refactor: address PR review feedback on test credentials

- Fix TEST_CRYPTO_KEY doc comment ("32-byte hex" → "32-character key string") - Rename confusing "real"/"fake" Anthropic constant names and values - Change TEST_STRIPE_KEY from "sk-live" to "sk_test_fake123" to avoid scanners - Use test_secrets_store() helper in orchestrator and http tool tests - Clarify config_round_trip.rs doc comment about integration test visibility Co-Authored-By: Claude Opus 4.6 <[email protected]>
refactor: replace real Telegram bot token with obviously fake test stub
2026-08-30 08:17:53 +00:00 · 2026-03-10 11:07:17 -07:00 · 2026-03-09 22:19:57 -07:00 · 2026-03-09 22:16:36 -07:00
131 changed files with 2518 additions and 10816 deletions
@@ -64,7 +64,7 @@ If the event needs custom UI (cards, badges, etc.), add styles. Follow the exist
 Identify where in the backend this event should be triggered. Common locations:
 - `src/agent/agent_loop.rs` - During message processing or tool execution
- `src/worker/job.rs` - During job execution
+- `src/agent/worker.rs` - During job execution
 - `src/agent/heartbeat.rs` - During periodic execution
 Use the existing pattern:
@@ -1,50 +0,0 @@
 ## Summary
 <!-- 2-5 bullet points: what changed and why -->
 -
 ## Change Type
 <!-- Check one -->
 - [ ] Bug fix
 - [ ] New feature
 - [ ] Refactor
 - [ ] Documentation
 - [ ] CI/Infrastructure
 - [ ] Security
 - [ ] Dependencies
 ## Linked Issue
 <!-- Closes #N, or "None" -->
 ## Validation
 <!-- How did you verify this works? -->
 - [ ] `cargo fmt`
 - [ ] `cargo clippy --all --benches --tests --examples --all-features`
 - [ ] Relevant tests pass: <!-- list specific tests -->
 - [ ] Manual testing: <!-- describe what you tested -->
 ## Security Impact
 <!-- Does this change affect: permissions, network calls, secrets, file access, tool execution, sandbox policy? If yes, describe. If no, write "None". -->
 ## Database Impact
 <!-- Does this add/modify migrations, change schema, or affect both PostgreSQL and libSQL? If yes, describe. If no, write "None". -->
 ## Blast Radius
 <!-- What subsystems does this touch? What could break? -->
 ## Rollback Plan
 <!-- How to revert if this causes problems? For Track C changes, this is mandatory. -->
 ---
 **Review track**: <!-- A (docs/tests/chore) | B (feature/refactor) | C (security/runtime/DB/CI) -->
@@ -144,8 +144,6 @@ jobs:
      - name: Patch manifests with WASM checksums
        if: ${{ needs.plan.outputs.publishing == 'true' }}
        shell: bash
        env:
          RELEASE_TAG: ${{ github.ref_name }}
        run: |
          CHECKSUMS="target/distrib/checksums.txt"
          if [ ! -f "$CHECKSUMS" ]; then
@@ -156,17 +154,12 @@ jobs:
          while IFS= read -r line; do
            sha256=$(echo "$line" | awk '{print $1}')
            filename=$(echo "$line" | awk '{print $2}')
-            # Strip -{version}-wasm32-wasip2.tar.gz to get the extension name.
+            name=$(echo "$filename" | sed 's/-wasm32-wasip2\.tar\.gz$//')
            # Use '.*' (greedy) so pre-release suffixes like -alpha.1 are consumed too.
            name=$(echo "$filename" | sed 's/-[0-9].*-wasm32-wasip2\.tar\.gz$//')
            url="https://github.com/nearai/ironclaw/releases/download/${RELEASE_TAG}/${filename}"
            for manifest in registry/tools/${name}.json registry/channels/${name}.json; do
              if [ -f "$manifest" ]; then
-                jq --arg sha "$sha256" --arg url "$url" \
+                jq --arg sha "$sha256" '.artifacts["wasm32-wasip2"].sha256 = $sha' "$manifest" > "${manifest}.tmp" && mv "${manifest}.tmp" "$manifest"
-                  '.artifacts["wasm32-wasip2"].sha256 = $sha | .artifacts["wasm32-wasip2"].url = $url' \
+                echo "Patched $manifest with sha256=$sha256"
                  "$manifest" > "${manifest}.tmp" && mv "${manifest}.tmp" "$manifest"
                echo "Patched $manifest with sha256=$sha256 url=$url"
              fi
            done
          done < "$CHECKSUMS"
@@ -275,41 +268,21 @@ jobs:
          for manifest in registry/tools/*.json registry/channels/*.json; do
            [ -f "$manifest" ] || continue
-            # file_stem: JSON filename without extension (e.g. "slack" for slack.json).
+            name=$(jq -r '.name' "$manifest")
            # Used for the bundle filename and CI manifest lookup, so patching always
            # finds the right file regardless of whether manifest.name matches the filename.
            file_stem=$(basename "$manifest" .json)
            # ext_name: the manifest's .name field (e.g. "slack-tool").
            # Used for file names *inside* the archive — the installer extracts by manifest.name.
            ext_name=$(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' "$manifest")
            ext_version=$(jq -r '.version // ""' "$manifest")
            if [ ! -d "$source_dir" ]; then
-              echo "::warning::Source dir '$source_dir' not found for '$file_stem', skipping"
+              echo "::warning::Source dir '$source_dir' not found for '$name', skipping"
              continue
            fi
-            # Skip rebuild if this exact version was already built and checksummed.
+            echo "=== Building $name from $source_dir ==="
            # Checks that (1) the manifest already has a sha256, and (2) the version
            # embedded in the existing artifact URL matches the current manifest version.
            # This ensures stable checksums: only rebuild when the source version changes.
            existing_sha=$(jq -r '.artifacts["wasm32-wasip2"].sha256 // ""' "$manifest")
            existing_url=$(jq -r '.artifacts["wasm32-wasip2"].url // ""' "$manifest")
            url_version=$(echo "$existing_url" | sed -n 's/.*-\([0-9].*\)-wasm32-wasip2\.tar\.gz$/\1/p')
            if [[ -n "$ext_version" && "$url_version" == "$ext_version" && -n "$existing_sha" ]]; then
              echo "=== Skipping $file_stem v$ext_version — already checksummed at $existing_url ==="
              continue
            fi
            echo "=== Building $file_stem ($ext_name) v$ext_version from $source_dir ==="
            # Build WASM component
            cargo component build --release --manifest-path "$source_dir/Cargo.toml" || {
-              echo "::warning::Build failed for '$file_stem', skipping"
+              echo "::warning::Build failed for '$name', skipping"
              continue
            }
@@ -325,36 +298,30 @@ jobs:
            done
            if [ -z "$wasm_path" ]; then
-              echo "::warning::No WASM output found for '$file_stem', skipping"
+              echo "::warning::No WASM output found for '$name', skipping"
              continue
            fi
-            # Archive contents use ext_name (manifest .name) — the installer extracts
+            # Copy files with standardized names for the archive
-            # files by manifest.name, so these must match even when file_stem differs.
