: String,
-    pub pass_rate: f64,
-    pub avg_score: f64,
-    pub total_tasks: usize,
-    pub completed_tasks: usize,
-    pub total_cost_usd: f64,
-    pub total_wall_time_ms: u64,
-    pub started_at: DateTime<Utc>,
-    pub finished_at: DateTime<Utc>,
-}
-
-impl RunResult {
-    /// Build aggregate from individual task results.
-    #[allow(clippy::too_many_arguments)]
-    pub fn from_tasks(
-        run_id: Uuid,
-        suite_id: &str,
-        config_label: &str,
-        model: &str,
-        commit_hash: &str,
-        total_tasks: usize,
-        tasks: &[TaskResult],
-        started_at: DateTime<Utc>,
-    ) -> Self {
-        let pass_count = tasks.iter().filter(|t| t.score.value >= 1.0).count();
-        let pass_rate = if tasks.is_empty() {
-            0.0
-        } else {
-            pass_count as f64 / tasks.len() as f64
-        };
-        let avg_score = if tasks.is_empty() {
-            0.0
-        } else {
-            tasks.iter().map(|t| t.score.value).sum::<f64>() / tasks.len() as f64
-        };
-        let total_cost: f64 = tasks.iter().map(|t| t.trace.estimated_cost_usd).sum();
-        let total_wall: u64 = tasks.iter().map(|t| t.trace.wall_time_ms).sum();
-
-        Self {
-            run_id,
-            suite_id: suite_id.to_string(),
-            config_label: config_label.to_string(),
-            model: model.to_string(),
-            commit_hash: commit_hash.to_string(),
-            pass_rate,
-            avg_score,
-            total_tasks,
-            completed_tasks: tasks.len(),
-            total_cost_usd: total_cost,
-            total_wall_time_ms: total_wall,
-            started_at,
-            finished_at: Utc::now(),
-        }
-    }
-}
-
-/// Append a single task result as one JSON line to the JSONL file.
-pub fn append_task_result(path: &Path, result: &TaskResult) -> Result<(), BenchError> {
-    let mut file = std::fs::OpenOptions::new()
-        .create(true)
-        .append(true)
-        .open(path)?;
-    let line = serde_json::to_string(result)?;
-    writeln!(file, "{line}")?;
-    Ok(())
-}
-
-/// Overwrite the JSONL file with the given results (used after scoring).
-pub fn write_task_results(path: &Path, results: &[TaskResult]) -> Result<(), BenchError> {
-    let mut file = std::fs::File::create(path)?;
-    for result in results {
-        let line = serde_json::to_string(result)?;
-        writeln!(file, "{line}")?;
-    }
-    Ok(())
-}
-
-/// Read all task results from a JSONL file.
-pub fn read_task_results(path: &Path) -> Result<Vec<TaskResult>, BenchError> {
-    if !path.exists() {
-        return Ok(Vec::new());
-    }
-    let file = std::fs::File::open(path)?;
-    let reader = std::io::BufReader::new(file);
-    let mut results = Vec::new();
-    for line in reader.lines() {
-        let line = line?;
-        let trimmed = line.trim();
-        if trimmed.is_empty() {
-            continue;
-        }
-        let result: TaskResult = serde_json::from_str(trimmed)?;
-        results.push(result);
-    }
-    Ok(results)
-}
-
-/// Write the aggregate run result as JSON.
-pub fn write_run_result(path: &Path, result: &RunResult) -> Result<(), BenchError> {
-    let json = serde_json::to_string_pretty(result)?;
-    std::fs::write(path, json)?;
-    Ok(())
-}
-
-/// Read the aggregate run result from JSON.
-pub fn read_run_result(path: &Path) -> Result<RunResult, BenchError> {
-    let json = std::fs::read_to_string(path)?;
-    let result: RunResult = serde_json::from_str(&json)?;
-    Ok(result)
-}
-
-/// Get the set of already-completed task IDs from a JSONL file (for resume).
-///
-/// Only includes tasks that have been scored (label != "pending"). Tasks that
-/// were written but not scored (e.g., from an interrupted run) will be re-executed.
-pub fn completed_task_ids(path: &Path) -> Result<HashSet<String>, BenchError> {
-    let results = read_task_results(path)?;
-    Ok(results
-        .into_iter()
-        .filter(|r| r.score.label != "pending")
-        .map(|r| r.task_id)
-        .collect())
-}
-
-/// Get the results directory for a specific run.
-pub fn run_dir(base: &Path, run_id: Uuid) -> PathBuf {
-    base.join(run_id.to_string())
-}
-
-/// Get the tasks JSONL path for a run.
-pub fn tasks_jsonl_path(base: &Path, run_id: Uuid) -> PathBuf {
-    run_dir(base, run_id).join("tasks.jsonl")
-}
-
-/// Get the run JSON path for a run.
-pub fn run_json_path(base: &Path, run_id: Uuid) -> PathBuf {
-    run_dir(base, run_id).join("run.json")
-}
-
-/// Find the latest run directory by the modification time of its `run.json`.
-///
-/// Falls back to `tasks.jsonl` mtime, then directory mtime. This avoids the
-/// issue where modifying files inside a directory doesn't update the directory's
-/// mtime on many filesystems.
-pub fn find_latest_run(base: &Path) -> Result<Option<Uuid>, BenchError> {
-    if !base.exists() {
-        return Ok(None);
-    }
-    let mut entries: Vec<_> = std::fs::read_dir(base)?
-        .filter_map(|e| e.ok())
-        .filter(|e| e.file_type().map(|ft| ft.is_dir()).unwrap_or(false))
-        .filter_map(|e| {
-            let name = e.file_name().to_string_lossy().to_string();
-            let uuid = Uuid::parse_str(&name).ok()?;
-            let dir_path = e.path();
-            // Prefer run.json mtime, fall back to tasks.jsonl, then directory
-            let modified = std::fs::metadata(dir_path.join("run.json"))
-                .and_then(|m| m.modified())
-                .or_else(|_| {
-                    std::fs::metadata(dir_path.join("tasks.jsonl")).and_then(|m| m.modified())
-                })
-                .or_else(|_| e.metadata().and_then(|m| m.modified()))
-                .ok()?;
-            Some((uuid, modified))
-        })
-        .collect();
-    entries.sort_by(|a, b| b.1.cmp(&a.1));
-    Ok(entries.first().map(|(uuid, _)| *uuid))
-}
-
-/// Print a summary table of task results.
-pub fn print_results_table(tasks: &[TaskResult], run: &RunResult) {
-    println!();
-    let commit_suffix = if run.commit_hash.is_empty() {
-        String::new()
-    } else {
-        format!(" | Commit: {}", run.commit_hash)
-    };
-    println!(
-        "Run: {} | Suite: {} | Model: {}{}",
-        run.run_id, run.suite_id, run.model, commit_suffix
-    );
-    println!(
-        "Pass rate: {:.1}% | Avg score: {:.3} | Tasks: {}/{} | Cost: ${:.4} | Time: {:.1}s",
-        run.pass_rate * 100.0,
-        run.avg_score,
-        run.completed_tasks,
-        run.total_tasks,
-        run.total_cost_usd,
-        run.total_wall_time_ms as f64 / 1000.0,
-    );
-    println!();
-
-    // Header
-    println!(
-        "{:<30} {:>6} {:>7} {:>8} {:>10} {:>6} {:>8}",
-        "Task ID", "Score", "Label", "Tokens", "Cost", "Turns", "Time"
-    );
-    println!("{}", "-".repeat(80));
-
-    for task in tasks {
-        let total_tokens = task.trace.input_tokens + task.trace.output_tokens;
-        let task_id_display = if task.task_id.len() > 28 {
-            let truncated: String = task.task_id.chars().take(25).collect();
-            format!("{truncated}...")
-        } else {
-            task.task_id.clone()
-        };
-        println!(
-            "{:<30} {:>6.3} {:>7} {:>8} {:>10.4} {:>6} {:>7.1}s",
-            task_id_display,
-            task.score.value,
-            task.score.label,
-            total_tokens,
-            task.trace.estimated_cost_usd,
-            task.trace.turns,
-            task.trace.wall_time_ms as f64 / 1000.0,
-        );
-    }
-    println!();
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_run_result_from_tasks() {
-        let tasks = vec![
-            TaskResult {
-                task_id: "t1".to_string(),
-                suite_id: "custom".to_string(),
-                score: BenchScore {
-                    value: 1.0,
-                    label: "pass".to_string(),
-                    details: None,
-                },
-                trace: Trace {
-                    wall_time_ms: 1000,
-                    llm_calls: 2,
-                    input_tokens: 100,
-                    output_tokens: 50,
-                    estimated_cost_usd: 0.01,
-                    tool_calls: vec![],
-                    turns: 1,
-                    hit_iteration_limit: false,
-                    hit_timeout: false,
-                },
-                response: "answer".to_string(),
-                started_at: Utc::now(),
-                finished_at: Utc::now(),
-                config_label: "default".to_string(),
-                error: None,
-            },
-            TaskResult {
-                task_id: "t2".to_string(),
-                suite_id: "custom".to_string(),
-                score: BenchScore {
-                    value: 0.0,
-                    label: "fail".to_string(),
-                    details: Some("wrong".to_string()),
-                },
-                trace: Trace {
-                    wall_time_ms: 2000,
-                    llm_calls: 3,
-                    input_tokens: 200,
-                    output_tokens: 100,
-                    estimated_cost_usd: 0.02,
-                    tool_calls: vec![],
-                    turns: 2,
-                    hit_iteration_limit: false,
-                    hit_timeout: false,
-                },
-                response: "wrong answer".to_string(),
-                started_at: Utc::now(),
-                finished_at: Utc::now(),
-                config_label: "default".to_string(),
-                error: None,
-            },
-        ];
-
-        let run = RunResult::from_tasks(
-            Uuid::new_v4(),
-            "custom",
-            "default",
-            "test-model",
-            "abc1234",
-            2,
-            &tasks,
-            Utc::now(),
-        );
-
-        assert_eq!(run.pass_rate, 0.5);
-        assert_eq!(run.avg_score, 0.5);
-        assert_eq!(run.total_tasks, 2);
-        assert_eq!(run.completed_tasks, 2);
-        assert!((run.total_cost_usd - 0.03).abs() < f64::EPSILON);
-        assert_eq!(run.total_wall_time_ms, 3000);
-    }
-
-    #[test]
-    fn test_jsonl_roundtrip() {
-        let dir = tempfile::tempdir().expect("temp dir");
-        let path = dir.path().join("tasks.jsonl");
-
-        let result = TaskResult {
-            task_id: "round-trip-test".to_string(),
-            suite_id: "custom".to_string(),
-            score: BenchScore::pass(),
-            trace: Trace {
-                wall_time_ms: 500,
-                llm_calls: 1,
-                input_tokens: 10,
-                output_tokens: 5,
-                estimated_cost_usd: 0.001,
-                tool_calls: vec![],
-                turns: 1,
-                hit_iteration_limit: false,
-                hit_timeout: false,
-            },
-            response: "hello".to_string(),
-            started_at: Utc::now(),
-            finished_at: Utc::now(),
-            config_label: "test".to_string(),
-            error: None,
-        };
-
-        append_task_result(&path, &result).expect("append");
-        append_task_result(&path, &result).expect("append");
-
-        let loaded = read_task_results(&path).expect("read");
-        assert_eq!(loaded.len(), 2);
-        assert_eq!(loaded[0].task_id, "round-trip-test");
-    }
-
-    #[test]
-    fn test_completed_task_ids() {
-        let dir = tempfile::tempdir().expect("temp dir");
-        let path = dir.path().join("tasks.jsonl");
-
-        let result = TaskResult {
-            task_id: "unique-id-1".to_string(),
-            suite_id: "custom".to_string(),
-            score: BenchScore::pass(),
-            trace: Trace {
-                wall_time_ms: 100,
-                llm_calls: 1,
-                input_tokens: 10,
-                output_tokens: 5,
-                estimated_cost_usd: 0.0,
-                tool_calls: vec![],
-                turns: 1,
-                hit_iteration_limit: false,
-                hit_timeout: false,
-            },
-            response: "x".to_string(),
-            started_at: Utc::now(),
-            finished_at: Utc::now(),
-            config_label: "test".to_string(),
-            error: None,
-        };
-        append_task_result(&path, &result).expect("append");
-
-        let ids = completed_task_ids(&path).expect("ids");
-        assert!(ids.contains("unique-id-1"));
-        assert!(!ids.contains("unique-id-2"));
-    }
-
-    #[test]
-    fn test_write_task_results_overwrites() {
-        let dir = tempfile::tempdir().expect("temp dir");
-        let path = dir.path().join("tasks.jsonl");
-
-        // Write initial "pending" result via append
-        let pending = TaskResult {
-            task_id: "t1".to_string(),
-            suite_id: "spot".to_string(),
-            score: BenchScore {
-                value: 0.0,
-                label: "pending".to_string(),
-                details: None,
-            },
-            trace: Trace {
-                wall_time_ms: 100,
-                llm_calls: 1,
-                input_tokens: 10,
-                output_tokens: 5,
-                estimated_cost_usd: 0.001,
-                tool_calls: vec![],
-                turns: 1,
-                hit_iteration_limit: false,
-                hit_timeout: false,
-            },
-            response: "42".to_string(),
-            started_at: Utc::now(),
-            finished_at: Utc::now(),
-            config_label: "default".to_string(),
-            error: None,
-        };
-        append_task_result(&path, &pending).expect("append");
-
-        // Verify pending score
-        let before = read_task_results(&path).expect("read");
-        assert_eq!(before.len(), 1);
-        assert_eq!(before[0].score.label, "pending");
-
-        // Overwrite with scored result
-        let mut scored = pending;
-        scored.score = BenchScore::pass();
-        write_task_results(&path, &[scored]).expect("write");
-
-        // Verify scored result replaced pending
-        let after = read_task_results(&path).expect("read");
-        assert_eq!(after.len(), 1);
-        assert_eq!(after[0].score.label, "pass");
-        assert_eq!(after[0].score.value, 1.0);
-    }
-}

benchmarks/src/runner.rs

@@ -1,550 +0,0 @@
-use std::collections::{HashMap, HashSet};
-use std::sync::Arc;
-use std::time::Instant;
-
-use chrono::Utc;
-use tokio::sync::Mutex;
-use uuid::Uuid;
-
-use ironclaw::agent::{Agent, AgentDeps};
-use ironclaw::channels::{ChannelManager, IncomingMessage};
-use ironclaw::config::AgentConfig;
-use ironclaw::llm::LlmProvider;
-use ironclaw::safety::SafetyLayer;
-use ironclaw::tools::ToolRegistry;
-
-use crate::channel::BenchChannel;
-use crate::config::{BenchConfig, MatrixEntry};
-use crate::error::BenchError;
-use crate::instrumented_llm::InstrumentedLlm;
-use crate::results::{
-    RunResult, TaskResult, Trace, append_task_result, completed_task_ids, run_dir, run_json_path,
-    tasks_jsonl_path, write_run_result, write_task_results,
-};
-use crate::suite::{BenchSuite, BenchTask, ConversationTurn, TaskSubmission, TurnRole};
-
-/// Parameters for running a single task in isolation.
-struct TaskRunParams<'a> {
-    task: &'a BenchTask,
-    suite_id: &'a str,
-    config_label: &'a str,
-    llm: Arc<dyn LlmProvider>,
-    safety: Arc<SafetyLayer>,
-    timeout: std::time::Duration,
-    additional_tools: &'a [Arc<dyn ironclaw::tools::Tool>],
-}
-
-/// Orchestrates benchmark execution: loads tasks, runs agent per task,
-/// scores results, writes JSONL output.
-pub struct BenchRunner {
-    suite: Arc<dyn BenchSuite>,
-    config: BenchConfig,
-    llm: Arc<dyn LlmProvider>,
-    safety: Arc<SafetyLayer>,
-}
-
-impl BenchRunner {
-    pub fn new(
-        suite: Box<dyn BenchSuite>,
-        config: BenchConfig,
-        llm: Arc<dyn LlmProvider>,
-        safety: Arc<SafetyLayer>,
-    ) -> Self {
-        Self {
-            suite: Arc::from(suite),
-            config,
-            llm,
-            safety,
-        }
-    }
-
-    /// Run the benchmark for one matrix entry.
-    ///
-    /// Returns the run_id for result retrieval.
-    pub async fn run(
-        &self,
-        matrix: &MatrixEntry,
-        sample: Option<usize>,
-        task_filter: Option<&[String]>,
-        tag_filter: Option<&[String]>,
-        resume_run_id: Option<Uuid>,
-    ) -> Result<Uuid, BenchError> {
-        let run_id = resume_run_id.unwrap_or_else(Uuid::new_v4);
-        let results_base = &self.config.results_dir;
-        let dir = run_dir(results_base, run_id);
-        std::fs::create_dir_all(&dir)?;
-
-        let jsonl_path = tasks_jsonl_path(results_base, run_id);
-        let json_path = run_json_path(results_base, run_id);
-
-        // Load completed task IDs for resume support
-        let completed: HashSet<String> = if resume_run_id.is_some() {
-            completed_task_ids(&jsonl_path)?
-        } else {
-            HashSet::new()
-        };
-
-        if !completed.is_empty() {
-            tracing::info!(
-                "Resuming run {}: {} tasks already completed",
-                run_id,
-                completed.len()
-            );
-        }
-
-        // Load all tasks once (used for both execution and scoring)
-        let all_tasks = self.suite.load_tasks().await?;
-        let task_index: HashMap<String, BenchTask> = all_tasks
-            .iter()
-            .map(|t| (t.id.clone(), t.clone()))
-            .collect();
-
-        // Filter tasks for execution
-        let mut tasks = all_tasks;
-
-        if let Some(ids) = task_filter {
-            let id_set: HashSet<&str> = ids.iter().map(|s| s.as_str()).collect();
-            tasks.retain(|t| id_set.contains(t.id.as_str()));
-        }
-
-        if let Some(tags) = tag_filter {
-            let tag_set: HashSet<&str> = tags.iter().map(|s| s.as_str()).collect();
-            tasks.retain(|t| t.tags.iter().any(|tag| tag_set.contains(tag.as_str())));
-        }
-
-        // Filter out already-completed tasks
-        tasks.retain(|t| !completed.contains(&t.id));
-
-        // Sample if requested
-        if let Some(n) = sample {
-            tasks.truncate(n);
-        }
-
-        let total_tasks = tasks.len() + completed.len();
-        let model_label = matrix.model.as_deref().unwrap_or(self.llm.model_name());
-        let commit_hash = git_short_hash();
-        tracing::info!(
-            "[{} @ {}] Running {} tasks for suite '{}' (run: {})",
-            model_label,
-            commit_hash,
-            tasks.len(),
-            self.suite.id(),
-            run_id
-        );
-
-        let started_at = Utc::now();
-        let all_results: Arc<Mutex<Vec<TaskResult>>> =
-            Arc::new(Mutex::new(Vec::with_capacity(tasks.len())));
-
-        if self.config.parallelism <= 1 {
-            // Sequential execution
-            let additional_tools = self.suite.additional_tools();
-            for (i, task) in tasks.iter().enumerate() {
-                tracing::info!(
-                    "[{}/{}] Running task: {}",
-                    i + 1 + completed.len(),
-                    total_tasks,
-                    task.id
-                );
-                if let Err(e) = self.suite.setup_task(task).await {
-                    tracing::warn!("setup_task failed for {}: {}", task.id, e);
-                    let result = make_error_result(
-                        task,
-                        self.suite.id(),
-                        &matrix.label,
-                        Utc::now(),
-                        &format!("setup_task failed: {e}"),
-                    );
-                    append_task_result(&jsonl_path, &result)?;
-                    all_results.lock().await.push(result);
-                    continue;
-                }
-                let params = TaskRunParams {
-                    task,
-                    suite_id: self.suite.id(),
-                    config_label: &matrix.label,
-                    llm: Arc::clone(&self.llm),
-                    safety: Arc::clone(&self.safety),
-                    timeout: task.timeout.unwrap_or(self.config.task_timeout),
-                    additional_tools: &additional_tools,
-                };
-                let result = run_task_isolated(params).await;
-                if let Err(e) = self.suite.teardown_task(task).await {
-                    tracing::warn!("teardown_task failed for {}: {}", task.id, e);
-                }
-                append_task_result(&jsonl_path, &result)?;
-                all_results.lock().await.push(result);
-            }
-        } else {
-            // Parallel execution with bounded concurrency
-            let semaphore = Arc::new(tokio::sync::Semaphore::new(self.config.parallelism));
-            let shared_tools: Arc<[Arc<dyn ironclaw::tools::Tool>]> =
-                Arc::from(self.suite.additional_tools());
-
-            let mut handles = Vec::new();
-            for (i, task) in tasks.into_iter().enumerate() {
-                let sem = Arc::clone(&semaphore);
-                let suite = Arc::clone(&self.suite);
-                let config_label = matrix.label.clone();
-                let llm = Arc::clone(&self.llm);
-                let safety = Arc::clone(&self.safety);
-                let timeout = task.timeout.unwrap_or(self.config.task_timeout);
-                let results_ref = Arc::clone(&all_results);
-                let completed_count = completed.len();
-                let total = total_tasks;
-                let additional_tools = Arc::clone(&shared_tools);
-
-                handles.push(tokio::spawn(async move {
-                    let _permit = match sem.acquire().await {
-                        Ok(p) => p,
-                        Err(_) => {
-                            tracing::error!("Semaphore closed for task {}", task.id);
-                            return;
-                        }
-                    };
-                    tracing::info!(
-                        "[{}/{}] Running task: {}",
-                        i + 1 + completed_count,
-                        total,
-                        task.id
-                    );
-                    if let Err(e) = suite.setup_task(&task).await {
-                        tracing::warn!("setup_task failed for {}: {}", task.id, e);
-                        let result = make_error_result(
-                            &task,
-                            suite.id(),
-                            &config_label,
-                            Utc::now(),
-                            &format!("setup_task failed: {e}"),
-                        );
-                        results_ref.lock().await.push(result);
-                        return;
-                    }
-                    let suite_id = suite.id().to_string();
-                    let params = TaskRunParams {
-                        task: &task,
-                        suite_id: &suite_id,
-                        config_label: &config_label,
-                        llm,
-                        safety,
-                        timeout,
-                        additional_tools: &additional_tools,
-                    };
-                    let result = run_task_isolated(params).await;
-                    if let Err(e) = suite.teardown_task(&task).await {
-                        tracing::warn!("teardown_task failed for {}: {}", task.id, e);
-                    }
-                    results_ref.lock().await.push(result);
-                }));
-            }
-
-            for handle in handles {
-                if let Err(e) = handle.await {
-                    tracing::error!("Task panicked: {}", e);
-                }
-            }
-
-            // Write all results to JSONL after parallel execution completes.
-            // This avoids the race condition of concurrent file appends.
-            let results = all_results.lock().await;
-            for result in results.iter() {
-                append_task_result(&jsonl_path, result)?;
-            }
-        }
-
-        // Score all results using the cached task index
-        let results = all_results.lock().await;
-        let mut scored: Vec<TaskResult> = Vec::with_capacity(results.len());
-        for result in results.iter() {
-            if let Some(task) = task_index.get(&result.task_id) {
-                let submission = TaskSubmission {
-                    response: result.response.clone(),
-                    conversation: vec![],
-                    tool_calls: result
-                        .trace
-                        .tool_calls
-                        .iter()
-                        .map(|tc| tc.name.clone())
-                        .collect(),
-                    error: result.error.clone(),
-                };
-                match self.suite.score(task, &submission).await {
-                    Ok(score) => {
-                        let mut scored_result = result.clone();
-                        scored_result.score = score;
-                        scored.push(scored_result);
-                    }
-                    Err(e) => {
-                        tracing::warn!("Scoring failed for {}: {}", result.task_id, e);
-                        scored.push(result.clone());
-                    }
-                }
-            } else {
-                scored.push(result.clone());
-            }
-        }
-
-        // Combine with any previously completed results for the aggregate
-        let mut all_for_aggregate = crate::results::read_task_results(&jsonl_path)?;
-        // De-duplicate (prefer the newer scored versions)
-        let scored_ids: HashSet<String> = scored.iter().map(|r| r.task_id.clone()).collect();
-        all_for_aggregate.retain(|r| !scored_ids.contains(&r.task_id));
-        all_for_aggregate.extend(scored);
-
-        // Rewrite JSONL with scored results so `results` command shows final scores
-        write_task_results(&jsonl_path, &all_for_aggregate)?;
-
-        let model_name = matrix.model.as_deref().unwrap_or(self.llm.model_name());
-
-        let run_result = RunResult::from_tasks(
-            run_id,
-            self.suite.id(),
-            &matrix.label,
-            model_name,
-            &commit_hash,
-            total_tasks,
-            &all_for_aggregate,
-            started_at,
-        );
-
-        write_run_result(&json_path, &run_result)?;
-
-        tracing::info!(
-            "[{} @ {}] Run {} complete: {:.1}% pass rate, {:.3} avg score, ${:.4} cost",
-            model_name,
-            commit_hash,
-            run_id,
-            run_result.pass_rate * 100.0,
-            run_result.avg_score,
-            run_result.total_cost_usd,
-        );
-
-        Ok(run_id)
-    }
-}
-
-/// Run a single benchmark task in complete isolation.
-///
-/// Creates a fresh Agent + BenchChannel + InstrumentedLlm for the task,
-/// injects the prompt, waits for the response, and returns the result.
-///
-/// # Current limitations
-///
-/// - **Single-turn only**: After the first assistant response, `/quit` is sent.
-///   Multi-turn suites (e.g., Tau-bench's `next_user_message()`) are not yet wired.
-/// - **Resources not injected**: `BenchTask.resources` (e.g., GAIA file attachments)
-///   are not included in the prompt or made available via the workspace.
-/// - **Conversation not captured**: `TaskSubmission.conversation` is always empty,
-///   which prevents multi-turn scoring hooks from working.
-async fn run_task_isolated(params: TaskRunParams<'_>) -> TaskResult {
-    let TaskRunParams {
-        task,
-        suite_id,
-        config_label,
-        llm,
-        safety,
-        timeout,
-        additional_tools,
-    } = params;
-
-    let started_at = Utc::now();
-    let start = Instant::now();
-
-    // Wrap LLM with instrumentation
-    let instrumented = Arc::new(InstrumentedLlm::new(llm));
-
-    // Create bench channel
-    let (bench_channel, msg_tx) = BenchChannel::new();
-    let capture = bench_channel.capture();
-
-    // Build tool registry
-    let tools = Arc::new(ToolRegistry::new());
-    tools.register_builtin_tools();
-
-    // Register additional suite-specific tools
-    for tool in additional_tools {
-        tools.register(Arc::clone(tool)).await;
-    }
-
-    // Build agent config (minimal, headless)
-    let agent_config = AgentConfig {
-        name: format!("bench-{}", task.id),
-        max_parallel_jobs: 1,
-        job_timeout: timeout,
-        stuck_threshold: timeout,
-        repair_check_interval: timeout + std::time::Duration::from_secs(999),
-        max_repair_attempts: 0,
-        use_planning: false,
-        session_idle_timeout: timeout,
-        allow_local_tools: true,
-        max_cost_per_day_cents: None,
-        max_actions_per_hour: None,
-    };
-
-    let cost_guard = Arc::new(ironclaw::agent::cost_guard::CostGuard::new(
-        ironclaw::agent::cost_guard::CostGuardConfig::default(),
-    ));
-
-    let deps = AgentDeps {
-        store: None,
-        llm: instrumented.clone() as Arc<dyn LlmProvider>,
-        cheap_llm: None,
-        safety,
-        tools,
-        workspace: None,
-        extension_manager: None,
-        skill_registry: None,
-        skills_config: ironclaw::config::SkillsConfig::default(),
-        hooks: Arc::new(ironclaw::hooks::HookRegistry::new()),
-        cost_guard,
-    };
-
-    let mut channels = ChannelManager::new();
-    channels.add(Box::new(bench_channel));
-
-    let agent = Agent::new(agent_config, deps, channels, None, None, None, None);
-
-    // Build the full prompt with context
-    let full_prompt = if let Some(ref ctx) = task.context {
-        format!("{}\n\nContext:\n{}", task.prompt, ctx)
-    } else {
-        task.prompt.clone()
-    };
-
-    // Inject the task prompt
-    let incoming = IncomingMessage::new("bench", "bench-user", &full_prompt);
-    if msg_tx.send(incoming).await.is_err() {
-        return make_error_result(
-            task,
-            suite_id,
-            config_label,
-            started_at,
-            "failed to send prompt",
-        );
-    }
-
-    // Record prompt in conversation
-    {
-        let mut cap = capture.lock().await;
-        cap.conversation.push(ConversationTurn {
-            role: TurnRole::User,
-            content: full_prompt,
-        });
-    }
-
-    // Run agent with timeout.
-    // After the first response, send /quit to end the session.
-    let quit_tx = msg_tx.clone();
-    let capture_for_quit = Arc::clone(&capture);
-    let quit_handle = tokio::spawn(async move {
-        // Poll for first response
-        loop {
-            tokio::time::sleep(std::time::Duration::from_millis(100)).await;
-            let cap = capture_for_quit.lock().await;
-            if !cap.responses.is_empty() {
-                break;
-            }
-        }
-        // Give a small grace period for any final status events
-        tokio::time::sleep(std::time::Duration::from_millis(200)).await;
-        let quit = IncomingMessage::new("bench", "bench-user", "/quit");
-        let _ = quit_tx.send(quit).await;
-    });
-
-    let agent_result = tokio::time::timeout(timeout, agent.run()).await;
-
-    quit_handle.abort();
-
-    let wall_time = start.elapsed();
-    let hit_timeout = agent_result.is_err();
-
-    if let Ok(Err(e)) = &agent_result {
-        tracing::warn!("Agent error for task {}: {}", task.id, e);
-    }
-
-    // Extract results from capture
-    let cap = capture.lock().await;
-    let response = cap.responses.last().cloned().unwrap_or_default();
-
-    let trace = Trace {
-        wall_time_ms: wall_time.as_millis() as u64,
-        llm_calls: instrumented.call_count(),
-        input_tokens: instrumented.total_input_tokens(),
-        output_tokens: instrumented.total_output_tokens(),
-        estimated_cost_usd: instrumented.estimated_cost(),
-        tool_calls: cap.tool_calls.clone(),
-        turns: cap.responses.len() as u32,
-        hit_iteration_limit: false,
-        hit_timeout,
-    };
-
-    let error = if hit_timeout {
-        Some(format!("timeout after {}s", timeout.as_secs()))
-    } else if let Ok(Err(e)) = &agent_result {
-        Some(e.to_string())
-    } else {
-        None
-    };
-
-    TaskResult {
-        task_id: task.id.clone(),
-        suite_id: suite_id.to_string(),
-        score: crate::suite::BenchScore {
-            value: 0.0,
-            label: "pending".to_string(),
-            details: None,
-        },
-        trace,
-        response,
-        started_at,
-        finished_at: Utc::now(),
-        config_label: config_label.to_string(),
-        error,
-    }
-}
-
-fn make_error_result(
-    task: &BenchTask,
-    suite_id: &str,
-    config_label: &str,
-    started_at: chrono::DateTime<Utc>,
-    reason: &str,
-) -> TaskResult {
-    TaskResult {
-        task_id: task.id.clone(),
-        suite_id: suite_id.to_string(),
-        score: crate::suite::BenchScore::fail(reason),
-        trace: Trace {
-            wall_time_ms: 0,
-            llm_calls: 0,
-            input_tokens: 0,
-            output_tokens: 0,
-            estimated_cost_usd: 0.0,
-            tool_calls: vec![],
-            turns: 0,
-            hit_iteration_limit: false,
-            hit_timeout: false,
-        },
-        response: String::new(),
-        started_at,
-        finished_at: Utc::now(),
-        config_label: config_label.to_string(),
-        error: Some(reason.to_string()),
-    }
-}
-
-/// Get the short git commit hash of HEAD, or "unknown" if not in a repo.
-fn git_short_hash() -> String {
-    std::process::Command::new("git")
-        .args(["rev-parse", "--short", "HEAD"])
-        .output()
-        .ok()
-        .and_then(|o| {
-            if o.status.success() {
-                Some(String::from_utf8_lossy(&o.stdout).trim().to_string())
-            } else {
-                None
-            }
-        })
-        .unwrap_or_else(|| "unknown".to_string())
-}

benchmarks/src/scoring.rs

@@ -1,113 +0,0 @@
-use regex::Regex;
-
-use crate::suite::BenchScore;
-
-/// Normalize an answer string for comparison: lowercase, trim whitespace,
-/// strip trailing punctuation, collapse internal whitespace.
-pub fn normalize_answer(s: &str) -> String {
-    let trimmed = s.trim().to_lowercase();
-    let collapsed: String = trimmed.split_whitespace().collect::<Vec<_>>().join(" ");
-    collapsed.trim_end_matches(['.', ',', ';', '!']).to_string()
-}
-
-/// Exact match after normalization.
-pub fn exact_match(expected: &str, actual: &str) -> BenchScore {
-    let norm_expected = normalize_answer(expected);
-    let norm_actual = normalize_answer(actual);
-    if norm_expected == norm_actual {
-        BenchScore::pass()
-    } else {
-        BenchScore::fail(format!(
-            "expected \"{norm_expected}\", got \"{norm_actual}\""
-        ))
-    }
-}
-
-/// Check if the actual answer contains the expected substring (normalized).
-pub fn contains_match(expected_substring: &str, actual: &str) -> BenchScore {
-    let norm_expected = normalize_answer(expected_substring);
-    let norm_actual = normalize_answer(actual);
-    if norm_actual.contains(&norm_expected) {
-        BenchScore::pass()
-    } else {
-        BenchScore::fail(format!("response does not contain \"{norm_expected}\""))
-    }
-}
-
-/// Check if the actual answer matches a regex pattern.
-pub fn regex_match(pattern: &str, actual: &str) -> BenchScore {
-    match Regex::new(pattern) {
-        Ok(re) => {
-            if re.is_match(actual) {
-                BenchScore::pass()
-            } else {
-                BenchScore::fail(format!("response does not match pattern /{pattern}/"))
-            }
-        }
-        Err(e) => BenchScore::fail(format!("invalid regex pattern: {e}")),
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_normalize_answer() {
-        assert_eq!(normalize_answer("  Hello   World.  "), "hello world");
-        assert_eq!(normalize_answer("Yes!"), "yes");
-        assert_eq!(normalize_answer("42"), "42");
-        assert_eq!(normalize_answer("  "), "");
-    }
-
-    #[test]
-    fn test_exact_match_pass() {
-        let score = exact_match("Hello World", "  hello   world.  ");
-        assert_eq!(score.value, 1.0);
-        assert_eq!(score.label, "pass");
-    }
-
-    #[test]
-    fn test_exact_match_fail() {
-        let score = exact_match("hello", "world");
-        assert_eq!(score.value, 0.0);
-        assert_eq!(score.label, "fail");
-    }
-
-    #[test]
-    fn test_contains_match_pass() {
-        let score = contains_match("world", "Hello World!");
-        assert_eq!(score.value, 1.0);
-    }
-
-    #[test]
-    fn test_contains_match_fail() {
-        let score = contains_match("xyz", "Hello World!");
-        assert_eq!(score.value, 0.0);
-    }
-
-    #[test]
-    fn test_regex_match_pass() {
-        let score = regex_match(r"\d{4}", "The year is 2024.");
-        assert_eq!(score.value, 1.0);
-    }
-
-    #[test]
-    fn test_regex_match_fail() {
-        let score = regex_match(r"\d{4}", "No numbers here.");
-        assert_eq!(score.value, 0.0);
-    }
-
-    #[test]
-    fn test_regex_match_invalid_pattern() {
-        let score = regex_match(r"[invalid", "anything");
-        assert_eq!(score.value, 0.0);
-        assert!(
-            score
-                .details
-                .as_deref()
-                .unwrap_or("")
-                .contains("invalid regex")
-        );
-    }
-}

benchmarks/src/suite.rs

@@ -1,154 +0,0 @@
-use std::sync::Arc;
-use std::time::Duration;
-
-use async_trait::async_trait;
-
-use crate::error::BenchError;
-
-/// A single task in a benchmark suite.
-#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
-pub struct BenchTask {
-    pub id: String,
-    pub prompt: String,
-    #[serde(default)]
-    pub context: Option<String>,
-    #[serde(default)]
-    pub resources: Vec<TaskResource>,
-    #[serde(default)]
-    pub tags: Vec<String>,
-    #[serde(default)]
-    pub expected_turns: Option<usize>,
-    #[serde(default)]
-    pub timeout: Option<Duration>,
-    #[serde(default)]
-    pub metadata: serde_json::Value,
-}
-
-/// A resource attached to a benchmark task (file, URL, etc.).
-#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
-pub struct TaskResource {
-    pub name: String,
-    pub path: String,
-    #[serde(default)]
-    pub resource_type: ResourceType,
-}
-
-#[derive(Debug, Clone, Default, serde::Serialize, serde::Deserialize)]
-#[serde(rename_all = "snake_case")]
-pub enum ResourceType {
-    #[default]
-    File,
-    Url,
-    Directory,
-}
-
-/// What the agent produced for scoring.
-#[derive(Debug, Clone)]
-#[allow(dead_code)]
-pub struct TaskSubmission {
-    pub response: String,
-    pub conversation: Vec<ConversationTurn>,
-    pub tool_calls: Vec<String>,
-    pub error: Option<String>,
-}
-
-/// A single turn in a multi-turn conversation.
-#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
-pub struct ConversationTurn {
-    pub role: TurnRole,
-    pub content: String,
-}
-
-#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
-#[serde(rename_all = "snake_case")]
-pub enum TurnRole {
-    User,
-    Assistant,
-    System,
-}
-
-/// Score for a single task.
-#[derive(Debug, Clone, PartialEq, serde::Serialize, serde::Deserialize)]
-pub struct BenchScore {
-    /// 0.0 to 1.0 (1.0 = perfect).
-    pub value: f64,
-    /// "pass" / "fail" / "partial".
-    pub label: String,
-    #[serde(default)]
-    pub details: Option<String>,
-}
-
-impl BenchScore {
-    pub fn pass() -> Self {
-        Self {
-            value: 1.0,
-            label: "pass".to_string(),
-            details: None,
-        }
-    }
-
-    pub fn fail(details: impl Into<String>) -> Self {
-        Self {
-            value: 0.0,
-            label: "fail".to_string(),
-            details: Some(details.into()),
-        }
-    }
-
-    pub fn partial(value: f64, details: impl Into<String>) -> Self {
-        Self {
-            value: value.clamp(0.0, 1.0),
-            label: "partial".to_string(),
-            details: Some(details.into()),
-        }
-    }
-}
-
-/// Trait for benchmark suite adapters.
-///
-/// Each suite (GAIA, Tau-bench, custom, etc.) implements this trait
-/// to provide task loading, scoring, and optional lifecycle hooks.
-#[async_trait]
-#[allow(dead_code)]
-pub trait BenchSuite: Send + Sync {
-    /// Human-readable name (e.g., "GAIA Validation").
-    fn name(&self) -> &str;
-
-    /// Machine ID (e.g., "gaia").
-    fn id(&self) -> &str;
-
-    /// Load all tasks from the suite's data source.
-    async fn load_tasks(&self) -> Result<Vec<BenchTask>, BenchError>;
-
-    /// Score the agent's submission against the expected answer.
-    async fn score(
-        &self,
-        task: &BenchTask,
-        submission: &TaskSubmission,
-    ) -> Result<BenchScore, BenchError>;
-
-    /// Optional: set up environment before running a task (clone repo, init DB, etc.).
-    async fn setup_task(&self, _task: &BenchTask) -> Result<(), BenchError> {
-        Ok(())
-    }
-
-    /// Optional: tear down environment after a task completes.
-    async fn teardown_task(&self, _task: &BenchTask) -> Result<(), BenchError> {
-        Ok(())
-    }
-
-    /// Optional: additional tools to register for this suite's tasks.
-    fn additional_tools(&self) -> Vec<Arc<dyn ironclaw::tools::Tool>> {
-        vec![]
-    }
-
-    /// Multi-turn: generate next simulated user message based on conversation so far.
-    /// Return `None` to end the conversation.
-    async fn next_user_message(
-        &self,
-        _task: &BenchTask,
-        _conversation: &[ConversationTurn],
-    ) -> Result<Option<String>, BenchError> {
-        Ok(None)
-    }
-}

build.rs

@@ -10,12 +10,17 @@
 //! Prerequisites: rustup target add wasm32-wasip2, cargo install wasm-tools
 
 use std::env;
-use std::path::PathBuf;
+use std::path::{Path, PathBuf};
 use std::process::Command;
 
 fn main() {
     let manifest_dir = env::var("CARGO_MANIFEST_DIR").unwrap();
     let root = PathBuf::from(&manifest_dir);
+
+    // ── Embed registry manifests ────────────────────────────────────────
+    embed_registry_catalog(&root);
+
+    // ── Build Telegram channel WASM ─────────────────────────────────────
     let channel_dir = root.join("channels-src/telegram");
     let wasm_out = channel_dir.join("telegram.wasm");
 
@@ -104,3 +109,89 @@ fn main() {
         }
     }
 }
+
+/// Collect all registry manifests into a single JSON blob at compile time.
+///
+/// Output: `$OUT_DIR/embedded_catalog.json` with structure:
+/// ```json
+/// { "tools": [...], "channels": [...], "bundles": {...} }
+/// ```
+fn embed_registry_catalog(root: &Path) {
+    use std::fs;
+
+    let registry_dir = root.join("registry");
+
+    // Rerun if the bundles file changes (per-file watches for tools/channels
+    // are emitted inside collect_json_files to track content changes reliably).
+    println!("cargo:rerun-if-changed=registry/_bundles.json");
+
+    let out_dir = PathBuf::from(env::var("OUT_DIR").unwrap());
+    let out_path = out_dir.join("embedded_catalog.json");
+
+    if !registry_dir.is_dir() {
+        // No registry dir: write empty catalog
+        fs::write(
+            &out_path,
+            r#"{"tools":[],"channels":[],"bundles":{"bundles":{}}}"#,
+        )
+        .unwrap();
+        return;
+    }
+
+    let mut tools = Vec::new();
+    let mut channels = Vec::new();
+
+    // Collect tool manifests
+    let tools_dir = registry_dir.join("tools");
+    if tools_dir.is_dir() {
+        collect_json_files(&tools_dir, &mut tools);
+    }
+
+    // Collect channel manifests
+    let channels_dir = registry_dir.join("channels");
+    if channels_dir.is_dir() {
+        collect_json_files(&channels_dir, &mut channels);
+    }
+
+    // Read bundles
+    let bundles_path = registry_dir.join("_bundles.json");
+    let bundles_raw = if bundles_path.is_file() {
+        fs::read_to_string(&bundles_path).unwrap_or_else(|_| r#"{"bundles":{}}"#.to_string())
+    } else {
+        r#"{"bundles":{}}"#.to_string()
+    };
+
+    // Build the combined JSON
+    let catalog = format!(
+        r#"{{"tools":[{}],"channels":[{}],"bundles":{}}}"#,
+        tools.join(","),
+        channels.join(","),
+        bundles_raw,
+    );
+
+    fs::write(&out_path, catalog).unwrap();
+}
+
+/// Read all .json files from a directory and push their raw contents into `out`.
+fn collect_json_files(dir: &Path, out: &mut Vec<String>) {
+    use std::fs;
+
+    let mut entries: Vec<_> = fs::read_dir(dir)
+        .unwrap()
+        .filter_map(|e| e.ok())
+        .filter(|e| {
+            e.path().is_file() && e.path().extension().and_then(|x| x.to_str()) == Some("json")
+        })
+        .collect();
+
+    // Sort for deterministic output
+    entries.sort_by_key(|e| e.file_name());
+
+    for entry in entries {
+        // Emit per-file watch so Cargo reruns when file contents change
+        println!("cargo:rerun-if-changed={}", entry.path().display());
+        if let Ok(content) = fs::read_to_string(entry.path()) {
+            out.push(content);
+        }
+    }
+}

channels-src/discord/Cargo.lock

@@ -0,0 +1,401 @@
+# This file is automatically @generated by Cargo.
+# It is not intended for manual editing.
+version = 4
+
+[[package]]
+name = "ahash"
+version = "0.8.12"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5a15f179cd60c4584b8a8c596927aadc462e27f2ca70c04e0071964a73ba7a75"
+dependencies = [
+ "cfg-if",
+ "once_cell",
+ "version_check",
+ "zerocopy",
+]
+
+[[package]]
+name = "anyhow"
+version = "1.0.102"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7f202df86484c868dbad7eaa557ef785d5c66295e41b460ef922eca0723b842c"
+
+[[package]]
+name = "bitflags"
+version = "2.11.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "843867be96c8daad0d758b57df9392b6d8d271134fce549de6ce169ff98a92af"
+
+[[package]]
+name = "cfg-if"
+version = "1.0.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9330f8b2ff13f34540b44e946ef35111825727b38d33286ef986142615121801"
+
+[[package]]
+name = "discord-channel"
+version = "0.1.0"
+dependencies = [
+ "serde",
+ "serde_json",
+ "wit-bindgen",
+]
+
+[[package]]
+name = "equivalent"
+version = "1.0.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "877a4ace8713b0bcf2a4e7eec82529c029f1d0619886d18145fea96c3ffe5c0f"
+
+[[package]]
+name = "hashbrown"
+version = "0.14.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e5274423e17b7c9fc20b6e7e208532f9b19825d82dfd615708b70edd83df41f1"
+dependencies = [
+ "ahash",
+]
+
+[[package]]
+name = "hashbrown"
+version = "0.16.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "841d1cc9bed7f9236f321df977030373f4a4163ae1a7dbfe1a51a2c1a51d9100"
+
+[[package]]
+name = "heck"
+version = "0.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2304e00983f87ffb38b55b444b5e3b60a884b5d30c0fca7d82fe33449bbe55ea"
+
+[[package]]
+name = "id-arena"
+version = "2.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3d3067d79b975e8844ca9eb072e16b31c3c1c36928edf9c6789548c524d0d954"
+
+[[package]]
+name = "indexmap"
+version = "2.13.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7714e70437a7dc3ac8eb7e6f8df75fd8eb422675fc7678aff7364301092b1017"
+dependencies = [
+ "equivalent",
+ "hashbrown 0.16.1",
+ "serde",
+ "serde_core",
+]
+
+[[package]]
+name = "itoa"
+version = "1.0.17"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "92ecc6618181def0457392ccd0ee51198e065e016d1d527a7ac1b6dc7c1f09d2"
+
+[[package]]
+name = "leb128"
+version = "0.2.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "884e2677b40cc8c339eaefcb701c32ef1fd2493d71118dc0ca4b6a736c93bd67"
+
+[[package]]
+name = "log"
+version = "0.4.29"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5e5032e24019045c762d3c0f28f5b6b8bbf38563a65908389bf7978758920897"
+
+[[package]]
+name = "memchr"
+version = "2.8.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f8ca58f447f06ed17d5fc4043ce1b10dd205e060fb3ce5b979b8ed8e59ff3f79"
+
+[[package]]
+name = "once_cell"
+version = "1.21.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "42f5e15c9953c5e4ccceeb2e7382a716482c34515315f7b03532b8b4e8393d2d"
+
+[[package]]
+name = "prettyplease"
+version = "0.2.37"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "479ca8adacdd7ce8f1fb39ce9ecccbfe93a3f1344b3d0d97f20bc0196208f62b"
+dependencies = [
+ "proc-macro2",
+ "syn",
+]
+
+[[package]]
+name = "proc-macro2"
+version = "1.0.106"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8fd00f0bb2e90d81d1044c2b32617f68fcb9fa3bb7640c23e9c748e53fb30934"
+dependencies = [
+ "unicode-ident",
+]
+
+[[package]]
+name = "quote"
+version = "1.0.44"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "21b2ebcf727b7760c461f091f9f0f539b77b8e87f2fd88131e7f1b433b3cece4"
+dependencies = [
+ "proc-macro2",
+]
+
+[[package]]
+name = "semver"
+version = "1.0.27"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d767eb0aabc880b29956c35734170f26ed551a859dbd361d140cdbeca61ab1e2"
+
+[[package]]
+name = "serde"
+version = "1.0.228"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9a8e94ea7f378bd32cbbd37198a4a91436180c5bb472411e48b5ec2e2124ae9e"
+dependencies = [
+ "serde_core",
+ "serde_derive",
+]
+
+[[package]]
+name = "serde_core"
+version = "1.0.228"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "41d385c7d4ca58e59fc732af25c3983b67ac852c1a25000afe1175de458b67ad"
+dependencies = [
+ "serde_derive",
+]
+
+[[package]]
+name = "serde_derive"
+version = "1.0.228"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d540f220d3187173da220f885ab66608367b6574e925011a9353e4badda91d79"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "serde_json"
+version = "1.0.149"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "83fc039473c5595ace860d8c4fafa220ff474b3fc6bfdb4293327f1a37e94d86"
+dependencies = [
+ "itoa",
+ "memchr",
+ "serde",
+ "serde_core",
+ "zmij",
+]
+
+[[package]]
+name = "smallvec"
+version = "1.15.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "67b1b7a3b5fe4f1376887184045fcf45c69e92af734b7aaddc05fb777b6fbd03"
+
+[[package]]
+name = "spdx"
+version = "0.10.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c3e17e880bafaeb362a7b751ec46bdc5b61445a188f80e0606e68167cd540fa3"
+dependencies = [
+ "smallvec",
+]
+
+[[package]]
+name = "syn"
+version = "2.0.117"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e665b8803e7b1d2a727f4023456bbbbe74da67099c585258af0ad9c5013b9b99"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "unicode-ident",
+]
+
+[[package]]
+name = "unicode-ident"
+version = "1.0.24"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e6e4313cd5fcd3dad5cafa179702e2b244f760991f45397d14d4ebf38247da75"
+
+[[package]]
+name = "unicode-xid"
+version = "0.2.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ebc1c04c71510c7f702b52b7c350734c9ff1295c464a03335b00bb84fc54f853"
+
+[[package]]
+name = "version_check"
+version = "0.9.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0b928f33d975fc6ad9f86c8f283853ad26bdd5b10b7f1542aa2fa15e2289105a"
+
+[[package]]
+name = "wasm-encoder"
+version = "0.220.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e913f9242315ca39eff82aee0e19ee7a372155717ff0eb082c741e435ce25ed1"
+dependencies = [
+ "leb128",
+ "wasmparser",
+]
+
+[[package]]
+name = "wasm-metadata"
+version = "0.220.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "185dfcd27fa5db2e6a23906b54c28199935f71d9a27a1a27b3a88d6fee2afae7"
+dependencies = [
+ "anyhow",
+ "indexmap",
+ "serde",
+ "serde_derive",
+ "serde_json",
+ "spdx",
+ "wasm-encoder",
+ "wasmparser",
+]
+
+[[package]]
+name = "wasmparser"
+version = "0.220.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8d07b6a3b550fefa1a914b6d54fc175dd11c3392da11eee604e6ffc759805d25"
+dependencies = [
+ "ahash",
+ "bitflags",
+ "hashbrown 0.14.5",
+ "indexmap",
+ "semver",
+]
+
+[[package]]
+name = "wit-bindgen"
+version = "0.36.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6a2b3e15cd6068f233926e7d8c7c588b2ec4fb7cc7bf3824115e7c7e2a8485a3"
+dependencies = [
+ "wit-bindgen-rt",
+ "wit-bindgen-rust-macro",
+]
+
+[[package]]
+name = "wit-bindgen-core"
+version = "0.36.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b632a5a0fa2409489bd49c9e6d99fcc61bb3d4ce9d1907d44662e75a28c71172"
+dependencies = [
+ "anyhow",
+ "heck",
+ "wit-parser",
+]
+
+[[package]]
+name = "wit-bindgen-rt"
+version = "0.36.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7947d0131c7c9da3f01dfde0ab8bd4c4cf3c5bd49b6dba0ae640f1fa752572ea"
+dependencies = [
+ "bitflags",
+]
+
+[[package]]
+name = "wit-bindgen-rust"
+version = "0.36.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4329de4186ee30e2ef30a0533f9b3c123c019a237a7c82d692807bf1b3ee2697"
+dependencies = [
+ "anyhow",
+ "heck",
+ "indexmap",
+ "prettyplease",
+ "syn",
+ "wasm-metadata",
+ "wit-bindgen-core",
+ "wit-component",
+]
+
+[[package]]
+name = "wit-bindgen-rust-macro"
+version = "0.36.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "177fb7ee1484d113b4792cc480b1ba57664bbc951b42a4beebe573502135b1fc"
+dependencies = [
+ "anyhow",
+ "prettyplease",
+ "proc-macro2",
+ "quote",
+ "syn",
+ "wit-bindgen-core",
+ "wit-bindgen-rust",
+]
+
+[[package]]
+name = "wit-component"
+version = "0.220.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b505603761ed400c90ed30261f44a768317348e49f1864e82ecdc3b2744e5627"
+dependencies = [
+ "anyhow",
+ "bitflags",
+ "indexmap",
+ "log",
+ "serde",
+ "serde_derive",
+ "serde_json",
+ "wasm-encoder",
+ "wasm-metadata",
+ "wasmparser",
+ "wit-parser",
+]
+
+[[package]]
+name = "wit-parser"
+version = "0.220.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ae2a7999ed18efe59be8de2db9cb2b7f84d88b27818c79353dfc53131840fe1a"
+dependencies = [
+ "anyhow",
+ "id-arena",
+ "indexmap",
+ "log",
+ "semver",
+ "serde",
+ "serde_derive",
+ "serde_json",
+ "unicode-xid",
+ "wasmparser",
+]
+
+[[package]]
+name = "zerocopy"
+version = "0.8.39"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "db6d35d663eadb6c932438e763b262fe1a70987f9ae936e60158176d710cae4a"
+dependencies = [
+ "zerocopy-derive",
+]
+
+[[package]]
+name = "zerocopy-derive"
+version = "0.8.39"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4122cd3169e94605190e77839c9a40d40ed048d305bfdc146e7df40ab0f3e517"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "zmij"
+version = "1.0.21"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b8848ee67ecc8aedbaf3e4122217aff892639231befc6a1b58d29fff4c2cabaa"

channels-src/discord/Cargo.toml

@@ -9,7 +9,7 @@ publish = false
 [dependencies]
 serde = { version = "1.0", features = ["derive"] }
 serde_json = "1.0"
-wit-bindgen = "0.41.0"
+wit-bindgen = "0.36"
 
 [lib]
 crate-type = ["cdylib"]
@@ -21,3 +21,5 @@ lto = true
 codegen-units = 1
 
 
+
+[workspace]

channels-src/discord/discord.capabilities.json

@@ -2,6 +2,15 @@
   "type": "channel",
   "name": "discord",
   "description": "Discord Gateway/Webhook channel for handling slash commands, buttons, and messages",
+  "setup": {
+    "required_secrets": [
+      {
+        "name": "discord_bot_token",
+        "prompt": "Enter your Discord Bot Token (from Developer Portal)",
+        "optional": false
+      }
+    ]
+  },
   "capabilities": {
     "http": {
       "allowlist": [
@@ -10,7 +19,7 @@
       "credentials": {
         "discord_bot_token": {
           "secret_name": "discord_bot_token",
-          "location": { "type": "header", "header_name": "Authorization", "prefix": "Bot " },
+          "location": { "type": "header", "name": "Authorization", "prefix": "Bot " },
           "host_patterns": ["discord.com"]
         }
       },
@@ -34,6 +43,9 @@
     }
   },
   "config": {
-    "require_signature_verification": true
+    "require_signature_verification": true,
+    "owner_id": null,
+    "dm_policy": "pairing",
+    "allow_from": []
   }
 }

channels-src/discord/src/lib.rs

@@ -124,12 +124,57 @@ struct DiscordMessageMetadata {
     thread_id: Option<String>,
 }
 
+/// Workspace path for persisting owner_id across WASM callbacks.
+const OWNER_ID_PATH: &str = "state/owner_id";
+/// Workspace path for persisting dm_policy across WASM callbacks.
+const DM_POLICY_PATH: &str = "state/dm_policy";
+/// Workspace path for persisting allow_from (JSON array) across WASM callbacks.
+const ALLOW_FROM_PATH: &str = "state/allow_from";
+/// Channel name for pairing store (used by pairing host APIs).
+const CHANNEL_NAME: &str = "discord";
+
+/// Channel configuration from capabilities file.
+#[derive(Debug, Deserialize)]
+struct DiscordConfig {
+    #[serde(default)]
+    #[allow(dead_code)]
+    require_signature_verification: bool,
+    #[serde(default)]
+    owner_id: Option<String>,
+    #[serde(default)]
+    dm_policy: Option<String>,
+    #[serde(default)]
+    allow_from: Option<Vec<String>>,
+}
+
 struct DiscordChannel;
 
 impl Guest for DiscordChannel {
-    fn on_start(_config_json: String) -> Result<ChannelConfig, String> {
+    fn on_start(config_json: String) -> Result<ChannelConfig, String> {
+        let config: DiscordConfig = serde_json::from_str(&config_json)
+            .map_err(|e| format!("Failed to parse config: {}", e))?;
+
         channel_host::log(channel_host::LogLevel::Info, "Discord channel starting");
 
+        // Persist owner_id so subsequent callbacks can read it
+        if let Some(ref owner_id) = config.owner_id {
+            let _ = channel_host::workspace_write(OWNER_ID_PATH, owner_id);
+            channel_host::log(
+                channel_host::LogLevel::Info,
+                &format!("Owner restriction enabled: user {}", owner_id),
+            );
+        } else {
+            let _ = channel_host::workspace_write(OWNER_ID_PATH, "");
+        }
+
+        // Persist dm_policy and allow_from for DM pairing
+        let dm_policy = config.dm_policy.as_deref().unwrap_or("pairing");
+        let _ = channel_host::workspace_write(DM_POLICY_PATH, dm_policy);
+
+        let allow_from_json = serde_json::to_string(&config.allow_from.unwrap_or_default())
+            .unwrap_or_else(|_| "[]".to_string());
+        let _ = channel_host::workspace_write(ALLOW_FROM_PATH, &allow_from_json);
+
         Ok(ChannelConfig {
             display_name: "Discord".to_string(),
             http_endpoints: vec![HttpEndpointConfig {
@@ -169,16 +214,21 @@ impl Guest for DiscordChannel {
 
             // Application Command (slash command)
             2 => {
-                handle_slash_command(&interaction);
-                json_response(
-                    200,
-                    serde_json::json!({
-                        "type": 5,
-                        "data": {
-                            "content": "🤔 Thinking..."
-                        }
-                    }),
-                )
+                if handle_slash_command(&interaction) {
+                    json_response(200, serde_json::json!({"type": 5}))
+                } else {
+                    // Permission denied — ephemeral response
+                    json_response(
+                        200,
+                        serde_json::json!({
+                            "type": 4,
+                            "data": {
+                                "content": "You are not authorized to use this bot.",
+                                "flags": 64
+                            }
+                        }),
+                    )
+                }
             }
 
             // Message Component (buttons, selects)
@@ -270,7 +320,8 @@ impl Guest for DiscordChannel {
     }
 }
 
-fn handle_slash_command(interaction: &DiscordInteraction) {
+/// Returns true if the message was emitted, false if permission denied.
+fn handle_slash_command(interaction: &DiscordInteraction) -> bool {
     let user = interaction
         .member
         .as_ref()
@@ -287,6 +338,22 @@ fn handle_slash_command(interaction: &DiscordInteraction) {
         })
         .unwrap_or_default();
 
+    // DM if no guild member context (only direct user field set)
+    let is_dm = interaction.member.is_none();
+
+    // Permission check
+    if !check_sender_permission(
+        &user_id,
+        Some(&user_name),
+        is_dm,
+        Some(&PairingReplyCtx {
+            application_id: interaction.application_id.clone(),
+            token: interaction.token.clone(),
+        }),
+    ) {
+        return false;
+    }
+
     let channel_id = interaction.channel_id.clone().unwrap_or_default();
 
     let command_name = interaction
@@ -322,14 +389,13 @@ fn handle_slash_command(interaction: &DiscordInteraction) {
                 channel_host::LogLevel::Error,
                 &format!("Failed to serialize metadata: {}", e),
             );
-            // Attempt to notify user of internal error
             let url = format!(
                 "https://discord.com/api/v10/webhooks/{}/{}",
                 interaction.application_id, interaction.token
             );
             let payload = serde_json::json!({
                 "content": "❌ Internal Error: Failed to process command metadata.",
-                "flags": 64 // Ephemeral
+                "flags": 64
             });
             let _ = channel_host::http_request(
                 "POST",
@@ -338,7 +404,7 @@ fn handle_slash_command(interaction: &DiscordInteraction) {
                 Some(&serde_json::to_vec(&payload).unwrap_or_default()),
                 None,
             );
-            return;
+            return true; // Error, but not a permission denial
         }
     };
 
@@ -349,10 +415,10 @@ fn handle_slash_command(interaction: &DiscordInteraction) {
         thread_id: None,
         metadata_json,
     });
+    true
 }
 
 fn handle_message_component(interaction: &DiscordInteraction, message: &DiscordMessage) {
-    // Check member first (for server contexts), then user (for DMs)
     let user = interaction
         .member
         .as_ref()
@@ -369,6 +435,11 @@ fn handle_message_component(interaction: &DiscordInteraction, message: &DiscordM
         })
         .unwrap_or_default();
 
+    let is_dm = interaction.member.is_none();
+    if !check_sender_permission(&user_id, Some(&user_name), is_dm, None) {
+        return;
+    }
+
     let channel_id = message.channel_id.clone();
 
     let metadata = DiscordMessageMetadata {
@@ -399,6 +470,145 @@ fn handle_message_component(interaction: &DiscordInteraction, message: &DiscordM
     });
 }
 
+// ============================================================================
+// Permission & Pairing
+// ============================================================================
+
+/// Context needed to send a pairing reply via Discord webhook followup.
+struct PairingReplyCtx {
+    application_id: String,
+    token: String,
+}
+
+/// Check if a sender is permitted to interact with the bot.
+/// Returns true if allowed, false if denied (pairing reply sent if applicable).
+fn check_sender_permission(
+    user_id: &str,
+    username: Option<&str>,
+    is_dm: bool,
+    reply_ctx: Option<&PairingReplyCtx>,
+) -> bool {
+    // 1. Owner check (highest priority, applies to all contexts)
+    let owner_id = channel_host::workspace_read(OWNER_ID_PATH).filter(|s| !s.is_empty());
+    if let Some(ref owner) = owner_id {
+        if user_id != owner {
+            channel_host::log(
+                channel_host::LogLevel::Debug,
+                &format!(
+                    "Dropping interaction from non-owner user {} (owner: {})",
+                    user_id, owner
+                ),
+            );
+            return false;
+        }
+        return true;
+    }
+
+    // 2. DM policy (only for DMs when no owner_id)
+    if !is_dm {
+        return true; // Guild interactions bypass DM policy
+    }
+
+    let dm_policy =
+        channel_host::workspace_read(DM_POLICY_PATH).unwrap_or_else(|| "pairing".to_string());
+
+    if dm_policy == "open" {
+        return true;
+    }
+
+    // 3. Build merged allow list: config allow_from + pairing store
+    let mut allowed: Vec<String> = channel_host::workspace_read(ALLOW_FROM_PATH)
+        .and_then(|s| serde_json::from_str(&s).ok())
+        .unwrap_or_default();
+
+    if let Ok(store_allowed) = channel_host::pairing_read_allow_from(CHANNEL_NAME) {
+        allowed.extend(store_allowed);
+    }
+
+    // 4. Check sender against allow list
+    let is_allowed = allowed.contains(&"*".to_string())
+        || allowed.contains(&user_id.to_string())
+        || username.is_some_and(|u| allowed.contains(&u.to_string()));
+
+    if is_allowed {
+        return true;
+    }
+
+    // 5. Not allowed — handle by policy
+    if dm_policy == "pairing" {
+        let meta = serde_json::json!({
+            "user_id": user_id,
+            "username": username,
+        })
+        .to_string();
+
+        match channel_host::pairing_upsert_request(CHANNEL_NAME, user_id, &meta) {
+            Ok(result) => {
+                channel_host::log(
+                    channel_host::LogLevel::Info,
+                    &format!(
+                        "Pairing request for user {}: code {}",
+                        user_id, result.code
+                    ),
+                );
+                if result.created {
+                    if let Some(ctx) = reply_ctx {
+                        let _ = send_pairing_reply(ctx, &result.code);
+                    }
+                }
+            }
+            Err(e) => {
+                channel_host::log(
+                    channel_host::LogLevel::Error,
+                    &format!("Pairing upsert failed: {}", e),
+                );
+            }
+        }
+    }
+    false
+}
+
+/// Send a pairing code as an ephemeral Discord followup message.
+fn send_pairing_reply(ctx: &PairingReplyCtx, code: &str) -> Result<(), String> {
+    let url = format!(
+        "https://discord.com/api/v10/webhooks/{}/{}",
+        ctx.application_id, ctx.token
+    );
+
+    let payload = serde_json::json!({
+        "content": format!(
+            "To pair with this bot, run: `ironclaw pairing approve discord {}`",
+            code
+        ),
+        "flags": 64 // Ephemeral — only visible to the sender
+    });
+
+    let payload_bytes =
+        serde_json::to_vec(&payload).map_err(|e| format!("Failed to serialize: {}", e))?;
+
+    let headers = serde_json::json!({"Content-Type": "application/json"});
+
+    let result = channel_host::http_request(
+        "POST",
+        &url,
+        &headers.to_string(),
+        Some(&payload_bytes),
+        None,
+    );
+
+    match result {
+        Ok(response) if response.status >= 200 && response.status < 300 => Ok(()),
+        Ok(response) => {
+            let body_str = String::from_utf8_lossy(&response.body);
+            Err(format!(
+                "Discord API error: {} - {}",
+                response.status, body_str
+            ))
+        }
+        Err(e) => Err(format!("HTTP request failed: {}", e)),
+    }
+}
+
 fn json_response(status: u16, value: serde_json::Value) -> OutgoingHttpResponse {
     let body = serde_json::to_vec(&value).unwrap_or_default();
     let headers = serde_json::json!({"Content-Type": "application/json"});

channels-src/slack/Cargo.toml

@@ -27,3 +27,5 @@ opt-level = "s"
 lto = true
 strip = true
 codegen-units = 1
+
+[workspace]

channels-src/slack/slack.capabilities.json

@@ -2,6 +2,20 @@
   "type": "channel",
   "name": "slack",
   "description": "Slack Events API channel for receiving and responding to Slack messages",
+  "setup": {
+    "required_secrets": [
+      {
+        "name": "slack_bot_token",
+        "prompt": "Enter your Slack Bot OAuth Token (xoxb-...)",
+        "optional": false
+      },
+      {
+        "name": "slack_signing_secret",
+        "prompt": "Enter your Slack Signing Secret (from App Credentials)",
+        "optional": false
+      }
+    ]
+  },
   "capabilities": {
     "http": {
       "allowlist": [
@@ -33,6 +47,9 @@
     }
   },
   "config": {
-    "signing_secret_name": "slack_signing_secret"
+    "signing_secret_name": "slack_signing_secret",
+    "owner_id": null,
+    "dm_policy": "pairing",
+    "allow_from": []
   }
 }

channels-src/slack/src/lib.rs

@@ -104,15 +104,31 @@ struct SlackPostMessageResponse {
     ts: Option<String>,
 }
 
+/// Workspace path for persisting owner_id across WASM callbacks.
+const OWNER_ID_PATH: &str = "state/owner_id";
+/// Workspace path for persisting dm_policy across WASM callbacks.
+const DM_POLICY_PATH: &str = "state/dm_policy";
+/// Workspace path for persisting allow_from (JSON array) across WASM callbacks.
+const ALLOW_FROM_PATH: &str = "state/allow_from";
+/// Channel name for pairing store (used by pairing host APIs).
+const CHANNEL_NAME: &str = "slack";
+
 /// Channel configuration from capabilities file.
 #[derive(Debug, Deserialize)]
 struct SlackConfig {
     /// Name of secret containing signing secret (for verification by host).
-    /// Parsed from config for forward compatibility; not yet used in WASM
-    /// (host handles signature verification).
     #[serde(default = "default_signing_secret_name")]
     #[allow(dead_code)]
     signing_secret_name: String,
+
+    #[serde(default)]
+    owner_id: Option<String>,
+
+    #[serde(default)]
+    dm_policy: Option<String>,
+
+    #[serde(default)]
+    allow_from: Option<Vec<String>>,
 }
 
 fn default_signing_secret_name() -> String {
@@ -123,12 +139,30 @@ struct SlackChannel;
 
 impl Guest for SlackChannel {
     fn on_start(config_json: String) -> Result<ChannelConfig, String> {
-        // Parse configuration
-        let _config: SlackConfig = serde_json::from_str(&config_json)
+        let config: SlackConfig = serde_json::from_str(&config_json)
             .map_err(|e| format!("Failed to parse config: {}", e))?;
 
         channel_host::log(channel_host::LogLevel::Info, "Slack channel starting");
 
+        // Persist owner_id so subsequent callbacks can read it
+        if let Some(ref owner_id) = config.owner_id {
+            let _ = channel_host::workspace_write(OWNER_ID_PATH, owner_id);
+            channel_host::log(
+                channel_host::LogLevel::Info,
+                &format!("Owner restriction enabled: user {}", owner_id),
+            );
+        } else {
+            let _ = channel_host::workspace_write(OWNER_ID_PATH, "");
+        }
+
+        // Persist dm_policy and allow_from for DM pairing
+        let dm_policy = config.dm_policy.as_deref().unwrap_or("pairing");
+        let _ = channel_host::workspace_write(DM_POLICY_PATH, dm_policy);
+
+        let allow_from_json = serde_json::to_string(&config.allow_from.unwrap_or_default())
+            .unwrap_or_else(|_| "[]".to_string());
+        let _ = channel_host::workspace_write(ALLOW_FROM_PATH, &allow_from_json);
+
         Ok(ChannelConfig {
             display_name: "Slack".to_string(),
             http_endpoints: vec![HttpEndpointConfig {
@@ -136,7 +170,7 @@ impl Guest for SlackChannel {
                 methods: vec!["POST".to_string()],
                 require_secret: true,
             }],
-            poll: None, // Slack uses push via webhooks, no polling needed
+            poll: None,
         })
     }
 
@@ -280,7 +314,7 @@ impl Guest for SlackChannel {
 /// Handle a Slack event and emit message if applicable.
 fn handle_slack_event(event: SlackEvent, team_id: Option<String>, _event_id: Option<String>) {
     match event.event_type.as_str() {
-        // Direct mention of the bot
+        // Direct mention of the bot (always in a channel, not a DM)
         "app_mention" => {
             if let (Some(user), Some(channel), Some(text), Some(ts)) = (
                 event.user,
@@ -288,6 +322,10 @@ fn handle_slack_event(event: SlackEvent, team_id: Option<String>, _event_id: Opt
                 event.text,
                 event.ts.clone(),
             ) {
+                // app_mention is always in a channel (not DM)
+                if !check_sender_permission(&user, &channel, false) {
+                    return;
+                }
                 emit_message(user, text, channel, event.thread_ts.or(Some(ts)), team_id);
             }
         }
@@ -307,6 +345,9 @@ fn handle_slack_event(event: SlackEvent, team_id: Option<String>, _event_id: Opt
             ) {
                 // Only process DMs (channel IDs starting with D)
                 if channel.starts_with('D') {
+                    if !check_sender_permission(&user, &channel, true) {
+                        return;
+                    }
                     emit_message(user, text, channel, event.thread_ts.or(Some(ts)), team_id);
                 }
             }
@@ -358,6 +399,126 @@ fn emit_message(
     });
 }
 
+// ============================================================================
+// Permission & Pairing
+// ============================================================================
+
+/// Check if a sender is permitted. Returns true if allowed.
+/// For pairing mode, sends a pairing code DM if denied.
+fn check_sender_permission(user_id: &str, channel_id: &str, is_dm: bool) -> bool {
+    // 1. Owner check (highest priority, applies to all contexts)
+    let owner_id = channel_host::workspace_read(OWNER_ID_PATH).filter(|s| !s.is_empty());
+    if let Some(ref owner) = owner_id {
+        if user_id != owner {
+            channel_host::log(
+                channel_host::LogLevel::Debug,
+                &format!(
+                    "Dropping message from non-owner user {} (owner: {})",
+                    user_id, owner
+                ),
+            );
+            return false;
+        }
+        return true;
+    }
+
+    // 2. DM policy (only for DMs when no owner_id)
+    if !is_dm {
+        return true; // Channel messages bypass DM policy
+    }
+
+    let dm_policy =
+        channel_host::workspace_read(DM_POLICY_PATH).unwrap_or_else(|| "pairing".to_string());
+
+    if dm_policy == "open" {
+        return true;
+    }
+
+    // 3. Build merged allow list: config allow_from + pairing store
+    let mut allowed: Vec<String> = channel_host::workspace_read(ALLOW_FROM_PATH)
+        .and_then(|s| serde_json::from_str(&s).ok())
+        .unwrap_or_default();
+
+    if let Ok(store_allowed) = channel_host::pairing_read_allow_from(CHANNEL_NAME) {
+        allowed.extend(store_allowed);
+    }
+
+    // 4. Check sender (Slack events only have user ID, not username)
+    let is_allowed =
+        allowed.contains(&"*".to_string()) || allowed.contains(&user_id.to_string());
+
+    if is_allowed {
+        return true;
+    }
+
+    // 5. Not allowed — handle by policy
+    if dm_policy == "pairing" {
+        let meta = serde_json::json!({
+            "user_id": user_id,
+            "channel_id": channel_id,
+        })
+        .to_string();
+
+        match channel_host::pairing_upsert_request(CHANNEL_NAME, user_id, &meta) {
+            Ok(result) => {
+                channel_host::log(
+                    channel_host::LogLevel::Info,
+                    &format!(
+                        "Pairing request for user {}: code {}",
+                        user_id, result.code
+                    ),
+                );
+                if result.created {
+                    let _ = send_pairing_reply(channel_id, &result.code);
+                }
+            }
+            Err(e) => {
+                channel_host::log(
+                    channel_host::LogLevel::Error,
+                    &format!("Pairing upsert failed: {}", e),
+                );
+            }
+        }
+    }
+    false
+}
+
+/// Send a pairing code message via Slack chat.postMessage.
+fn send_pairing_reply(channel_id: &str, code: &str) -> Result<(), String> {
+    let payload = serde_json::json!({
+        "channel": channel_id,
+        "text": format!(
+            "To pair with this bot, run: `ironclaw pairing approve slack {}`",
+            code
+        ),
+    });
+
+    let payload_bytes =
+        serde_json::to_vec(&payload).map_err(|e| format!("Failed to serialize: {}", e))?;
+
+    let headers = serde_json::json!({"Content-Type": "application/json"});
+
+    let result = channel_host::http_request(
+        "POST",
+        "https://slack.com/api/chat.postMessage",
+        &headers.to_string(),
+        Some(&payload_bytes),
+        None,
+    );
+
+    match result {
+        Ok(response) if response.status == 200 => Ok(()),
+        Ok(response) => {
+            let body_str = String::from_utf8_lossy(&response.body);
+            Err(format!(
+                "Slack API error: {} - {}",
+                response.status, body_str
+            ))
+        }
+        Err(e) => Err(format!("HTTP request failed: {}", e)),
+    }
+}
+
 /// Strip leading bot mention from text.
 fn strip_bot_mention(text: &str) -> String {
     // Slack mentions look like <@U12345678>

channels-src/telegram/src/lib.rs

@@ -76,6 +76,9 @@ struct TelegramMessage {
     #[serde(default)]
     caption: Option<String>,
 
+    /// Voice message.
+    voice: Option<TelegramVoice>,
+
     /// Original message if this is a reply.
     reply_to_message: Option<Box<TelegramMessage>>,
 
@@ -139,6 +142,36 @@ struct MessageEntity {
     user: Option<TelegramUser>,
 }
 
+/// Telegram Voice object.
+/// https://core.telegram.org/bots/api#voice
+#[derive(Debug, Deserialize)]
+struct TelegramVoice {
+    /// Identifier for this file, which can be used to download the file.
+    file_id: String,
+
+    /// Duration of the audio in seconds.
+    duration: u32,
+
+    /// MIME type of the file.
+    #[serde(default)]
+    mime_type: Option<String>,
+
+    /// File size in bytes.
+    #[serde(default)]
+    file_size: Option<i64>,
+}
+
+/// Telegram File object returned by getFile.
+/// https://core.telegram.org/bots/api#file
+#[derive(Debug, Deserialize)]
+struct TelegramFile {
+    /// Identifier for this file.
+    file_id: String,
+
+    /// File path for downloading. Use https://api.telegram.org/file/bot<token>/<file_path>.
+    file_path: Option<String>,
+}
+
 /// Telegram API response wrapper.
 #[derive(Debug, Deserialize)]
 struct TelegramApiResponse<T> {
@@ -867,6 +900,87 @@ fn send_pairing_reply(chat_id: i64, code: &str) -> Result<(), String> {
     }
 }
 