+            cp "$wasm_path" "target/wasm-bundles/${name}.wasm"
            cp "$wasm_path" "target/wasm-bundles/${ext_name}.wasm"
            caps_path="$source_dir/$caps_file"
            if [ -f "$caps_path" ]; then
-              cp "$caps_path" "target/wasm-bundles/${ext_name}.capabilities.json"
+              cp "$caps_path" "target/wasm-bundles/${name}.capabilities.json"
            else
-              echo "::warning::No capabilities file at '$caps_path' for '$file_stem'"
+              echo "::warning::No capabilities file at '$caps_path' for '$name'"
            fi
-            # Bundle filename uses file_stem so CI patching can find the manifest by
+            # Create tar.gz bundle
-            # filename (e.g. slack-0.1.0-wasm32-wasip2.tar.gz → registry/tools/slack.json).
+            bundle="target/wasm-bundles/${name}-wasm32-wasip2.tar.gz"
-            bundle="target/wasm-bundles/${file_stem}-${ext_version}-wasm32-wasip2.tar.gz"
+            (cd target/wasm-bundles && if [ -f "${name}.capabilities.json" ]; then tar czf "${name}-wasm32-wasip2.tar.gz" "${name}.wasm" "${name}.capabilities.json"; else tar czf "${name}-wasm32-wasip2.tar.gz" "${name}.wasm"; fi)
            (cd target/wasm-bundles && if [ -f "${ext_name}.capabilities.json" ]; then
              tar czf "${file_stem}-${ext_version}-wasm32-wasip2.tar.gz" "${ext_name}.wasm" "${ext_name}.capabilities.json"
            else
              tar czf "${file_stem}-${ext_version}-wasm32-wasip2.tar.gz" "${ext_name}.wasm"
            fi)
            # Compute SHA256
            sha256=$(sha256sum "$bundle" | cut -d' ' -f1)
-            echo "$sha256  ${file_stem}-${ext_version}-wasm32-wasip2.tar.gz" >> target/wasm-bundles/checksums.txt
+            echo "$sha256  ${name}-wasm32-wasip2.tar.gz" >> target/wasm-bundles/checksums.txt
            # Clean up intermediate files
-            rm -f "target/wasm-bundles/${ext_name}.wasm" "target/wasm-bundles/${ext_name}.capabilities.json"
+            rm -f "target/wasm-bundles/${name}.wasm" "target/wasm-bundles/${name}.capabilities.json"
            echo "  -> $bundle ($sha256)"
          done
@@ -460,10 +427,8 @@ jobs:
        with:
          name: artifacts-wasm-extensions
          path: target/wasm-bundles/
-      - name: Patch manifests with SHA256 and version-pinned URL
+      - name: Patch manifests with SHA256
        shell: bash
        env:
          RELEASE_TAG: ${{ github.ref_name }}
        run: |
          CHECKSUMS="target/wasm-bundles/checksums.txt"
          if [ ! -f "$CHECKSUMS" ]; then
@@ -474,17 +439,12 @@ jobs:
          while IFS= read -r line; do
            sha256=$(echo "$line" | awk '{print $1}')
            filename=$(echo "$line" | awk '{print $2}')
-            # Strip -{version}-wasm32-wasip2.tar.gz to get the extension name.
+            name=$(echo "$filename" | sed 's/-wasm32-wasip2\.tar\.gz$//')
            # Use '.*' (greedy) so pre-release suffixes like -alpha.1 are consumed too.
            name=$(echo "$filename" | sed 's/-[0-9].*-wasm32-wasip2\.tar\.gz$//')
            url="https://github.com/nearai/ironclaw/releases/download/${RELEASE_TAG}/${filename}"
            for manifest in registry/tools/${name}.json registry/channels/${name}.json; do
              if [ -f "$manifest" ]; then
-                jq --arg sha "$sha256" --arg url "$url" \
+                jq --arg sha "$sha256" '.artifacts["wasm32-wasip2"].sha256 = $sha' "$manifest" > "${manifest}.tmp" && mv "${manifest}.tmp" "$manifest"
-                  '.artifacts["wasm32-wasip2"].sha256 = $sha | .artifacts["wasm32-wasip2"].url = $url' \
+                echo "Patched $manifest with sha256=$sha256"
                  "$manifest" > "${manifest}.tmp" && mv "${manifest}.tmp" "$manifest"
                echo "Patched $manifest with sha256=$sha256 url=$url"
              fi
            done
          done < "$CHECKSUMS"
@@ -501,8 +461,8 @@ jobs:
            git commit -m "chore: update WASM artifact SHA256 checksums [skip ci]"
            git push origin "$BRANCH"
            gh pr create \
-              --title "chore: update WASM artifact checksums and version-pinned URLs" \
+              --title "chore: update WASM artifact SHA256 checksums" \
-              --body "Auto-generated by release CI. Updates SHA256 checksums and version-pinned artifact URLs in registry manifests to match the released WASM artifacts. Only extensions whose version changed since the last release are included." \
+              --body "Auto-generated by release CI. Updates SHA256 checksums in registry manifests to match the released WASM artifacts." \
              --base main \
              --head "$BRANCH"
          fi
@@ -28,4 +28,3 @@ trace_*.json
 # Local Claude Code settings (machine-specific, should not be committed)
 .claude/settings.local.json
 .worktrees/
@@ -7,90 +7,6 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [Unreleased]
 ## [0.18.0](https://github.com/nearai/ironclaw/compare/v0.17.0...v0.18.0) - 2026-03-11
 ### Other
 - Merge pull request #907 from nearai/staging-promote/b0214fef-22930316561
 - promote staging to main (2026-03-10 15:19 UTC) ([#865](https://github.com/nearai/ironclaw/pull/865))
 - Merge pull request #830 from nearai/staging-promote/3a2989d0-22888378864
 - update WASM artifact SHA256 checksums [skip ci] ([#876](https://github.com/nearai/ironclaw/pull/876))
 ## [0.17.0](https://github.com/nearai/ironclaw/compare/v0.16.1...v0.17.0) - 2026-03-10
 ### Added
 - *(llm)* per-provider unsupported parameter filtering (#749, #728) ([#809](https://github.com/nearai/ironclaw/pull/809))
 - persist user_id in save_job and expose job_id on routine runs ([#709](https://github.com/nearai/ironclaw/pull/709))