+// ============================================================================
+// Voice File Download
+// ============================================================================
+
+/// Download a voice file from Telegram by file_id.
+///
+/// 1. Call getFile to get the file_path.
+/// 2. Download the file bytes from /file/bot{TOKEN}/{file_path}.
+fn download_voice_file(file_id: &str) -> Result<Vec<u8>, String> {
+    // Reject file_id containing curly braces to prevent credential placeholder
+    // injection (e.g., a malicious file_id like "{OPENAI_API_KEY}" would be
+    // interpreted by the host-side credential injector).
+    if file_id.contains('{') || file_id.contains('}') {
+        return Err("invalid file_id: contains forbidden characters".to_string());
+    }
+
+    // Step 1: Call getFile to get file_path
+    // Double braces `{{...}}` produce a literal `{TELEGRAM_BOT_TOKEN}` placeholder
+    // in the URL, which the host-side credential injector replaces with the real token.
+    let get_file_url = format!(
+        "https://api.telegram.org/bot{{TELEGRAM_BOT_TOKEN}}/getFile?file_id={}",
+        file_id
+    );
+
+    let headers = serde_json::json!({});
+    let result = channel_host::http_request("GET", &get_file_url, &headers.to_string(), None, None);
+
+    let response = result.map_err(|e| format!("getFile request failed: {}", e))?;
+
+    if response.status != 200 {
+        let body_str = String::from_utf8_lossy(&response.body);
+        return Err(format!("getFile returned {}: {}", response.status, body_str));
+    }
+
+    let api_response: TelegramApiResponse<TelegramFile> =
+        serde_json::from_slice(&response.body)
+            .map_err(|e| format!("Failed to parse getFile response: {}", e))?;
+
+    if !api_response.ok {
+        return Err(format!(
+            "getFile API error: {}",
+            api_response
+                .description
+                .unwrap_or_else(|| "unknown".to_string())
+        ));
+    }
+
+    let file = api_response
+        .result
+        .ok_or_else(|| "getFile returned no result".to_string())?;
+
+    let file_path = file
+        .file_path
+        .ok_or_else(|| "getFile returned no file_path".to_string())?;
+
+    // Sanitize file_path against credential placeholder injection
+    if file_path.contains('{') || file_path.contains('}') {
+        return Err("invalid file_path: contains forbidden characters".to_string());
+    }
+
+    // Step 2: Download the actual file bytes
+    let download_url = format!(
+        "https://api.telegram.org/file/bot{{TELEGRAM_BOT_TOKEN}}/{}",
+        file_path
+    );
+
+    let result =
+        channel_host::http_request("GET", &download_url, &headers.to_string(), None, None);
+
+    let response = result.map_err(|e| format!("File download failed: {}", e))?;
+
+    if response.status != 200 {
+        return Err(format!(
+            "File download returned status {}",
+            response.status
+        ));
+    }
+
+    Ok(response.body)
+}
+
 // ============================================================================
 // Update Handling
 // ============================================================================
@@ -886,6 +1000,9 @@ fn handle_update(update: TelegramUpdate) {
 
 /// Process a single message.
 fn handle_message(message: TelegramMessage) {
+    // Check for voice note first (voice-only messages have no text)
+    let is_voice = message.voice.is_some();
+
     // Use text or caption (for media messages)
     let content = message
         .text
@@ -893,7 +1010,8 @@ fn handle_message(message: TelegramMessage) {
         .or_else(|| message.caption.filter(|c| !c.is_empty()))
         .unwrap_or_default();
 
-    if content.is_empty() {
+    // Allow voice notes through even when content is empty
+    if content.is_empty() && !is_voice {
         return;
     }
 
@@ -1038,19 +1156,65 @@ fn handle_message(message: TelegramMessage) {
         },
     );
 
+    // Handle voice notes: download and attach audio bytes.
+    // Note: download is synchronous (two HTTP roundtrips to Telegram API).
+    // This blocks the WASM execution for the current polling tick.
+    let mut attachments = Vec::new();
+    let mut voice_download_failed = false;
+    if let Some(ref voice) = message.voice {
+        channel_host::log(
+            channel_host::LogLevel::Info,
+            &format!(
+                "Voice note from user {} (duration: {}s, file_id: {})",
+                from.id, voice.duration, voice.file_id
+            ),
+        );
+
+        match download_voice_file(&voice.file_id) {
+            Ok(audio_bytes) => {
+                channel_host::log(
+                    channel_host::LogLevel::Info,
+                    &format!("Downloaded voice file: {} bytes", audio_bytes.len()),
+                );
+                attachments.push(channel_host::Attachment {
+                    kind: channel_host::AttachmentKind::Audio,
+                    mime_type: voice
+                        .mime_type
+                        .clone()
+                        .unwrap_or_else(|| "audio/ogg".to_string()),
+                    data: audio_bytes,
+                    filename: Some(format!("voice_{}.ogg", voice.file_id)),
+                    duration_secs: Some(voice.duration),
+                });
+            }
+            Err(e) => {
+                channel_host::log(
+                    channel_host::LogLevel::Error,
+                    &format!("Failed to download voice file: {}", e),
+                );
+                voice_download_failed = true;
+            }
+        }
+    }
+
     // Determine what to emit to the agent.
+    // - Voice notes: use "[Voice note]" as content (transcription happens host-side)
     // - `/start` (no args): emit a welcome placeholder so the agent greets the user
     // - Other bare `/commands` (e.g. /interrupt, /help): pass the raw command through
     //   so Submission::parse() can handle it
     // - Commands with args (e.g. `/start hello`): cleaned_text already has the args
     // - Plain text: pass through as-is
     let trimmed_content = content.trim();
-    let content_to_emit = if trimmed_content.eq_ignore_ascii_case("/start") {
+    let content_to_emit = if is_voice && voice_download_failed && content.is_empty() {
+        "[Voice note: download failed]".to_string()
+    } else if is_voice && content.is_empty() {
+        "[Voice note]".to_string()
+    } else if trimmed_content.eq_ignore_ascii_case("/start") {
         "[User started the bot]".to_string()
     } else if cleaned_text.is_empty() && trimmed_content.starts_with('/') {
         // Bare control command like /interrupt, /stop, /help — pass through raw
         trimmed_content.to_string()
-    } else if cleaned_text.is_empty() {
+    } else if cleaned_text.is_empty() && !is_voice {
         return;
     } else {
         cleaned_text
@@ -1063,6 +1227,7 @@ fn handle_message(message: TelegramMessage) {
         content: content_to_emit,
         thread_id: None, // Telegram doesn't have threads in the same way
         metadata_json,
+        attachments,
     });
 
     channel_host::log(

channels-src/telegram/telegram.capabilities.json

@@ -1 +1,55 @@
-{"type":"channel","name":"telegram","description":"Telegram Bot API channel for receiving and responding to Telegram messages","capabilities":{"http":{"allowlist":[{"host":"api.telegram.org","path_prefix":"/bot"}],"credentials":{"telegram_bot":{"secret_name":"telegram_bot_token","location":{"type":"url_path","placeholder":"{TELEGRAM_BOT_TOKEN}"},"host_patterns":["api.telegram.org"]}},"rate_limit":{"requests_per_minute":30,"requests_per_hour":1000}},"secrets":{"allowed_names":["telegram_*"]},"channel":{"allowed_paths":["/webhook/telegram"],"allow_polling":true,"min_poll_interval_ms":30000,"workspace_prefix":"channels/telegram/","emit_rate_limit":{"messages_per_minute":100,"messages_per_hour":5000}}},"config":{"bot_username":null,"owner_id":null,"respond_to_all_group_messages":false,"polling_enabled":false,"poll_interval_ms":30000,"dm_policy":"pairing","allow_from":[]}}
+{
+  "type": "channel",
+  "name": "telegram",
+  "description": "Telegram Bot API channel for receiving and responding to Telegram messages",
+  "setup": {
+    "required_secrets": [
+      {
+        "name": "telegram_bot_token",
+        "prompt": "Enter your Telegram Bot API token (from @BotFather)",
+        "optional": false
+      }
+    ]
+  },
+  "capabilities": {
+    "http": {
+      "allowlist": [
+        { "host": "api.telegram.org", "path_prefix": "/bot" },
+        { "host": "api.telegram.org", "path_prefix": "/file/bot" }
+      ],
+      "credentials": {
+        "telegram_bot": {
+          "secret_name": "telegram_bot_token",
+          "location": { "type": "url_path", "placeholder": "{TELEGRAM_BOT_TOKEN}" },
+          "host_patterns": ["api.telegram.org"]
+        }
+      },
+      "rate_limit": {
+        "requests_per_minute": 30,
+        "requests_per_hour": 1000
+      }
+    },
+    "secrets": {
+      "allowed_names": ["telegram_*"]
+    },
+    "channel": {
+      "allowed_paths": ["/webhook/telegram"],
+      "allow_polling": true,
+      "min_poll_interval_ms": 30000,
+      "workspace_prefix": "channels/telegram/",
+      "emit_rate_limit": {
+        "messages_per_minute": 100,
+        "messages_per_hour": 5000
+      }
+    }
+  },
+  "config": {
+    "bot_username": null,
+    "owner_id": null,
+    "respond_to_all_group_messages": false,
+    "polling_enabled": false,
+    "poll_interval_ms": 30000,
+    "dm_policy": "pairing",
+    "allow_from": []
+  }
+}

channels-src/whatsapp/Cargo.toml

@@ -16,3 +16,5 @@ serde_json = "1"
 opt-level = "s"
 lto = true
 strip = true
+
+[workspace]

channels-src/whatsapp/src/lib.rs

@@ -226,6 +226,15 @@ struct WhatsAppMessageMetadata {
     timestamp: String,
 }
 
+/// Workspace path for persisting owner_id across WASM callbacks.
+const OWNER_ID_PATH: &str = "state/owner_id";
+/// Workspace path for persisting dm_policy across WASM callbacks.
+const DM_POLICY_PATH: &str = "state/dm_policy";
+/// Workspace path for persisting allow_from (JSON array) across WASM callbacks.
+const ALLOW_FROM_PATH: &str = "state/allow_from";
+/// Channel name for pairing store (used by pairing host APIs).
+const CHANNEL_NAME: &str = "whatsapp";
+
 /// Channel configuration from capabilities file.
 #[derive(Debug, Deserialize)]
 struct WhatsAppConfig {
@@ -236,6 +245,15 @@ struct WhatsAppConfig {
     /// Whether to reply to the original message (thread context)
     #[serde(default = "default_reply_to_message")]
     reply_to_message: bool,
+
+    #[serde(default)]
+    owner_id: Option<String>,
+
+    #[serde(default)]
+    dm_policy: Option<String>,
+
+    #[serde(default)]
+    allow_from: Option<Vec<String>>,
 }
 
 fn default_api_version() -> String {
@@ -264,6 +282,9 @@ impl Guest for WhatsAppChannel {
                 WhatsAppConfig {
                     api_version: default_api_version(),
                     reply_to_message: default_reply_to_message(),
+                    owner_id: None,
+                    dm_policy: None,
+                    allow_from: None,
                 }
             }
         };
@@ -279,6 +300,24 @@ impl Guest for WhatsAppChannel {
         // Persist api_version in workspace so on_respond() can read it
         let _ = channel_host::workspace_write("channels/whatsapp/api_version", &config.api_version);
 
+        // Persist permission config for handle_message
+        if let Some(ref owner_id) = config.owner_id {
+            let _ = channel_host::workspace_write(OWNER_ID_PATH, owner_id);
+            channel_host::log(
+                channel_host::LogLevel::Info,
+                &format!("Owner restriction enabled: user {}", owner_id),
+            );
+        } else {
+            let _ = channel_host::workspace_write(OWNER_ID_PATH, "");
+        }
+
+        let dm_policy = config.dm_policy.as_deref().unwrap_or("pairing");
+        let _ = channel_host::workspace_write(DM_POLICY_PATH, dm_policy);
+
+        let allow_from_json = serde_json::to_string(&config.allow_from.unwrap_or_default())
+            .unwrap_or_else(|_| "[]".to_string());
+        let _ = channel_host::workspace_write(ALLOW_FROM_PATH, &allow_from_json);
+
         // WhatsApp Cloud API is webhook-only, no polling available
         Ok(ChannelConfig {
             display_name: "WhatsApp".to_string(),
@@ -604,6 +643,15 @@ fn handle_message(
     // Look up sender's name from contacts
     let user_name = contact_names.get(&message.from).cloned();
 
+    // Permission check (WhatsApp is always DM)
+    if !check_sender_permission(
+        &message.from,
+        user_name.as_deref(),
+        phone_number_id,
+    ) {
+        return;
+    }
+
     // Build metadata for response routing
     // This is critical - the response handler uses this to know where to send
     let metadata = WhatsAppMessageMetadata {
@@ -637,6 +685,149 @@ fn handle_message(
 // Utilities
 // ============================================================================
 
+// ============================================================================
+// Permission & Pairing
+// ============================================================================
+
+/// Check if a sender is permitted. Returns true if allowed.
+/// WhatsApp is always 1-to-1 (DM), so dm_policy always applies.
+fn check_sender_permission(
+    sender_phone: &str,
+    user_name: Option<&str>,
+    phone_number_id: &str,
+) -> bool {
+    // 1. Owner check (highest priority)
+    let owner_id = channel_host::workspace_read(OWNER_ID_PATH).filter(|s| !s.is_empty());
+    if let Some(ref owner) = owner_id {
+        if sender_phone != owner {
+            channel_host::log(
+                channel_host::LogLevel::Debug,
+                &format!(
+                    "Dropping message from non-owner {} (owner: {})",
+                    sender_phone, owner
+                ),
+            );
+            return false;
+        }
+        return true;
+    }
+
+    // 2. DM policy (WhatsApp is always DM)
+    let dm_policy =
+        channel_host::workspace_read(DM_POLICY_PATH).unwrap_or_else(|| "pairing".to_string());
+
+    if dm_policy == "open" {
+        return true;
+    }
+
+    // 3. Build merged allow list
+    let mut allowed: Vec<String> = channel_host::workspace_read(ALLOW_FROM_PATH)
+        .and_then(|s| serde_json::from_str(&s).ok())
+        .unwrap_or_default();
+
+    if let Ok(store_allowed) = channel_host::pairing_read_allow_from(CHANNEL_NAME) {
+        allowed.extend(store_allowed);
+    }
+
+    // 4. Check sender (phone number or name)
+    let is_allowed = allowed.contains(&"*".to_string())
+        || allowed.contains(&sender_phone.to_string())
+        || user_name.is_some_and(|u| allowed.contains(&u.to_string()));
+
+    if is_allowed {
+        return true;
+    }
+
+    // 5. Not allowed — handle by policy
+    if dm_policy == "pairing" {
+        let meta = serde_json::json!({
+            "phone": sender_phone,
+            "name": user_name,
+        })
+        .to_string();
+
+        match channel_host::pairing_upsert_request(CHANNEL_NAME, sender_phone, &meta) {
+            Ok(result) => {
+                channel_host::log(
+                    channel_host::LogLevel::Info,
+                    &format!(
+                        "Pairing request for {}: code {}",
+                        sender_phone, result.code
+                    ),
+                );
+                if result.created {
+                    let _ = send_pairing_reply(sender_phone, phone_number_id, &result.code);
+                }
+            }
+            Err(e) => {
+                channel_host::log(
+                    channel_host::LogLevel::Error,
+                    &format!("Pairing upsert failed: {}", e),
+                );
+            }
+        }
+    }
+    false
+}
+
+/// Send a pairing code message via WhatsApp Cloud API.
+fn send_pairing_reply(
+    recipient_phone: &str,
+    phone_number_id: &str,
+    code: &str,
+) -> Result<(), String> {
+    let api_version = channel_host::workspace_read("channels/whatsapp/api_version")
+        .filter(|s| !s.is_empty())
+        .unwrap_or_else(|| "v18.0".to_string());
+
+    let url = format!(
+        "https://graph.facebook.com/{}/{}/messages",
+        api_version, phone_number_id
+    );
+
+    let payload = serde_json::json!({
+        "messaging_product": "whatsapp",
+        "recipient_type": "individual",
+        "to": recipient_phone,
+        "type": "text",
+        "text": {
+            "preview_url": false,
+            "body": format!(
+                "To pair with this bot, run: ironclaw pairing approve whatsapp {}",
+                code
+            )
+        }
+    });
+
+    let payload_bytes =
+        serde_json::to_vec(&payload).map_err(|e| format!("Failed to serialize: {}", e))?;
+
+    let headers = serde_json::json!({
+        "Content-Type": "application/json",
+        "Authorization": "Bearer {WHATSAPP_ACCESS_TOKEN}"
+    });
+
+    let result = channel_host::http_request(
+        "POST",
+        &url,
+        &headers.to_string(),
+        Some(&payload_bytes),
+        None,
+    );
+
+    match result {
+        Ok(response) if response.status >= 200 && response.status < 300 => Ok(()),
+        Ok(response) => {
+            let body_str = String::from_utf8_lossy(&response.body);
+            Err(format!(
+                "WhatsApp API error: {} - {}",
+                response.status, body_str
+            ))
+        }
+        Err(e) => Err(format!("HTTP request failed: {}", e)),
+    }
+}
+
 /// Create a JSON HTTP response.
 fn json_response(status: u16, value: serde_json::Value) -> OutgoingHttpResponse {
     let body = serde_json::to_vec(&value).unwrap_or_default();

channels-src/whatsapp/whatsapp.capabilities.json

@@ -48,6 +48,9 @@
   },
   "config": {
     "api_version": "v18.0",
-    "reply_to_message": true
+    "reply_to_message": true,
+    "owner_id": null,
+    "dm_policy": "pairing",
+    "allow_from": []
   }
 }

deploy/env.example

@@ -2,12 +2,15 @@
 # Do not use placeholder passwords in production.
 DATABASE_URL=postgres://ironclaw:CHANGE_ME@localhost:5432/ironclaw
 
-# NEAR AI
-NEARAI_SESSION_TOKEN=CHANGE_ME
+# NEAR AI Cloud (API key auth, Chat Completions API)
+# Get an API key from https://cloud.near.ai
+NEARAI_API_KEY=CHANGE_ME
 NEARAI_MODEL=claude-3-5-sonnet-20241022
-NEARAI_BASE_URL=https://private.near.ai
-NEARAI_AUTH_URL=https://private.near.ai
-NEARAI_API_MODE=chat_completions
+NEARAI_BASE_URL=https://cloud-api.near.ai
+
+# Or use NEAR AI Chat (session token auth, Responses API):
+# NEARAI_SESSION_TOKEN=sess_...
+# NEARAI_BASE_URL=https://private.near.ai
 
 # Agent
 AGENT_NAME=ironclaw

docs/LLM_PROVIDERS.md

@@ -0,0 +1,172 @@
+# LLM Provider Configuration
+
+IronClaw defaults to NEAR AI for model access, but supports any OpenAI-compatible
+endpoint as well as Anthropic and Ollama directly. This guide covers the most common
+configurations.
+
+## Provider Overview
+
+| Provider | Backend value | Requires API key | Notes |
+|---|---|---|---|
+| NEAR AI | `nearai` | OAuth (browser) | Default; multi-model |
+| Anthropic | `anthropic` | `ANTHROPIC_API_KEY` | Claude models |
+| OpenAI | `openai` | `OPENAI_API_KEY` | GPT models |
+| Ollama | `ollama` | No | Local inference |
+| OpenRouter | `openai_compatible` | `LLM_API_KEY` | 300+ models |
+| Together AI | `openai_compatible` | `LLM_API_KEY` | Fast inference |
+| Fireworks AI | `openai_compatible` | `LLM_API_KEY` | Fast inference |
+| vLLM / LiteLLM | `openai_compatible` | Optional | Self-hosted |
+| LM Studio | `openai_compatible` | No | Local GUI |
+
+---
+
+## NEAR AI (default)
+
+No additional configuration required. On first run, `ironclaw onboard` opens a browser
+for OAuth authentication. Credentials are saved to `~/.ironclaw/session.json`.
+
+```env
+NEARAI_MODEL=claude-3-5-sonnet-20241022
+NEARAI_BASE_URL=https://private.near.ai
+```
+
+---
+
+## Anthropic (Claude)
+
+```env
+LLM_BACKEND=anthropic
+ANTHROPIC_API_KEY=sk-ant-...
+```
+
+Popular models: `claude-sonnet-4-20250514`, `claude-3-5-sonnet-20241022`, `claude-3-5-haiku-20241022`
+
+---
+
+## OpenAI (GPT)
+
+```env
+LLM_BACKEND=openai
+OPENAI_API_KEY=sk-...
+```
+
+Popular models: `gpt-4o`, `gpt-4o-mini`, `o3-mini`
+
+---
+
+## Ollama (local)
+
+Install Ollama from [ollama.com](https://ollama.com), pull a model, then:
+
+```env
+LLM_BACKEND=ollama
+OLLAMA_MODEL=llama3.2
+# OLLAMA_BASE_URL=http://localhost:11434   # default
+```
+
+Pull a model first: `ollama pull llama3.2`
+
+---
+
+## OpenAI-Compatible Endpoints
+
+All providers below use `LLM_BACKEND=openai_compatible`. Set `LLM_BASE_URL` to the
+provider's OpenAI-compatible endpoint and `LLM_API_KEY` to your API key.
+
+### OpenRouter
+
+[OpenRouter](https://openrouter.ai) routes to 300+ models from a single API key.
+
+```env
+LLM_BACKEND=openai_compatible
+LLM_BASE_URL=https://openrouter.ai/api/v1
+LLM_API_KEY=sk-or-...
+LLM_MODEL=anthropic/claude-sonnet-4
+```
+
+Popular OpenRouter model IDs:
+
+| Model | ID |
+|---|---|
+| Claude Sonnet 4 | `anthropic/claude-sonnet-4` |
+| GPT-4o | `openai/gpt-4o` |
+| Llama 4 Maverick | `meta-llama/llama-4-maverick` |
+| Gemini 2.0 Flash | `google/gemini-2.0-flash-001` |
+| Mistral Small | `mistralai/mistral-small-3.1-24b-instruct` |
+
+Browse all models at [openrouter.ai/models](https://openrouter.ai/models).
+
+### Together AI
+
+[Together AI](https://www.together.ai) provides fast inference for open-source models.
+
+```env
+LLM_BACKEND=openai_compatible
+LLM_BASE_URL=https://api.together.xyz/v1
+LLM_API_KEY=...
+LLM_MODEL=meta-llama/Llama-3.3-70B-Instruct-Turbo
+```
+
+Popular Together AI model IDs:
+
+| Model | ID |
+|---|---|
+| Llama 3.3 70B | `meta-llama/Llama-3.3-70B-Instruct-Turbo` |
+| DeepSeek R1 | `deepseek-ai/DeepSeek-R1` |
+| Qwen 2.5 72B | `Qwen/Qwen2.5-72B-Instruct-Turbo` |
+
+### Fireworks AI
+
+[Fireworks AI](https://fireworks.ai) offers fast inference with compound AI system support.
+
+```env
+LLM_BACKEND=openai_compatible
+LLM_BASE_URL=https://api.fireworks.ai/inference/v1
+LLM_API_KEY=fw_...
+LLM_MODEL=accounts/fireworks/models/llama4-maverick-instruct-basic
+```
+
+### vLLM / LiteLLM (self-hosted)
+
+For self-hosted inference servers:
+
+```env
+LLM_BACKEND=openai_compatible
+LLM_BASE_URL=http://localhost:8000/v1
+LLM_API_KEY=token-abc123        # set to any string if auth is not configured
+LLM_MODEL=meta-llama/Llama-3.1-8B-Instruct
+```
+
+LiteLLM proxy (forwards to any backend, including Bedrock, Vertex, Azure):
+
+```env
+LLM_BACKEND=openai_compatible
+LLM_BASE_URL=http://localhost:4000/v1
+LLM_API_KEY=sk-...
+LLM_MODEL=gpt-4o                 # as configured in litellm config.yaml
+```
+
+### LM Studio (local GUI)
+
+Start LM Studio's local server, then:
+
+```env
+LLM_BACKEND=openai_compatible
+LLM_BASE_URL=http://localhost:1234/v1
+LLM_MODEL=llama-3.2-3b-instruct-q4_K_M
+# LLM_API_KEY is not required for LM Studio
+```
+
+---
+
+## Using the Setup Wizard
+
+Instead of editing `.env` manually, run the onboarding wizard:
+
+```bash
+ironclaw onboard
+```
+
+Select **"OpenAI-compatible"** for OpenRouter, Together AI, Fireworks, vLLM, LiteLLM,
+or LM Studio. You will be prompted for the base URL and (optionally) an API key.
+The model name is configured in the following step.

registry/_bundles.json

@@ -0,0 +1,42 @@
+{
+  "bundles": {
+    "google": {
+      "display_name": "Google Suite",
+      "description": "Gmail, Calendar, Drive, Docs, Sheets, Slides",
+      "extensions": [
+        "tools/gmail",
+        "tools/google-calendar",
+        "tools/google-docs",
+        "tools/google-drive",
+        "tools/google-sheets",
+        "tools/google-slides"
+      ],
+      "shared_auth": "google_oauth_token"
+    },
+    "messaging": {
+      "display_name": "Messaging Channels",
+      "description": "Discord, Telegram, Slack, and WhatsApp channels",
+      "extensions": [
+        "channels/discord",
+        "channels/telegram",
+        "channels/slack",
+        "channels/whatsapp"
+      ],
+      "shared_auth": null
+    },
+    "default": {
+      "display_name": "Recommended Set",
+      "description": "Core tools and channels for a productive setup",
+      "extensions": [
+        "tools/github",
+        "tools/gmail",
+        "tools/google-calendar",
+        "tools/google-drive",
+        "tools/slack",
+        "channels/telegram",
+        "channels/slack"
+      ],
+      "shared_auth": null
+    }
+  }
+}

registry/channels/discord.json

@@ -0,0 +1,31 @@
+{
+  "name": "discord",
+  "display_name": "Discord",
+  "kind": "channel",
+  "version": "0.1.0",
+  "description": "Discord Gateway/Webhook channel for slash commands, buttons, and messages",
+  "keywords": ["messaging", "chat", "discord", "bot"],
+
+  "source": {
+    "dir": "channels-src/discord",
+    "capabilities": "discord.capabilities.json",
+    "crate_name": "discord-channel"
+  },
+
+  "artifacts": {
+    "wasm32-wasip2": {
+      "url": "https://github.com/nearai/ironclaw/releases/latest/download/discord-wasm32-wasip2.tar.gz",
+      "sha256": null
+    }
+  },
+
+  "auth_summary": {
+    "method": "manual",
+    "provider": "Discord",
+    "secrets": ["discord_bot_token"],
+    "shared_auth": null,
+    "setup_url": "https://discord.com/developers/applications"
+  },
+
+  "tags": ["messaging"]
+}

registry/channels/slack.json

@@ -0,0 +1,31 @@
+{
+  "name": "slack",
+  "display_name": "Slack",
+  "kind": "channel",
+  "version": "0.1.0",
+  "description": "Slack Events API channel for receiving and responding to Slack messages",
+  "keywords": ["messaging", "chat", "workspace", "slack"],
+
+  "source": {
+    "dir": "channels-src/slack",
+    "capabilities": "slack.capabilities.json",
+    "crate_name": "slack-channel"
+  },
+
+  "artifacts": {
+    "wasm32-wasip2": {
+      "url": "https://github.com/nearai/ironclaw/releases/latest/download/slack-wasm32-wasip2.tar.gz",
+      "sha256": null
+    }
+  },
+
+  "auth_summary": {
+    "method": "manual",
+    "provider": "Slack",
+    "secrets": ["slack_bot_token", "slack_signing_secret"],
+    "shared_auth": null,
+    "setup_url": "https://api.slack.com/apps"
+  },
+
+  "tags": ["default", "messaging"]
+}

registry/channels/telegram.json

@@ -0,0 +1,31 @@
+{
+  "name": "telegram",
+  "display_name": "Telegram",
+  "kind": "channel",
+  "version": "0.1.0",
+  "description": "Telegram Bot API channel for receiving and responding to messages",
+  "keywords": ["messaging", "bot", "chat", "telegram"],
+
+  "source": {
+    "dir": "channels-src/telegram",
+    "capabilities": "telegram.capabilities.json",
+    "crate_name": "telegram-channel"
+  },
+
+  "artifacts": {
+    "wasm32-wasip2": {
+      "url": "https://github.com/nearai/ironclaw/releases/latest/download/telegram-wasm32-wasip2.tar.gz",
+      "sha256": null
+    }
+  },
+
+  "auth_summary": {
+    "method": "manual",
+    "provider": "Telegram",
+    "secrets": ["telegram_bot_token"],
+    "shared_auth": null,
+    "setup_url": "https://t.me/BotFather"
+  },
+
+  "tags": ["default", "messaging"]
+}

registry/channels/whatsapp.json

@@ -0,0 +1,31 @@
+{
+  "name": "whatsapp",
+  "display_name": "WhatsApp",
+  "kind": "channel",
+  "version": "0.1.0",
+  "description": "WhatsApp Cloud API channel for receiving and responding to messages",
+  "keywords": ["messaging", "chat", "whatsapp", "meta"],
+
+  "source": {
+    "dir": "channels-src/whatsapp",
+    "capabilities": "whatsapp.capabilities.json",
+    "crate_name": "whatsapp-channel"
+  },
+
+  "artifacts": {
+    "wasm32-wasip2": {
+      "url": "https://github.com/nearai/ironclaw/releases/latest/download/whatsapp-wasm32-wasip2.tar.gz",
+      "sha256": null
+    }
+  },
+
+  "auth_summary": {
+    "method": "manual",
+    "provider": "Meta",
+    "secrets": ["whatsapp_access_token", "whatsapp_verify_token"],
+    "shared_auth": null,
+    "setup_url": "https://developers.facebook.com/apps/"
+  },
+
+  "tags": ["messaging"]
+}

registry/tools/github.json

@@ -0,0 +1,31 @@
+{
+  "name": "github",
+  "display_name": "GitHub",
+  "kind": "tool",
+  "version": "0.1.0",
+  "description": "GitHub integration for issues, PRs, repos, and code search",
+  "keywords": ["git", "code", "issues", "pull-requests", "repositories"],
+
+  "source": {
+    "dir": "tools-src/github",
+    "capabilities": "github-tool.capabilities.json",
+    "crate_name": "github-tool"
+  },
+
+  "artifacts": {
+    "wasm32-wasip2": {
+      "url": "https://github.com/nearai/ironclaw/releases/latest/download/github-wasm32-wasip2.tar.gz",
+      "sha256": null
+    }
+  },
+
+  "auth_summary": {
+    "method": "manual",
+    "provider": "GitHub",
+    "secrets": ["github_token"],
+    "shared_auth": null,
+    "setup_url": "https://github.com/settings/tokens"
+  },
+
+  "tags": ["default", "development"]
+}

registry/tools/gmail.json

@@ -0,0 +1,31 @@
+{
+  "name": "gmail",
+  "display_name": "Gmail",
+  "kind": "tool",
+  "version": "0.1.0",
+  "description": "Read, send, and manage Gmail messages and threads",
+  "keywords": ["email", "google", "mail", "messaging"],
+
+  "source": {
+    "dir": "tools-src/gmail",
+    "capabilities": "gmail-tool.capabilities.json",
+    "crate_name": "gmail-tool"
+  },
+
+  "artifacts": {
+    "wasm32-wasip2": {
+      "url": "https://github.com/nearai/ironclaw/releases/latest/download/gmail-wasm32-wasip2.tar.gz",
+      "sha256": null
+    }
+  },
+
+  "auth_summary": {
+    "method": "oauth",
+    "provider": "Google",
+    "secrets": ["google_oauth_token"],
+    "shared_auth": "google_oauth_token",
+    "setup_url": "https://console.cloud.google.com/apis/credentials"
+  },
+
+  "tags": ["default", "google", "messaging"]
+}

registry/tools/google-calendar.json

@@ -0,0 +1,31 @@
+{
+  "name": "google-calendar",
+  "display_name": "Google Calendar",
+  "kind": "tool",
+  "version": "0.1.0",
+  "description": "Create, read, update, and delete Google Calendar events",
+  "keywords": ["calendar", "google", "scheduling", "events"],
+
+  "source": {
+    "dir": "tools-src/google-calendar",
+    "capabilities": "google-calendar-tool.capabilities.json",
+    "crate_name": "google-calendar-tool"
+  },
+
+  "artifacts": {
+    "wasm32-wasip2": {
+      "url": "https://github.com/nearai/ironclaw/releases/latest/download/google-calendar-wasm32-wasip2.tar.gz",
+      "sha256": null
+    }
+  },
+
+  "auth_summary": {
+    "method": "oauth",
+    "provider": "Google",
+    "secrets": ["google_oauth_token"],
+    "shared_auth": "google_oauth_token",
+    "setup_url": "https://console.cloud.google.com/apis/credentials"
+  },
+
+  "tags": ["default", "google", "productivity"]
+}

registry/tools/google-docs.json

@@ -0,0 +1,31 @@
+{
+  "name": "google-docs",
+  "display_name": "Google Docs",
+  "kind": "tool",
+  "version": "0.1.0",
+  "description": "Create and edit Google Docs documents",
+  "keywords": ["documents", "google", "writing", "docs"],
+
+  "source": {
+    "dir": "tools-src/google-docs",
+    "capabilities": "google-docs-tool.capabilities.json",
+    "crate_name": "google-docs-tool"
+  },
+
+  "artifacts": {
+    "wasm32-wasip2": {
+      "url": "https://github.com/nearai/ironclaw/releases/latest/download/google-docs-wasm32-wasip2.tar.gz",
+      "sha256": null
+    }
+  },
+
+  "auth_summary": {
+    "method": "oauth",
+    "provider": "Google",
+    "secrets": ["google_oauth_token"],
+    "shared_auth": "google_oauth_token",
+    "setup_url": "https://console.cloud.google.com/apis/credentials"
+  },
+
+  "tags": ["google", "productivity"]
+}

registry/tools/google-drive.json

@@ -0,0 +1,31 @@
+{
+  "name": "google-drive",
+  "display_name": "Google Drive",
+  "kind": "tool",
+  "version": "0.1.0",
+  "description": "Upload, download, search, and manage Google Drive files and folders",
+  "keywords": ["storage", "google", "files", "drive"],
+
+  "source": {
+    "dir": "tools-src/google-drive",
+    "capabilities": "google-drive-tool.capabilities.json",
+    "crate_name": "google-drive-tool"
+  },
+
+  "artifacts": {
+    "wasm32-wasip2": {
+      "url": "https://github.com/nearai/ironclaw/releases/latest/download/google-drive-wasm32-wasip2.tar.gz",
+      "sha256": null
+    }
+  },
+
+  "auth_summary": {
+    "method": "oauth",
+    "provider": "Google",
+    "secrets": ["google_oauth_token"],
+    "shared_auth": "google_oauth_token",
+    "setup_url": "https://console.cloud.google.com/apis/credentials"
+  },
+
+  "tags": ["default", "google", "storage"]
+}

registry/tools/google-sheets.json

@@ -0,0 +1,31 @@
+{
+  "name": "google-sheets",
+  "display_name": "Google Sheets",
+  "kind": "tool",
+  "version": "0.1.0",
+  "description": "Read and write Google Sheets spreadsheet data",
+  "keywords": ["spreadsheets", "google", "data", "sheets"],
+
+  "source": {
+    "dir": "tools-src/google-sheets",
+    "capabilities": "google-sheets-tool.capabilities.json",
+    "crate_name": "google-sheets-tool"
+  },
+
+  "artifacts": {
+    "wasm32-wasip2": {
+      "url": "https://github.com/nearai/ironclaw/releases/latest/download/google-sheets-wasm32-wasip2.tar.gz",
+      "sha256": null
+    }
+  },
+
+  "auth_summary": {
+    "method": "oauth",
+    "provider": "Google",
+    "secrets": ["google_oauth_token"],
+    "shared_auth": "google_oauth_token",
+    "setup_url": "https://console.cloud.google.com/apis/credentials"
+  },
+
+  "tags": ["google", "productivity"]
+}

registry/tools/google-slides.json

@@ -0,0 +1,31 @@
+{
+  "name": "google-slides",
+  "display_name": "Google Slides",
+  "kind": "tool",
+  "version": "0.1.0",
+  "description": "Create and edit Google Slides presentations",
+  "keywords": ["presentations", "google", "slides"],
+
+  "source": {
+    "dir": "tools-src/google-slides",
+    "capabilities": "google-slides-tool.capabilities.json",
+    "crate_name": "google-slides-tool"
+  },
+
+  "artifacts": {
+    "wasm32-wasip2": {
+      "url": "https://github.com/nearai/ironclaw/releases/latest/download/google-slides-wasm32-wasip2.tar.gz",
+      "sha256": null
+    }
+  },
+
+  "auth_summary": {
+    "method": "oauth",
+    "provider": "Google",
+    "secrets": ["google_oauth_token"],
+    "shared_auth": "google_oauth_token",
+    "setup_url": "https://console.cloud.google.com/apis/credentials"
+  },
+
+  "tags": ["google", "productivity"]
+}

registry/tools/okta.json

@@ -0,0 +1,31 @@
+{
+  "name": "okta",
+  "display_name": "Okta",
+  "kind": "tool",
+  "version": "0.1.0",
+  "description": "Okta SSO for user profile, app catalog, and SSO launch links",
+  "keywords": ["sso", "identity", "authentication", "okta"],
+
+  "source": {
+    "dir": "tools-src/okta",
+    "capabilities": "okta-tool.capabilities.json",
+    "crate_name": "okta-tool"
+  },
+
+  "artifacts": {
+    "wasm32-wasip2": {
+      "url": "https://github.com/nearai/ironclaw/releases/latest/download/okta-wasm32-wasip2.tar.gz",
+      "sha256": null
+    }
+  },
+
+  "auth_summary": {
+    "method": "oauth",
+    "provider": "Okta",
+    "secrets": ["okta_oauth_token"],
+    "shared_auth": null,
+    "setup_url": "https://developer.okta.com/docs/guides/implement-oauth-for-okta/main/"
+  },
+
+  "tags": ["identity"]
+}

registry/tools/slack.json

@@ -0,0 +1,31 @@
+{
+  "name": "slack",
+  "display_name": "Slack",
+  "kind": "tool",
+  "version": "0.1.0",
+  "description": "Post messages, read channels, and manage conversations via Slack API",
+  "keywords": ["messaging", "chat", "workspace"],
+
+  "source": {
+    "dir": "tools-src/slack",
+    "capabilities": "slack-tool.capabilities.json",
+    "crate_name": "slack-tool"
+  },
+
+  "artifacts": {
+    "wasm32-wasip2": {
+      "url": "https://github.com/nearai/ironclaw/releases/latest/download/slack-wasm32-wasip2.tar.gz",
+      "sha256": null
+    }
+  },
+
+  "auth_summary": {
+    "method": "oauth",
+    "provider": "Slack",
+    "secrets": ["slack_bot_token"],
+    "shared_auth": null,
+    "setup_url": "https://api.slack.com/apps"
+  },
+
+  "tags": ["default", "messaging"]
+}

registry/tools/telegram.json

@@ -0,0 +1,31 @@
+{
+  "name": "telegram",
+  "display_name": "Telegram",
+  "kind": "tool",
+  "version": "0.1.0",
+  "description": "Telegram user-mode integration via MTProto for messages and contacts",
+  "keywords": ["messaging", "chat", "telegram", "mtproto"],
+
+  "source": {
+    "dir": "tools-src/telegram",
+    "capabilities": "telegram-tool.capabilities.json",
+    "crate_name": "telegram-tool"
+  },
+
+  "artifacts": {
+    "wasm32-wasip2": {
+      "url": "https://github.com/nearai/ironclaw/releases/latest/download/telegram-wasm32-wasip2.tar.gz",
+      "sha256": null
+    }
+  },
+
+  "auth_summary": {
+    "method": "manual",
+    "provider": "Telegram",
+    "secrets": ["telegram_api_id", "telegram_api_hash"],
+    "shared_auth": null,
+    "setup_url": "https://my.telegram.org/apps"
+  },
+
+  "tags": ["messaging"]
+}

src/agent/agent_loop.rs

@@ -85,6 +85,7 @@ pub struct Agent {
     pub(super) session_manager: Arc<SessionManager>,
     pub(super) context_monitor: ContextMonitor,
     pub(super) heartbeat_config: Option<HeartbeatConfig>,
+    pub(super) hygiene_config: Option<crate::config::HygieneConfig>,
     pub(super) routine_config: Option<RoutineConfig>,
 }
 
@@ -93,11 +94,13 @@ impl Agent {
     ///
     /// Optionally accepts pre-created `ContextManager` and `SessionManager` for sharing
     /// with external components (job tools, web gateway). Creates new ones if not provided.
+    #[allow(clippy::too_many_arguments)]
     pub fn new(
         config: AgentConfig,
         deps: AgentDeps,
         channels: ChannelManager,
         heartbeat_config: Option<HeartbeatConfig>,
+        hygiene_config: Option<crate::config::HygieneConfig>,
         routine_config: Option<RoutineConfig>,
         context_manager: Option<Arc<ContextManager>>,
         session_manager: Option<Arc<SessionManager>>,
@@ -127,6 +130,7 @@ impl Agent {
             session_manager,
             context_monitor: ContextMonitor::new(),
             heartbeat_config,
+            hygiene_config,
             routine_config,
         }
     }
@@ -354,14 +358,18 @@ impl Agent {
                         }
                     });
 
-                    tracing::info!(
-                        "Heartbeat enabled with {}s interval",
-                        hb_config.interval_secs
-                    );
+                    let hygiene = self
+                        .hygiene_config
+                        .as_ref()
+                        .map(|h| h.to_workspace_config())
+                        .unwrap_or_default();
+
                     Some(spawn_heartbeat(
                         config,
+                        hygiene,
                         workspace.clone(),
                         self.cheap_llm().clone(),
+                        self.safety().clone(),
                         Some(notify_tx),
                     ))
                 } else {
@@ -674,7 +682,15 @@ impl Agent {
 
         // Convert SubmissionResult to response string
         match result? {
-            SubmissionResult::Response { content } => Ok(Some(content)),
+            SubmissionResult::Response { content } => {
+                // Suppress silent replies (e.g. from group chat "nothing to say" responses)
+                if crate::llm::is_silent_reply(&content) {
+                    tracing::debug!("Suppressing silent reply token");
+                    Ok(None)
+                } else {
+                    Ok(Some(content))
+                }
+            }
             SubmissionResult::Ok { message } => Ok(message),
             SubmissionResult::Error { message } => Ok(Some(format!("Error: {}", message))),
             SubmissionResult::Interrupted => Ok(Some("Interrupted.".into())),

src/agent/commands.rs

@@ -13,7 +13,7 @@ use crate::agent::submission::SubmissionResult;
 use crate::agent::{Agent, MessageIntent};
 use crate::channels::{IncomingMessage, StatusUpdate};
 use crate::error::Error;
-use crate::llm::ChatMessage;
+use crate::llm::{ChatMessage, Reasoning};
 
 impl Agent {
     /// Handle job-related intents without turn tracking.
@@ -232,8 +232,10 @@ impl Agent {
 
         let runner = crate::agent::HeartbeatRunner::new(
             crate::agent::HeartbeatConfig::default(),
+            crate::workspace::hygiene::HygieneConfig::default(),
             workspace.clone(),
             self.llm().clone(),
+            self.safety().clone(),
         );
 
         match runner.check_heartbeat().await {
@@ -294,10 +296,11 @@ impl Agent {
             .with_max_tokens(512)
             .with_temperature(0.3);
 
-        match self.llm().complete(request).await {
-            Ok(response) => Ok(SubmissionResult::response(format!(
+        let reasoning = Reasoning::new(self.llm().clone(), self.safety().clone());
+        match reasoning.complete(request).await {
+            Ok((text, _usage)) => Ok(SubmissionResult::response(format!(
                 "Thread Summary:\n\n{}",
-                response.content.trim()
+                text.trim()
             ))),
             Err(e) => Ok(SubmissionResult::error(format!("Summarize failed: {}", e))),
         }
@@ -341,10 +344,11 @@ impl Agent {
             .with_max_tokens(512)
             .with_temperature(0.5);
 
-        match self.llm().complete(request).await {
-            Ok(response) => Ok(SubmissionResult::response(format!(
+        let reasoning = Reasoning::new(self.llm().clone(), self.safety().clone());
+        match reasoning.complete(request).await {
+            Ok((text, _usage)) => Ok(SubmissionResult::response(format!(
                 "Suggested Next Steps:\n\n{}",
-                response.content.trim()
+                text.trim()
             ))),
             Err(e) => Ok(SubmissionResult::error(format!("Suggest failed: {}", e))),
         }

src/agent/compaction.rs

@@ -12,7 +12,8 @@ use chrono::Utc;
 use crate::agent::context_monitor::{CompactionStrategy, ContextBreakdown};
 use crate::agent::session::Thread;
 use crate::error::Error;
-use crate::llm::{ChatMessage, CompletionRequest, LlmProvider};
+use crate::llm::{ChatMessage, CompletionRequest, LlmProvider, Reasoning};
+use crate::safety::SafetyLayer;
 use crate::workspace::Workspace;
 
 /// Result of a compaction operation.
@@ -33,12 +34,13 @@ pub struct CompactionResult {
 /// Compacts conversation context to stay within limits.
 pub struct ContextCompactor {
     llm: Arc<dyn LlmProvider>,
+    safety: Arc<SafetyLayer>,
 }
 
 impl ContextCompactor {
     /// Create a new context compactor.
-    pub fn new(llm: Arc<dyn LlmProvider>) -> Self {
-        Self { llm }
+    pub fn new(llm: Arc<dyn LlmProvider>, safety: Arc<SafetyLayer>) -> Self {
+        Self { llm, safety }
     }
 
     /// Compact a thread's context using the given strategy.
@@ -105,7 +107,16 @@ impl ContextCompactor {
 
         // Write to workspace if available
         let summary_written = if let Some(ws) = workspace {
-            self.write_summary_to_workspace(ws, &summary).await.is_ok()
+            match self.write_summary_to_workspace(ws, &summary).await {
+                Ok(()) => true,
+                Err(e) => {
+                    tracing::warn!(
+                        "Compaction summary write failed (turns will still be truncated): {}",
+                        e
+                    );
+                    false
+                }
+            }
         } else {
             false
         };
@@ -157,7 +168,16 @@ impl ContextCompactor {
         let content = format_turns_for_storage(old_turns);
 
         // Write to workspace
-        let written = self.write_context_to_workspace(ws, &content).await.is_ok();
+        let written = match self.write_context_to_workspace(ws, &content).await {
+            Ok(()) => true,
+            Err(e) => {
+                tracing::warn!(
+                    "Compaction context write failed (turns will still be truncated): {}",
+                    e
+                );
+                false
+            }
+        };
 
         // Truncate
         thread.truncate_turns(keep_recent);
@@ -213,8 +233,9 @@ Be brief but capture all important details. Use bullet points."#,
             .with_max_tokens(1024)
             .with_temperature(0.3);
 
-        let response = self.llm.complete(request).await?;
-        Ok(response.content)
+        let reasoning = Reasoning::new(self.llm.clone(), self.safety.clone());
+        let (text, _) = reasoning.complete(request).await?;
+        Ok(text)
     }
 
     /// Write a summary to the workspace daily log.

src/agent/cost_guard.rs

@@ -4,7 +4,7 @@
 //! to prevent runaway agents from burning through API credits. Especially
 //! important for daemon/heartbeat modes where the agent acts autonomously.
 
-use std::collections::VecDeque;
+use std::collections::{HashMap, VecDeque};
 use std::sync::atomic::{AtomicBool, Ordering};
 use std::time::Instant;
 
@@ -53,6 +53,14 @@ impl std::fmt::Display for CostLimitExceeded {
     }
 }
 
+/// Per-model token usage counters.
+#[derive(Debug, Clone, Default)]
+pub struct ModelTokens {
+    pub input_tokens: u64,
+    pub output_tokens: u64,
+    pub cost: Decimal,
+}
+
 /// Tracks costs and action rates, enforcing configurable limits.
 ///
 /// Thread-safe; designed to be shared via `Arc<CostGuard>`.
@@ -67,6 +75,9 @@ pub struct CostGuard {
 
     /// Flag set when daily budget is exceeded to short-circuit checks.
     budget_exceeded: AtomicBool,
+
+    /// Per-model token usage since startup.
+    model_tokens: Mutex<HashMap<String, ModelTokens>>,
 }
 
 struct DailyCost {
@@ -85,6 +96,7 @@ impl CostGuard {
             }),
             action_window: Mutex::new(VecDeque::new()),
             budget_exceeded: AtomicBool::new(false),
+            model_tokens: Mutex::new(HashMap::new()),
         }
     }
 
@@ -192,6 +204,15 @@ impl CostGuard {
             window.push_back(Instant::now());
         }
 
+        // Track per-model token usage
+        {
+            let mut tokens = self.model_tokens.lock().await;
+            let entry = tokens.entry(model.to_string()).or_default();
+            entry.input_tokens += u64::from(input_tokens);
+            entry.output_tokens += u64::from(output_tokens);
+            entry.cost += cost;
+        }
+
         cost
     }
 
@@ -215,6 +236,11 @@ impl CostGuard {
         }
         window.len() as u64
     }
+
+    /// Per-model token usage since startup.
+    pub async fn model_usage(&self) -> HashMap<String, ModelTokens> {
+        self.model_tokens.lock().await.clone()
+    }
 }
 
 /// Convert a Decimal USD amount to whole cents (truncated).
@@ -336,4 +362,37 @@ mod tests {
         assert!(rate.to_string().contains("101 actions"));
         assert!(rate.to_string().contains("100 allowed"));
     }
+
+    #[tokio::test]
+    async fn test_model_usage_per_model_tracking() {
+        let guard = CostGuard::new(CostGuardConfig::default());
+
+        // Initially empty
+        assert!(guard.model_usage().await.is_empty());
+
+        // Record calls for two different models
+        guard.record_llm_call("gpt-4o", 1000, 500).await;
+        guard.record_llm_call("gpt-4o", 2000, 1000).await;
+        guard
+            .record_llm_call("claude-3-5-sonnet-20241022", 500, 200)
+            .await;
+
+        let usage = guard.model_usage().await;
+        assert_eq!(usage.len(), 2);
+
+        let gpt = usage.get("gpt-4o").expect("gpt-4o should be tracked");
+        assert_eq!(gpt.input_tokens, 3000);
+        assert_eq!(gpt.output_tokens, 1500);
+        assert!(gpt.cost > Decimal::ZERO);
+
+        let claude = usage
+            .get("claude-3-5-sonnet-20241022")
+            .expect("claude should be tracked");
+        assert_eq!(claude.input_tokens, 500);
+        assert_eq!(claude.output_tokens, 200);
+        assert!(claude.cost > Decimal::ZERO);
+
+        // Costs should differ since models have different pricing
+        assert_ne!(gpt.cost, claude.cost);
+    }
 }

src/agent/heartbeat.rs

@@ -29,8 +29,10 @@ use std::time::Duration;
 use tokio::sync::mpsc;
 
 use crate::channels::OutgoingResponse;
-use crate::llm::{ChatMessage, CompletionRequest, FinishReason, LlmProvider};
+use crate::llm::{ChatMessage, CompletionRequest, LlmProvider, Reasoning};
+use crate::safety::SafetyLayer;
 use crate::workspace::Workspace;
+use crate::workspace::hygiene::HygieneConfig;
 
 /// Configuration for the heartbeat runner.
 #[derive(Debug, Clone)]
@@ -96,8 +98,10 @@ pub enum HeartbeatResult {
 /// Heartbeat runner for proactive periodic execution.
 pub struct HeartbeatRunner {
     config: HeartbeatConfig,
+    hygiene_config: HygieneConfig,
     workspace: Arc<Workspace>,
     llm: Arc<dyn LlmProvider>,
+    safety: Arc<SafetyLayer>,
     response_tx: Option<mpsc::Sender<OutgoingResponse>>,
     consecutive_failures: u32,
 }
@@ -106,13 +110,17 @@ impl HeartbeatRunner {
     /// Create a new heartbeat runner.
     pub fn new(
         config: HeartbeatConfig,
+        hygiene_config: HygieneConfig,
         workspace: Arc<Workspace>,
         llm: Arc<dyn LlmProvider>,
+        safety: Arc<SafetyLayer>,
     ) -> Self {
         Self {
             config,
+            hygiene_config,
             workspace,
             llm,
+            safety,
             response_tx: None,
             consecutive_failures: 0,
         }
@@ -145,6 +153,22 @@ impl HeartbeatRunner {
         loop {
             interval.tick().await;
 
+            // Run memory hygiene in the background so it never delays the
+            // heartbeat checklist. Failures are logged inside run_if_due.
+            let hygiene_workspace = Arc::clone(&self.workspace);
+            let hygiene_config = self.hygiene_config.clone();
+            tokio::spawn(async move {
+                let report =
+                    crate::workspace::hygiene::run_if_due(&hygiene_workspace, &hygiene_config)
+                        .await;
+                if report.had_work() {
+                    tracing::info!(
+                        daily_logs_deleted = report.daily_logs_deleted,
+                        "heartbeat: memory hygiene deleted stale documents"
+                    );
+                }
+            });
+
             match self.check_heartbeat().await {
                 HeartbeatResult::Ok => {
                     tracing::debug!("Heartbeat OK");
@@ -238,25 +262,18 @@ impl HeartbeatRunner {
             .with_max_tokens(max_tokens)
             .with_temperature(0.3);
 
-        let response = match self.llm.complete(request).await {
+        let reasoning = Reasoning::new(self.llm.clone(), self.safety.clone());
+        let (content, _usage) = match reasoning.complete(request).await {
             Ok(r) => r,
             Err(e) => return HeartbeatResult::Failed(format!("LLM call failed: {}", e)),
         };
 
-        let content = response.content.trim();
+        let content = content.trim();
 
         // Guard against empty content. Reasoning models (e.g. GLM-4.7) may
         // burn all output tokens on chain-of-thought and return content: null.
         if content.is_empty() {
-            return if response.finish_reason == FinishReason::Length {
-                HeartbeatResult::Failed(
-                    "LLM response was truncated (finish_reason=length) with no content. \
-                     The model may have exhausted its token budget on reasoning."
-                        .to_string(),
-                )
-            } else {
-                HeartbeatResult::Failed("LLM returned empty content.".to_string())
-            };
+            return HeartbeatResult::Failed("LLM returned empty content.".to_string());
         }
 
         // Check if nothing needs attention
@@ -332,11 +349,13 @@ fn strip_html_comments(content: &str) -> String {
 /// Returns a handle that can be used to stop the runner.
 pub fn spawn_heartbeat(
     config: HeartbeatConfig,
+    hygiene_config: HygieneConfig,
     workspace: Arc<Workspace>,
     llm: Arc<dyn LlmProvider>,
+    safety: Arc<SafetyLayer>,
     response_tx: Option<mpsc::Sender<OutgoingResponse>>,
 ) -> tokio::task::JoinHandle<()> {
-    let mut runner = HeartbeatRunner::new(config, workspace, llm);
+    let mut runner = HeartbeatRunner::new(config, hygiene_config, workspace, llm, safety);
     if let Some(tx) = response_tx {
         runner = runner.with_response_channel(tx);
     }

src/agent/mod.rs

@@ -44,6 +44,6 @@ pub use self_repair::{BrokenTool, RepairResult, RepairTask, SelfRepair, StuckJob
 pub use session::{PendingApproval, PendingAuth, Session, Thread, ThreadState, Turn, TurnState};
 pub use session_manager::SessionManager;
 pub use submission::{Submission, SubmissionParser, SubmissionResult};
-pub use task::{Task, TaskContext, TaskHandler, TaskOutput, TaskStatus};
+pub use task::{Task, TaskContext, TaskHandler, TaskOutput};
 pub use undo::{Checkpoint, UndoManager};
 pub use worker::{Worker, WorkerDeps};

src/agent/routine.rs

@@ -26,6 +26,8 @@ use chrono::{DateTime, Utc};
 use serde::{Deserialize, Serialize};
 use uuid::Uuid;
 
+use crate::error::RoutineError;
+
 /// A routine is a named, persistent, user-owned task with a trigger and an action.
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct Routine {
@@ -86,13 +88,16 @@ impl Trigger {
     }
 
     /// Parse a trigger from its DB representation.
-    pub fn from_db(trigger_type: &str, config: serde_json::Value) -> Result<Self, String> {
+    pub fn from_db(trigger_type: &str, config: serde_json::Value) -> Result<Self, RoutineError> {
         match trigger_type {
             "cron" => {
                 let schedule = config
                     .get("schedule")
                     .and_then(|v| v.as_str())
-                    .ok_or("cron trigger missing 'schedule'")?
+                    .ok_or_else(|| RoutineError::MissingField {
+                        context: "cron trigger".into(),
+                        field: "schedule".into(),
+                    })?
                     .to_string();
                 Ok(Trigger::Cron { schedule })
             }
@@ -100,7 +105,10 @@ impl Trigger {
                 let pattern = config
                     .get("pattern")
                     .and_then(|v| v.as_str())
-                    .ok_or("event trigger missing 'pattern'")?
+                    .ok_or_else(|| RoutineError::MissingField {
+                        context: "event trigger".into(),
+                        field: "pattern".into(),
+                    })?
                     .to_string();
                 let channel = config
                     .get("channel")
@@ -120,7 +128,9 @@ impl Trigger {
                 Ok(Trigger::Webhook { path, secret })
             }
             "manual" => Ok(Trigger::Manual),
-            other => Err(format!("unknown trigger type: {other}")),
+            other => Err(RoutineError::UnknownTriggerType {
+                trigger_type: other.to_string(),
+            }),
         }
     }
 
@@ -186,13 +196,16 @@ impl RoutineAction {
     }
 
     /// Parse an action from its DB representation.
-    pub fn from_db(action_type: &str, config: serde_json::Value) -> Result<Self, String> {
+    pub fn from_db(action_type: &str, config: serde_json::Value) -> Result<Self, RoutineError> {
         match action_type {
             "lightweight" => {
                 let prompt = config
                     .get("prompt")
                     .and_then(|v| v.as_str())
-                    .ok_or("lightweight action missing 'prompt'")?
+                    .ok_or_else(|| RoutineError::MissingField {
+                        context: "lightweight action".into(),
+                        field: "prompt".into(),
+                    })?
                     .to_string();
                 let context_paths = config
                     .get("context_paths")
@@ -217,12 +230,18 @@ impl RoutineAction {
                 let title = config
                     .get("title")
                     .and_then(|v| v.as_str())
-                    .ok_or("full_job action missing 'title'")?
+                    .ok_or_else(|| RoutineError::MissingField {
+                        context: "full_job action".into(),
+                        field: "title".into(),
+                    })?
                     .to_string();
                 let description = config
                     .get("description")
                     .and_then(|v| v.as_str())
-                    .ok_or("full_job action missing 'description'")?
+                    .ok_or_else(|| RoutineError::MissingField {
+                        context: "full_job action".into(),
+                        field: "description".into(),
+                    })?
                     .to_string();
                 let max_iterations = config
                     .get("max_iterations")
@@ -235,7 +254,9 @@ impl RoutineAction {
                     max_iterations,
                 })
             }
-            other => Err(format!("unknown action type: {other}")),
+            other => Err(RoutineError::UnknownActionType {
+                action_type: other.to_string(),
+            }),
         }
     }
 
@@ -334,14 +355,16 @@ impl std::fmt::Display for RunStatus {
 }
 
 impl FromStr for RunStatus {
-    type Err = String;
+    type Err = RoutineError;
     fn from_str(s: &str) -> Result<Self, Self::Err> {
         match s {
             "running" => Ok(RunStatus::Running),
             "ok" => Ok(RunStatus::Ok),
             "attention" => Ok(RunStatus::Attention),
             "failed" => Ok(RunStatus::Failed),
-            other => Err(format!("unknown run status: {other}")),
+            other => Err(RoutineError::UnknownRunStatus {
+                status: other.to_string(),
+            }),
         }
     }
 }
@@ -370,9 +393,11 @@ pub fn content_hash(content: &str) -> u64 {
 }
 
 /// Parse a cron expression and compute the next fire time from now.
-pub fn next_cron_fire(schedule: &str) -> Result<Option<DateTime<Utc>>, String> {
+pub fn next_cron_fire(schedule: &str) -> Result<Option<DateTime<Utc>>, RoutineError> {
     let cron_schedule =
-        cron::Schedule::from_str(schedule).map_err(|e| format!("invalid cron: {e}"))?;
+        cron::Schedule::from_str(schedule).map_err(|e| RoutineError::InvalidCron {
+            reason: e.to_string(),
+        })?;
     Ok(cron_schedule.upcoming(Utc).next())
 }
 

src/agent/routine_engine.rs

@@ -25,6 +25,7 @@ use crate::agent::routine::{
 use crate::channels::{IncomingMessage, OutgoingResponse};
 use crate::config::RoutineConfig;
 use crate::db::Database;
+use crate::error::RoutineError;
 use crate::llm::{ChatMessage, CompletionRequest, FinishReason, LlmProvider};
 use crate::workspace::Workspace;
 
@@ -174,23 +175,26 @@ impl RoutineEngine {
     }
 
     /// Fire a routine manually (from tool call or CLI).
-    pub async fn fire_manual(&self, routine_id: Uuid) -> Result<Uuid, String> {
+    pub async fn fire_manual(&self, routine_id: Uuid) -> Result<Uuid, RoutineError> {
         let routine = self
             .store
             .get_routine(routine_id)
             .await
-            .map_err(|e| format!("DB error: {e}"))?
-            .ok_or_else(|| format!("routine {routine_id} not found"))?;
+            .map_err(|e| RoutineError::Database {
+                reason: e.to_string(),
+            })?
+            .ok_or(RoutineError::NotFound { id: routine_id })?;
 
         if !routine.enabled {
-            return Err(format!("routine '{}' is disabled", routine.name));
+            return Err(RoutineError::Disabled {
+                name: routine.name.clone(),
+            });
         }
 
         if !self.check_concurrent(&routine).await {
-            return Err(format!(
-                "routine '{}' already at max concurrent runs",
-                routine.name
-            ));
+            return Err(RoutineError::MaxConcurrent {
+                name: routine.name.clone(),
+            });
         }
 
         let run_id = Uuid::new_v4();
@@ -209,7 +213,9 @@ impl RoutineEngine {
         };
 
         if let Err(e) = self.store.create_routine_run(&run).await {
-            return Err(format!("failed to create run record: {e}"));
+            return Err(RoutineError::Database {
+                reason: format!("failed to create run record: {e}"),
+            });
         }
 
         // Execute inline for manual triggers (caller wants to wait)
@@ -313,13 +319,27 @@ async fn execute_routine(ctx: EngineContext, routine: Routine, run: RoutineRun)
             max_tokens,
         } => execute_lightweight(&ctx, &routine, prompt, context_paths, *max_tokens).await,
         RoutineAction::FullJob { description, .. } => {
-            // Full job mode: for now, execute as lightweight with the description
-            // as prompt. Full scheduler integration will come as a follow-up.
-            tracing::info!(
+            // Full job mode: scheduler integration not yet implemented.
+            // Execute as lightweight and prepend a warning to the summary.
+            tracing::warn!(
                 routine = %routine.name,
-                "FullJob mode executing as lightweight (scheduler integration pending)"
+                "FullJob mode not yet implemented; falling back to lightweight execution"
             );
-            execute_lightweight(&ctx, &routine, description, &[], ctx.max_lightweight_tokens).await
+            match execute_lightweight(&ctx, &routine, description, &[], ctx.max_lightweight_tokens)
+                .await
+            {
+                Ok((status, summary, tokens)) => {
+                    let warning = "[Note: FullJob mode is not yet implemented. This routine ran as \
+                         a single LLM call without tool access. Configure as 'lightweight' \
+                         or wait for full scheduler integration.]";
+                    let summary = match summary {
+                        Some(s) => Some(format!("{warning}\n\n{s}")),
+                        None => Some(warning.to_string()),
+                    };
+                    Ok((status, summary, tokens))
+                }
+                Err(e) => Err(e),
+            }
         }
     };
 
@@ -331,7 +351,7 @@ async fn execute_routine(ctx: EngineContext, routine: Routine, run: RoutineRun)
         Ok(execution) => execution,
         Err(e) => {
             tracing::error!(routine = %routine.name, "Execution failed: {}", e);
-            (RunStatus::Failed, Some(e), None)
+            (RunStatus::Failed, Some(e.to_string()), None)
         }
     };
 
@@ -384,6 +404,20 @@ async fn execute_routine(ctx: EngineContext, routine: Routine, run: RoutineRun)
     .await;
 }
 
+/// Sanitize a routine name for use in workspace paths.
+/// Only keeps alphanumeric, dash, and underscore characters; replaces everything else.
+fn sanitize_routine_name(name: &str) -> String {
+    name.chars()
+        .map(|c| {
+            if c.is_ascii_alphanumeric() || c == '-' || c == '_' {
+                c
+            } else {
+                '_'
+            }
+        })
+        .collect()
+}
+
 /// Execute a lightweight routine (single LLM call).
 async fn execute_lightweight(
     ctx: &EngineContext,
@@ -391,7 +425,7 @@ async fn execute_lightweight(
     prompt: &str,
     context_paths: &[String],
     max_tokens: u32,
-) -> Result<(RunStatus, Option<String>, Option<i32>), String> {
+) -> Result<(RunStatus, Option<String>, Option<i32>), RoutineError> {
     // Load context from workspace
     let mut context_parts = Vec::new();
     for path in context_paths {
@@ -408,8 +442,9 @@ async fn execute_lightweight(
         }
     }
 
-    // Load routine state from workspace
-    let state_path = format!("routines/{}/state.md", routine.name);
+    // Load routine state from workspace (name sanitized to prevent path traversal)
+    let safe_name = sanitize_routine_name(&routine.name);
+    let state_path = format!("routines/{safe_name}/state.md");
     let state_content = match ctx.workspace.read(&state_path).await {
         Ok(doc) => Some(doc.content),
         Err(_) => None,
@@ -469,7 +504,9 @@ async fn execute_lightweight(
         .llm
         .complete(request)
         .await
-        .map_err(|e| format!("LLM call failed: {e}"))?;
+        .map_err(|e| RoutineError::LlmFailed {
+            reason: e.to_string(),
+        })?;
 
     let content = response.content.trim();
     let tokens_used = Some((response.input_tokens + response.output_tokens) as i32);
@@ -477,13 +514,9 @@ async fn execute_lightweight(
     // Empty content guard (same as heartbeat)
     if content.is_empty() {
         return if response.finish_reason == FinishReason::Length {
-            Err(
-                "LLM response truncated (finish_reason=length) with no content. \
-                 Model may have exhausted token budget on reasoning."
-                    .to_string(),
-            )
+            Err(RoutineError::TruncatedResponse)
         } else {
-            Err("LLM returned empty content.".to_string())
+            Err(RoutineError::EmptyResponse)
         };
     }
 

src/agent/scheduler.rs

@@ -136,7 +136,9 @@ impl Scheduler {
             });
 
             // Start the worker
-            let _ = tx.send(WorkerMessage::Start).await;
+            if tx.send(WorkerMessage::Start).await.is_err() {
+                tracing::error!(job_id = %job_id, "Worker died before receiving Start message");
+            }
 
             // Insert while still holding the write lock
             jobs.insert(job_id, ScheduledJob { handle, tx });
@@ -355,7 +357,7 @@ impl Scheduler {
             .into());
         }
 
-        if tool.requires_approval() {
+        if tool.requires_approval(&params).is_required() {
             return Err(crate::error::ToolError::AuthRequired {
                 name: tool_name.to_string(),
             }
@@ -418,10 +420,16 @@ impl Scheduler {
             // Update job state
             self.context_manager
                 .update_context(job_id, |ctx| {
-                    let _ = ctx.transition_to(
+                    if let Err(e) = ctx.transition_to(
                         JobState::Cancelled,
                         Some("Stopped by scheduler".to_string()),
-                    );
+                    ) {
+                        tracing::warn!(
+                            job_id = %job_id,
+                            error = %e,
+                            "Failed to transition job to Cancelled state"
+                        );
+                    }
                 })
                 .await?;
 

src/agent/self_repair.rs

@@ -66,12 +66,14 @@ pub trait SelfRepair: Send + Sync {
 /// Default self-repair implementation.
 pub struct DefaultSelfRepair {
     context_manager: Arc<ContextManager>,
-    #[allow(dead_code)] // Will be used for time-based stuck detection
+    // TODO: use for time-based stuck detection (currently only max_repair_attempts is checked)
+    #[allow(dead_code)]
     stuck_threshold: Duration,
     max_repair_attempts: u32,
     store: Option<Arc<dyn Database>>,
     builder: Option<Arc<dyn SoftwareBuilder>>,
-    #[allow(dead_code)] // Will be used for tool hot-reload after repair
+    // TODO: use for tool hot-reload after repair
+    #[allow(dead_code)]
     tools: Option<Arc<ToolRegistry>>,
 }
 
@@ -93,15 +95,15 @@ impl DefaultSelfRepair {
     }
 
     /// Add a Store for tool failure tracking.
-    #[allow(dead_code)] // Public API for configuring repair with persistence
-    pub fn with_store(mut self, store: Arc<dyn Database>) -> Self {
+    #[allow(dead_code)] // TODO: wire up in main.rs when persistence is needed
+    pub(crate) fn with_store(mut self, store: Arc<dyn Database>) -> Self {
         self.store = Some(store);
         self
     }
 
     /// Add a Builder and ToolRegistry for automatic tool repair.
-    #[allow(dead_code)] // Public API for enabling automatic tool repair
-    pub fn with_builder(
+    #[allow(dead_code)] // TODO: wire up in main.rs when auto-repair is needed
+    pub(crate) fn with_builder(
         mut self,
         builder: Arc<dyn SoftwareBuilder>,
         tools: Arc<ToolRegistry>,

src/agent/session.rs

@@ -70,10 +70,9 @@ impl Session {
     pub fn create_thread(&mut self) -> &mut Thread {
         let thread = Thread::new(self.id);
         let thread_id = thread.id;
-        self.threads.insert(thread_id, thread);
         self.active_thread = Some(thread_id);
         self.last_active_at = Utc::now();
-        self.threads.get_mut(&thread_id).expect("just inserted")
+        self.threads.entry(thread_id).or_insert(thread)
     }
 
     /// Get the active thread.
@@ -88,10 +87,19 @@ impl Session {
 
     /// Get or create the active thread.
     pub fn get_or_create_thread(&mut self) -> &mut Thread {
-        if self.active_thread.is_none() {
-            self.create_thread();
+        match self.active_thread {
+            None => self.create_thread(),
+            Some(id) => {
+                if self.threads.contains_key(&id) {
+                    // Safe: contains_key confirmed the entry exists.
+                    self.threads.get_mut(&id).unwrap()
+                } else {
+                    // Stale active_thread ID: create a new thread, which
+                    // updates self.active_thread to the new thread's ID.
+                    self.create_thread()
+                }
+            }
         }
-        self.active_thread_mut().expect("just created")
     }
 
     /// Switch to a different thread.
@@ -177,10 +185,6 @@ pub struct Thread {
     /// Pending auth token request (thread is in auth mode).
     #[serde(default)]
     pub pending_auth: Option<PendingAuth>,
-    /// Last NEAR AI response ID for response chaining. Persisted to DB
-    /// metadata so we can resume chaining across restarts.
-    #[serde(default)]
-    pub last_response_id: Option<String>,
 }
 
 impl Thread {
@@ -197,7 +201,6 @@ impl Thread {
             metadata: serde_json::Value::Null,
             pending_approval: None,
             pending_auth: None,
-            last_response_id: None,
         }
     }
 
@@ -214,7 +217,6 @@ impl Thread {
             metadata: serde_json::Value::Null,
             pending_approval: None,
             pending_auth: None,
-            last_response_id: None,
         }
     }
 
@@ -240,7 +242,8 @@ impl Thread {
         self.turns.push(turn);
         self.state = ThreadState::Processing;
         self.updated_at = Utc::now();
-        self.turns.last_mut().expect("just pushed")
+        // turn_number was len() before push, so it's a valid index after push
+        &mut self.turns[turn_number]
     }
 
     /// Complete the current turn with a response.
@@ -353,8 +356,10 @@ impl Thread {
                 if let Some(next) = iter.peek()
                     && next.role == crate::llm::Role::Assistant
                 {
-                    let response = iter.next().expect("peeked");
-                    turn.complete(&response.content);
+                    // iter.next() is guaranteed Some after a successful peek()
+                    if let Some(response) = iter.next() {
+                        turn.complete(&response.content);
+                    }
                 }
 
                 self.turns.push(turn);
@@ -852,7 +857,6 @@ mod tests {
 
         thread.start_turn("hello");
         thread.complete_turn("world");
-        thread.last_response_id = Some("resp_abc123".to_string());
 
         let json = serde_json::to_string(&thread).unwrap();
         let restored: Thread = serde_json::from_str(&json).unwrap();
@@ -862,7 +866,6 @@ mod tests {
         assert_eq!(restored.turns.len(), 1);
         assert_eq!(restored.turns[0].user_input, "hello");
         assert_eq!(restored.turns[0].response, Some("world".to_string()));
-        assert_eq!(restored.last_response_id, Some("resp_abc123".to_string()));
     }
 
     #[test]

src/agent/session_manager.rs

@@ -13,6 +13,9 @@ use crate::agent::session::Session;
 use crate::agent::undo::UndoManager;
 use crate::hooks::HookRegistry;
 
+/// Warn when session count exceeds this threshold.
+const SESSION_COUNT_WARNING_THRESHOLD: usize = 1000;
+
 /// Key for mapping external thread IDs to internal ones.
 #[derive(Clone, Hash, Eq, PartialEq)]
 struct ThreadKey {
@@ -68,6 +71,14 @@ impl SessionManager {
         let session = Arc::new(Mutex::new(new_session));
         sessions.insert(user_id.to_string(), Arc::clone(&session));
 
+        if sessions.len() >= SESSION_COUNT_WARNING_THRESHOLD && sessions.len() % 100 == 0 {
+            tracing::warn!(
+                "High session count: {} active sessions. \
+                 Pruning runs every 10 minutes; consider reducing session_idle_timeout.",
+                sessions.len()
+            );
+        }
+
         // Fire OnSessionStart hook (fire-and-forget)
         if let Some(ref hooks) = self.hooks {
             let hooks = hooks.clone();

src/agent/submission.rs

@@ -234,6 +234,7 @@ impl Submission {
     }
 
     /// Create an approval submission.
+    #[cfg(test)]
     pub fn approval(request_id: Uuid, approved: bool) -> Self {
         Self::ExecApproval {
             request_id,
@@ -243,6 +244,7 @@ impl Submission {
     }
 
     /// Create an "always approve" submission.
+    #[cfg(test)]
     pub fn always_approve(request_id: Uuid) -> Self {
         Self::ExecApproval {
             request_id,
@@ -252,26 +254,31 @@ impl Submission {
     }
 
     /// Create an interrupt submission.
+    #[cfg(test)]
     pub fn interrupt() -> Self {
         Self::Interrupt
     }
 
     /// Create a compact submission.
+    #[cfg(test)]
     pub fn compact() -> Self {
         Self::Compact
     }
 
     /// Create an undo submission.
+    #[cfg(test)]
     pub fn undo() -> Self {
         Self::Undo
     }
 
     /// Create a redo submission.
+    #[cfg(test)]
     pub fn redo() -> Self {
         Self::Redo
     }
 
     /// Check if this submission starts a new turn.
+    #[cfg(test)]
     pub fn starts_turn(&self) -> bool {
         matches!(self, Self::UserInput { .. })
     }
@@ -340,6 +347,7 @@ impl SubmissionResult {
     }
 
     /// Create an OK result.
+    #[cfg(test)]
     pub fn ok() -> Self {
         Self::Ok { message: None }
     }

src/agent/task.rs

@@ -29,6 +29,7 @@ impl TaskOutput {
     }
 
     /// Create a text result.
+    #[cfg(test)]
     pub fn text(text: impl Into<String>, duration: Duration) -> Self {
         Self {
             result: serde_json::Value::String(text.into()),
@@ -37,6 +38,7 @@ impl TaskOutput {
     }
 