 - *(ci)* chained promotion PRs with multi-agent Claude review ([#776](https://github.com/nearai/ironclaw/pull/776))
 - add background sandbox reaper for orphaned Docker containers ([#634](https://github.com/nearai/ironclaw/pull/634))
 - *(wasm)* lazy schema injection on WASM tool errors ([#638](https://github.com/nearai/ironclaw/pull/638))
 - add AWS Bedrock LLM provider via native Converse API ([#713](https://github.com/nearai/ironclaw/pull/713))
 - full image support across all channels ([#725](https://github.com/nearai/ironclaw/pull/725))
 - *(skills)* exclude_keywords veto in skill activation scoring ([#688](https://github.com/nearai/ironclaw/pull/688))
 - *(mcp)* transport abstraction, stdio/UDS transports, and OAuth fixes ([#721](https://github.com/nearai/ironclaw/pull/721))
 - add PID-based gateway lock to prevent multiple instances ([#717](https://github.com/nearai/ironclaw/pull/717))
 - configurable LLM request timeout via LLM_REQUEST_TIMEOUT_SECS ([#615](https://github.com/nearai/ironclaw/pull/615)) ([#630](https://github.com/nearai/ironclaw/pull/630))
 - *(timezone)* add timezone-aware session context ([#671](https://github.com/nearai/ironclaw/pull/671))
 - *(setup)* Anthropic OAuth onboarding with setup-token support ([#384](https://github.com/nearai/ironclaw/pull/384))
 - *(llm)* add Google Gemini, AWS Bedrock, io.net, Mistral, Yandex, and Cloudflare WS AI providers ([#676](https://github.com/nearai/ironclaw/pull/676))
 - unified thread model for web gateway ([#607](https://github.com/nearai/ironclaw/pull/607))
 - WASM channel attachments with LLM pipeline integration ([#596](https://github.com/nearai/ironclaw/pull/596))
 - enable Anthropic prompt caching via automatic cache_control injection ([#660](https://github.com/nearai/ironclaw/pull/660))
 - *(routines)* approval context for autonomous job execution ([#577](https://github.com/nearai/ironclaw/pull/577))
 - *(llm)* declarative provider registry ([#618](https://github.com/nearai/ironclaw/pull/618))
 - *(gateway)* show IronClaw version in status popover [skip-regression-check] ([#636](https://github.com/nearai/ironclaw/pull/636))
 - Wire memory hygiene retention policy into heartbeat loop ([#629](https://github.com/nearai/ironclaw/pull/629))
 ### Fixed
 - *(ci)* run fmt + clippy on staging PRs, skip Windows clippy [skip-regression-check] ([#802](https://github.com/nearai/ironclaw/pull/802))
 - *(ci)* clean up staging pipeline — remove hacks, skip redundant checks [skip-regression-check] ([#794](https://github.com/nearai/ironclaw/pull/794))
 - *(ci)* secrets can't be used in step if conditions [skip-regression-check] ([#787](https://github.com/nearai/ironclaw/pull/787))
 - prevent irreversible context loss when compaction archive write fails ([#754](https://github.com/nearai/ironclaw/pull/754))
 - button styles ([#637](https://github.com/nearai/ironclaw/pull/637))
 - *(mcp)* JSON-RPC spec compliance — flexible id, correct notification format ([#685](https://github.com/nearai/ironclaw/pull/685))
 - preserve tool-call history across thread hydration ([#568](https://github.com/nearai/ironclaw/pull/568)) ([#670](https://github.com/nearai/ironclaw/pull/670))
 - CLI commands ignore runtime DATABASE_BACKEND when both features compiled ([#740](https://github.com/nearai/ironclaw/pull/740))
 - *(web)* prevent fetch error when hostname is an IP address in TEE check ([#672](https://github.com/nearai/ironclaw/pull/672))
 - add timezone conversion support to time tool ([#687](https://github.com/nearai/ironclaw/pull/687))
 - standardize libSQL timestamps as RFC 3339 UTC ([#683](https://github.com/nearai/ironclaw/pull/683))
 - *(docker)* bind postgres to localhost only ([#686](https://github.com/nearai/ironclaw/pull/686))
 - *(repl)* skip /quit on EOF when stdin is not a TTY ([#724](https://github.com/nearai/ironclaw/pull/724))
 - *(web)* prevent Enter key from sending message during IME composition ([#715](https://github.com/nearai/ironclaw/pull/715))
 - *(config)* init_secrets no longer overwrites entire config ([#726](https://github.com/nearai/ironclaw/pull/726))
 - *(cli)* status command ignores config.toml and settings.json ([#354](https://github.com/nearai/ironclaw/pull/354)) ([#734](https://github.com/nearai/ironclaw/pull/734))
 - *(setup)* preserve model name when re-running onboarding with same provider ([#600](https://github.com/nearai/ironclaw/pull/600)) ([#694](https://github.com/nearai/ironclaw/pull/694))
 - *(setup)* initialize secrets crypto for env-var security option ([#666](https://github.com/nearai/ironclaw/pull/666)) ([#706](https://github.com/nearai/ironclaw/pull/706))
 - persist /model selection across restarts ([#707](https://github.com/nearai/ironclaw/pull/707))
 - *(routines)* resolve message tool channel/target from per-job metadata ([#708](https://github.com/nearai/ironclaw/pull/708))
 - sanitize HTML error bodies from MCP servers to prevent web UI white screen ([#263](https://github.com/nearai/ironclaw/pull/263)) ([#656](https://github.com/nearai/ironclaw/pull/656))
 - prevent Instant duration overflow on Windows ([#657](https://github.com/nearai/ironclaw/pull/657)) ([#664](https://github.com/nearai/ironclaw/pull/664))
 - enable libsql remote + tls features for Turso cloud sync ([#587](https://github.com/nearai/ironclaw/pull/587))
 - *(tests)* replace hardcoded /tmp paths with tempdir + add 300 unit tests ([#659](https://github.com/nearai/ironclaw/pull/659))
 - *(llm)* nudge LLM when it expresses tool intent without calling tools ([#653](https://github.com/nearai/ironclaw/pull/653))
 - *(llm)* report zero cost for OpenRouter free-tier models ([#463](https://github.com/nearai/ironclaw/pull/463)) ([#613](https://github.com/nearai/ironclaw/pull/613))