     /// Create an empty success result.
+    #[cfg(test)]
     pub fn empty(duration: Duration) -> Self {
         Self {
             result: serde_json::Value::Null,
@@ -130,6 +132,7 @@ impl Task {
     }
 
     /// Create a new Job task with a specific ID.
+    #[cfg(test)]
     pub fn job_with_id(id: Uuid, title: impl Into<String>, description: impl Into<String>) -> Self {
         Self::Job {
             id,
@@ -152,6 +155,7 @@ impl Task {
     }
 
     /// Create a new Background task.
+    #[cfg(test)]
     pub fn background(handler: std::sync::Arc<dyn TaskHandler>) -> Self {
         Self::Background {
             id: Uuid::new_v4(),
@@ -160,6 +164,7 @@ impl Task {
     }
 
     /// Create a new Background task with a specific ID.
+    #[cfg(test)]
     pub fn background_with_id(id: Uuid, handler: std::sync::Arc<dyn TaskHandler>) -> Self {
         Self::Background { id, handler }
     }
@@ -174,6 +179,7 @@ impl Task {
     }
 
     /// Get the parent ID for sub-tasks.
+    #[cfg(test)]
     pub fn parent_id(&self) -> Option<Uuid> {
         match self {
             Self::Job { .. } => None,
@@ -225,6 +231,7 @@ impl fmt::Debug for Task {
 }
 
 /// Status of a scheduled task.
+#[cfg(test)]
 #[derive(Debug, Clone, Copy, PartialEq, Eq)]
 pub enum TaskStatus {
     /// Task is queued waiting for execution.

src/agent/thread_ops.rs

@@ -6,11 +6,14 @@
 use std::sync::Arc;
 
 use tokio::sync::Mutex;
+use tokio::task::JoinSet;
 use uuid::Uuid;
 
 use crate::agent::Agent;
 use crate::agent::compaction::ContextCompactor;
-use crate::agent::dispatcher::{AgenticLoopResult, detect_auth_awaiting, parse_auth_result};
+use crate::agent::dispatcher::{
+    AgenticLoopResult, check_auth_required, execute_chat_tool_standalone, parse_auth_result,
+};
 use crate::agent::session::{PendingApproval, Session, ThreadState};
 use crate::agent::submission::SubmissionResult;
 use crate::channels::{IncomingMessage, StatusUpdate};
@@ -84,20 +87,6 @@ impl Agent {
             thread.restore_from_messages(chat_messages);
         }
 
-        // Restore response chain from conversation metadata
-        if let Some(store) = self.store()
-            && let Ok(Some(metadata)) = store.get_conversation_metadata(thread_uuid).await
-            && let Some(rid) = metadata
-                .get("last_response_id")
-                .and_then(|v| v.as_str())
-                .map(String::from)
-        {
-            thread.last_response_id = Some(rid.clone());
-            self.llm()
-                .seed_response_chain(&thread_uuid.to_string(), rid);
-            tracing::debug!("Restored response chain for thread {}", thread_uuid);
-        }
-
         // Insert into session and register with session manager
         {
             let mut sess = session.lock().await;
@@ -225,7 +214,7 @@ impl Agent {
                     )
                     .await;
 
-                let compactor = ContextCompactor::new(self.llm().clone());
+                let compactor = ContextCompactor::new(self.llm().clone(), self.safety().clone());
                 if let Err(e) = compactor
                     .compact(thread, strategy, self.workspace().map(|w| w.as_ref()))
                     .await
@@ -275,7 +264,7 @@ impl Agent {
 
         // Run the agentic tool execution loop
         let result = self
-            .run_agentic_loop(message, session.clone(), thread_id, turn_messages, false)
+            .run_agentic_loop(message, session.clone(), thread_id, turn_messages)
             .await;
 
         // Re-acquire lock and check if interrupted
@@ -322,7 +311,6 @@ impl Agent {
                 };
 
                 thread.complete_turn(&response);
-                self.persist_response_chain(thread);
                 let _ = self
                     .channels
                     .send_status(
@@ -332,8 +320,10 @@ impl Agent {
                     )
                     .await;
 
-                // Fire-and-forget: persist turn to DB
-                self.persist_turn(thread_id, &message.user_id, content, Some(&response));
+                // Persist turn to DB before returning so the write
+                // completes even if the process shuts down right after.
+                self.persist_turn(thread_id, &message.user_id, content, Some(&response))
+                    .await;
 
                 Ok(SubmissionResult::response(response))
             }
@@ -363,15 +353,16 @@ impl Agent {
                 thread.fail_turn(e.to_string());
 
                 // Persist the user message even on failure
-                self.persist_turn(thread_id, &message.user_id, content, None);
+                self.persist_turn(thread_id, &message.user_id, content, None)
+                    .await;
 
                 Ok(SubmissionResult::error(e.to_string()))
             }
         }
     }
 
-    /// Fire-and-forget: persist a turn (user message + optional assistant response) to the DB.
-    pub(super) fn persist_turn(
+    /// Persist a turn (user message + optional assistant response) to the DB.
+    pub(super) async fn persist_turn(
         &self,
         thread_id: Uuid,
         user_id: &str,
@@ -383,70 +374,29 @@ impl Agent {
             None => return,
         };
 
-        let user_id = user_id.to_string();
-        let user_input = user_input.to_string();
-        let response = response.map(String::from);
+        if let Err(e) = store
+            .ensure_conversation(thread_id, "gateway", user_id, None)
+            .await
+        {
+            tracing::warn!("Failed to ensure conversation {}: {}", thread_id, e);
+            return;
+        }
 
-        tokio::spawn(async move {
-            if let Err(e) = store
-                .ensure_conversation(thread_id, "gateway", &user_id, None)
+        if let Err(e) = store
+            .add_conversation_message(thread_id, "user", user_input)
+            .await
+        {
+            tracing::warn!("Failed to persist user message: {}", e);
+            return;
+        }
+
+        if let Some(resp) = response
+            && let Err(e) = store
+                .add_conversation_message(thread_id, "assistant", resp)
                 .await
-            {
-                tracing::warn!("Failed to ensure conversation {}: {}", thread_id, e);
-                return;
-            }
-
-            if let Err(e) = store
-                .add_conversation_message(thread_id, "user", &user_input)
-                .await
-            {
-                tracing::warn!("Failed to persist user message: {}", e);
-                return;
-            }
-
-            if let Some(ref resp) = response
-                && let Err(e) = store
-                    .add_conversation_message(thread_id, "assistant", resp)
-                    .await
-            {
-                tracing::warn!("Failed to persist assistant message: {}", e);
-            }
-        });
-    }
-
-    /// Sync the provider's response chain ID to the thread and DB metadata.
-    ///
-    /// Call after a successful agentic loop to persist the latest
-    /// `previous_response_id` so chaining survives restarts.
-    pub(super) fn persist_response_chain(&self, thread: &mut crate::agent::session::Thread) {
-        let tid = thread.id.to_string();
-        let response_id = match self.llm().get_response_chain_id(&tid) {
-            Some(rid) => rid,
-            None => return,
-        };
-
-        // Update in-memory thread
-        thread.last_response_id = Some(response_id.clone());
-
-        // Fire-and-forget DB write
-        let store = match self.store() {
-            Some(s) => Arc::clone(s),
-            None => return,
-        };
-        let thread_id = thread.id;
-        tokio::spawn(async move {
-            let val = serde_json::json!(response_id);
-            if let Err(e) = store
-                .update_conversation_metadata_field(thread_id, "last_response_id", &val)
-                .await
-            {
-                tracing::warn!(
-                    "Failed to persist response chain for thread {}: {}",
-                    thread_id,
-                    e
-                );
-            }
-        });
+        {
+            tracing::warn!("Failed to persist assistant message: {}", e);
+        }
     }
 
     pub(super) async fn process_undo(
@@ -559,7 +509,7 @@ impl Agent {
                 crate::agent::context_monitor::CompactionStrategy::Summarize { keep_recent: 5 },
             );
 
-        let compactor = ContextCompactor::new(self.llm().clone());
+        let compactor = ContextCompactor::new(self.llm().clone(), self.safety().clone());
         match compactor
             .compact(thread, strategy, self.workspace().map(|w| w.as_ref()))
             .await
@@ -608,8 +558,8 @@ impl Agent {
         approved: bool,
         always: bool,
     ) -> Result<SubmissionResult, Error> {
-        // Get thread state and pending approval
-        let (_thread_state, pending) = {
+        // Get pending approval for this thread
+        let pending = {
             let mut sess = session.lock().await;
             let thread = sess
                 .threads
@@ -620,8 +570,7 @@ impl Agent {
                 return Ok(SubmissionResult::error("No pending approval request."));
             }
 
-            let pending = thread.take_pending_approval();
-            (thread.state, pending)
+            thread.take_pending_approval()
         };
 
         let pending = match pending {
@@ -734,29 +683,17 @@ impl Agent {
             // If tool_auth returned awaiting_token, enter auth mode and
             // return instructions directly (skip agentic loop continuation).
             if let Some((ext_name, instructions)) =
-                detect_auth_awaiting(&pending.tool_name, &tool_result)
+                check_auth_required(&pending.tool_name, &tool_result)
             {
-                let auth_data = parse_auth_result(&tool_result);
-                {
-                    let mut sess = session.lock().await;
-                    if let Some(thread) = sess.threads.get_mut(&thread_id) {
-                        thread.enter_auth_mode(ext_name.clone());
-                        thread.complete_turn(&instructions);
-                    }
-                }
-                let _ = self
-                    .channels
-                    .send_status(
-                        &message.channel,
-                        StatusUpdate::AuthRequired {
-                            extension_name: ext_name,
-                            instructions: Some(instructions.clone()),
-                            auth_url: auth_data.auth_url,
-                            setup_url: auth_data.setup_url,
-                        },
-                        &message.metadata,
-                    )
-                    .await;
+                self.handle_auth_intercept(
+                    &session,
+                    thread_id,
+                    message,
+                    &tool_result,
+                    ext_name,
+                    instructions.clone(),
+                )
+                .await;
                 return Ok(SubmissionResult::response(instructions));
             }
 
@@ -798,89 +735,165 @@ impl Agent {
                     .await;
             }
 
-            let mut deferred_queue = std::collections::VecDeque::from(deferred_tool_calls);
-            while let Some(tc) = deferred_queue.pop_front() {
-                // Re-check approval for each deferred tool call
-                if let Some(tool) = self.tools().get(&tc.name).await
-                    && tool.requires_approval()
-                {
-                    let is_auto_approved = {
-                        let sess = session.lock().await;
-                        let mut approved = sess.is_tool_auto_approved(&tc.name);
-                        if approved && tool.requires_approval_for(&tc.arguments) {
-                            approved = false;
+            // === Phase 1: Preflight (sequential) ===
+            // Walk deferred tools checking approval. Collect runnable
+            // tools; stop at the first that needs approval.
+            let mut runnable: Vec<crate::llm::ToolCall> = Vec::new();
+            let mut approval_needed: Option<(
+                usize,
+                crate::llm::ToolCall,
+                Arc<dyn crate::tools::Tool>,
+            )> = None;
+
+            for (idx, tc) in deferred_tool_calls.iter().enumerate() {
+                if let Some(tool) = self.tools().get(&tc.name).await {
+                    use crate::tools::ApprovalRequirement;
+                    let needs_approval = match tool.requires_approval(&tc.arguments) {
+                        ApprovalRequirement::Never => false,
+                        ApprovalRequirement::UnlessAutoApproved => {
+                            let sess = session.lock().await;
+                            !sess.is_tool_auto_approved(&tc.name)
                         }
-                        approved
+                        ApprovalRequirement::Always => true,
                     };
 
-                    if !is_auto_approved {
-                        let new_pending = PendingApproval {
-                            request_id: Uuid::new_v4(),
-                            tool_name: tc.name.clone(),
-                            parameters: tc.arguments.clone(),
-                            description: tool.description().to_string(),
-                            tool_call_id: tc.id.clone(),
-                            context_messages: context_messages.clone(),
-                            deferred_tool_calls: deferred_queue.iter().cloned().collect(),
-                        };
-
-                        let request_id = new_pending.request_id;
-                        let tool_name = new_pending.tool_name.clone();
-                        let description = new_pending.description.clone();
-                        let parameters = new_pending.parameters.clone();
-
-                        {
-                            let mut sess = session.lock().await;
-                            if let Some(thread) = sess.threads.get_mut(&thread_id) {
-                                thread.await_approval(new_pending);
-                            }
-                        }
-
-                        let _ = self
-                            .channels
-                            .send_status(
-                                &message.channel,
-                                StatusUpdate::Status("Awaiting approval".into()),
-                                &message.metadata,
-                            )
-                            .await;
-
-                        return Ok(SubmissionResult::NeedApproval {
-                            request_id,
-                            tool_name,
-                            description,
-                            parameters,
-                        });
+                    if needs_approval {
+                        approval_needed = Some((idx, tc.clone(), tool));
+                        break; // remaining tools stay deferred
                     }
                 }
 
-                let _ = self
-                    .channels
-                    .send_status(
-                        &message.channel,
-                        StatusUpdate::ToolStarted {
-                            name: tc.name.clone(),
-                        },
-                        &message.metadata,
-                    )
-                    .await;
+                runnable.push(tc.clone());
+            }
 
-                let deferred_result = self
-                    .execute_chat_tool(&tc.name, &tc.arguments, &job_ctx)
-                    .await;
+            // === Phase 2: Parallel execution ===
+            let exec_results: Vec<(crate::llm::ToolCall, Result<String, Error>)> = if runnable.len()
+                <= 1
+            {
+                // Single tool (or none): execute inline
+                let mut results = Vec::new();
+                for tc in &runnable {
+                    let _ = self
+                        .channels
+                        .send_status(
+                            &message.channel,
+                            StatusUpdate::ToolStarted {
+                                name: tc.name.clone(),
+                            },
+                            &message.metadata,
+                        )
+                        .await;
 
-                let _ = self
-                    .channels
-                    .send_status(
-                        &message.channel,
-                        StatusUpdate::ToolCompleted {
-                            name: tc.name.clone(),
-                            success: deferred_result.is_ok(),
-                        },
-                        &message.metadata,
-                    )
-                    .await;
+                    let result = self
+                        .execute_chat_tool(&tc.name, &tc.arguments, &job_ctx)
+                        .await;
 
+                    let _ = self
+                        .channels
+                        .send_status(
+                            &message.channel,
+                            StatusUpdate::ToolCompleted {
+                                name: tc.name.clone(),
+                                success: result.is_ok(),
+                            },
+                            &message.metadata,
+                        )
+                        .await;
+
+                    results.push((tc.clone(), result));
+                }
+                results
+            } else {
+                // Multiple tools: execute in parallel via JoinSet
+                let mut join_set = JoinSet::new();
+                let runnable_count = runnable.len();
+
+                for (spawn_idx, tc) in runnable.iter().enumerate() {
+                    let tools = self.tools().clone();
+                    let safety = self.safety().clone();
+                    let channels = self.channels.clone();
+                    let job_ctx = job_ctx.clone();
+                    let tc = tc.clone();
+                    let channel = message.channel.clone();
+                    let metadata = message.metadata.clone();
+
+                    join_set.spawn(async move {
+                        let _ = channels
+                            .send_status(
+                                &channel,
+                                StatusUpdate::ToolStarted {
+                                    name: tc.name.clone(),
+                                },
+                                &metadata,
+                            )
+                            .await;
+
+                        let result = execute_chat_tool_standalone(
+                            &tools,
+                            &safety,
+                            &tc.name,
+                            &tc.arguments,
+                            &job_ctx,
+                        )
+                        .await;
+
+                        let _ = channels
+                            .send_status(
+                                &channel,
+                                StatusUpdate::ToolCompleted {
+                                    name: tc.name.clone(),
+                                    success: result.is_ok(),
+                                },
+                                &metadata,
+                            )
+                            .await;
+
+                        (spawn_idx, tc, result)
+                    });
+                }
+
+                // Collect and reorder by original index
+                let mut ordered: Vec<Option<(crate::llm::ToolCall, Result<String, Error>)>> =
+                    (0..runnable_count).map(|_| None).collect();
+                while let Some(join_result) = join_set.join_next().await {
+                    match join_result {
+                        Ok((idx, tc, result)) => {
+                            ordered[idx] = Some((tc, result));
+                        }
+                        Err(e) => {
+                            if e.is_panic() {
+                                tracing::error!("Deferred tool execution task panicked: {}", e);
+                            } else {
+                                tracing::error!("Deferred tool execution task cancelled: {}", e);
+                            }
+                        }
+                    }
+                }
+
+                // Fill panicked slots with error results
+                ordered
+                    .into_iter()
+                    .enumerate()
+                    .map(|(i, opt)| {
+                        opt.unwrap_or_else(|| {
+                            let tc = runnable[i].clone();
+                            let err: Error = crate::error::ToolError::ExecutionFailed {
+                                name: tc.name.clone(),
+                                reason: "Task failed during execution".to_string(),
+                            }
+                            .into();
+                            (tc, Err(err))
+                        })
+                    })
+                    .collect()
+            };
+
+            // === Phase 3: Post-flight (sequential, in original order) ===
+            // Process all results before any conditional return so every
+            // tool result is recorded in the session audit trail.
+            let mut deferred_auth: Option<String> = None;
+
+            for (tc, deferred_result) in exec_results {
                 if let Ok(ref output) = deferred_result
                     && !output.is_empty()
                 {
@@ -910,32 +923,21 @@ impl Agent {
                     }
                 }
 
-                // Auth detection for deferred tools
-                if let Some((ext_name, instructions)) =
-                    detect_auth_awaiting(&tc.name, &deferred_result)
+                // Auth detection — defer return until all results are recorded
+                if deferred_auth.is_none()
+                    && let Some((ext_name, instructions)) =
+                        check_auth_required(&tc.name, &deferred_result)
                 {
-                    let auth_data = parse_auth_result(&deferred_result);
-                    {
-                        let mut sess = session.lock().await;
-                        if let Some(thread) = sess.threads.get_mut(&thread_id) {
-                            thread.enter_auth_mode(ext_name.clone());
-                            thread.complete_turn(&instructions);
-                        }
-                    }
-                    let _ = self
-                        .channels
-                        .send_status(
-                            &message.channel,
-                            StatusUpdate::AuthRequired {
-                                extension_name: ext_name,
-                                instructions: Some(instructions.clone()),
-                                auth_url: auth_data.auth_url,
-                                setup_url: auth_data.setup_url,
-                            },
-                            &message.metadata,
-                        )
-                        .await;
-                    return Ok(SubmissionResult::response(instructions));
+                    self.handle_auth_intercept(
+                        &session,
+                        thread_id,
+                        message,
+                        &deferred_result,
+                        ext_name,
+                        instructions.clone(),
+                    )
+                    .await;
+                    deferred_auth = Some(instructions);
                 }
 
                 let deferred_content = match deferred_result {
@@ -953,9 +955,55 @@ impl Agent {
                 context_messages.push(ChatMessage::tool_result(&tc.id, &tc.name, deferred_content));
             }
 
+            // Return auth response after all results are recorded
+            if let Some(instructions) = deferred_auth {
+                return Ok(SubmissionResult::response(instructions));
+            }
+
+            // Handle approval if a tool needed it
+            if let Some((approval_idx, tc, tool)) = approval_needed {
+                let new_pending = PendingApproval {
+                    request_id: Uuid::new_v4(),
+                    tool_name: tc.name.clone(),
+                    parameters: tc.arguments.clone(),
+                    description: tool.description().to_string(),
+                    tool_call_id: tc.id.clone(),
+                    context_messages: context_messages.clone(),
+                    deferred_tool_calls: deferred_tool_calls[approval_idx + 1..].to_vec(),
+                };
+
+                let request_id = new_pending.request_id;
+                let tool_name = new_pending.tool_name.clone();
+                let description = new_pending.description.clone();
+                let parameters = new_pending.parameters.clone();
+
+                {
+                    let mut sess = session.lock().await;
+                    if let Some(thread) = sess.threads.get_mut(&thread_id) {
+                        thread.await_approval(new_pending);
+                    }
+                }
+
+                let _ = self
+                    .channels
+                    .send_status(
+                        &message.channel,
+                        StatusUpdate::Status("Awaiting approval".into()),
+                        &message.metadata,
+                    )
+                    .await;
+
+                return Ok(SubmissionResult::NeedApproval {
+                    request_id,
+                    tool_name,
+                    description,
+                    parameters,
+                });
+            }
+
             // Continue the agentic loop (a tool was already executed this turn)
             let result = self
-                .run_agentic_loop(message, session.clone(), thread_id, context_messages, true)
+                .run_agentic_loop(message, session.clone(), thread_id, context_messages)
                 .await;
 
             // Handle the result
@@ -967,8 +1015,12 @@ impl Agent {
 
             match result {
                 Ok(AgenticLoopResult::Response(response)) => {
+                    let user_input = thread.last_turn().map(|t| t.user_input.clone());
                     thread.complete_turn(&response);
-                    self.persist_response_chain(thread);
+                    if let Some(input) = user_input {
+                        self.persist_turn(thread_id, &message.user_id, &input, Some(&response))
+                            .await;
+                    }
                     let _ = self
                         .channels
                         .send_status(
@@ -1003,16 +1055,32 @@ impl Agent {
                     })
                 }
                 Err(e) => {
+                    let user_input = thread.last_turn().map(|t| t.user_input.clone());
                     thread.fail_turn(e.to_string());
+                    if let Some(input) = user_input {
+                        self.persist_turn(thread_id, &message.user_id, &input, None)
+                            .await;
+                    }
                     Ok(SubmissionResult::error(e.to_string()))
                 }
             }
         } else {
-            // Rejected - clear approval and return to idle
+            // Rejected - complete the turn with a rejection message and persist
+            let rejection = format!(
+                "Tool '{}' was rejected. The agent will not execute this tool.\n\n\
+                 You can continue the conversation or try a different approach.",
+                pending.tool_name
+            );
             {
                 let mut sess = session.lock().await;
                 if let Some(thread) = sess.threads.get_mut(&thread_id) {
+                    let user_input = thread.last_turn().map(|t| t.user_input.clone());
                     thread.clear_pending_approval();
+                    thread.complete_turn(&rejection);
+                    if let Some(input) = user_input {
+                        self.persist_turn(thread_id, &message.user_id, &input, Some(&rejection))
+                            .await;
+                    }
                 }
             }
 
@@ -1025,14 +1093,52 @@ impl Agent {
                 )
                 .await;
 
-            Ok(SubmissionResult::response(format!(
-                "Tool '{}' was rejected. The agent will not execute this tool.\n\n\
-                 You can continue the conversation or try a different approach.",
-                pending.tool_name
-            )))
+            Ok(SubmissionResult::response(rejection))
         }
     }
 
+    /// Handle an auth-required result from a tool execution.
+    ///
+    /// Enters auth mode on the thread, completes + persists the turn,
+    /// and sends the AuthRequired status to the channel.
+    /// Returns the instructions string for the caller to wrap in a response.
+    async fn handle_auth_intercept(
+        &self,
+        session: &Arc<Mutex<Session>>,
+        thread_id: Uuid,
+        message: &IncomingMessage,
+        tool_result: &Result<String, Error>,
+        ext_name: String,
+        instructions: String,
+    ) {
+        let auth_data = parse_auth_result(tool_result);
+        {
+            let mut sess = session.lock().await;
+            if let Some(thread) = sess.threads.get_mut(&thread_id) {
+                let user_input = thread.last_turn().map(|t| t.user_input.clone());
+                thread.enter_auth_mode(ext_name.clone());
+                thread.complete_turn(&instructions);
+                if let Some(input) = user_input {
+                    self.persist_turn(thread_id, &message.user_id, &input, Some(&instructions))
+                        .await;
+                }
+            }
+        }
+        let _ = self
+            .channels
+            .send_status(
+                &message.channel,
+                StatusUpdate::AuthRequired {
+                    extension_name: ext_name,
+                    instructions: Some(instructions.clone()),
+                    auth_url: auth_data.auth_url,
+                    setup_url: auth_data.setup_url,
+                },
+                &message.metadata,
+            )
+            .await;
+    }
+
     /// Handle an auth token submitted while the thread is in auth mode.
     ///
     /// The token goes directly to the extension manager's credential store,

src/agent/undo.rs

@@ -67,6 +67,7 @@ impl UndoManager {
     }
 
     /// Create with a custom checkpoint limit.
+    #[cfg(test)]
     pub fn with_max_checkpoints(mut self, max: usize) -> Self {
         self.max_checkpoints = max;
         self
@@ -126,6 +127,7 @@ impl UndoManager {
     }
 
     /// Pop the last checkpoint from the undo stack.
+    #[cfg(test)]
     pub fn pop_undo(&mut self) -> Option<Checkpoint> {
         self.undo_stack.pop_back()
     }
@@ -178,6 +180,7 @@ impl UndoManager {
     }
 
     /// Get a checkpoint by ID.
+    #[cfg(test)]
     pub fn get_checkpoint(&self, id: Uuid) -> Option<&Checkpoint> {
         self.undo_stack
             .iter()
@@ -186,6 +189,7 @@ impl UndoManager {
     }
 
     /// List all available checkpoints (for UI display).
+    #[cfg(test)]
     pub fn list_checkpoints(&self) -> Vec<&Checkpoint> {
         self.undo_stack.iter().collect()
     }

src/agent/worker.rs

@@ -3,8 +3,8 @@
 use std::sync::Arc;
 use std::time::Duration;
 
-use futures::future::join_all;
 use tokio::sync::mpsc;
+use tokio::task::JoinSet;
 use uuid::Uuid;
 
 use crate::agent::scheduler::WorkerMessage;
@@ -18,6 +18,7 @@ use crate::llm::{
 };
 use crate::safety::SafetyLayer;
 use crate::tools::ToolRegistry;
+use crate::tools::rate_limiter::RateLimitResult;
 
 /// Shared dependencies for worker execution.
 ///
@@ -292,19 +293,21 @@ Report when the job is complete or if you encounter issues you cannot resolve."#
                                 tool_calls.clone(),
                             ));
 
-                        for tc in tool_calls {
-                            let result = self.execute_tool(&tc.name, &tc.arguments).await;
-
-                            // Create synthetic selection for process_tool_result
-                            let selection = ToolSelection {
+                        // Convert ToolCalls to ToolSelections and execute in parallel
+                        let selections: Vec<ToolSelection> = tool_calls
+                            .iter()
+                            .map(|tc| ToolSelection {
                                 tool_name: tc.name.clone(),
                                 parameters: tc.arguments.clone(),
                                 reasoning: String::new(),
                                 alternatives: vec![],
                                 tool_call_id: tc.id.clone(),
-                            };
+                            })
+                            .collect();
 
-                            self.process_tool_result(reason_ctx, &selection, result)
+                        let results = self.execute_tools_parallel(&selections).await;
+                        for (selection, result) in selections.iter().zip(results) {
+                            self.process_tool_result(reason_ctx, selection, result.result)
                                 .await?;
                         }
                     }
@@ -347,24 +350,71 @@ Report when the job is complete or if you encounter issues you cannot resolve."#
         }
     }
 
-    /// Execute multiple tools in parallel.
+    /// Execute multiple tools in parallel using a JoinSet.
+    ///
+    /// Each task is tagged with its original index so results are returned
+    /// in the same order as `selections`, regardless of completion order.
     async fn execute_tools_parallel(&self, selections: &[ToolSelection]) -> Vec<ToolExecResult> {
-        let futures: Vec<_> = selections
-            .iter()
-            .map(|selection| {
-                let tool_name = selection.tool_name.clone();
-                let params = selection.parameters.clone();
-                let deps = self.deps.clone();
-                let job_id = self.job_id;
+        let count = selections.len();
 
-                async move {
-                    let result = Self::execute_tool_inner(&deps, job_id, &tool_name, &params).await;
-                    ToolExecResult { result }
+        // Short-circuit for single tool: execute directly without JoinSet overhead
+        if count <= 1 {
+            let mut results = Vec::with_capacity(count);
+            for selection in selections {
+                let result = Self::execute_tool_inner(
+                    &self.deps,
+                    self.job_id,
+                    &selection.tool_name,
+                    &selection.parameters,
+                )
+                .await;
+                results.push(ToolExecResult { result });
+            }
+            return results;
+        }
+
+        let mut join_set = JoinSet::new();
+
+        for (idx, selection) in selections.iter().enumerate() {
+            let deps = self.deps.clone();
+            let job_id = self.job_id;
+            let tool_name = selection.tool_name.clone();
+            let params = selection.parameters.clone();
+            join_set.spawn(async move {
+                let result = Self::execute_tool_inner(&deps, job_id, &tool_name, &params).await;
+                (idx, ToolExecResult { result })
+            });
+        }
+
+        // Collect and reorder by original index
+        let mut results: Vec<Option<ToolExecResult>> = (0..count).map(|_| None).collect();
+        while let Some(join_result) = join_set.join_next().await {
+            match join_result {
+                Ok((idx, exec_result)) => results[idx] = Some(exec_result),
+                Err(e) => {
+                    if e.is_panic() {
+                        tracing::error!("Tool execution task panicked: {}", e);
+                    } else {
+                        tracing::error!("Tool execution task cancelled: {}", e);
+                    }
                 }
-            })
-            .collect();
+            }
+        }
 
-        join_all(futures).await
+        // Fill any panicked slots with error results
+        results
+            .into_iter()
+            .enumerate()
+            .map(|(i, opt)| {
+                opt.unwrap_or_else(|| ToolExecResult {
+                    result: Err(crate::error::ToolError::ExecutionFailed {
+                        name: selections[i].tool_name.clone(),
+                        reason: "Task failed during execution".to_string(),
+                    }
+                    .into()),
+                })
+            })
+            .collect()
     }
 
     /// Inner tool execution logic that can be called from both single and parallel paths.
@@ -383,16 +433,31 @@ Report when the job is complete or if you encounter issues you cannot resolve."#
                 })?;
 
         // Tools requiring approval are blocked in autonomous jobs
-        if tool.requires_approval() {
+        if tool.requires_approval(params).is_required() {
             return Err(crate::error::ToolError::AuthRequired {
                 name: tool_name.to_string(),
             }
             .into());
         }
 
-        // Fetch job context early so we have the real user_id for hooks
+        // Fetch job context early so we have the real user_id for hooks and rate limiting
         let job_ctx = deps.context_manager.get_context(job_id).await?;
 
+        // Check per-tool rate limit before running hooks or executing (cheaper check first)
+        if let Some(config) = tool.rate_limit_config()
+            && let RateLimitResult::Limited { retry_after, .. } = deps
+                .tools
+                .rate_limiter()
+                .check_and_record(&job_ctx.user_id, tool_name, &config)
+                .await
+        {
+            return Err(crate::error::ToolError::RateLimited {
+                name: tool_name.to_string(),
+                retry_after: Some(retry_after),
+            }
+            .into());
+        }
+
         // Run BeforeToolCall hook
         let params = {
             use crate::hooks::{HookError, HookEvent, HookOutcome};
@@ -505,7 +570,8 @@ Report when the job is complete or if you encounter issues you cannot resolve."#
                 let output_str = serde_json::to_string_pretty(&output.result)
                     .ok()
                     .map(|s| deps.safety.sanitize_tool_output(tool_name, &s).content);
-                deps.context_manager
+                match deps
+                    .context_manager
                     .update_memory(job_id, |mem| {
                         let rec = mem.create_action(tool_name, params.clone()).succeed(
                             output_str.clone(),
@@ -516,30 +582,52 @@ Report when the job is complete or if you encounter issues you cannot resolve."#
                         rec
                     })
                     .await
-                    .ok()
+                {
+                    Ok(rec) => Some(rec),
+                    Err(e) => {
+                        tracing::warn!(job_id = %job_id, tool = tool_name, "Failed to record action in memory: {e}");
+                        None
+                    }
+                }
+            }
+            Ok(Err(e)) => {
+                match deps
+                    .context_manager
+                    .update_memory(job_id, |mem| {
+                        let rec = mem
+                            .create_action(tool_name, params.clone())
+                            .fail(e.to_string(), elapsed);
+                        mem.record_action(rec.clone());
+                        rec
+                    })
+                    .await
+                {
+                    Ok(rec) => Some(rec),
+                    Err(e) => {
+                        tracing::warn!(job_id = %job_id, tool = tool_name, "Failed to record action in memory: {e}");
+                        None
+                    }
+                }
+            }
+            Err(_) => {
+                match deps
+                    .context_manager
+                    .update_memory(job_id, |mem| {
+                        let rec = mem
+                            .create_action(tool_name, params.clone())
+                            .fail("Execution timeout", elapsed);
+                        mem.record_action(rec.clone());
+                        rec
+                    })
+                    .await
+                {
+                    Ok(rec) => Some(rec),
+                    Err(e) => {
+                        tracing::warn!(job_id = %job_id, tool = tool_name, "Failed to record action in memory: {e}");
+                        None
+                    }
+                }
             }
-            Ok(Err(e)) => deps
-                .context_manager
-                .update_memory(job_id, |mem| {
-                    let rec = mem
-                        .create_action(tool_name, params.clone())
-                        .fail(e.to_string(), elapsed);
-                    mem.record_action(rec.clone());
-                    rec
-                })
-                .await
-                .ok(),
-            Err(_) => deps
-                .context_manager
-                .update_memory(job_id, |mem| {
-                    let rec = mem
-                        .create_action(tool_name, params.clone())
-                        .fail("Execution timeout", elapsed);
-                    mem.record_action(rec.clone());
-                    rec
-                })
-                .await
-                .ok(),
         };
 
         // Persist action to database (fire-and-forget)
@@ -800,6 +888,102 @@ mod tests {
     use crate::llm::ToolSelection;
     use crate::util::llm_signals_completion;
 
+    use super::*;
+    use crate::config::SafetyConfig;
+    use crate::context::JobContext;
+    use crate::llm::{
+        CompletionRequest, CompletionResponse, LlmProvider, ToolCompletionRequest,
+        ToolCompletionResponse,
+    };
+    use crate::safety::SafetyLayer;
+    use crate::tools::{Tool, ToolError, ToolOutput};
+
+    /// A test tool that sleeps for a configurable duration before returning.
+    struct SlowTool {
+        tool_name: String,
+        delay: Duration,
+    }
+
+    #[async_trait::async_trait]
+    impl Tool for SlowTool {
+        fn name(&self) -> &str {
+            &self.tool_name
+        }
+        fn description(&self) -> &str {
+            "Test tool with configurable delay"
+        }
+        fn parameters_schema(&self) -> serde_json::Value {
+            serde_json::json!({"type": "object", "properties": {}})
+        }
+        async fn execute(
+            &self,
+            _params: serde_json::Value,
+            _ctx: &JobContext,
+        ) -> Result<ToolOutput, ToolError> {
+            let start = std::time::Instant::now();
+            tokio::time::sleep(self.delay).await;
+            Ok(ToolOutput::text(
+                format!("done_{}", self.tool_name),
+                start.elapsed(),
+            ))
+        }
+        fn requires_sanitization(&self) -> bool {
+            false
+        }
+    }
+
+    /// Stub LLM provider (never called in these tests).
+    struct StubLlm;
+
+    #[async_trait::async_trait]
+    impl LlmProvider for StubLlm {
+        fn model_name(&self) -> &str {
+            "stub"
+        }
+        fn cost_per_token(&self) -> (rust_decimal::Decimal, rust_decimal::Decimal) {
+            (rust_decimal::Decimal::ZERO, rust_decimal::Decimal::ZERO)
+        }
+        async fn complete(
+            &self,
+            _req: CompletionRequest,
+        ) -> Result<CompletionResponse, crate::error::LlmError> {
+            unimplemented!("stub")
+        }
+        async fn complete_with_tools(
+            &self,
+            _req: ToolCompletionRequest,
+        ) -> Result<ToolCompletionResponse, crate::error::LlmError> {
+            unimplemented!("stub")
+        }
+    }
+
+    /// Build a Worker wired to a ToolRegistry containing the given tools.
+    async fn make_worker(tools: Vec<Arc<dyn Tool>>) -> Worker {
+        let registry = ToolRegistry::new();
+        for t in tools {
+            registry.register(t).await;
+        }
+
+        let cm = Arc::new(crate::context::ContextManager::new(5));
+        let job_id = cm.create_job("test", "test job").await.unwrap();
+
+        let deps = WorkerDeps {
+            context_manager: cm,
+            llm: Arc::new(StubLlm),
+            safety: Arc::new(SafetyLayer::new(&SafetyConfig {
+                max_output_length: 100_000,
+                injection_check_enabled: false,
+            })),
+            tools: Arc::new(registry),
+            store: None,
+            hooks: Arc::new(crate::hooks::HookRegistry::new()),
+            timeout: Duration::from_secs(30),
+            use_planning: false,
+        };
+
+        Worker::new(job_id, deps)
+    }
+
     #[test]
     fn test_tool_selection_preserves_call_id() {
         let selection = ToolSelection {
@@ -876,4 +1060,119 @@ mod tests {
             "The tool returned: TASK_COMPLETE signal"
         ));
     }
+
+    #[tokio::test]
+    async fn test_parallel_speedup() {
+        // 3 tools each sleeping 200ms should finish in roughly 200ms (parallel),
+        // not ~600ms (sequential).
+        let tools: Vec<Arc<dyn Tool>> = (0..3)
+            .map(|i| {
+                Arc::new(SlowTool {
+                    tool_name: format!("slow_{}", i),
+                    delay: Duration::from_millis(200),
+                }) as Arc<dyn Tool>
+            })
+            .collect();
+
+        let worker = make_worker(tools).await;
+
+        let selections: Vec<ToolSelection> = (0..3)
+            .map(|i| ToolSelection {
+                tool_name: format!("slow_{}", i),
+                parameters: serde_json::json!({}),
+                reasoning: String::new(),
+                alternatives: vec![],
+                tool_call_id: format!("call_{}", i),
+            })
+            .collect();
+
+        let start = std::time::Instant::now();
+        let results = worker.execute_tools_parallel(&selections).await;
+        let elapsed = start.elapsed();
+
+        assert_eq!(results.len(), 3);
+        for r in &results {
+            assert!(r.result.is_ok(), "Tool should succeed");
+        }
+        // Parallel should complete well under the sequential 600ms threshold.
+        assert!(
+            elapsed < Duration::from_millis(500),
+            "Parallel execution took {:?}, expected < 500ms",
+            elapsed
+        );
+    }
+
+    #[tokio::test]
+    async fn test_result_ordering_preserved() {
+        // Tools with different delays finish in different order.
+        // Results must be returned in the original request order.
+        let tools: Vec<Arc<dyn Tool>> = vec![
+            Arc::new(SlowTool {
+                tool_name: "tool_a".into(),
+                delay: Duration::from_millis(300),
+            }),
+            Arc::new(SlowTool {
+                tool_name: "tool_b".into(),
+                delay: Duration::from_millis(100),
+            }),
+            Arc::new(SlowTool {
+                tool_name: "tool_c".into(),
+                delay: Duration::from_millis(200),
+            }),
+        ];
+
+        let worker = make_worker(tools).await;
+
+        let selections = vec![
+            ToolSelection {
+                tool_name: "tool_a".into(),
+                parameters: serde_json::json!({}),
+                reasoning: String::new(),
+                alternatives: vec![],
+                tool_call_id: "call_a".into(),
+            },
+            ToolSelection {
+                tool_name: "tool_b".into(),
+                parameters: serde_json::json!({}),
+                reasoning: String::new(),
+                alternatives: vec![],
+                tool_call_id: "call_b".into(),
+            },
+            ToolSelection {
+                tool_name: "tool_c".into(),
+                parameters: serde_json::json!({}),
+                reasoning: String::new(),
+                alternatives: vec![],
+                tool_call_id: "call_c".into(),
+            },
+        ];
+
+        let results = worker.execute_tools_parallel(&selections).await;
+
+        // Results must be in same order as selections, not completion order.
+        assert!(results[0].result.as_ref().unwrap().contains("done_tool_a"));
+        assert!(results[1].result.as_ref().unwrap().contains("done_tool_b"));
+        assert!(results[2].result.as_ref().unwrap().contains("done_tool_c"));
+    }
+
+    #[tokio::test]
+    async fn test_missing_tool_produces_error_not_panic() {
+        // If a tool doesn't exist, the result slot should contain an error.
+        let worker = make_worker(vec![]).await;
+
+        let selections = vec![ToolSelection {
+            tool_name: "nonexistent_tool".into(),
+            parameters: serde_json::json!({}),
+            reasoning: String::new(),
+            alternatives: vec![],
+            tool_call_id: "call_x".into(),
+        }];
+
+        let results = worker.execute_tools_parallel(&selections).await;
+        assert_eq!(results.len(), 1);
+        assert!(
+            results[0].result.is_err(),
+            "Missing tool should produce an error, not a panic"
+        );
+    }
 }

src/app.rs

@@ -200,9 +200,13 @@ impl AppBuilder {
 
         self.session.attach_store(db.clone(), "default").await;
 
-        if let Err(e) = db.cleanup_stale_sandbox_jobs().await {
-            tracing::warn!("Failed to cleanup stale sandbox jobs: {}", e);
-        }
+        // Fire-and-forget housekeeping — no need to block startup.
+        let db_cleanup = db.clone();
+        tokio::spawn(async move {
+            if let Err(e) = db_cleanup.cleanup_stale_sandbox_jobs().await {
+                tracing::warn!("Failed to cleanup stale sandbox jobs: {}", e);
+            }
+        });
 
         self.db = Some(db);
         Ok(())
@@ -285,94 +289,14 @@ impl AppBuilder {
 
     /// Phase 3: Initialize LLM provider chain.
     ///
-    /// Creates the primary provider, then wraps with failover, circuit
-    /// breaker, and response cache as configured.
+    /// Delegates to `build_provider_chain` which applies all decorators
+    /// (retry, smart routing, failover, circuit breaker, response cache).
     #[allow(clippy::type_complexity)]
     pub fn init_llm(
         &self,
     ) -> Result<(Arc<dyn LlmProvider>, Option<Arc<dyn LlmProvider>>), anyhow::Error> {
-        use crate::llm::{
-            CachedProvider, CircuitBreakerConfig, CircuitBreakerProvider, CooldownConfig,
-            FailoverProvider, ResponseCacheConfig, create_cheap_llm_provider, create_llm_provider,
-            create_llm_provider_with_config,
-        };
-
-        let llm = create_llm_provider(&self.config.llm, self.session.clone())?;
-        tracing::info!("LLM provider initialized: {}", llm.model_name());
-
-        // Wrap in failover if a fallback model is configured
-        let llm: Arc<dyn LlmProvider> = if let Some(fallback_model) =
-            self.config.llm.nearai.fallback_model.as_ref()
-        {
-            if fallback_model == &self.config.llm.nearai.model {
-                tracing::warn!(
-                    "fallback_model is the same as primary model, failover may not be effective"
-                );
-            }
-            let mut fallback_config = self.config.llm.nearai.clone();
-            fallback_config.model = fallback_model.clone();
-            let fallback = create_llm_provider_with_config(&fallback_config, self.session.clone())?;
-            tracing::info!(
-                primary = %llm.model_name(),
-                fallback = %fallback.model_name(),
-                "LLM failover enabled"
-            );
-            let cooldown_config = CooldownConfig {
-                cooldown_duration: std::time::Duration::from_secs(
-                    self.config.llm.nearai.failover_cooldown_secs,
-                ),
-                failure_threshold: self.config.llm.nearai.failover_cooldown_threshold,
-            };
-            Arc::new(FailoverProvider::with_cooldown(
-                vec![llm, fallback],
-                cooldown_config,
-            )?)
-        } else {
-            llm
-        };
-
-        // Wrap in circuit breaker if configured
-        let llm: Arc<dyn LlmProvider> =
-            if let Some(threshold) = self.config.llm.nearai.circuit_breaker_threshold {
-                let cb_config = CircuitBreakerConfig {
-                    failure_threshold: threshold,
-                    recovery_timeout: std::time::Duration::from_secs(
-                        self.config.llm.nearai.circuit_breaker_recovery_secs,
-                    ),
-                    ..CircuitBreakerConfig::default()
-                };
-                tracing::info!(
-                    threshold,
-                    recovery_secs = self.config.llm.nearai.circuit_breaker_recovery_secs,
-                    "LLM circuit breaker enabled"
-                );
-                Arc::new(CircuitBreakerProvider::new(llm, cb_config))
-            } else {
-                llm
-            };
-
-        // Wrap in response cache if configured
-        let llm: Arc<dyn LlmProvider> = if self.config.llm.nearai.response_cache_enabled {
-            let rc_config = ResponseCacheConfig {
-                ttl: std::time::Duration::from_secs(self.config.llm.nearai.response_cache_ttl_secs),
-                max_entries: self.config.llm.nearai.response_cache_max_entries,
-            };
-            tracing::info!(
-                ttl_secs = self.config.llm.nearai.response_cache_ttl_secs,
-                max_entries = self.config.llm.nearai.response_cache_max_entries,
-                "LLM response cache enabled"
-            );
-            Arc::new(CachedProvider::new(llm, rc_config))
-        } else {
-            llm
-        };
-
-        // Cheap LLM for lightweight tasks
-        let cheap_llm = create_cheap_llm_provider(&self.config.llm, self.session.clone())?;
-        if let Some(ref cheap) = cheap_llm {
-            tracing::info!("Cheap LLM provider initialized: {}", cheap.model_name());
-        }
-
+        let (llm, cheap_llm) =
+            crate::llm::build_provider_chain(&self.config.llm, self.session.clone())?;
         Ok((llm, cheap_llm))
     }
 
@@ -396,7 +320,6 @@ impl AppBuilder {
 
         let tools = Arc::new(ToolRegistry::new());
         tools.register_builtin_tools();
-        tracing::info!("Registered {} built-in tools", tools.count());
 
         // Create embeddings provider if configured
         let embeddings: Option<Arc<dyn EmbeddingProvider>> = if self.config.embeddings.enabled {
@@ -656,11 +579,44 @@ impl AppBuilder {
 
         tokio::join!(wasm_tools_future, mcp_servers_future);
 
-        // Create extension manager
-        let extension_manager = if let Some(ref secrets) = self.secrets_store {
+        // Load registry catalog entries for extension discovery
+        let catalog_entries = match crate::registry::RegistryCatalog::load_or_embedded() {
+            Ok(catalog) => {
+                let entries: Vec<_> = catalog
+                    .all()
+                    .iter()
+                    .map(|m| m.to_registry_entry())
+                    .collect();
+                tracing::info!(
+                    count = entries.len(),
+                    "Loaded registry catalog entries for extension discovery"
+                );
+                entries
+            }
+            Err(e) => {
+                tracing::warn!("Failed to load registry catalog: {}", e);
+                Vec::new()
+            }
+        };
+
+        // Create extension manager. Use ephemeral in-memory secrets if no
+        // persistent store is configured (listing/install/activate still work).
+        let ext_secrets: Arc<dyn crate::secrets::SecretsStore + Send + Sync> = if let Some(ref s) =
+            self.secrets_store
+        {
+            Arc::clone(s)
+        } else {
+            use crate::secrets::{InMemorySecretsStore, SecretsCrypto};
+            let ephemeral_key =
+                secrecy::SecretString::from(crate::secrets::keychain::generate_master_key_hex());
+            let crypto = Arc::new(SecretsCrypto::new(ephemeral_key).expect("ephemeral crypto"));
+            tracing::debug!("Using ephemeral in-memory secrets store for extension manager");
+            Arc::new(InMemorySecretsStore::new(crypto))
+        };
+        let extension_manager = {
             let manager = Arc::new(ExtensionManager::new(
                 Arc::clone(&mcp_session_manager),
-                Arc::clone(secrets),
+                ext_secrets,
                 Arc::clone(tools),
                 Some(Arc::clone(hooks)),
                 wasm_tool_runtime.clone(),
@@ -669,24 +625,19 @@ impl AppBuilder {
                 self.config.tunnel.public_url.clone(),
                 "default".to_string(),
                 self.db.clone(),
+                catalog_entries.clone(),
             ));
             tools.register_extension_tools(Arc::clone(&manager));
             tracing::info!("Extension manager initialized with in-chat discovery tools");
             Some(manager)
-        } else {
-            tracing::debug!(
-                "Extension manager not available (no secrets store). \
-                 Extension tools won't be registered."
-            );
-            None
         };
 
-        // Register dev tools if local tools are enabled
-        if self.config.agent.allow_local_tools {
+        // register_builder_tool() already calls register_dev_tools() internally,
+        // so only register them here when the builder didn't already do it.
+        let builder_registered_dev_tools = self.config.builder.enabled
+            && (self.config.agent.allow_local_tools || !self.config.sandbox.enabled);
+        if self.config.agent.allow_local_tools && !builder_registered_dev_tools {
             tools.register_dev_tools();
-            tracing::info!(
-                "Local tools enabled (allow_local_tools=true), dev tools registered directly"
-            );
         }
 
         Ok((mcp_session_manager, wasm_tool_runtime, extension_manager))
@@ -709,9 +660,6 @@ impl AppBuilder {
         // Seed workspace and backfill embeddings
         if let Some(ref ws) = workspace {
             match ws.seed_if_empty().await {
-                Ok(count) if count > 0 => {
-                    tracing::info!("Workspace seeded with {} core files", count);
-                }
                 Ok(_) => {}
                 Err(e) => {
                     tracing::warn!("Failed to seed workspace: {}", e);
@@ -719,15 +667,18 @@ impl AppBuilder {
             }
 
             if embeddings.is_some() {
-                match ws.backfill_embeddings().await {
-                    Ok(count) if count > 0 => {
-                        tracing::info!("Backfilled embeddings for {} chunks", count);
+                let ws_bg = Arc::clone(ws);
+                tokio::spawn(async move {
+                    match ws_bg.backfill_embeddings().await {
+                        Ok(count) if count > 0 => {
+                            tracing::info!("Backfilled embeddings for {} chunks", count);
+                        }
+                        Ok(_) => {}
+                        Err(e) => {
+                            tracing::warn!("Failed to backfill embeddings: {}", e);
+                        }
                     }
-                    Ok(_) => {}
-                    Err(e) => {
-                        tracing::warn!("Failed to backfill embeddings: {}", e);
-                    }
-                }
+                });
             }
         }
 

src/bootstrap.rs

@@ -103,7 +103,67 @@ pub fn save_bootstrap_env(vars: &[(&str, &str)]) -> std::io::Result<()> {
         let escaped = value.replace('\\', "\\\\").replace('"', "\\\"");
         content.push_str(&format!("{}=\"{}\"\n", key, escaped));
     }
-    std::fs::write(&path, content)
+    std::fs::write(&path, &content)?;
+    restrict_file_permissions(&path)?;
+    Ok(())
+}
+
+/// Update or add a single variable in `~/.ironclaw/.env`, preserving existing content.
+///
+/// Unlike `save_bootstrap_env` (which overwrites the entire file), this
+/// reads the current `.env`, replaces the line for `key` if it exists,
+/// or appends it otherwise. Use this when writing a single bootstrap var
+/// outside the wizard (which manages the full set via `save_bootstrap_env`).
+pub fn upsert_bootstrap_var(key: &str, value: &str) -> std::io::Result<()> {
+    let path = ironclaw_env_path();
+    if let Some(parent) = path.parent() {
+        std::fs::create_dir_all(parent)?;
+    }
+
+    let escaped = value.replace('\\', "\\\\").replace('"', "\\\"");
+    let new_line = format!("{}=\"{}\"", key, escaped);
+    let prefix = format!("{}=", key);
+
+    let existing = std::fs::read_to_string(&path).unwrap_or_default();
+
+    let mut found = false;
+    let mut result = String::new();
+    for line in existing.lines() {
+        if line.starts_with(&prefix) {
+            if !found {
+                result.push_str(&new_line);
+                result.push('\n');
+                found = true;
+            }
+            // Skip duplicate lines for this key
+            continue;
+        }
+        result.push_str(line);
+        result.push('\n');
+    }
+
+    if !found {
+        result.push_str(&new_line);
+        result.push('\n');
+    }
+
+    std::fs::write(&path, result)?;
+    restrict_file_permissions(&path)?;
+    Ok(())
+}
+
+/// Set restrictive file permissions (0o600) on Unix systems.
+///
+/// The `.env` file may contain database credentials and API keys,
+/// so it should only be readable by the owner.
+fn restrict_file_permissions(_path: &std::path::Path) -> std::io::Result<()> {
+    #[cfg(unix)]
+    {
+        use std::os::unix::fs::PermissionsExt;
+        let perms = std::fs::Permissions::from_mode(0o600);
+        std::fs::set_permissions(_path, perms)?;
+    }
+    Ok(())
 }
 
 /// Write `DATABASE_URL` to `~/.ironclaw/.env`.

src/channels/channel.rs

@@ -9,6 +9,32 @@ use uuid::Uuid;
 
 use crate::error::ChannelError;
 
+/// Kind of attachment carried on an incoming message.
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub enum AttachmentKind {
+    /// Audio content (voice notes, audio files).
+    Audio,
+    /// Image content (photos, screenshots).
+    Image,
+    /// Document content (PDFs, files).
+    Document,
+}
+
+/// Binary attachment on a message (e.g., voice note, photo).
+#[derive(Debug, Clone)]
+pub struct Attachment {
+    /// What kind of content this is.
+    pub kind: AttachmentKind,
+    /// MIME type (e.g., "audio/ogg", "image/jpeg").
+    pub mime_type: String,
+    /// Raw bytes of the attachment.
+    pub data: Vec<u8>,
+    /// Optional filename.
+    pub filename: Option<String>,
+    /// Duration in seconds (for audio/video).
+    pub duration_secs: Option<u32>,
+}
+
 /// A message received from an external channel.
 #[derive(Debug, Clone)]
 pub struct IncomingMessage {
@@ -28,6 +54,8 @@ pub struct IncomingMessage {
     pub received_at: DateTime<Utc>,
     /// Channel-specific metadata.
     pub metadata: serde_json::Value,
+    /// Binary attachments (voice notes, images, etc.).
+    pub attachments: Vec<Attachment>,
 }
 
 impl IncomingMessage {
@@ -46,6 +74,7 @@ impl IncomingMessage {
             thread_id: None,
             received_at: Utc::now(),
             metadata: serde_json::Value::Null,
+            attachments: Vec::new(),
         }
     }
 
@@ -66,6 +95,12 @@ impl IncomingMessage {
         self.user_name = Some(name.into());
         self
     }
+
+    /// Set attachments.
+    pub fn with_attachments(mut self, attachments: Vec<Attachment>) -> Self {
+        self.attachments = attachments;
+        self
+    }
 }
 
 /// Stream of incoming messages.

src/channels/mod.rs

@@ -35,7 +35,10 @@ pub mod wasm;
 pub mod web;
 mod webhook_server;
 
-pub use channel::{Channel, IncomingMessage, MessageStream, OutgoingResponse, StatusUpdate};
+pub use channel::{
+    Attachment, AttachmentKind, Channel, IncomingMessage, MessageStream, OutgoingResponse,
+    StatusUpdate,
+};
 pub use http::HttpChannel;
 pub use manager::ChannelManager;
 pub use repl::ReplChannel;

src/channels/repl.rs

@@ -15,6 +15,7 @@
 //! - `/compact` - Compact the context
 //! - `/new` - Start a new thread
 //! - `yes`/`no`/`always` - Respond to tool approval prompts
+//! - `Esc` - Interrupt current operation
 
 use std::borrow::Cow;
 use std::io::{self, Write};
@@ -28,7 +29,10 @@ use rustyline::error::ReadlineError;
 use rustyline::highlight::Highlighter;
 use rustyline::hint::Hinter;
 use rustyline::validate::Validator;
-use rustyline::{CompletionType, Editor, Helper};
+use rustyline::{
+    Cmd as ReadlineCmd, CompletionType, ConditionalEventHandler, Editor, Event, EventContext,
+    EventHandler, Helper, KeyCode, KeyEvent, Modifiers, RepeatCount,
+};
 use termimad::MadSkin;
 use tokio::sync::mpsc;
 use tokio_stream::wrappers::ReceiverStream;
@@ -121,6 +125,23 @@ impl Highlighter for ReplHelper {
 impl Validator for ReplHelper {}
 impl Helper for ReplHelper {}
 
+struct EscInterruptHandler {
+    triggered: Arc<AtomicBool>,
+}
+
+impl ConditionalEventHandler for EscInterruptHandler {
+    fn handle(
+        &self,
+        _evt: &Event,
+        _n: RepeatCount,
+        _positive: bool,
+        _ctx: &EventContext,
+    ) -> Option<ReadlineCmd> {
+        self.triggered.store(true, Ordering::Relaxed);
+        Some(ReadlineCmd::Interrupt)
+    }
+}
+
 /// Build a termimad skin with our color scheme.
 fn make_skin() -> MadSkin {
     let mut skin = MadSkin::default();
@@ -247,6 +268,7 @@ fn print_help() {
     println!("  {c}/compact{r}           {d}compact context window{r}");
     println!("  {c}/new{r}               {d}new conversation thread{r}");
     println!("  {c}/interrupt{r}         {d}stop current operation{r}");
+    println!("  {c}esc{r}                {d}stop current operation{r}");
     println!();
     println!("  {h}Approval responses{r}");
     println!("  {c}yes{r} ({c}y{r})            {d}approve tool execution{r}");
@@ -274,6 +296,7 @@ impl Channel for ReplChannel {
         let single_message = self.single_message.clone();
         let debug_mode = Arc::clone(&self.debug_mode);
         let suppress_banner = Arc::clone(&self.suppress_banner);
+        let esc_interrupt_triggered_for_thread = Arc::new(AtomicBool::new(false));
 
         std::thread::spawn(move || {
             // Single message mode: send it and return
@@ -301,6 +324,13 @@ impl Channel for ReplChannel {
 
             rl.set_helper(Some(ReplHelper));
 
+            rl.bind_sequence(
+                KeyEvent(KeyCode::Esc, Modifiers::NONE),
+                EventHandler::Conditional(Box::new(EscInterruptHandler {
+                    triggered: Arc::clone(&esc_interrupt_triggered_for_thread),
+                })),
+            );
+
             // Load history
             let hist_path = history_path();
             if let Some(parent) = hist_path.parent() {
@@ -360,9 +390,16 @@ impl Channel for ReplChannel {
                         }
                     }
                     Err(ReadlineError::Interrupted) => {
-                        // Ctrl+C: send /interrupt
-                        let msg = IncomingMessage::new("repl", "default", "/interrupt");
-                        if tx.blocking_send(msg).is_err() {
+                        if esc_interrupt_triggered_for_thread.swap(false, Ordering::Relaxed) {
+                            // Esc: interrupt current operation and keep REPL open.
+                            let msg = IncomingMessage::new("repl", "default", "/interrupt");
+                            if tx.blocking_send(msg).is_err() {
+                                break;
+                            }
+                        } else {
+                            // Ctrl+C (VINTR): request graceful shutdown.
+                            let msg = IncomingMessage::new("repl", "default", "/quit");
+                            let _ = tx.blocking_send(msg);
                             break;
                         }
                     }

src/channels/wasm/bundled.rs

@@ -20,6 +20,7 @@ const CARGO_MANIFEST_DIR: &str = env!("CARGO_MANIFEST_DIR");
 const KNOWN_CHANNELS: &[(&str, &str)] = &[
     ("telegram", "telegram_channel"),
     ("slack", "slack_channel"),
+    ("discord", "discord_channel"),
     ("whatsapp", "whatsapp_channel"),
 ];
 
@@ -42,6 +43,10 @@ fn channels_src_dir() -> PathBuf {
 
 /// Locate the build artifacts for a channel.
 ///
+/// Checks two layouts:
+/// 1. **Flat** (Docker/packaged): `<channels_src>/<name>/<name>.wasm`
+/// 2. **Build tree** (dev): `<channels_src>/<name>/target/wasm32-wasip2/release/<crate_name>.wasm`
+///
 /// Returns (wasm_path, capabilities_path) or an error if files are missing.
 fn locate_channel_artifacts(name: &str) -> Result<(PathBuf, PathBuf), String> {
     let (_, crate_name) = KNOWN_CHANNELS
@@ -52,31 +57,34 @@ fn locate_channel_artifacts(name: &str) -> Result<(PathBuf, PathBuf), String> {
     let src_dir = channels_src_dir();
     let channel_dir = src_dir.join(name);
 
-    let wasm_path = channel_dir
+    let caps_path = channel_dir.join(format!("{}.capabilities.json", name));
+
+    // Check flat layout first (Docker/packaged deployments)
+    let flat_wasm = channel_dir.join(format!("{}.wasm", name));
+    if flat_wasm.exists() && caps_path.exists() {
+        return Ok((flat_wasm, caps_path));
+    }
+
+    // Fall back to build tree layout (dev builds)
+    let build_wasm = channel_dir
         .join("target/wasm32-wasip2/release")
         .join(format!("{}.wasm", crate_name));
 
-    let caps_path = channel_dir.join(format!("{}.capabilities.json", name));
-
-    if !wasm_path.exists() {
-        return Err(format!(
-            "Channel '{}' WASM not found at {}. Build it first:\n  \
-             cd {} && cargo build --target wasm32-wasip2 --release",
-            name,
-            wasm_path.display(),
-            channel_dir.display()
-        ));
+    if build_wasm.exists() && caps_path.exists() {
+        return Ok((build_wasm, caps_path));
     }
 
-    if !caps_path.exists() {
-        return Err(format!(
-            "Channel '{}' capabilities not found at {}",
-            name,
-            caps_path.display()
-        ));
-    }
-
-    Ok((wasm_path, caps_path))
+    Err(format!(
+        "Channel '{}' WASM not found. Checked:\n  \
+         - {} (flat/packaged)\n  \
+         - {} (build tree)\n  \
+         Build it first:\n  \
+         cd {} && cargo build --target wasm32-wasip2 --release",
+        name,
+        flat_wasm.display(),
+        build_wasm.display(),
+        channel_dir.display()
+    ))
 }
 
 /// Install a channel from build artifacts into the channels directory.
@@ -130,10 +138,11 @@ mod tests {
     use super::*;
 
     #[test]
-    fn test_known_channels_includes_all_three() {
+    fn test_known_channels_includes_all_four() {
         let names = bundled_channel_names();
         assert!(names.contains(&"telegram"));
         assert!(names.contains(&"slack"));
+        assert!(names.contains(&"discord"));
         assert!(names.contains(&"whatsapp"));
     }
 

src/channels/wasm/host.rs

@@ -7,6 +7,7 @@
 
 use std::time::{SystemTime, UNIX_EPOCH};
 
+use crate::channels::channel::Attachment;
 use crate::channels::wasm::capabilities::{ChannelCapabilities, EmitRateLimitConfig};
 use crate::channels::wasm::error::WasmChannelError;
 use crate::tools::wasm::{HostState, LogLevel};
@@ -17,6 +18,9 @@ const MAX_EMITS_PER_EXECUTION: usize = 100;
 /// Maximum message content size (64 KB).
 const MAX_MESSAGE_CONTENT_SIZE: usize = 64 * 1024;
 
+/// Maximum size for a single attachment (10 MB).
+const MAX_ATTACHMENT_SIZE: usize = 10 * 1024 * 1024;
+
 /// A message emitted by a WASM channel to be sent to the agent.
 #[derive(Debug, Clone)]
 pub struct EmittedMessage {
@@ -37,6 +41,9 @@ pub struct EmittedMessage {
 
     /// Timestamp when the message was emitted.
     pub emitted_at_millis: u64,
+
+    /// Binary attachments (voice notes, images, etc.).
+    pub attachments: Vec<Attachment>,
 }
 
 impl EmittedMessage {
@@ -52,6 +59,7 @@ impl EmittedMessage {
                 .duration_since(UNIX_EPOCH)
                 .map(|d| d.as_millis() as u64)
                 .unwrap_or(0),
+            attachments: Vec::new(),
         }
     }
 
@@ -72,6 +80,12 @@ impl EmittedMessage {
         self.metadata_json = metadata_json.into();
         self
     }
+
+    /// Set attachments.
+    pub fn with_attachments(mut self, attachments: Vec<Attachment>) -> Self {
+        self.attachments = attachments;
+        self
+    }
 }
 
 /// A pending workspace write operation.
@@ -168,7 +182,7 @@ impl ChannelHostState {
     ///
     /// Messages are queued and delivered after callback execution completes.
     /// Rate limiting is enforced per-execution and globally.
-    pub fn emit_message(&mut self, msg: EmittedMessage) -> Result<(), WasmChannelError> {
+    pub fn emit_message(&mut self, mut msg: EmittedMessage) -> Result<(), WasmChannelError> {
         // Check per-execution limit
         if !self.emit_enabled {
             self.emits_dropped += 1;
@@ -186,6 +200,22 @@ impl ChannelHostState {
             return Ok(());
         }
 
+        // Validate attachment sizes — drop only oversized attachments, not the whole message
+        msg.attachments.retain(|attachment| {
+            if attachment.data.len() > MAX_ATTACHMENT_SIZE {
+                tracing::warn!(
+                    channel = %self.channel_name,
+                    size = attachment.data.len(),
+                    max = MAX_ATTACHMENT_SIZE,
+                    mime = %attachment.mime_type,
+                    "Attachment too large, dropping attachment (message still delivered)"
+                );
+                false
+            } else {
+                true
+            }
+        });
+
         // Validate message content size
         if msg.content.len() > MAX_MESSAGE_CONTENT_SIZE {
             tracing::warn!(
@@ -300,6 +330,51 @@ impl ChannelHostState {
     }
 }
 
+/// In-memory workspace store for WASM channels.
+///
+/// Persists workspace writes across callback invocations within a single
+/// channel lifetime. This allows WASM channels to maintain state (e.g.,
+/// Telegram polling offsets) between poll ticks without requiring a
+/// full database-backed workspace.
+///
+/// Uses `std::sync::RwLock` (not tokio) because WASM execution runs
+/// inside `spawn_blocking`.
+pub struct ChannelWorkspaceStore {
+    data: std::sync::RwLock<std::collections::HashMap<String, String>>,
+}
+
+impl ChannelWorkspaceStore {
+    /// Create a new empty workspace store.
+    pub fn new() -> Self {
+        Self {
+            data: std::sync::RwLock::new(std::collections::HashMap::new()),
+        }
+    }
+
+    /// Commit pending writes from a callback execution into the store.
+    pub fn commit_writes(&self, writes: &[PendingWorkspaceWrite]) {
+        if writes.is_empty() {
+            return;
+        }
+        if let Ok(mut data) = self.data.write() {
+            for write in writes {
+                tracing::debug!(
+                    path = %write.path,
+                    content_len = write.content.len(),
+                    "Committing workspace write to channel store"
+                );
+                data.insert(write.path.clone(), write.content.clone());
+            }
+        }
+    }
+}
+
+impl crate::tools::wasm::WorkspaceReader for ChannelWorkspaceStore {
+    fn read(&self, path: &str) -> Option<String> {
+        self.data.read().ok()?.get(path).cloned()
+    }
+}
+
 /// Rate limiter for channel message emission.
 ///
 /// Tracks emission rates across multiple executions.
@@ -497,4 +572,56 @@ mod tests {
 
         assert_eq!(state.channel_name(), "telegram");
     }
+
+    #[test]
+    fn test_channel_workspace_store_commit_and_read() {
+        use crate::channels::wasm::host::{ChannelWorkspaceStore, PendingWorkspaceWrite};
+        use crate::tools::wasm::WorkspaceReader;
+
+        let store = ChannelWorkspaceStore::new();
+
+        // Initially empty
+        assert!(store.read("channels/telegram/offset").is_none());
+
+        // Commit some writes
+        let writes = vec![
+            PendingWorkspaceWrite {
+                path: "channels/telegram/offset".to_string(),
+                content: "103".to_string(),
+            },
+            PendingWorkspaceWrite {
+                path: "channels/telegram/state.json".to_string(),
+                content: r#"{"ok":true}"#.to_string(),
+            },
+        ];
+        store.commit_writes(&writes);
+
+        // Should be readable
+        assert_eq!(
+            store.read("channels/telegram/offset"),
+            Some("103".to_string())
+        );
+        assert_eq!(
+            store.read("channels/telegram/state.json"),
+            Some(r#"{"ok":true}"#.to_string())
+        );
+
+        // Overwrite a value
+        let writes2 = vec![PendingWorkspaceWrite {
+            path: "channels/telegram/offset".to_string(),
+            content: "200".to_string(),
+        }];
+        store.commit_writes(&writes2);
+        assert_eq!(
+            store.read("channels/telegram/offset"),
+            Some("200".to_string())
+        );
+
+        // Empty writes are a no-op
+        store.commit_writes(&[]);
+        assert_eq!(
+            store.read("channels/telegram/offset"),
+            Some("200".to_string())
+        );
+    }
 }

src/channels/wasm/router.rs

@@ -488,7 +488,7 @@ mod tests {
         let prepared = Arc::new(PreparedChannelModule {
             name: name.to_string(),
             description: format!("Test channel: {}", name),
-            component_bytes: Vec::new(),
+            component: None,
             limits: ResourceLimits::default(),
         });
 

src/channels/wasm/runtime.rs

@@ -68,38 +68,51 @@ impl WasmChannelRuntimeConfig {
 }
 
 /// A compiled WASM channel component ready for instantiation.
-#[derive(Debug)]
+///
+/// Stores the pre-compiled `Component` directly so instantiation
+/// doesn't require recompilation.
 pub struct PreparedChannelModule {
     /// Channel name.
     pub name: String,
     /// Channel description.
     pub description: String,
-    /// Compiled component bytes (public for testing, otherwise use component_bytes()).
-    pub(crate) component_bytes: Vec<u8>,
+    /// Pre-compiled component (cheaply cloneable via internal Arc).
+    pub(crate) component: Option<wasmtime::component::Component>,
     /// Resource limits for this channel.
     pub limits: ResourceLimits,
 }
 
 impl PreparedChannelModule {
-    /// Get the compiled component bytes.
-    pub fn component_bytes(&self) -> &[u8] {
-        &self.component_bytes
+    /// Get the pre-compiled component for instantiation.
+    pub fn component(&self) -> Option<&wasmtime::component::Component> {
+        self.component.as_ref()
     }
 
     /// Create a PreparedChannelModule for testing purposes.
     ///
-    /// Creates a module with no actual WASM bytes, suitable for testing
+    /// Creates a module with no actual WASM component, suitable for testing
     /// channel infrastructure without requiring a real WASM component.
     pub fn for_testing(name: impl Into<String>, description: impl Into<String>) -> Self {
         Self {
             name: name.into(),
             description: description.into(),
-            component_bytes: Vec::new(),
+            component: None,
             limits: ResourceLimits::default(),
         }
     }
 }
 
+impl std::fmt::Debug for PreparedChannelModule {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.debug_struct("PreparedChannelModule")
+            .field("name", &self.name)
+            .field("description", &self.description)
+            .field("has_component", &self.component.is_some())
+            .field("limits", &self.limits)
+            .finish()
+    }
+}
+
 /// WASM channel runtime.
 ///
 /// Manages the Wasmtime engine and a cache of prepared channel modules.
@@ -137,6 +150,13 @@ impl WasmChannelRuntime {
         // Disable debug info in production
         wasmtime_config.debug_info(false);
 
+        // Enable persistent compilation cache. Wasmtime serializes compiled native
+        // code to disk (~/.cache/wasmtime by default), so subsequent startups
+        // deserialize instead of recompiling — typically 10-50x faster.
+        if let Err(e) = wasmtime_config.cache_config_load_default() {
+            tracing::warn!("Failed to enable wasmtime compilation cache: {}", e);
+        }
+
         let engine = Engine::new(&wasmtime_config).map_err(|e| {
             WasmChannelError::Config(format!("Failed to create Wasmtime engine: {}", e))
         })?;
@@ -183,13 +203,13 @@ impl WasmChannelRuntime {
         // Compile in blocking task (Wasmtime compilation is synchronous)
         let prepared = tokio::task::spawn_blocking(move || {
             // Validate and compile the component
-            let _component = wasmtime::component::Component::new(&engine, &wasm_bytes)
+            let component = wasmtime::component::Component::new(&engine, &wasm_bytes)
                 .map_err(|e| WasmChannelError::Compilation(e.to_string()))?;
 
             Ok::<_, WasmChannelError>(PreparedChannelModule {
                 name: name.clone(),
                 description: desc,
-                component_bytes: wasm_bytes,
+                component: Some(component),
                 limits: limits.unwrap_or(default_limits),
             })
         })

src/channels/wasm/wrapper.rs

@@ -37,12 +37,15 @@ use tokio::sync::{RwLock, mpsc, oneshot};
 use tokio_stream::wrappers::ReceiverStream;
 use uuid::Uuid;
 use wasmtime::Store;
-use wasmtime::component::{Component, Linker};
+use wasmtime::component::Linker;
 use wasmtime_wasi::{ResourceTable, WasiCtx, WasiCtxBuilder, WasiView};
 
+use crate::channels::channel::{Attachment, AttachmentKind};
 use crate::channels::wasm::capabilities::ChannelCapabilities;
 use crate::channels::wasm::error::WasmChannelError;
-use crate::channels::wasm::host::{ChannelEmitRateLimiter, ChannelHostState, EmittedMessage};
+use crate::channels::wasm::host::{
+    ChannelEmitRateLimiter, ChannelHostState, ChannelWorkspaceStore, EmittedMessage,
+};
 use crate::channels::wasm::router::RegisteredEndpoint;
 use crate::channels::wasm::runtime::{PreparedChannelModule, WasmChannelRuntime};
 use crate::channels::wasm::schema::ChannelConfig;
@@ -434,6 +437,7 @@ impl near::agent::channel_host::Host for ChannelStoreData {
             user_id = %msg.user_id,
             user_name = ?msg.user_name,
             content_len = msg.content.len(),
+            attachment_count = msg.attachments.len(),
             "WASM emit_message called"
         );
 
@@ -446,6 +450,31 @@ impl near::agent::channel_host::Host for ChannelStoreData {
         }
         emitted = emitted.with_metadata(msg.metadata_json);
 
+        // Convert WIT attachments to Rust types
+        if !msg.attachments.is_empty() {
+            let attachments = msg
+                .attachments
+                .into_iter()
+                .map(|a| {
+                    let kind = match a.kind {
+                        near::agent::channel_host::AttachmentKind::Audio => AttachmentKind::Audio,
+                        near::agent::channel_host::AttachmentKind::Image => AttachmentKind::Image,
+                        near::agent::channel_host::AttachmentKind::Document => {
+                            AttachmentKind::Document
+                        }
+                    };
+                    Attachment {
+                        kind,
+                        mime_type: a.mime_type,
+                        data: a.data,
+                        filename: a.filename,
+                        duration_secs: a.duration_secs,
+                    }
+                })
+                .collect();
+            emitted = emitted.with_attachments(attachments);
+        }
+
         match self.host_state.emit_message(emitted) {
             Ok(()) => {
                 tracing::info!("Message emitted to host state successfully");
@@ -547,6 +576,13 @@ pub struct WasmChannel {
 
     /// Pairing store for DM pairing (guest access control).
     pairing_store: Arc<PairingStore>,
+
+    /// In-memory workspace store persisting writes across callback invocations.
+    /// Ensures WASM channels can maintain state (e.g., polling offsets) between ticks.
+    workspace_store: Arc<ChannelWorkspaceStore>,
+
+    /// Optional transcription middleware for audio attachments.
+    transcription_middleware: Option<Arc<crate::transcription::TranscriptionMiddleware>>,
 }
 
 impl WasmChannel {
@@ -577,9 +613,19 @@ impl WasmChannel {
             credentials: Arc::new(RwLock::new(HashMap::new())),
             typing_task: RwLock::new(None),
             pairing_store,
+            workspace_store: Arc::new(ChannelWorkspaceStore::new()),
+            transcription_middleware: None,
         }
     }
 
+    /// Set the transcription middleware for audio attachment processing.
+    pub fn set_transcription_middleware(
+        &mut self,
+        middleware: Arc<crate::transcription::TranscriptionMiddleware>,
+    ) {
+        self.transcription_middleware = Some(middleware);
+    }
+
     /// Update the channel config before starting.
     ///
     /// Merges the provided values into the existing config JSON.
@@ -634,6 +680,26 @@ impl WasmChannel {
         self.endpoints.read().await.clone()
     }
 