 - reliable network tests and improved tool error messages ([#626](https://github.com/nearai/ironclaw/pull/626))
 - *(wasm)* use per-engine cache dirs on Windows to avoid file lock error ([#624](https://github.com/nearai/ironclaw/pull/624))
 - *(libsql)* support flexible embedding dimensions ([#534](https://github.com/nearai/ironclaw/pull/534))
 ### Other
 - Restructure CLAUDE.md into modular rules + add pr-shepherd command ([#750](https://github.com/nearai/ironclaw/pull/750))
 - make src/llm/ self-contained for crate extraction ([#767](https://github.com/nearai/ironclaw/pull/767))
 - add simplified Chinese (zh-CN) README translation ([#488](https://github.com/nearai/ironclaw/pull/488))
 - *(job)* cover job tool validation and state transitions ([#681](https://github.com/nearai/ironclaw/pull/681))
 - *(agent)* wire TestRig job tools through the scheduler ([#716](https://github.com/nearai/ironclaw/pull/716))
 - Fix single-message mode to exit after one turn when background channels are enabled ([#719](https://github.com/nearai/ironclaw/pull/719))
 - remove dead code ([#648](https://github.com/nearai/ironclaw/pull/648)) ([#703](https://github.com/nearai/ironclaw/pull/703))
 - add reviewer-feedback guardrails (CLAUDE.md, pre-commit hook, skill) ([#665](https://github.com/nearai/ironclaw/pull/665))
 - update WASM artifact SHA256 checksums [skip ci] ([#631](https://github.com/nearai/ironclaw/pull/631))
 - add explanatory comments to coverage workflow ([#610](https://github.com/nearai/ironclaw/pull/610))
 - build system prompt once per turn, skip tools on force-text ([#583](https://github.com/nearai/ironclaw/pull/583))
 - add comprehensive subdirectory CLAUDE.md files and update root ([#589](https://github.com/nearai/ironclaw/pull/589))
 - Improve test infrastructure: StubChannel, gateway helpers, security tests, search edge cases ([#623](https://github.com/nearai/ironclaw/pull/623))
 - *(workspace)* regression test for document_path in search results ([#509](https://github.com/nearai/ironclaw/pull/509))
 ### Added
 - AWS Bedrock LLM provider via native Converse API with IAM and SSO auth support (feature-gated: `--features bedrock`)
@@ -99,8 +99,7 @@ src/
 │   └── job_manager.rs  # Container lifecycle (create, stop, cleanup)
 │
 ├── worker/             # Runs inside Docker containers
-│   ├── container.rs    # Container worker runtime (ContainerDelegate + shared agentic loop)
+│   ├── runtime.rs      # Worker execution loop (tool calls, LLM)
 │   ├── job.rs          # Background job worker (JobDelegate + shared agentic loop)
 │   ├── claude_bridge.rs # Claude Code bridge (spawns claude CLI)
 │   └── proxy_llm.rs    # LlmProvider that proxies through orchestrator
 │
@@ -1,34 +1,5 @@
 # Contributing
 ## Getting Started
 ```bash
 git clone https://github.com/nearai/ironclaw.git
 cd ironclaw
 ./scripts/dev-setup.sh
 ```
 This installs the Rust toolchain, WASM targets, git hooks, and runs initial checks.
 ## Development Workflow
 ```bash
 cargo fmt                                                    # format
 cargo clippy --all --benches --tests --examples --all-features  # lint (zero warnings)
 cargo test                                                   # unit tests
 cargo test --features integration                            # + PostgreSQL tests
 ```
 ## Code Style
 - Zero clippy warnings policy
 - No `.unwrap()` or `.expect()` in production code (tests are fine)
 - Use `thiserror` for error types, map errors with context
 - Prefer `crate::` for cross-module imports
 - Comments for non-obvious logic only
 See `CLAUDE.md` for full style guidelines.
 ## Feature Parity Requirement
 When your change affects a tracked capability, update `FEATURE_PARITY.md` in the same branch.
@@ -38,23 +9,3 @@ When your change affects a tracked capability, update `FEATURE_PARITY.md` in the
 1. Review the relevant parity rows in `FEATURE_PARITY.md`.
 2. Update status/notes if behavior changed.
 3. Include the `FEATURE_PARITY.md` diff in your commit when applicable.
 ## Review Tracks
 All PRs follow a risk-based review process:
 | Track | Scope | Requirements |
 |-------|-------|-------------|
 | **A** | Docs, tests, chore, dependency bumps | 1 approval + CI green |
 | **B** | Features, refactors, new tools/channels | 1 approval + CI green + test evidence |
 | **C** | Security (`src/safety/`, `src/secrets/`), runtime (`src/agent/`, `src/worker/`), database schema, CI workflows | 2 approvals + rollback plan documented |
 Select the appropriate track in the PR template based on what your changes touch.
 ## Database Changes
 IronClaw uses dual-backend persistence (PostgreSQL + libSQL). All new persistence features must support both backends. See `src/db/CLAUDE.md`.
 ## Adding Dependencies
 Run `cargo deny check` before adding new dependencies to verify license compatibility and check for known advisories.
@@ -63,12 +63,12 @@ These files account for the vast majority of the coverage gap:
 | `src/main.rs` | 740 | 522 | 29.4% | 485 |
 | `src/channels/web/handlers/jobs.rs` | 513 | 456 | 11.1% | 430 |
 | `src/tools/builder/core.rs` | 524 | 456 | 13.0% | 429 |
-| `src/worker/job.rs` | 1,078 | 467 | 56.7% | 413 |
+| `src/agent/worker.rs` | 1,078 | 467 | 56.7% | 413 |
 | `src/channels/web/handlers/chat.rs` | 564 | 417 | 26.1% | 388 |
 | `src/tools/wasm/wrapper.rs` | 1,005 | 436 | 56.6% | 385 |
 | `src/channels/signal.rs` | 1,814 | 472 | 74.0% | 381 |
 | `src/tools/mcp/auth.rs` | 472 | 378 | 19.9% | 354 |
-| `src/worker/container.rs` | 350 | 330 | 5.7% | 312 |
+| `src/worker/runtime.rs` | 350 | 330 | 5.7% | 312 |
 | `src/tools/builtin/job.rs` | 1,014 | 359 | 64.6% | 308 |
 | `src/cli/mcp.rs` | 322 | 319 | 0.9% | 302 |
 | `src/cli/oauth_defaults.rs` | 730 | 335 | 54.1% | 298 |
@@ -346,7 +346,7 @@ Test slash commands through the agent loop.