+    /// Inject the workspace store as the reader into a capabilities clone.
+    ///
+    /// Ensures `workspace_read` capability is present with the store as its reader,
+    /// so WASM callbacks can read previously written workspace state.
+    fn inject_workspace_reader(
+        capabilities: &ChannelCapabilities,
+        store: &Arc<ChannelWorkspaceStore>,
+    ) -> ChannelCapabilities {
+        let mut caps = capabilities.clone();
+        let ws_cap = caps
+            .tool_capabilities
+            .workspace_read
+            .get_or_insert_with(|| crate::tools::wasm::WorkspaceCapability {
+                allowed_prefixes: Vec::new(),
+                reader: None,
+            });
+        ws_cap.reader = Some(Arc::clone(store) as Arc<dyn crate::tools::wasm::WorkspaceReader>);
+        caps
+    }
+
     /// Add channel host functions to the linker using generated bindings.
     ///
     /// Uses the wasmtime::component::bindgen! generated `add_to_linker` function
@@ -698,9 +764,13 @@ impl WasmChannel {
     ) -> Result<SandboxedChannel, WasmChannelError> {
         let engine = runtime.engine();
 
-        // Compile the component (uses cached bytes)
-        let component = Component::new(engine, prepared.component_bytes())
-            .map_err(|e| WasmChannelError::Compilation(e.to_string()))?;
+        // Use the pre-compiled component (no recompilation needed)
+        let component = prepared
+            .component()
+            .ok_or_else(|| {
+                WasmChannelError::Compilation("No compiled component available".to_string())
+            })?
+            .clone();
 
         // Create linker and add host functions
         let mut linker = Linker::new(engine);
@@ -751,7 +821,7 @@ impl WasmChannel {
     /// Returns the channel configuration for HTTP endpoint registration.
     async fn call_on_start(&self) -> Result<ChannelConfig, WasmChannelError> {
         // If no WASM bytes, return default config (for testing)
-        if self.prepared.component_bytes.is_empty() {
+        if self.prepared.component().is_none() {
             tracing::info!(
                 channel = %self.name,
                 "WASM channel on_start called (no WASM module, returning defaults)"
@@ -765,12 +835,13 @@ impl WasmChannel {
 
         let runtime = Arc::clone(&self.runtime);
         let prepared = Arc::clone(&self.prepared);
-        let capabilities = self.capabilities.clone();
+        let capabilities = Self::inject_workspace_reader(&self.capabilities, &self.workspace_store);
         let config_json = self.config_json.read().await.clone();
         let timeout = self.runtime.config().callback_timeout;
         let channel_name = self.name.clone();
         let credentials = self.get_credentials().await;
         let pairing_store = self.pairing_store.clone();
+        let workspace_store = self.workspace_store.clone();
 
         // Execute in blocking task with timeout
         let result = tokio::time::timeout(timeout, async move {
@@ -801,8 +872,13 @@ impl WasmChannel {
                     }
                 };
 
-                let host_state =
+                let mut host_state =
                     Self::extract_host_state(&mut store, &prepared.name, &capabilities);
+
+                // Commit pending workspace writes to the persistent store
+                let pending_writes = host_state.take_pending_writes();
+                workspace_store.commit_writes(&pending_writes);
+
                 Ok((config, host_state))
             })
             .await
@@ -885,7 +961,7 @@ impl WasmChannel {
         );
 
         // If no WASM bytes, return 200 OK (for testing)
-        if self.prepared.component_bytes.is_empty() {
+        if self.prepared.component().is_none() {
             tracing::debug!(
                 channel = %self.name,
                 method = method,
@@ -897,10 +973,11 @@ impl WasmChannel {
 
         let runtime = Arc::clone(&self.runtime);
         let prepared = Arc::clone(&self.prepared);
-        let capabilities = self.capabilities.clone();
+        let capabilities = Self::inject_workspace_reader(&self.capabilities, &self.workspace_store);
         let timeout = self.runtime.config().callback_timeout;
         let credentials = self.get_credentials().await;
         let pairing_store = self.pairing_store.clone();
+        let workspace_store = self.workspace_store.clone();
 
         // Prepare request data
         let method = method.to_string();
@@ -940,8 +1017,13 @@ impl WasmChannel {
                     .map_err(|e| Self::map_wasm_error(e, &prepared.name, prepared.limits.fuel))?;
 
                 let response = convert_http_response(wit_response);
-                let host_state =
+                let mut host_state =
                     Self::extract_host_state(&mut store, &prepared.name, &capabilities);
+
+                // Commit pending workspace writes to the persistent store
+                let pending_writes = host_state.take_pending_writes();
+                workspace_store.commit_writes(&pending_writes);
+
                 Ok((response, host_state))
             })
             .await
@@ -979,7 +1061,7 @@ impl WasmChannel {
     /// Called periodically if polling is configured.
     pub async fn call_on_poll(&self) -> Result<(), WasmChannelError> {
         // If no WASM bytes, do nothing (for testing)
-        if self.prepared.component_bytes.is_empty() {
+        if self.prepared.component().is_none() {
             tracing::debug!(
                 channel = %self.name,
                 "WASM channel on_poll called (no WASM module)"
@@ -989,11 +1071,12 @@ impl WasmChannel {
 
         let runtime = Arc::clone(&self.runtime);
         let prepared = Arc::clone(&self.prepared);
-        let capabilities = self.capabilities.clone();
+        let capabilities = Self::inject_workspace_reader(&self.capabilities, &self.workspace_store);
         let timeout = self.runtime.config().callback_timeout;
         let channel_name = self.name.clone();
         let credentials = self.get_credentials().await;
         let pairing_store = self.pairing_store.clone();
+        let workspace_store = self.workspace_store.clone();
 
         // Execute in blocking task with timeout
         let result = tokio::time::timeout(timeout, async move {
@@ -1013,8 +1096,13 @@ impl WasmChannel {
                     .call_on_poll(&mut store)
                     .map_err(|e| Self::map_wasm_error(e, &prepared.name, prepared.limits.fuel))?;
 
-                let host_state =
+                let mut host_state =
                     Self::extract_host_state(&mut store, &prepared.name, &capabilities);
+
+                // Commit pending workspace writes to the persistent store
+                let pending_writes = host_state.take_pending_writes();
+                workspace_store.commit_writes(&pending_writes);
+
                 Ok(((), host_state))
             })
             .await
@@ -1073,7 +1161,7 @@ impl WasmChannel {
         );
 
         // If no WASM bytes, do nothing (for testing)
-        if self.prepared.component_bytes.is_empty() {
+        if self.prepared.component().is_none() {
             tracing::debug!(
                 channel = %self.name,
                 message_id = %message_id,
@@ -1191,7 +1279,7 @@ impl WasmChannel {
         metadata: &serde_json::Value,
     ) -> Result<(), WasmChannelError> {
         // If no WASM bytes, do nothing (for testing)
-        if self.prepared.component_bytes.is_empty() {
+        if self.prepared.component().is_none() {
             return Ok(());
         }
 
@@ -1262,7 +1350,7 @@ impl WasmChannel {
         timeout: Duration,
         wit_update: wit_channel::StatusUpdate,
     ) -> Result<(), WasmChannelError> {
-        if prepared.component_bytes.is_empty() {
+        if prepared.component().is_none() {
             return Ok(());
         }
 
@@ -1445,27 +1533,18 @@ impl WasmChannel {
                 });
             }
 
-            // Convert to IncomingMessage
-            let mut msg = IncomingMessage::new(&self.name, &emitted.user_id, &emitted.content);
+            let msg = Self::convert_emitted_to_incoming(
+                &self.name,
+                emitted,
+                self.transcription_middleware.as_deref(),
+            )
+            .await;
 
-            if let Some(name) = emitted.user_name {
-                msg = msg.with_user_name(name);
-            }
-
-            if let Some(thread_id) = emitted.thread_id {
-                msg = msg.with_thread(thread_id);
-            }
-
-            // Parse metadata JSON
-            if let Ok(metadata) = serde_json::from_str(&emitted.metadata_json) {
-                msg = msg.with_metadata(metadata);
-            }
-
-            // Send to stream
+            // Send to stream (log post-transcription state intentionally)
             tracing::info!(
                 channel = %self.name,
-                user_id = %emitted.user_id,
-                content_len = emitted.content.len(),
+                user_id = %msg.user_id,
+                content_len = msg.content.len(),
                 "Sending emitted message to agent"
             );
 
@@ -1501,6 +1580,8 @@ impl WasmChannel {
         let credentials = self.credentials.clone();
         let pairing_store = self.pairing_store.clone();
         let callback_timeout = self.runtime.config().callback_timeout;
+        let workspace_store = self.workspace_store.clone();
+        let transcription_middleware = self.transcription_middleware.clone();
 
         tokio::spawn(async move {
             let mut interval_timer = tokio::time::interval(interval);
@@ -1523,6 +1604,7 @@ impl WasmChannel {
                             &credentials,
                             pairing_store.clone(),
                             callback_timeout,
+                            &workspace_store,
                         ).await;
 
                         match result {
@@ -1534,6 +1616,7 @@ impl WasmChannel {
                                         emitted_messages,
                                         &message_tx,
                                         &rate_limiter,
+                                        transcription_middleware.as_deref(),
                                     ).await {
                                         tracing::warn!(
                                             channel = %channel_name,
@@ -1565,7 +1648,10 @@ impl WasmChannel {
 
     /// Execute a single poll callback with a fresh WASM instance.
     ///
-    /// Returns any emitted messages from the callback.
+    /// Returns any emitted messages from the callback. Pending workspace writes
+    /// are committed to the shared `ChannelWorkspaceStore` so state persists
+    /// across poll ticks (e.g., Telegram polling offset).
+    #[allow(clippy::too_many_arguments)]
     async fn execute_poll(
         channel_name: &str,
         runtime: &Arc<WasmChannelRuntime>,
@@ -1574,9 +1660,10 @@ impl WasmChannel {
         credentials: &RwLock<HashMap<String, String>>,
         pairing_store: Arc<PairingStore>,
         timeout: Duration,
+        workspace_store: &Arc<ChannelWorkspaceStore>,
     ) -> Result<Vec<EmittedMessage>, WasmChannelError> {
         // Skip if no WASM bytes (testing mode)
-        if prepared.component_bytes.is_empty() {
+        if prepared.component().is_none() {
             tracing::debug!(
                 channel = %channel_name,
                 "WASM channel on_poll called (no WASM module)"
@@ -1586,9 +1673,10 @@ impl WasmChannel {
 
         let runtime = Arc::clone(runtime);
         let prepared = Arc::clone(prepared);
-        let capabilities = capabilities.clone();
+        let capabilities = Self::inject_workspace_reader(capabilities, workspace_store);
         let credentials_snapshot = credentials.read().await.clone();
         let channel_name_owned = channel_name.to_string();
+        let workspace_store = Arc::clone(workspace_store);
 
         // Execute in blocking task with timeout
         let result = tokio::time::timeout(timeout, async move {
@@ -1608,8 +1696,13 @@ impl WasmChannel {
                     .call_on_poll(&mut store)
                     .map_err(|e| Self::map_wasm_error(e, &prepared.name, prepared.limits.fuel))?;
 
-                let host_state =
+                let mut host_state =
                     Self::extract_host_state(&mut store, &prepared.name, &capabilities);
+
+                // Commit pending workspace writes to the persistent store
+                let pending_writes = host_state.take_pending_writes();
+                workspace_store.commit_writes(&pending_writes);
+
                 Ok(host_state)
             })
             .await
@@ -1647,6 +1740,7 @@ impl WasmChannel {
         messages: Vec<EmittedMessage>,
         message_tx: &RwLock<Option<mpsc::Sender<IncomingMessage>>>,
         rate_limiter: &RwLock<ChannelEmitRateLimiter>,
+        transcription_middleware: Option<&crate::transcription::TranscriptionMiddleware>,
     ) -> Result<(), WasmChannelError> {
         tracing::info!(
             channel = %channel_name,
@@ -1678,27 +1772,15 @@ impl WasmChannel {
                 });
             }
 
-            // Convert to IncomingMessage
-            let mut msg = IncomingMessage::new(channel_name, &emitted.user_id, &emitted.content);
+            let msg =
+                Self::convert_emitted_to_incoming(channel_name, emitted, transcription_middleware)
+                    .await;
 
-            if let Some(name) = emitted.user_name {
-                msg = msg.with_user_name(name);
-            }
-
-            if let Some(thread_id) = emitted.thread_id {
-                msg = msg.with_thread(thread_id);
-            }
-
-            // Parse metadata JSON
-            if let Ok(metadata) = serde_json::from_str(&emitted.metadata_json) {
-                msg = msg.with_metadata(metadata);
-            }
-
-            // Send to stream
+            // Send to stream (log post-transcription state intentionally)
             tracing::info!(
                 channel = %channel_name,
-                user_id = %emitted.user_id,
-                content_len = emitted.content.len(),
+                user_id = %msg.user_id,
+                content_len = msg.content.len(),
                 "Sending polled message to agent"
             );
 
@@ -1718,6 +1800,65 @@ impl WasmChannel {
 
         Ok(())
     }
+
+    /// Convert an `EmittedMessage` to an `IncomingMessage`, applying transcription
+    /// middleware with a timeout if available.
+    ///
+    /// Shared by both `process_emitted_messages` (HTTP callback path) and
+    /// `dispatch_emitted_messages` (polling path) to avoid duplication.
+    async fn convert_emitted_to_incoming(
+        channel_name: &str,
+        emitted: EmittedMessage,
+        transcription_middleware: Option<&crate::transcription::TranscriptionMiddleware>,
+    ) -> IncomingMessage {
+        // Save user_id before partial moves for potential timeout fallback
+        let user_id = emitted.user_id.clone();
+
+        let mut msg = IncomingMessage::new(channel_name, &emitted.user_id, &emitted.content);
+
+        if let Some(name) = emitted.user_name {
+            msg = msg.with_user_name(name);
+        }
+
+        if let Some(thread_id) = emitted.thread_id {
+            msg = msg.with_thread(thread_id);
+        }
+
+        // Parse metadata JSON
+        if let Ok(metadata) = serde_json::from_str(&emitted.metadata_json) {
+            msg = msg.with_metadata(metadata);
+        }
+
+        // Carry attachments through (moves emitted.attachments into msg)
+        if !emitted.attachments.is_empty() {
+            msg = msg.with_attachments(emitted.attachments);
+        }
+
+        // Apply transcription middleware with a 30-second timeout to prevent
+        // a slow/hanging provider from blocking the message pipeline indefinitely.
+        if let Some(middleware) = transcription_middleware {
+            match tokio::time::timeout(std::time::Duration::from_secs(30), middleware.process(msg))
+                .await
+            {
+                Ok(processed) => return processed,
+                Err(_) => {
+                    tracing::error!(
+                        channel = %channel_name,
+                        "Transcription timed out after 30s, delivering message without transcript"
+                    );
+                    // Timeout: `msg` was moved into the timed-out future, so
+                    // reconstruct a fallback message from the saved user_id.
+                    return IncomingMessage::new(
+                        channel_name,
+                        &user_id,
+                        "[Voice note: transcription timed out]",
+                    );
+                }
+            }
+        }
+
+        msg
+    }
 }
 
 #[async_trait]
@@ -2149,7 +2290,7 @@ mod tests {
         let prepared = Arc::new(PreparedChannelModule {
             name: "test".to_string(),
             description: "Test channel".to_string(),
-            component_bytes: Vec::new(),
+            component: None,
             limits: ResourceLimits::default(),
         });
 
@@ -2214,7 +2355,7 @@ mod tests {
 
     #[tokio::test]
     async fn test_execute_poll_no_wasm_returns_empty() {
-        // When there's no WASM module (empty component_bytes), execute_poll
+        // When there's no WASM module (None component), execute_poll
         // should return an empty vector of messages
         let config = WasmChannelRuntimeConfig::for_testing();
         let runtime = Arc::new(WasmChannelRuntime::new(config).unwrap());
@@ -2222,7 +2363,7 @@ mod tests {
         let prepared = Arc::new(PreparedChannelModule {
             name: "poll-test".to_string(),
             description: "Test channel".to_string(),
-            component_bytes: Vec::new(), // No WASM bytes
+            component: None, // No WASM module
             limits: ResourceLimits::default(),
         });
 
@@ -2230,6 +2371,8 @@ mod tests {
         let credentials = Arc::new(tokio::sync::RwLock::new(std::collections::HashMap::new()));
         let timeout = std::time::Duration::from_secs(5);
 
+        let workspace_store = Arc::new(crate::channels::wasm::host::ChannelWorkspaceStore::new());
+
         let result = WasmChannel::execute_poll(
             "poll-test",
             &runtime,
@@ -2238,6 +2381,7 @@ mod tests {
             &credentials,
             Arc::new(PairingStore::new()),
             timeout,
+            &workspace_store,
         )
         .await;
 
@@ -2268,6 +2412,7 @@ mod tests {
             messages,
             &message_tx,
             &rate_limiter,
+            None,
         )
         .await;
 
@@ -2306,6 +2451,7 @@ mod tests {
             messages,
             &message_tx,
             &rate_limiter,
+            None,
         )
         .await;
 
@@ -2321,7 +2467,7 @@ mod tests {
         let prepared = Arc::new(PreparedChannelModule {
             name: "poll-channel".to_string(),
             description: "Polling test channel".to_string(),
-            component_bytes: Vec::new(),
+            component: None,
             limits: ResourceLimits::default(),
         });
 

src/channels/web/log_layer.rs

@@ -22,7 +22,9 @@ use std::sync::{Arc, Mutex};
 use serde::Serialize;
 use tokio::sync::broadcast;
 use tracing::field::{Field, Visit};
-use tracing_subscriber::Layer;
+use tracing_subscriber::layer::SubscriberExt;
+use tracing_subscriber::util::SubscriberInitExt;
+use tracing_subscriber::{EnvFilter, Layer, reload};
 
 use crate::safety::LeakDetector;
 
@@ -102,6 +104,115 @@ impl Default for LogBroadcaster {
     }
 }
 
+/// Handle for changing the tracing `EnvFilter` at runtime.
+///
+/// Wraps a `reload::Handle` so the gateway can switch between log levels
+/// (e.g. `ironclaw=debug`) without restarting the process.
+pub struct LogLevelHandle {
+    handle: reload::Handle<EnvFilter, tracing_subscriber::Registry>,
+    current_level: Mutex<String>,
+    base_filter: String,
+}
+
+impl LogLevelHandle {
+    pub fn new(
+        handle: reload::Handle<EnvFilter, tracing_subscriber::Registry>,
+        initial_level: String,
+        base_filter: String,
+    ) -> Self {
+        Self {
+            handle,
+            current_level: Mutex::new(initial_level),
+            base_filter,
+        }
+    }
+
+    /// Change the `ironclaw=<level>` directive at runtime.
+    ///
+    /// `level` must be one of: trace, debug, info, warn, error.
+    pub fn set_level(&self, level: &str) -> Result<(), String> {
+        const VALID: &[&str] = &["trace", "debug", "info", "warn", "error"];
+        let level = level.to_lowercase();
+        if !VALID.contains(&level.as_str()) {
+            return Err(format!(
+                "invalid level '{}', must be one of: {}",
+                level,
+                VALID.join(", ")
+            ));
+        }
+
+        let filter_str = if self.base_filter.is_empty() {
+            format!("ironclaw={}", level)
+        } else {
+            format!("ironclaw={},{}", level, self.base_filter)
+        };
+
+        let new_filter = EnvFilter::new(&filter_str);
+        self.handle
+            .reload(new_filter)
+            .map_err(|e| format!("failed to reload filter: {}", e))?;
+
+        if let Ok(mut current) = self.current_level.lock() {
+            *current = level;
+        }
+        Ok(())
+    }
+
+    /// Returns the current ironclaw log level (e.g. "info", "debug").
+    pub fn current_level(&self) -> String {
+        self.current_level
+            .lock()
+            .map(|l| l.clone())
+            .unwrap_or_else(|_| "info".to_string())
+    }
+}
+
+/// Initialise the tracing subscriber with a reloadable `EnvFilter`.
+///
+/// Returns the `LogLevelHandle` so callers can swap the filter at runtime.
+/// The fmt layer and `WebLogLayer` are attached alongside the reloadable filter.
+pub fn init_tracing(log_broadcaster: Arc<LogBroadcaster>) -> Arc<LogLevelHandle> {
+    let raw_filter =
+        std::env::var("RUST_LOG").unwrap_or_else(|_| "ironclaw=info,tower_http=warn".to_string());
+
+    // Split into the ironclaw directive and "everything else" (base_filter).
+    let mut ironclaw_level = String::from("info");
+    let mut base_parts: Vec<&str> = Vec::new();
+
+    for part in raw_filter.split(',') {
+        let trimmed = part.trim();
+        if trimmed.starts_with("ironclaw=") {
+            if let Some(lvl) = trimmed.strip_prefix("ironclaw=") {
+                ironclaw_level = lvl.to_string();
+            }
+        } else if !trimmed.is_empty() {
+            base_parts.push(trimmed);
+        }
+    }
+    let base_filter = base_parts.join(",");
+
+    let env_filter = EnvFilter::new(&raw_filter);
+    let (reload_layer, reload_handle) = reload::Layer::new(env_filter);
+
+    let handle = Arc::new(LogLevelHandle::new(
+        reload_handle,
+        ironclaw_level,
+        base_filter,
+    ));
+
+    tracing_subscriber::registry()
+        .with(reload_layer)
+        .with(
+            tracing_subscriber::fmt::layer()
+                .with_target(false)
+                .with_writer(crate::tracing_fmt::TruncatingStderr::default()),
+        )
+        .with(WebLogLayer::new(log_broadcaster))
+        .init();
+
+    handle
+}
+
 /// Visitor that extracts the `message` field and all extra key-value
 /// fields from a tracing event.
 ///

src/channels/web/mod.rs

@@ -41,7 +41,7 @@ use crate::skills::registry::SkillRegistry;
 use crate::tools::ToolRegistry;
 use crate::workspace::Workspace;
 
-use self::log_layer::LogBroadcaster;
+use self::log_layer::{LogBroadcaster, LogLevelHandle};
 
 use self::server::GatewayState;
 use self::sse::SseManager;
@@ -76,6 +76,7 @@ impl GatewayChannel {
             workspace: None,
             session_manager: None,
             log_broadcaster: None,
+            log_level_handle: None,
             extension_manager: None,
             tool_registry: None,
             store: None,
@@ -88,6 +89,9 @@ impl GatewayChannel {
             skill_registry: None,
             skill_catalog: None,
             chat_rate_limiter: server::RateLimiter::new(30, 60),
+            registry_entries: Vec::new(),
+            cost_guard: None,
+            startup_time: std::time::Instant::now(),
         });
 
         Self {
@@ -105,6 +109,7 @@ impl GatewayChannel {
             workspace: self.state.workspace.clone(),
             session_manager: self.state.session_manager.clone(),
             log_broadcaster: self.state.log_broadcaster.clone(),
+            log_level_handle: self.state.log_level_handle.clone(),
             extension_manager: self.state.extension_manager.clone(),
             tool_registry: self.state.tool_registry.clone(),
             store: self.state.store.clone(),
@@ -117,6 +122,9 @@ impl GatewayChannel {
             skill_registry: self.state.skill_registry.clone(),
             skill_catalog: self.state.skill_catalog.clone(),
             chat_rate_limiter: server::RateLimiter::new(30, 60),
+            registry_entries: self.state.registry_entries.clone(),
+            cost_guard: self.state.cost_guard.clone(),
+            startup_time: self.state.startup_time,
         };
         mutate(&mut new_state);
         self.state = Arc::new(new_state);
@@ -140,6 +148,12 @@ impl GatewayChannel {
         self
     }
 
+    /// Inject the log level handle for runtime log level control.
+    pub fn with_log_level_handle(mut self, h: Arc<LogLevelHandle>) -> Self {
+        self.rebuild_state(|s| s.log_level_handle = Some(h));
+        self
+    }
+
     /// Inject the extension manager for the extensions API.
     pub fn with_extension_manager(mut self, em: Arc<ExtensionManager>) -> Self {
         self.rebuild_state(|s| s.extension_manager = Some(em));
@@ -198,6 +212,18 @@ impl GatewayChannel {
         self
     }
 
+    /// Inject registry catalog entries for the available extensions API.
+    pub fn with_registry_entries(mut self, entries: Vec<crate::extensions::RegistryEntry>) -> Self {
+        self.rebuild_state(|s| s.registry_entries = entries);
+        self
+    }
+
+    /// Inject the cost guard for token/cost tracking in the status popover.
+    pub fn with_cost_guard(mut self, cg: Arc<crate::agent::cost_guard::CostGuard>) -> Self {
+        self.rebuild_state(|s| s.cost_guard = Some(cg));
+        self
+    }
+
     /// Get the auth token (for printing to console on startup).
     pub fn auth_token(&self) -> &str {
         &self.auth_token

src/channels/web/server.rs

@@ -13,7 +13,7 @@ use axum::{
     http::{StatusCode, header},
     middleware,
     response::{
-        Html, IntoResponse,
+        IntoResponse,
         sse::{Event, KeepAlive, Sse},
     },
     routing::{get, post},
@@ -122,6 +122,8 @@ pub struct GatewayState {
     pub session_manager: Option<Arc<SessionManager>>,
     /// Log broadcaster for the logs SSE endpoint.
     pub log_broadcaster: Option<Arc<LogBroadcaster>>,
+    /// Handle for changing the tracing log level at runtime.
+    pub log_level_handle: Option<Arc<crate::channels::web::log_layer::LogLevelHandle>>,
     /// Extension manager for extension management API.
     pub extension_manager: Option<Arc<ExtensionManager>>,
     /// Tool registry for listing registered tools.
@@ -146,6 +148,13 @@ pub struct GatewayState {
     pub skill_catalog: Option<Arc<crate::skills::catalog::SkillCatalog>>,
     /// Rate limiter for chat endpoints (30 messages per 60 seconds).
     pub chat_rate_limiter: RateLimiter,
+    /// Registry catalog entries for the available extensions API.
+    /// Populated at startup from `registry/` manifests, independent of extension manager.
+    pub registry_entries: Vec<crate::extensions::RegistryEntry>,
+    /// Cost guard for token/cost tracking.
+    pub cost_guard: Option<Arc<crate::agent::cost_guard::CostGuard>>,
+    /// Server startup time for uptime calculation.
+    pub startup_time: std::time::Instant,
 }
 
 /// Start the gateway HTTP server.
@@ -204,9 +213,15 @@ pub async fn start_server(
         .route("/api/jobs/{id}/files/read", get(job_files_read_handler))
         // Logs
         .route("/api/logs/events", get(logs_events_handler))
+        .route("/api/logs/level", get(logs_level_get_handler))
+        .route(
+            "/api/logs/level",
+            axum::routing::put(logs_level_set_handler),
+        )
         // Extensions
         .route("/api/extensions", get(extensions_list_handler))
         .route("/api/extensions/tools", get(extensions_tools_handler))
+        .route("/api/extensions/registry", get(extensions_registry_handler))
         .route("/api/extensions/install", post(extensions_install_handler))
         .route(
             "/api/extensions/{name}/activate",
@@ -216,6 +231,16 @@ pub async fn start_server(
             "/api/extensions/{name}/remove",
             post(extensions_remove_handler),
         )
+        .route(
+            "/api/extensions/{name}/setup",
+            get(extensions_setup_handler).post(extensions_setup_submit_handler),
+        )
+        // Pairing
+        .route("/api/pairing/{channel}", get(pairing_list_handler))
+        .route(
+            "/api/pairing/{channel}/approve",
+            post(pairing_approve_handler),
+        )
         // Routines
         .route("/api/routines", get(routines_list_handler))
         .route("/api/routines/summary", get(routines_summary_handler))
@@ -337,20 +362,32 @@ pub async fn start_server(
 
 // --- Static file handlers ---
 
-async fn index_handler() -> Html<&'static str> {
-    Html(include_str!("static/index.html"))
+async fn index_handler() -> impl IntoResponse {
+    (
+        [
+            (header::CONTENT_TYPE, "text/html; charset=utf-8"),
+            (header::CACHE_CONTROL, "no-cache"),
+        ],
+        include_str!("static/index.html"),
+    )
 }
 
 async fn css_handler() -> impl IntoResponse {
     (
-        [(header::CONTENT_TYPE, "text/css")],
+        [
+            (header::CONTENT_TYPE, "text/css"),
+            (header::CACHE_CONTROL, "no-cache"),
+        ],
         include_str!("static/style.css"),
     )
 }
 
 async fn js_handler() -> impl IntoResponse {
     (
-        [(header::CONTENT_TYPE, "application/javascript")],
+        [
+            (header::CONTENT_TYPE, "application/javascript"),
+            (header::CACHE_CONTROL, "no-cache"),
+        ],
         include_str!("static/app.js"),
     )
 }
@@ -557,9 +594,13 @@ pub async fn clear_auth_mode(state: &GatewayState) {
 async fn chat_events_handler(
     State(state): State<Arc<GatewayState>>,
 ) -> Result<impl IntoResponse, (StatusCode, String)> {
-    state.sse.subscribe().ok_or((
+    let sse = state.sse.subscribe().ok_or((
         StatusCode::SERVICE_UNAVAILABLE,
         "Too many connections".to_string(),
+    ))?;
+    Ok((
+        [("X-Accel-Buffering", "no"), ("Cache-Control", "no-cache")],
+        sse,
     ))
 }
 
@@ -1585,10 +1626,7 @@ async fn job_files_read_handler(
 
 async fn logs_events_handler(
     State(state): State<Arc<GatewayState>>,
-) -> Result<
-    Sse<impl futures::Stream<Item = Result<Event, Infallible>> + Send + 'static>,
-    (StatusCode, String),
-> {
+) -> Result<impl IntoResponse, (StatusCode, String)> {
     let broadcaster = state.log_broadcaster.as_ref().ok_or((
         StatusCode::SERVICE_UNAVAILABLE,
         "Log broadcaster not available".to_string(),
@@ -1601,25 +1639,60 @@ async fn logs_events_handler(
 
     let history_stream = futures::stream::iter(history).map(|entry| {
         let data = serde_json::to_string(&entry).unwrap_or_default();
-        Ok(Event::default().event("log").data(data))
+        Ok::<_, Infallible>(Event::default().event("log").data(data))
     });
 
     let live_stream = tokio_stream::wrappers::BroadcastStream::new(rx)
         .filter_map(|result| result.ok())
         .map(|entry| {
             let data = serde_json::to_string(&entry).unwrap_or_default();
-            Ok(Event::default().event("log").data(data))
+            Ok::<_, Infallible>(Event::default().event("log").data(data))
         });
 
     let stream = history_stream.chain(live_stream);
 
-    Ok(Sse::new(stream).keep_alive(
-        KeepAlive::new()
-            .interval(std::time::Duration::from_secs(30))
-            .text(""),
+    Ok((
+        [("X-Accel-Buffering", "no"), ("Cache-Control", "no-cache")],
+        Sse::new(stream).keep_alive(
+            KeepAlive::new()
+                .interval(std::time::Duration::from_secs(30))
+                .text(""),
+        ),
     ))
 }
 
+async fn logs_level_get_handler(
+    State(state): State<Arc<GatewayState>>,
+) -> Result<Json<serde_json::Value>, (StatusCode, String)> {
+    let handle = state.log_level_handle.as_ref().ok_or((
+        StatusCode::SERVICE_UNAVAILABLE,
+        "Log level control not available".to_string(),
+    ))?;
+    Ok(Json(serde_json::json!({ "level": handle.current_level() })))
+}
+
+async fn logs_level_set_handler(
+    State(state): State<Arc<GatewayState>>,
+    Json(body): Json<serde_json::Value>,
+) -> Result<Json<serde_json::Value>, (StatusCode, String)> {
+    let handle = state.log_level_handle.as_ref().ok_or((
+        StatusCode::SERVICE_UNAVAILABLE,
+        "Log level control not available".to_string(),
+    ))?;
+
+    let level = body
+        .get("level")
+        .and_then(|v| v.as_str())
+        .ok_or((StatusCode::BAD_REQUEST, "missing 'level' field".to_string()))?;
+
+    handle
+        .set_level(level)
+        .map_err(|e| (StatusCode::BAD_REQUEST, e))?;
+
+    tracing::info!("Log level changed to '{}'", handle.current_level());
+    Ok(Json(serde_json::json!({ "level": handle.current_level() })))
+}
+
 // --- Extension handlers ---
 
 async fn extensions_list_handler(
@@ -1645,6 +1718,7 @@ async fn extensions_list_handler(
             authenticated: ext.authenticated,
             active: ext.active,
             tools: ext.tools,
+            needs_setup: ext.needs_setup,
         })
         .collect();
 
@@ -1675,10 +1749,30 @@ async fn extensions_install_handler(
     State(state): State<Arc<GatewayState>>,
     Json(req): Json<InstallExtensionRequest>,
 ) -> Result<Json<ActionResponse>, (StatusCode, String)> {
-    let ext_mgr = state.extension_manager.as_ref().ok_or((
-        StatusCode::NOT_IMPLEMENTED,
-        "Extension manager not available (secrets store required)".to_string(),
-    ))?;
+    // When extension manager isn't available, check registry entries for a helpful message
+    let Some(ext_mgr) = state.extension_manager.as_ref() else {
+        // Look up the entry in the catalog to give a specific error
+        if let Some(entry) = state.registry_entries.iter().find(|e| e.name == req.name) {
+            let msg = match &entry.source {
+                crate::extensions::ExtensionSource::WasmBuildable { .. } => {
+                    format!(
+                        "'{}' requires building from source. \
+                         Run `ironclaw registry install {}` from the CLI.",
+                        req.name, req.name
+                    )
+                }
+                _ => format!(
+                    "Extension manager not available (secrets store required). \
+                     Configure DATABASE_URL or a secrets backend to enable installation of '{}'.",
+                    req.name
+                ),
+            };
+            return Ok(Json(ActionResponse::fail(msg)));
+        }
+        return Ok(Json(ActionResponse::fail(
+            "Extension manager not available (secrets store required)".to_string(),
+        )));
+    };
 
     let kind_hint = req.kind.as_deref().and_then(|k| match k {
         "mcp_server" => Some(crate::extensions::ExtensionKind::McpServer),
@@ -1827,6 +1921,160 @@ async fn extensions_remove_handler(
     }
 }
 
+async fn extensions_registry_handler(
+    State(state): State<Arc<GatewayState>>,
+    Query(params): Query<RegistrySearchQuery>,
+) -> Json<RegistrySearchResponse> {
+    let query = params.query.unwrap_or_default();
+    let query_lower = query.to_lowercase();
+    let tokens: Vec<&str> = query_lower.split_whitespace().collect();
+
+    // Filter registry entries by query (or return all if empty)
+    let matching: Vec<&crate::extensions::RegistryEntry> = if tokens.is_empty() {
+        state.registry_entries.iter().collect()
+    } else {
+        state
+            .registry_entries
+            .iter()
+            .filter(|e| {
+                let name = e.name.to_lowercase();
+                let display = e.display_name.to_lowercase();
+                let desc = e.description.to_lowercase();
+                tokens.iter().any(|t| {
+                    name.contains(t)
+                        || display.contains(t)
+                        || desc.contains(t)
+                        || e.keywords.iter().any(|k| k.to_lowercase().contains(t))
+                })
+            })
+            .collect()
+    };
+
+    // Cross-reference with installed extensions by (name, kind) to avoid
+    // false positives when the same name exists as different kinds.
+    let installed: std::collections::HashSet<(String, String)> =
+        if let Some(ext_mgr) = state.extension_manager.as_ref() {
+            ext_mgr
+                .list(None)
+                .await
+                .unwrap_or_default()
+                .into_iter()
+                .map(|ext| (ext.name, ext.kind.to_string()))
+                .collect()
+        } else {
+            std::collections::HashSet::new()
+        };
+
+    let entries = matching
+        .into_iter()
+        .map(|e| {
+            let kind_str = e.kind.to_string();
+            RegistryEntryInfo {
+                name: e.name.clone(),
+                display_name: e.display_name.clone(),
+                installed: installed.contains(&(e.name.clone(), kind_str.clone())),
+                kind: kind_str,
+                description: e.description.clone(),
+                keywords: e.keywords.clone(),
+            }
+        })
+        .collect();
+
+    Json(RegistrySearchResponse { entries })
+}
+
+async fn extensions_setup_handler(
+    State(state): State<Arc<GatewayState>>,
+    Path(name): Path<String>,
+) -> Result<Json<ExtensionSetupResponse>, (StatusCode, String)> {
+    let ext_mgr = state.extension_manager.as_ref().ok_or((
+        StatusCode::NOT_IMPLEMENTED,
+        "Extension manager not available (secrets store required)".to_string(),
+    ))?;
+
+    let secrets = ext_mgr
+        .get_setup_schema(&name)
+        .await
+        .map_err(|e| (StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))?;
+
+    let kind = ext_mgr
+        .list(None)
+        .await
+        .ok()
+        .and_then(|list| list.into_iter().find(|e| e.name == name))
+        .map(|e| e.kind.to_string())
+        .unwrap_or_default();
+
+    Ok(Json(ExtensionSetupResponse {
+        name,
+        kind,
+        secrets,
+    }))
+}
+
+async fn extensions_setup_submit_handler(
+    State(state): State<Arc<GatewayState>>,
+    Path(name): Path<String>,
+    Json(req): Json<ExtensionSetupRequest>,
+) -> Result<Json<ActionResponse>, (StatusCode, String)> {
+    let ext_mgr = state.extension_manager.as_ref().ok_or((
+        StatusCode::NOT_IMPLEMENTED,
+        "Extension manager not available (secrets store required)".to_string(),
+    ))?;
+
+    match ext_mgr.save_setup_secrets(&name, &req.secrets).await {
+        Ok(message) => Ok(Json(ActionResponse::ok(message))),
+        Err(e) => Ok(Json(ActionResponse::fail(e.to_string()))),
+    }
+}
+
+// --- Pairing handlers ---
+
+async fn pairing_list_handler(
+    Path(channel): Path<String>,
+) -> Result<Json<PairingListResponse>, (StatusCode, String)> {
+    let store = crate::pairing::PairingStore::new();
+    let requests = store
+        .list_pending(&channel)
+        .map_err(|e| (StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))?;
+
+    let infos = requests
+        .into_iter()
+        .map(|r| PairingRequestInfo {
+            code: r.code,
+            sender_id: r.id,
+            meta: r.meta,
+            created_at: r.created_at,
+        })
+        .collect();
+
+    Ok(Json(PairingListResponse {
+        channel,
+        requests: infos,
+    }))
+}
+
+async fn pairing_approve_handler(
+    Path(channel): Path<String>,
+    Json(req): Json<PairingApproveRequest>,
+) -> Result<Json<ActionResponse>, (StatusCode, String)> {
+    let store = crate::pairing::PairingStore::new();
+    match store.approve(&channel, &req.code) {
+        Ok(Some(approved)) => Ok(Json(ActionResponse::ok(format!(
+            "Pairing approved for sender '{}'",
+            approved.id
+        )))),
+        Ok(None) => Ok(Json(ActionResponse::fail(
+            "Invalid or expired pairing code".to_string(),
+        ))),
+        Err(crate::pairing::PairingStoreError::ApproveRateLimited) => Err((
+            StatusCode::TOO_MANY_REQUESTS,
+            "Too many failed approve attempts; try again later".to_string(),
+        )),
+        Err(e) => Ok(Json(ActionResponse::fail(e.to_string()))),
+    }
+}
+
 // --- Skills handlers ---
 
 async fn skills_list_handler(
@@ -2526,18 +2774,57 @@ async fn gateway_status_handler(
         .map(|t| t.connection_count())
         .unwrap_or(0);
 