 ### Trace: Worker Multi-Turn Execution
-**Covers:** `worker/job.rs` (+413 lines), `agent/agent_loop.rs` (+207 lines)
+**Covers:** `agent/worker.rs` (+413 lines), `agent/agent_loop.rs` (+207 lines)
 Test multi-turn tool calling, error recovery, and completion flows.
@@ -769,7 +769,7 @@ HTTP proxy for container network access.
 - `test_proxy_connect_tunnel` -- HTTPS CONNECT method handling
 - `test_proxy_logging` -- request/response logging
-### `src/worker/container.rs` -- 5.7% -> 95% (+312 lines)
+### `src/worker/runtime.rs` -- 5.7% -> 95% (+312 lines)
 Worker execution loop (runs inside containers).
@@ -3350,7 +3350,7 @@ dependencies = [
 [[package]]
 name = "ironclaw"
-version = "0.18.0"
+version = "0.16.1"
 dependencies = [
 "aes-gcm",
 "aho-corasick",
@@ -14,12 +14,11 @@ exclude = [
    "tools-src/google-slides",
    "tools-src/slack",
    "tools-src/telegram",
    "fuzz",
 ]
 [package]
 name = "ironclaw"
-version = "0.18.0"
+version = "0.16.1"
 edition = "2024"
 rust-version = "1.92"
 description = "Secure personal AI assistant that protects your data and expands its capabilities on the fly"
@@ -215,14 +214,10 @@ bedrock = ["dep:aws-config", "dep:aws-sdk-bedrockruntime", "dep:aws-smithy-types
 name = "html_to_markdown"
 required-features = ["html-to-markdown"]
 [profile.release]
 strip = true          # Remove debug symbols from release binaries
 # The profile that 'cargo dist' will build with
 [profile.dist]
 inherits = "release"
-lto = "fat"           # Full cross-crate LTO (slow build, better codegen)
+lto = "thin"
 codegen-units = 1     # Single codegen unit for maximum optimization
 # Config for 'dist'
 [workspace.metadata.dist]
@@ -46,14 +46,14 @@ This document tracks feature parity between IronClaw (Rust implementation) and O
 | Bonjour/mDNS discovery | ✅ | ❌ | |
 | Tailscale integration | ✅ | ❌ | |
 | Health check endpoints | ✅ | ✅ | /api/health + /api/gateway/status + /healthz + /readyz, with channel-backed readiness probes |
-| `doctor` diagnostics | ✅ | 🚧 | 16 checks: settings, LLM, DB, embeddings, routines, gateway, MCP, skills, secrets, service, Docker daemon, tunnel binaries |
+| `doctor` diagnostics | ✅ | ❌ | |
 | Agent event broadcast | ✅ | 🚧 | SSE broadcast manager exists (SseManager) but tool/job-state events not fully wired |
 | Channel health monitor | ✅ | ❌ | Auto-restart with configurable interval |
 | Presence system | ✅ | ❌ | Beacons on connect, system presence for agents |
 | Trusted-proxy auth mode | ✅ | ❌ | Header-based auth for reverse proxies |
 | APNs push pipeline | ✅ | ❌ | Wake disconnected iOS nodes via push |
 | Oversized payload guard | ✅ | 🚧 | HTTP webhook has 64KB body limit + Content-Length check; no chat.history cap |
-| Pre-prompt context diagnostics | ✅ | 🚧 | Token breakdown logged before LLM call (conversational dispatcher path); other LLM entry points not yet covered |
+| Pre-prompt context diagnostics | ✅ | ❌ | Context size logging before prompt |
 ### Owner: _Unassigned_
@@ -175,7 +175,7 @@ This document tracks feature parity between IronClaw (Rust implementation) and O
 | `message send` | ✅ | ❌ | P2 | Send to channels |
 | `browser` | ✅ | ❌ | P3 | Browser automation |
 | `sandbox` | ✅ | ✅ | - | WASM sandbox |
-| `doctor` | ✅ | 🚧 | P2 | 16 subsystem checks |
+| `doctor` | ✅ | ❌ | P2 | Diagnostics |
 | `logs` | ✅ | ❌ | P3 | Query logs |
 | `update` | ✅ | ❌ | P3 | Self-update |
 | `completion` | ✅ | ✅ | - | Shell completion |
@@ -1,40 +0,0 @@
 [package]
 name = "ironclaw-fuzz"
 version = "0.0.0"
 publish = false
 edition = "2021"
 [package.metadata]
 cargo-fuzz = true
 [dependencies]
 libfuzzer-sys = "0.4"
 serde_json = "1"
 [dependencies.ironclaw]
 path = ".."
 [[bin]]
 name = "fuzz_safety_sanitizer"
 path = "fuzz_targets/fuzz_safety_sanitizer.rs"
 doc = false
 [[bin]]
 name = "fuzz_safety_validator"
 path = "fuzz_targets/fuzz_safety_validator.rs"
 doc = false
 [[bin]]
 name = "fuzz_leak_detector"
 path = "fuzz_targets/fuzz_leak_detector.rs"
 doc = false
 [[bin]]
 name = "fuzz_tool_params"
 path = "fuzz_targets/fuzz_tool_params.rs"
 doc = false
 [[bin]]
 name = "fuzz_config_env"
 path = "fuzz_targets/fuzz_config_env.rs"
 doc = false
@@ -1,43 +0,0 @@
 # IronClaw Fuzz Targets
 Fuzz testing for security-critical input parsing paths using [cargo-fuzz](https://github.com/rust-fuzz/cargo-fuzz) (libFuzzer).
 ## Targets
 | Target | What it exercises |
 |--------|-------------------|
 | `fuzz_safety_sanitizer` | Prompt injection pattern detection (Aho-Corasick + regex) |
 | `fuzz_safety_validator` | Input validation (length, encoding, forbidden patterns) |
 | `fuzz_leak_detector` | Secret leak detection (API keys, tokens, credentials) |
 | `fuzz_tool_params` | Tool parameter and schema JSON validation |
 | `fuzz_config_env` | SafetyLayer end-to-end (sanitize, validate, policy check) |
 ## Setup
 ```bash
 cargo install cargo-fuzz
 rustup install nightly
 ```
 ## Running
 ```bash
 # Run a specific target (runs until stopped or crash found)
 cargo +nightly fuzz run fuzz_safety_sanitizer
 # Run with a time limit (5 minutes)
 cargo +nightly fuzz run fuzz_leak_detector -- -max_total_time=300
 # Run all targets for 60 seconds each
 for target in fuzz_safety_sanitizer fuzz_safety_validator fuzz_leak_detector fuzz_tool_params fuzz_config_env; do
    echo "==> $target"
    cargo +nightly fuzz run "$target" -- -max_total_time=60
 done
 ```
 ## Adding New Targets
 1. Create `fuzz/fuzz_targets/fuzz_<name>.rs` following the existing pattern
 2. Add a `[[bin]]` entry in `fuzz/Cargo.toml`
 3. Create `fuzz/corpus/fuzz_<name>/` for seed inputs
 4. Exercise real IronClaw code paths, not just generic serde
@@ -1,55 +0,0 @@
 #![no_main]
 use libfuzzer_sys::fuzz_target;
 use ironclaw::safety::{LeakDetector, Sanitizer, Validator};
 fuzz_target!(|data: &[u8]| {
    if let Ok(input) = std::str::from_utf8(data) {
        // Exercise Sanitizer: detect and neutralize prompt injection attempts.
        let sanitizer = Sanitizer::new();
        let sanitized = sanitizer.sanitize(input);
        // The sanitized content must never be empty when input is non-empty,
        // because sanitization wraps/escapes rather than deleting.
        if !input.is_empty() {
            assert!(
                !sanitized.content.is_empty(),
                "sanitize() produced empty content for non-empty input"
            );
        }
        // If no modification occurred, content must equal input.
        if !sanitized.was_modified {
            assert_eq!(sanitized.content, input);
        }
        // Exercise Validator: input validation (length, encoding, patterns).
        let validator = Validator::new();
        let result = validator.validate(input);
        // ValidationResult must always be well-formed: if valid, no errors.
        if result.is_valid {
            assert!(
                result.errors.is_empty(),
                "valid result should have no errors"
            );
        }
        // Exercise LeakDetector: secret detection (API keys, tokens, etc.).
        let detector = LeakDetector::new();
        let scan = detector.scan(input);
        // scan_and_clean must not panic and must return valid UTF-8.
        let cleaned = detector.scan_and_clean(input);
        if let Ok(ref clean_str) = cleaned {
            // Cleaned output must never be longer than original + redaction markers.
            // At minimum it should be valid UTF-8 (guaranteed by String type).
            let _ = clean_str.len();
        }
        // If scan found no matches, scan_and_clean should return the input unchanged.
        if scan.matches.is_empty() {
            if let Ok(ref clean_str) = cleaned {
                assert_eq!(
                    clean_str, input,
                    "scan_and_clean changed content despite no matches"
                );
            }
        }
    }
 });
@@ -1,23 +0,0 @@
 #![no_main]
 use libfuzzer_sys::fuzz_target;
 use ironclaw::safety::LeakDetector;
 fuzz_target!(|data: &[u8]| {
    if let Ok(s) = std::str::from_utf8(data) {
        let detector = LeakDetector::new();
        // Exercise scan path
        let result = detector.scan(s);
        // Invariant: if should_block, there must be matches
        if result.should_block {
            assert!(!result.matches.is_empty());
        }
        // Invariant: match locations must be valid
        for m in &result.matches {
            assert!(m.location.end <= s.len());
        }
        // Exercise scan_and_clean path
        let _ = detector.scan_and_clean(s);
    }
 });
@@ -1,23 +0,0 @@
 #![no_main]
 use libfuzzer_sys::fuzz_target;
 use ironclaw::safety::Sanitizer;
 fuzz_target!(|data: &[u8]| {
    if let Ok(s) = std::str::from_utf8(data) {
        let sanitizer = Sanitizer::new();
        // Exercise the main sanitization path
        let result = sanitizer.sanitize(s);
        // Verify invariant: warnings should have valid ranges
        for w in &result.warnings {
            assert!(w.location.end <= s.len());
        }
        // Verify invariant: critical severity triggers modification
        let has_critical = result.warnings.iter().any(|w| {
            w.severity == ironclaw::safety::Severity::Critical
        });
        if has_critical {
            assert!(result.was_modified);
        }
    }
 });
@@ -1,21 +0,0 @@
 #![no_main]
 use libfuzzer_sys::fuzz_target;
 use ironclaw::safety::Validator;
 fuzz_target!(|data: &[u8]| {
    if let Ok(s) = std::str::from_utf8(data) {
        let validator = Validator::new();
        // Exercise input validation
        let result = validator.validate(s);
        // Invariant: empty input is always invalid
        if s.is_empty() {
            assert!(!result.is_valid);
        }
        // Exercise tool parameter validation with arbitrary JSON
        if let Ok(value) = serde_json::from_str::<serde_json::Value>(s) {
            let _ = validator.validate_tool_params(&value);
        }
    }
 });
@@ -1,22 +0,0 @@
 #![no_main]
 use libfuzzer_sys::fuzz_target;
 use ironclaw::safety::Validator;
 use ironclaw::tools::validate_tool_schema;
 fuzz_target!(|data: &[u8]| {
    if let Ok(s) = std::str::from_utf8(data) {
        // Try parsing as JSON and validating as tool parameters
        if let Ok(value) = serde_json::from_str::<serde_json::Value>(s) {
            // Exercise Validator::validate_tool_params with arbitrary JSON
            let validator = Validator::new();
            let result = validator.validate_tool_params(&value);
            // Invariant: result should always be well-formed
            if !result.is_valid {
                assert!(!result.errors.is_empty());
            }
            // Exercise validate_tool_schema with arbitrary JSON as a schema
            let _ = validate_tool_schema(&value, "fuzz");
        }
    }
 });
@@ -1,7 +0,0 @@
 -- Add token budget tracking columns to agent_jobs.
 --
 -- Tracks max_tokens (configured limit per job) and total_tokens_used (running total)
 -- to enforce job-level token budgets and prevent budget bypass via user-supplied metadata.