+    let uptime_secs = state.startup_time.elapsed().as_secs();
+
+    let (daily_cost, actions_this_hour, model_usage) = if let Some(ref cg) = state.cost_guard {
+        let cost = cg.daily_spend().await;
+        let actions = cg.actions_this_hour().await;
+        let usage = cg.model_usage().await;
+        let models: Vec<ModelUsageEntry> = usage
+            .into_iter()
+            .map(|(model, tokens)| ModelUsageEntry {
+                model,
+                input_tokens: tokens.input_tokens,
+                output_tokens: tokens.output_tokens,
+                cost: format!("{:.6}", tokens.cost),
+            })
+            .collect();
+        (Some(format!("{:.4}", cost)), Some(actions), Some(models))
+    } else {
+        (None, None, None)
+    };
+
     Json(GatewayStatusResponse {
         sse_connections,
         ws_connections,
         total_connections: sse_connections + ws_connections,
+        uptime_secs,
+        daily_cost,
+        actions_this_hour,
+        model_usage,
     })
 }
 
+#[derive(serde::Serialize)]
+struct ModelUsageEntry {
+    model: String,
+    input_tokens: u64,
+    output_tokens: u64,
+    cost: String,
+}
+
 #[derive(serde::Serialize)]
 struct GatewayStatusResponse {
     sse_connections: u64,
     ws_connections: u64,
     total_connections: u64,
+    uptime_secs: u64,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    daily_cost: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    actions_this_hour: Option<u64>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    model_usage: Option<Vec<ModelUsageEntry>>,
 }
 
 #[cfg(test)]

src/channels/web/static/app.js

@@ -29,16 +29,25 @@ function authenticate() {
       sessionStorage.setItem('ironclaw_token', token);
       document.getElementById('auth-screen').style.display = 'none';
       document.getElementById('app').style.display = 'flex';
-      // Strip token from URL so it's not visible in the address bar
+      // Strip token and log_level from URL so they're not visible in the address bar
       const cleaned = new URL(window.location);
+      const urlLogLevel = cleaned.searchParams.get('log_level');
       cleaned.searchParams.delete('token');
+      cleaned.searchParams.delete('log_level');
       window.history.replaceState({}, '', cleaned.pathname + cleaned.search);
       connectSSE();
       connectLogSSE();
       startGatewayStatusPolling();
+      checkTeeStatus();
       loadThreads();
       loadMemoryTree();
       loadJobs();
+      // Apply URL log_level param if present, otherwise just sync the dropdown
+      if (urlLogLevel) {
+        setServerLogLevel(urlLogLevel);
+      } else {
+        loadServerLogLevel();
+      }
     })
     .catch(() => {
       sessionStorage.removeItem('ironclaw_token');
@@ -1167,19 +1176,46 @@ function applyLogFilters() {
   }
 }
 
+// --- Server-side log level control ---
+
+function setServerLogLevel(level) {
+  apiFetch('/api/logs/level', {
+    method: 'PUT',
+    headers: { 'Content-Type': 'application/json' },
+    body: JSON.stringify({ level: level }),
+  })
+    .then(r => r.json())
+    .then(data => {
+      document.getElementById('logs-server-level').value = data.level;
+    })
+    .catch(err => console.error('Failed to set server log level:', err));
+}
+
+function loadServerLogLevel() {
+  apiFetch('/api/logs/level')
+    .then(r => r.json())
+    .then(data => {
+      document.getElementById('logs-server-level').value = data.level;
+    })
+    .catch(() => {}); // ignore if not available
+}
+
 // --- Extensions ---
 
 function loadExtensions() {
   const extList = document.getElementById('extensions-list');
+  const wasmList = document.getElementById('available-wasm-list');
+  const mcpList = document.getElementById('mcp-servers-list');
   const toolsTbody = document.getElementById('tools-tbody');
   const toolsEmpty = document.getElementById('tools-empty');
 
-  // Fetch both in parallel
+  // Fetch all three in parallel
   Promise.all([
     apiFetch('/api/extensions').catch(() => ({ extensions: [] })),
     apiFetch('/api/extensions/tools').catch(() => ({ tools: [] })),
-  ]).then(([extData, toolData]) => {
-    // Render extensions
+    apiFetch('/api/extensions/registry').catch(function(err) { console.warn('registry fetch failed:', err); return { entries: [] }; }),
+  ]).then(([extData, toolData, registryData]) => {
+    // Render installed extensions
     if (extData.extensions.length === 0) {
       extList.innerHTML = '<div class="empty-state">No extensions installed</div>';
     } else {
@@ -1189,6 +1225,31 @@ function loadExtensions() {
       }
     }
 
+    // Split registry entries by kind
+    var wasmEntries = registryData.entries.filter(function(e) { return e.kind !== 'mcp_server' && !e.installed; });
+    var mcpEntries = registryData.entries.filter(function(e) { return e.kind === 'mcp_server'; });
+
+    // Available WASM extensions
+    if (wasmEntries.length === 0) {
+      wasmList.innerHTML = '<div class="empty-state">No additional WASM extensions available</div>';
+    } else {
+      wasmList.innerHTML = '';
+      for (const entry of wasmEntries) {
+        wasmList.appendChild(renderAvailableExtensionCard(entry));
+      }
+    }
+
+    // MCP servers (show both installed and uninstalled)
+    if (mcpEntries.length === 0) {
+      mcpList.innerHTML = '<div class="empty-state">No MCP servers available</div>';
+    } else {
+      mcpList.innerHTML = '';
+      for (const entry of mcpEntries) {
+        var installedExt = extData.extensions.find(function(e) { return e.name === entry.name; });
+        mcpList.appendChild(renderMcpServerCard(entry, installedExt));
+      }
+    }
+
     // Render tools
     if (toolData.tools.length === 0) {
       toolsTbody.innerHTML = '';
@@ -1202,6 +1263,148 @@ function loadExtensions() {
   });
 }
 
+function renderAvailableExtensionCard(entry) {
+  const card = document.createElement('div');
+  card.className = 'ext-card ext-available';
+
+  const header = document.createElement('div');
+  header.className = 'ext-header';
+
+  const name = document.createElement('span');
+  name.className = 'ext-name';
+  name.textContent = entry.display_name;
+  header.appendChild(name);
+
+  const kind = document.createElement('span');
+  kind.className = 'ext-kind kind-' + entry.kind;
+  kind.textContent = entry.kind;
+  header.appendChild(kind);
+
+  card.appendChild(header);
+
+  const desc = document.createElement('div');
+  desc.className = 'ext-desc';
+  desc.textContent = entry.description;
+  card.appendChild(desc);
+
+  if (entry.keywords && entry.keywords.length > 0) {
+    const kw = document.createElement('div');
+    kw.className = 'ext-keywords';
+    kw.textContent = entry.keywords.join(', ');
+    card.appendChild(kw);
+  }
+
+  const actions = document.createElement('div');
+  actions.className = 'ext-actions';
+
+  const installBtn = document.createElement('button');
+  installBtn.className = 'btn-ext install';
+  installBtn.textContent = 'Install';
+  installBtn.addEventListener('click', function() {
+    installBtn.disabled = true;
+    installBtn.textContent = 'Installing...';
+    apiFetch('/api/extensions/install', {
+      method: 'POST',
+      body: { name: entry.name, kind: entry.kind },
+    }).then(function(res) {
+      if (res.success) {
+        showToast('Installed ' + entry.display_name, 'success');
+      } else {
+        showToast('Install: ' + (res.message || 'unknown error'), 'error');
+      }
+      loadExtensions();
+    }).catch(function(err) {
+      showToast('Install failed: ' + err.message, 'error');
+      loadExtensions();
+    });
+  });
+  actions.appendChild(installBtn);
+
+  card.appendChild(actions);
+  return card;
+}
+
+function renderMcpServerCard(entry, installedExt) {
+  var card = document.createElement('div');
+  card.className = 'ext-card' + (installedExt ? '' : ' ext-available');
+
+  var header = document.createElement('div');
+  header.className = 'ext-header';
+
+  var name = document.createElement('span');
+  name.className = 'ext-name';
+  name.textContent = entry.display_name;
+  header.appendChild(name);
+
+  var kind = document.createElement('span');
+  kind.className = 'ext-kind kind-mcp_server';
+  kind.textContent = 'mcp_server';
+  header.appendChild(kind);
+
+  if (installedExt) {
+    var authDot = document.createElement('span');
+    authDot.className = 'ext-auth-dot ' + (installedExt.authenticated ? 'authed' : 'unauthed');
+    authDot.title = installedExt.authenticated ? 'Authenticated' : 'Not authenticated';
+    header.appendChild(authDot);
+  }
+
+  card.appendChild(header);
+
+  var desc = document.createElement('div');
+  desc.className = 'ext-desc';
+  desc.textContent = entry.description;
+  card.appendChild(desc);
+
+  var actions = document.createElement('div');
+  actions.className = 'ext-actions';
+
+  if (installedExt) {
+    if (!installedExt.active) {
+      var activateBtn = document.createElement('button');
+      activateBtn.className = 'btn-ext activate';
+      activateBtn.textContent = 'Activate';
+      activateBtn.addEventListener('click', function() { activateExtension(installedExt.name); });
+      actions.appendChild(activateBtn);
+    } else {
+      var activeLabel = document.createElement('span');
+      activeLabel.className = 'ext-active-label';
+      activeLabel.textContent = 'Active';
+      actions.appendChild(activeLabel);
+    }
+    var removeBtn = document.createElement('button');
+    removeBtn.className = 'btn-ext remove';
+    removeBtn.textContent = 'Remove';
+    removeBtn.addEventListener('click', function() { removeExtension(installedExt.name); });
+    actions.appendChild(removeBtn);
+  } else {
+    var installBtn = document.createElement('button');
+    installBtn.className = 'btn-ext install';
+    installBtn.textContent = 'Install';
+    installBtn.addEventListener('click', function() {
+      installBtn.disabled = true;
+      installBtn.textContent = 'Installing...';
+      apiFetch('/api/extensions/install', {
+        method: 'POST',
+        body: { name: entry.name, kind: entry.kind },
+      }).then(function(res) {
+        if (res.success) {
+          showToast('Installed ' + entry.display_name, 'success');
+        } else {
+          showToast('Install: ' + (res.message || 'unknown error'), 'error');
+        }
+        loadExtensions();
+      }).catch(function(err) {
+        showToast('Install failed: ' + err.message, 'error');
+        loadExtensions();
+      });
+    });
+    actions.appendChild(installBtn);
+  }
+
+  card.appendChild(actions);
+  return card;
+}
+
 function renderExtensionCard(ext) {
   const card = document.createElement('div');
   card.className = 'ext-card';
@@ -1252,11 +1455,18 @@ function renderExtensionCard(ext) {
   actions.className = 'ext-actions';
 
   if (!ext.active) {
-    const activateBtn = document.createElement('button');
-    activateBtn.className = 'btn-ext activate';
-    activateBtn.textContent = 'Activate';
-    activateBtn.addEventListener('click', () => activateExtension(ext.name));
-    actions.appendChild(activateBtn);
+    if (ext.kind === 'wasm_channel') {
+      const restartLabel = document.createElement('span');
+      restartLabel.className = 'ext-restart-label';
+      restartLabel.textContent = 'Restart to activate';
+      actions.appendChild(restartLabel);
+    } else {
+      const activateBtn = document.createElement('button');
+      activateBtn.className = 'btn-ext activate';
+      activateBtn.textContent = 'Activate';
+      activateBtn.addEventListener('click', () => activateExtension(ext.name));
+      actions.appendChild(activateBtn);
+    }
   } else {
     const activeLabel = document.createElement('span');
     activeLabel.className = 'ext-active-label';
@@ -1264,6 +1474,14 @@ function renderExtensionCard(ext) {
     actions.appendChild(activeLabel);
   }
 
+  if (ext.needs_setup) {
+    const configBtn = document.createElement('button');
+    configBtn.className = 'btn-ext configure';
+    configBtn.textContent = ext.authenticated ? 'Reconfigure' : 'Configure';
+    configBtn.addEventListener('click', () => showConfigureModal(ext.name));
+    actions.appendChild(configBtn);
+  }
+
   const removeBtn = document.createElement('button');
   removeBtn.className = 'btn-ext remove';
   removeBtn.textContent = 'Remove';
@@ -1271,6 +1489,15 @@ function renderExtensionCard(ext) {
   actions.appendChild(removeBtn);
 
   card.appendChild(actions);
+
+  // For active WASM channels, check for pending pairing requests
+  if (ext.active && ext.kind === 'wasm_channel') {
+    const pairingSection = document.createElement('div');
+    pairingSection.className = 'ext-pairing';
+    card.appendChild(pairingSection);
+    loadPairingRequests(ext.name, pairingSection);
+  }
+
   return card;
 }
 
@@ -1286,7 +1513,7 @@ function activateExtension(name) {
         showToast('Opening authentication for ' + name, 'info');
         window.open(res.auth_url, '_blank');
       } else if (res.awaiting_token) {
-        showToast(res.instructions || 'Please provide an API token for ' + name, 'info');
+        showConfigureModal(name);
       } else {
         showToast('Activate failed: ' + res.message, 'error');
       }
@@ -1309,6 +1536,189 @@ function removeExtension(name) {
     .catch((err) => showToast('Remove failed: ' + err.message, 'error'));
 }
 
+function showConfigureModal(name) {
+  apiFetch('/api/extensions/' + encodeURIComponent(name) + '/setup')
+    .then((setup) => {
+      if (!setup.secrets || setup.secrets.length === 0) {
+        showToast('No configuration needed for ' + name, 'info');
+        return;
+      }
+      renderConfigureModal(name, setup.secrets);
+    })
+    .catch((err) => showToast('Failed to load setup: ' + err.message, 'error'));
+}
+
+function renderConfigureModal(name, secrets) {
+  closeConfigureModal();
+  const overlay = document.createElement('div');
+  overlay.className = 'configure-overlay';
+  overlay.addEventListener('click', (e) => {
+    if (e.target === overlay) closeConfigureModal();
+  });
+
+  const modal = document.createElement('div');
+  modal.className = 'configure-modal';
+
+  const header = document.createElement('h3');
+  header.textContent = 'Configure ' + name;
+  modal.appendChild(header);
+
+  const form = document.createElement('div');
+  form.className = 'configure-form';
+
+  const fields = [];
+  for (const secret of secrets) {
+    const field = document.createElement('div');
+    field.className = 'configure-field';
+
+    const label = document.createElement('label');
+    label.textContent = secret.prompt;
+    if (secret.optional) {
+      const opt = document.createElement('span');
+      opt.className = 'field-optional';
+      opt.textContent = ' (optional)';
+      label.appendChild(opt);
+    }
+    field.appendChild(label);
+
+    const inputRow = document.createElement('div');
+    inputRow.className = 'configure-input-row';
+
+    const input = document.createElement('input');
+    input.type = 'password';
+    input.name = secret.name;
+    input.placeholder = secret.provided ? '(already set — leave empty to keep)' : '';
+    input.addEventListener('keydown', (e) => {
+      if (e.key === 'Enter') submitConfigureModal(name, fields);
+    });
+    inputRow.appendChild(input);
+
+    if (secret.provided) {
+      const badge = document.createElement('span');
+      badge.className = 'field-provided';
+      badge.textContent = 'Set';
+      inputRow.appendChild(badge);
+    }
+    if (secret.auto_generate && !secret.provided) {
+      const hint = document.createElement('span');
+      hint.className = 'field-autogen';
+      hint.textContent = 'Auto-generated if empty';
+      inputRow.appendChild(hint);
+    }
+
+    field.appendChild(inputRow);
+    form.appendChild(field);
+    fields.push({ name: secret.name, input: input });
+  }
+
+  modal.appendChild(form);
+
+  const actions = document.createElement('div');
+  actions.className = 'configure-actions';
+
+  const submitBtn = document.createElement('button');
+  submitBtn.className = 'btn-ext activate';
+  submitBtn.textContent = 'Save';
+  submitBtn.addEventListener('click', () => submitConfigureModal(name, fields));
+  actions.appendChild(submitBtn);
+
+  const cancelBtn = document.createElement('button');
+  cancelBtn.className = 'btn-ext remove';
+  cancelBtn.textContent = 'Cancel';
+  cancelBtn.addEventListener('click', closeConfigureModal);
+  actions.appendChild(cancelBtn);
+
+  modal.appendChild(actions);
+  overlay.appendChild(modal);
+  document.body.appendChild(overlay);
+
+  if (fields.length > 0) fields[0].input.focus();
+}
+
+function submitConfigureModal(name, fields) {
+  const secrets = {};
+  for (const f of fields) {
+    if (f.input.value.trim()) {
+      secrets[f.name] = f.input.value.trim();
+    }
+  }
+
+  apiFetch('/api/extensions/' + encodeURIComponent(name) + '/setup', {
+    method: 'POST',
+    body: { secrets },
+  })
+    .then((res) => {
+      closeConfigureModal();
+      if (res.success) {
+        showToast(res.message, 'success');
+      } else {
+        showToast(res.message || 'Configuration failed', 'error');
+      }
+      loadExtensions();
+    })
+    .catch((err) => {
+      showToast('Configuration failed: ' + err.message, 'error');
+    });
+}
+
+function closeConfigureModal() {
+  const existing = document.querySelector('.configure-overlay');
+  if (existing) existing.remove();
+}
+
+// --- Pairing ---
+
+function loadPairingRequests(channel, container) {
+  apiFetch('/api/pairing/' + encodeURIComponent(channel))
+    .then(data => {
+      container.innerHTML = '';
+      if (!data.requests || data.requests.length === 0) return;
+
+      const heading = document.createElement('div');
+      heading.className = 'pairing-heading';
+      heading.textContent = 'Pending pairing requests';
+      container.appendChild(heading);
+
+      data.requests.forEach(req => {
+        const row = document.createElement('div');
+        row.className = 'pairing-row';
+
+        const code = document.createElement('span');
+        code.className = 'pairing-code';
+        code.textContent = req.code;
+        row.appendChild(code);
+
+        const sender = document.createElement('span');
+        sender.className = 'pairing-sender';
+        sender.textContent = 'from ' + req.sender_id;
+        row.appendChild(sender);
+
+        const btn = document.createElement('button');
+        btn.className = 'btn-ext activate';
+        btn.textContent = 'Approve';
+        btn.addEventListener('click', () => approvePairing(channel, req.code, container));
+        row.appendChild(btn);
+
+        container.appendChild(row);
+      });
+    })
+    .catch(() => {});
+}
+
+function approvePairing(channel, code, container) {
+  apiFetch('/api/pairing/' + encodeURIComponent(channel) + '/approve', {
+    method: 'POST',
+    body: { code },
+  }).then(res => {
+    if (res.success) {
+      showToast('Pairing approved', 'success');
+      loadPairingRequests(channel, container);
+    } else {
+      showToast(res.message || 'Approve failed', 'error');
+    }
+  }).catch(err => showToast('Error: ' + err.message, 'error'));
+}
+
 // --- Jobs ---
 
 let currentJobId = null;
@@ -2049,13 +2459,72 @@ function startGatewayStatusPolling() {
   gatewayStatusInterval = setInterval(fetchGatewayStatus, 30000);
 }
 
+function formatTokenCount(n) {
+  if (n == null || n === 0) return '0';
+  if (n >= 1000000) return (n / 1000000).toFixed(1) + 'M';
+  if (n >= 1000) return (n / 1000).toFixed(1) + 'k';
+  return '' + n;
+}
+
+function formatCost(costStr) {
+  if (!costStr) return '$0.00';
+  var n = parseFloat(costStr);
+  if (n < 0.01) return '$' + n.toFixed(4);
+  return '$' + n.toFixed(2);
+}
+
+function shortModelName(model) {
+  // Strip provider prefix and shorten common model names
+  var m = model.indexOf('/') >= 0 ? model.split('/').pop() : model;
+  // Shorten dated suffixes
+  m = m.replace(/-20\d{6}$/, '');
+  return m;
+}
+
 function fetchGatewayStatus() {
-  apiFetch('/api/gateway/status').then((data) => {
-    const popover = document.getElementById('gateway-popover');
-    popover.innerHTML = '<div class="gw-stat"><span>SSE clients</span><span>' + (data.sse_clients || 0) + '</span></div>'
-      + '<div class="gw-stat"><span>Log clients</span><span>' + (data.log_clients || 0) + '</span></div>'
-      + '<div class="gw-stat"><span>Uptime</span><span>' + formatDuration(data.uptime_secs) + '</span></div>';
-  }).catch(() => {});
+  apiFetch('/api/gateway/status').then(function(data) {
+    var popover = document.getElementById('gateway-popover');
+    var html = '';
+
+    // Connection info
+    html += '<div class="gw-section-label">Connections</div>';
+    html += '<div class="gw-stat"><span>SSE</span><span>' + (data.sse_connections || 0) + '</span></div>';
+    html += '<div class="gw-stat"><span>WebSocket</span><span>' + (data.ws_connections || 0) + '</span></div>';
+    html += '<div class="gw-stat"><span>Uptime</span><span>' + formatDuration(data.uptime_secs) + '</span></div>';
+
+    // Cost tracker
+    if (data.daily_cost != null) {
+      html += '<div class="gw-divider"></div>';
+      html += '<div class="gw-section-label">Cost Today</div>';
+      html += '<div class="gw-stat"><span>Spent</span><span>' + formatCost(data.daily_cost) + '</span></div>';
+      if (data.actions_this_hour != null) {
+        html += '<div class="gw-stat"><span>Actions/hr</span><span>' + data.actions_this_hour + '</span></div>';
+      }
+    }
+
+    // Per-model token usage
+    if (data.model_usage && data.model_usage.length > 0) {
+      html += '<div class="gw-divider"></div>';
+      html += '<div class="gw-section-label">Token Usage</div>';
+      data.model_usage.sort(function(a, b) {
+        return (b.input_tokens + b.output_tokens) - (a.input_tokens + a.output_tokens);
+      });
+      for (var i = 0; i < data.model_usage.length; i++) {
+        var m = data.model_usage[i];
+        var name = escapeHtml(shortModelName(m.model));
+        html += '<div class="gw-model-row">'
+          + '<span class="gw-model-name">' + name + '</span>'
+          + '<span class="gw-model-cost">' + escapeHtml(formatCost(m.cost)) + '</span>'
+          + '</div>';
+        html += '<div class="gw-token-detail">'
+          + '<span>in: ' + formatTokenCount(m.input_tokens) + '</span>'
+          + '<span>out: ' + formatTokenCount(m.output_tokens) + '</span>'
+          + '</div>';
+      }
+    }
+
+    popover.innerHTML = html;
+  }).catch(function() {});
 }
 
 // Show/hide popover on hover
@@ -2066,34 +2535,160 @@ document.getElementById('gateway-status-trigger').addEventListener('mouseleave',
   document.getElementById('gateway-popover').classList.remove('visible');
 });
 
+// --- TEE attestation ---
+
+let teeInfo = null;
+let teeReportCache = null;
+let teeReportLoading = false;
+
+function teeApiBase() {
+  var parts = window.location.hostname.split('.');
+  if (parts.length < 2) return null;
+  var domain = parts.slice(1).join('.');
+  return window.location.protocol + '//api.' + domain;
+}
+
+function teeInstanceName() {
+  return window.location.hostname.split('.')[0];
+}
+
+function checkTeeStatus() {
+  var base = teeApiBase();
+  if (!base) return;
+  var name = teeInstanceName();
+  fetch(base + '/instances/' + encodeURIComponent(name) + '/attestation').then(function(res) {
+    if (!res.ok) throw new Error(res.status);
+    return res.json();
+  }).then(function(data) {
+    teeInfo = data;
+    document.getElementById('tee-shield').style.display = 'flex';
+  }).catch(function() {});
+}
+
+function fetchTeeReport() {
+  if (teeReportCache) {
+    renderTeePopover(teeReportCache);
+    return;
+  }
+  if (teeReportLoading) return;
+  teeReportLoading = true;
+  var base = teeApiBase();
+  if (!base) return;
+  var popover = document.getElementById('tee-popover');
+  popover.innerHTML = '<div class="tee-popover-loading">Loading attestation report...</div>';
+  fetch(base + '/attestation/report').then(function(res) {
+    if (!res.ok) throw new Error(res.status);
+    return res.json();
+  }).then(function(data) {
+    teeReportCache = data;
+    renderTeePopover(data);
+  }).catch(function() {
+    popover.innerHTML = '<div class="tee-popover-loading">Could not load attestation report</div>';
+  }).finally(function() {
+    teeReportLoading = false;
+  });
+}
+
+function renderTeePopover(report) {
+  var popover = document.getElementById('tee-popover');
+  var digest = (teeInfo && teeInfo.image_digest) || 'N/A';
+  var fingerprint = report.tls_certificate_fingerprint || 'N/A';
+  var reportData = report.report_data || '';
+  var vmConfig = report.vm_config || 'N/A';
+  var truncated = reportData.length > 32 ? reportData.slice(0, 32) + '...' : reportData;
+  popover.innerHTML = '<div class="tee-popover-title">'
+    + '<svg width="14" height="14" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M12 22s8-4 8-10V5l-8-3-8 3v7c0 6 8 10 8 10z"/></svg>'
+    + 'TEE Attestation</div>'
+    + '<div class="tee-field"><div class="tee-field-label">Image Digest</div>'
+    + '<div class="tee-field-value">' + escapeHtml(digest) + '</div></div>'
+    + '<div class="tee-field"><div class="tee-field-label">TLS Certificate Fingerprint</div>'
+    + '<div class="tee-field-value">' + escapeHtml(fingerprint) + '</div></div>'
+    + '<div class="tee-field"><div class="tee-field-label">Report Data</div>'
+    + '<div class="tee-field-value">' + escapeHtml(truncated) + '</div></div>'
+    + '<div class="tee-field"><div class="tee-field-label">VM Config</div>'
+    + '<div class="tee-field-value">' + escapeHtml(vmConfig) + '</div></div>'
+    + '<div class="tee-popover-actions">'
+    + '<button class="tee-btn-copy" onclick="copyTeeReport()">Copy Full Report</button></div>';
+}
+
+function copyTeeReport() {
+  if (!teeReportCache) return;
+  var combined = Object.assign({}, teeReportCache, teeInfo || {});
+  navigator.clipboard.writeText(JSON.stringify(combined, null, 2)).then(function() {
+    showToast('Attestation report copied', 'success');
+  }).catch(function() {
+    showToast('Failed to copy report', 'error');
+  });
+}
+
+document.getElementById('tee-shield').addEventListener('mouseenter', function() {
+  fetchTeeReport();
+  document.getElementById('tee-popover').classList.add('visible');
+});
+document.getElementById('tee-shield').addEventListener('mouseleave', function() {
+  document.getElementById('tee-popover').classList.remove('visible');
+});
+
 // --- Extension install ---
 
-function installExtension() {
-  const name = document.getElementById('ext-install-name').value.trim();
+function installWasmExtension() {
+  var name = document.getElementById('wasm-install-name').value.trim();
   if (!name) {
     showToast('Extension name is required', 'error');
     return;
   }
-  const url = document.getElementById('ext-install-url').value.trim();
-  const kind = document.getElementById('ext-install-kind').value;
+  var url = document.getElementById('wasm-install-url').value.trim();
+  if (!url) {
+    showToast('URL to .tar.gz bundle is required', 'error');
+    return;
+  }
 
   apiFetch('/api/extensions/install', {
     method: 'POST',
-    body: { name, url: url || undefined, kind },
-  }).then((res) => {
+    body: { name: name, url: url, kind: 'wasm_tool' },
+  }).then(function(res) {
     if (res.success) {
       showToast('Installed ' + name, 'success');
-      document.getElementById('ext-install-name').value = '';
-      document.getElementById('ext-install-url').value = '';
+      document.getElementById('wasm-install-name').value = '';
+      document.getElementById('wasm-install-url').value = '';
       loadExtensions();
     } else {
       showToast('Install failed: ' + (res.message || 'unknown error'), 'error');
     }
-  }).catch((err) => {
+  }).catch(function(err) {
     showToast('Install failed: ' + err.message, 'error');
   });
 }
 
+function addMcpServer() {
+  var name = document.getElementById('mcp-install-name').value.trim();
+  if (!name) {
+    showToast('Server name is required', 'error');
+    return;
+  }
+  var url = document.getElementById('mcp-install-url').value.trim();
+  if (!url) {
+    showToast('MCP server URL is required', 'error');
+    return;
+  }
+
+  apiFetch('/api/extensions/install', {
+    method: 'POST',
+    body: { name: name, url: url, kind: 'mcp_server' },
+  }).then(function(res) {
+    if (res.success) {
+      showToast('Added MCP server ' + name, 'success');
+      document.getElementById('mcp-install-name').value = '';
+      document.getElementById('mcp-install-url').value = '';
+      loadExtensions();
+    } else {
+      showToast('Failed to add MCP server: ' + (res.message || 'unknown error'), 'error');
+    }
+  }).catch(function(err) {
+    showToast('Failed to add MCP server: ' + err.message, 'error');
+  });
+}
+
 // --- Keyboard shortcuts ---
 
 document.addEventListener('keydown', (e) => {
@@ -2101,10 +2696,10 @@ document.addEventListener('keydown', (e) => {
   const tag = (e.target.tagName || '').toLowerCase();
   const inInput = tag === 'input' || tag === 'textarea';
 
-  // Mod+1-6: switch tabs
-  if (mod && e.key >= '1' && e.key <= '6') {
+  // Mod+1-5: switch tabs
+  if (mod && e.key >= '1' && e.key <= '5') {
     e.preventDefault();
-    const tabs = ['chat', 'memory', 'jobs', 'routines', 'logs', 'extensions'];
+    const tabs = ['chat', 'memory', 'jobs', 'routines', 'extensions'];
     const idx = parseInt(e.key) - 1;
     if (tabs[idx]) switchTab(tabs[idx]);
     return;

src/channels/web/static/index.html

@@ -4,6 +4,9 @@
   <meta charset="UTF-8">
   <meta name="viewport" content="width=device-width, initial-scale=1.0">
   <title>IronClaw</title>
+  <link rel="preconnect" href="https://fonts.googleapis.com">
+  <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
+  <link href="https://fonts.googleapis.com/css2?family=DM+Sans:wght@400;500;600;700&family=IBM+Plex+Mono:wght@400;500;600&display=swap" rel="stylesheet">
   <link rel="stylesheet" href="/style.css">
   <script
     src="https://cdn.jsdelivr.net/npm/[email protected]/lib/marked.umd.min.js"
@@ -36,10 +39,17 @@
       <button class="active" data-tab="chat">Chat</button>
       <button data-tab="memory">Memory</button>
       <button data-tab="jobs">Jobs</button>
-      <button data-tab="logs">Logs</button>
       <button data-tab="routines">Routines</button>
       <button data-tab="extensions">Extensions</button>
       <div class="spacer"></div>
+      <button class="status-logs-btn" data-tab="logs" title="Logs">Logs</button>
+      <div class="tee-shield" id="tee-shield" style="display:none" title="Running in a Trusted Execution Environment">
+        <svg width="16" height="16" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round">
+          <path d="M12 22s8-4 8-10V5l-8-3-8 3v7c0 6 8 10 8 10z"/>
+        </svg>
+        <span id="tee-shield-label">TEE Verified</span>
+        <div class="tee-popover" id="tee-popover"></div>
+      </div>
       <div class="status" id="gateway-status-trigger">
         <div class="dot" id="sse-dot"></div>
         <span id="sse-status">Connected</span>
@@ -127,6 +137,12 @@
     <div class="tab-panel" id="tab-logs">
       <div class="logs-container">
         <div class="logs-toolbar">
+          <select id="logs-server-level" onchange="setServerLogLevel(this.value)" title="Server-side log level (changes what the server emits)">
+            <option value="error">Server: ERROR</option>
+            <option value="warn">Server: WARN</option>
+            <option value="info" selected>Server: INFO</option>
+            <option value="debug">Server: DEBUG</option>
+          </select>
           <select id="logs-level-filter">
             <option value="all">All Levels</option>
             <option value="ERROR">Error</option>
@@ -172,34 +188,42 @@
     <!-- Extensions Tab -->
     <div class="tab-panel" id="tab-extensions">
       <div class="extensions-container">
-        <div class="extensions-section">
-          <h3>Install Extension</h3>
-          <div class="ext-install-form" id="ext-install-form">
-            <input type="text" id="ext-install-name" placeholder="Extension name (required)">
-            <input type="text" id="ext-install-url" placeholder="URL (optional)">
-            <select id="ext-install-kind">
-              <option value="mcp_server">MCP Server</option>
-              <option value="wasm_tool">WASM Tool</option>
-              <option value="wasm_channel">WASM Channel</option>
-            </select>
-            <button onclick="installExtension()">Install</button>
-          </div>
-        </div>
         <div class="extensions-section">
           <h3>Installed Extensions</h3>
           <div class="extensions-list" id="extensions-list">
             <div class="empty-state">Loading extensions...</div>
           </div>
         </div>
+        <div class="extensions-section" id="available-wasm-section">
+          <h3>Available WASM Extensions</h3>
+          <div class="extensions-list" id="available-wasm-list">
+            <div class="empty-state">Loading...</div>
+          </div>
+        </div>
+        <div class="extensions-section">
+          <h3>Install WASM Extension</h3>
+          <div class="ext-install-form">
+            <input type="text" id="wasm-install-name" placeholder="Extension name">
+            <input type="text" id="wasm-install-url" placeholder="URL to .tar.gz bundle">
+            <button onclick="installWasmExtension()">Install</button>
+          </div>
+        </div>
+        <div class="extensions-section">
+          <h3>MCP Servers</h3>
+          <div class="extensions-list" id="mcp-servers-list">
+            <div class="empty-state">Loading...</div>
+          </div>
+          <h4>Add Custom MCP Server</h4>
+          <div class="ext-install-form">
+            <input type="text" id="mcp-install-name" placeholder="Server name">
+            <input type="text" id="mcp-install-url" placeholder="MCP server URL (https://...)">
+            <button onclick="addMcpServer()">Add</button>
+          </div>
+        </div>
         <div class="extensions-section">
           <h3>Registered Tools</h3>
           <table class="tools-table" id="tools-table">
-            <thead>
-              <tr>
-                <th>Name</th>
-                <th>Description</th>
-              </tr>
-            </thead>
+            <thead><tr><th>Name</th><th>Description</th></tr></thead>
             <tbody id="tools-tbody"></tbody>
           </table>
           <div class="empty-state" id="tools-empty" style="display:none">No tools registered</div>

src/channels/web/types.rs

@@ -346,6 +346,9 @@ pub struct ExtensionInfo {
     pub authenticated: bool,
     pub active: bool,
     pub tools: Vec<String>,
+    /// Whether this extension has configurable secrets (setup schema).
+    #[serde(default)]
+    pub needs_setup: bool,
 }
 
 #[derive(Debug, Serialize)]
@@ -371,6 +374,31 @@ pub struct InstallExtensionRequest {
     pub kind: Option<String>,
 }
 
+// --- Extension Setup ---
+
+#[derive(Debug, Serialize)]
+pub struct ExtensionSetupResponse {
+    pub name: String,
+    pub kind: String,
+    pub secrets: Vec<SecretFieldInfo>,
+}
+
+#[derive(Debug, Serialize)]
+pub struct SecretFieldInfo {
+    pub name: String,
+    pub prompt: String,
+    pub optional: bool,
+    /// Whether this secret is already stored.
+    pub provided: bool,
+    /// Whether the secret will be auto-generated if left empty.
+    pub auto_generate: bool,
+}
+
+#[derive(Debug, Deserialize)]
+pub struct ExtensionSetupRequest {
+    pub secrets: std::collections::HashMap<String, String>,
+}
+
 #[derive(Debug, Serialize)]
 pub struct ActionResponse {
     pub success: bool,
@@ -408,6 +436,50 @@ impl ActionResponse {
     }
 }
 
+// --- Registry ---
+
+#[derive(Debug, Serialize)]
+pub struct RegistryEntryInfo {
+    pub name: String,
+    pub display_name: String,
+    pub kind: String,
+    pub description: String,
+    pub keywords: Vec<String>,
+    pub installed: bool,
+}
+
+#[derive(Debug, Serialize)]
+pub struct RegistrySearchResponse {
+    pub entries: Vec<RegistryEntryInfo>,
+}
+
+#[derive(Debug, Deserialize)]
+pub struct RegistrySearchQuery {
+    pub query: Option<String>,
+}
+
+// --- Pairing ---
+
+#[derive(Debug, Serialize)]
+pub struct PairingListResponse {
+    pub channel: String,
+    pub requests: Vec<PairingRequestInfo>,
+}
+
+#[derive(Debug, Serialize)]
+pub struct PairingRequestInfo {
+    pub code: String,
+    pub sender_id: String,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub meta: Option<serde_json::Value>,
+    pub created_at: String,
+}
+
+#[derive(Debug, Deserialize)]
+pub struct PairingApproveRequest {
+    pub code: String,
+}
+
 // --- Skills ---
 
 #[derive(Debug, Serialize)]

src/channels/web/ws.rs

@@ -477,6 +477,7 @@ mod tests {
             workspace: None,
             session_manager: None,
             log_broadcaster: None,
+            log_level_handle: None,
             extension_manager: None,
             tool_registry: None,
             store: None,
@@ -489,6 +490,9 @@ mod tests {
             skill_registry: None,
             skill_catalog: None,
             chat_rate_limiter: crate::channels::web::server::RateLimiter::new(30, 60),
+            registry_entries: Vec::new(),
+            cost_guard: None,
+            startup_time: std::time::Instant::now(),
         }
     }
 }