 ALTER TABLE agent_jobs ADD COLUMN max_tokens BIGINT NOT NULL DEFAULT 0;
 ALTER TABLE agent_jobs ADD COLUMN total_tokens_used BIGINT NOT NULL DEFAULT 0;
@@ -9,9 +9,8 @@
    "api_key_required": true,
    "base_url_env": "OPENAI_BASE_URL",
    "model_env": "OPENAI_MODEL",
-    "default_model": "gpt-5-mini",
+    "default_model": "gpt-4o",
    "description": "OpenAI GPT models (direct API)",
    "unsupported_params": ["temperature"],
    "setup": {
      "kind": "api_key",
      "secret_name": "llm_openai_api_key",
@@ -87,7 +86,6 @@
    "model_env": "TINFOIL_MODEL",
    "default_model": "kimi-k2-5",
    "description": "Tinfoil private inference (hardware-attested TEE)",
    "unsupported_params": ["temperature"],
    "setup": {
      "kind": "api_key",
      "secret_name": "llm_tinfoil_api_key",
@@ -19,7 +19,7 @@
  "artifacts": {
    "wasm32-wasip2": {
      "url": "https://github.com/nearai/ironclaw/releases/latest/download/discord-wasm32-wasip2.tar.gz",
-      "sha256": null
+      "sha256": "030707431717bca3411a48f311c6ab5f92a45c747de26cafe4f6e3e23a8b3b2d"
    }
  },
  "auth_summary": {
@@ -19,7 +19,7 @@
  "artifacts": {
    "wasm32-wasip2": {
      "url": "https://github.com/nearai/ironclaw/releases/latest/download/slack-wasm32-wasip2.tar.gz",
-      "sha256": null
+      "sha256": "6ed36077b67ac70a041f06f760f93ba79b33269885413c3c3f2c8c87ee60807e"
    }
  },
  "auth_summary": {
@@ -19,7 +19,7 @@
  "artifacts": {
    "wasm32-wasip2": {
      "url": "https://github.com/nearai/ironclaw/releases/latest/download/telegram-wasm32-wasip2.tar.gz",
-      "sha256": null
+      "sha256": "98c86895a9c4b0a1e19fe8a47f1ccbfe7e972e112b05e584bc897130dc32283a"
    }
  },
  "auth_summary": {
@@ -19,7 +19,7 @@
  "artifacts": {
    "wasm32-wasip2": {
      "url": "https://github.com/nearai/ironclaw/releases/latest/download/whatsapp-wasm32-wasip2.tar.gz",
-      "sha256": null
+      "sha256": "bd35cad18d87292ea8d2f52db9b514ed9f814a414de910f59073d475c26c4c14"
    }
  },
  "auth_summary": {
@@ -20,7 +20,7 @@
  "artifacts": {
    "wasm32-wasip2": {
      "url": "https://github.com/nearai/ironclaw/releases/latest/download/github-wasm32-wasip2.tar.gz",
-      "sha256": null
+      "sha256": "6fcd32719a4ff15641a4b50fff8984686550f0c491dce60518f4126857d0c544"
    }
  },
  "auth_summary": {
@@ -19,7 +19,7 @@
  "artifacts": {
    "wasm32-wasip2": {
      "url": "https://github.com/nearai/ironclaw/releases/latest/download/gmail-wasm32-wasip2.tar.gz",
-      "sha256": null
+      "sha256": "023da7000b17568bf0e64b2e5013c8a042b2f323c85f1632339231c73d500e39"
    }
  },
  "auth_summary": {
@@ -19,7 +19,7 @@
  "artifacts": {
    "wasm32-wasip2": {
      "url": "https://github.com/nearai/ironclaw/releases/latest/download/google-calendar-wasm32-wasip2.tar.gz",
-      "sha256": null
+      "sha256": "fc42277b65881d6e9bcc5403dc54c7f5b3ddeaaaf04617fce2c5da05d76325f0"
    }
  },
  "auth_summary": {
@@ -19,7 +19,7 @@
  "artifacts": {
    "wasm32-wasip2": {
      "url": "https://github.com/nearai/ironclaw/releases/latest/download/google-docs-wasm32-wasip2.tar.gz",
-      "sha256": null
+      "sha256": "385c04abd1e6b8011ccc330e1f4bd7ce58577e488959b51594aa04eb26cbe7cc"
    }
  },
  "auth_summary": {
@@ -19,7 +19,7 @@
  "artifacts": {
    "wasm32-wasip2": {
      "url": "https://github.com/nearai/ironclaw/releases/latest/download/google-drive-wasm32-wasip2.tar.gz",
-      "sha256": null
+      "sha256": "1b107d575a5d52cc8c76d9a681802190f4373fb485f7f54f445533f097fa37c0"
    }
  },
  "auth_summary": {
@@ -19,7 +19,7 @@
  "artifacts": {
    "wasm32-wasip2": {
      "url": "https://github.com/nearai/ironclaw/releases/latest/download/google-sheets-wasm32-wasip2.tar.gz",
-      "sha256": null
+      "sha256": "c4f6b1e8c5126ac2c8a4b98e4283a3afa32223d2488fc3c3a609758c0c9beb90"
    }
  },
  "auth_summary": {
@@ -18,7 +18,7 @@
  "artifacts": {
    "wasm32-wasip2": {
      "url": "https://github.com/nearai/ironclaw/releases/latest/download/google-slides-wasm32-wasip2.tar.gz",
-      "sha256": null
+      "sha256": "7110b8565340c888e51f99e9c013bf4de8f8a7f7b33bace00eb8fc47831ff20b"
    }
  },
  "auth_summary": {
@@ -18,7 +18,7 @@
  "artifacts": {
    "wasm32-wasip2": {
      "url": "https://github.com/nearai/ironclaw/releases/latest/download/slack-tool-wasm32-wasip2.tar.gz",
-      "sha256": null
+      "sha256": "6ed36077b67ac70a041f06f760f93ba79b33269885413c3c3f2c8c87ee60807e"
    }
  },
  "auth_summary": {
@@ -19,7 +19,7 @@
  "artifacts": {
    "wasm32-wasip2": {
      "url": "https://github.com/nearai/ironclaw/releases/latest/download/telegram-mtproto-wasm32-wasip2.tar.gz",
-      "sha256": null
+      "sha256": "98c86895a9c4b0a1e19fe8a47f1ccbfe7e972e112b05e584bc897130dc32283a"
    }
  },
  "auth_summary": {
@@ -19,7 +19,7 @@
  "artifacts": {
    "wasm32-wasip2": {
      "url": "https://github.com/nearai/ironclaw/releases/latest/download/web-search-wasm32-wasip2.tar.gz",
-      "sha256": null
+      "sha256": "66cb2b9b00652385e9f30f17c74902b9222c17c53e9d3bd1ef42f5cab705bcf6"
    }
  },
  "auth_summary": {
@@ -1,80 +0,0 @@
 ---
 name: ironclaw-workflow-orchestrator
 description: "Install and operate a full GitHub issue-to-merge workflow in IronClaw using event-driven and cron routines. Use when setting up or tuning autonomous project orchestration: issue intake, planning, maintainer feedback handling, branch/PR execution, CI/comment follow-up, batched staging review every 8 hours, and memory updates from merge outcomes."
 ---
 # IronClaw Workflow Orchestrator
 ## Overview
 Use this skill to install and maintain a complete project workflow as routines, not core code changes. It maps GitHub webhook events plus scheduled checks into plan/update/implement/review/merge loops with explicit staging-batch analysis.
 ## Workflow
 1. Gather workflow parameters.
 2. Verify runtime prerequisites.
 3. Install or update routine set from templates.
 4. Run a dry test with `event_emit`.
 5. Monitor outcomes and tune prompts/filters.
 ## Parameters
 Collect these values before creating routines:
 - `repository`: `owner/repo` (required)
 - `maintainers`: GitHub handles allowed to trigger implement/replan actions
 - `staging_branch`: default `staging`
 - `main_branch`: default `main`
 - `batch_interval_hours`: default `8`
 - `implementation_label`: default `autonomous-impl`
 ## Prerequisites
 Before installing routines, verify:
 - Routines system enabled.
 - GitHub tool authenticated (for issue/PR/comment/status operations).
 - Events are emitted via `event_emit` tool calls (a future HTTP webhook ingestion endpoint is planned but not yet available).
 ## Install Procedure
 1. Open [`workflow-routines.md`](references/workflow-routines.md).
 2. For each template block:
 - replace placeholders (`{{repository}}`, `{{maintainers}}`, branch names)
 - call `routine_create`
 3. If a routine already exists:
 - use `routine_update` instead of creating duplicates
 - keep names stable so long-lived metrics/history stay intact
 4. Confirm install with `routine_list` and `routine_history`.
 ## Routine Set
 Install these routines:
 - `wf-issue-plan`: on `issue.opened` or `issue.reopened`, generate implementation plan comment/checklist.
 - `wf-maintainer-comment-gate`: on maintainer comments, decide update-plan vs start implementation.
 - `wf-pr-monitor-loop`: on PR open/sync/review-comment/review, address feedback and refresh branch.
 - `wf-ci-fix-loop`: on CI status/check failures, apply fixes and push updates.
 - `wf-staging-batch-review`: every 8h, review ready PRs, merge into staging, run deep batch correctness analysis, fix findings, then merge staging -> main.
 - `wf-learning-memory`: on merged PRs, extract mistakes/lessons and write to shared memory.
 ## Event Filters
 Prefer top-level filters for stability:
 - `repository` (string)
 - `sender` (string)
 - `issue_number` / `pr_number`
 - `ci_status`, `ci_conclusion`
 - `review_state`, `comment_author`
 Use narrow filters to avoid accidental triggers across repos.
 ## Operating Rules
 - All implementation work must occur on non-main branches.
 - PR loop must resolve both human and AI review comments.
 - On conflicts with `origin/main`, refresh branch before continuing.
 - Staging-batch routine is the only path for bulk correctness verification before mainline merge.
 - Memory update routine runs only after successful merge.
 ## Validation
 After install, run:
 1. `event_emit` with a synthetic `issue.opened` payload for the target repo.
 2. Confirm at least one routine fired.
 3. Check corresponding `routine_history` entries.
 4. Confirm no unrelated routines fired.
 ## When To Update Templates
 Update this skill when:
 - GitHub event names/payload fields change.
 - Team review policy changes (e.g., staging cadence, maintainer gates).
 - New CI policy requires different failure routing.
@@ -1,4 +0,0 @@
 interface:
  display_name: "IronClaw Workflow Orchestrator"
  short_description: "Install and run event-driven GitHub workflow routines"
  default_prompt: "Set up the full issue-to-merge workflow using routines and event triggers."
@@ -1,128 +0,0 @@
 # Workflow Routine Templates
 Replace `{{...}}` placeholders before use.
 ## 1) Issue -> Plan
 ```json
 {
  "name": "wf-issue-plan",
  "description": "Create implementation plan when a new issue arrives",
  "trigger_type": "system_event",
  "event_source": "github",
  "event_type": "issue.opened",
  "event_filters": {
    "repository": "{{repository}}"
  },
  "action_type": "full_job",
  "prompt": "For issue #{{issue_number}} in {{repository}}, produce a concrete implementation plan with milestones, edge cases, and tests. Post/update an issue comment with the plan.",
  "cooldown_secs": 30
 }
 ```
 ## 2) Maintainer Comment Gate (Update Plan vs Implement)
 Trigger per-maintainer by creating one routine per handle, or maintain a shared author convention.
 ```json
 {
  "name": "wf-maintainer-comment-gate-{{maintainer}}",
  "description": "React to maintainer guidance comments on issues/PRs",
  "trigger_type": "system_event",
  "event_source": "github",
  "event_type": "pr.comment.created",
  "event_filters": {
    "repository": "{{repository}}",
    "comment_author": "{{maintainer}}"
  },
  "action_type": "full_job",
  "prompt": "Read the maintainer comment and decide: update plan or start/continue implementation. If plan changes are requested, edit the plan artifact first. If implementation is requested, continue on the feature branch and update PR status/comment.",
  "cooldown_secs": 20
 }
 ```
 ## 3) PR Monitor Loop
 ```json
 {
  "name": "wf-pr-monitor-loop",
  "description": "Keep PR healthy: address review comments and refresh branch",
  "trigger_type": "system_event",
  "event_source": "github",
  "event_type": "pr.synchronize",
  "event_filters": {
    "repository": "{{repository}}"
  },
  "action_type": "full_job",
  "prompt": "For PR #{{pr_number}}, collect open review comments and unresolved threads, apply fixes, push branch updates, and summarize remaining blockers. If conflict with {{main_branch}}, rebase/merge from origin/{{main_branch}} and resolve safely.",
  "cooldown_secs": 20
 }
 ```
 ## 4) CI Failure Fix Loop
 ```json
 {
  "name": "wf-ci-fix-loop",
  "description": "Fix failing CI checks on active PRs",
  "trigger_type": "system_event",
  "event_source": "github",
  "event_type": "ci.check_run.completed",
  "event_filters": {
    "repository": "{{repository}}",
    "ci_conclusion": "failure"
  },
  "action_type": "full_job",
  "prompt": "Find failing check details for PR #{{pr_number}}, implement minimal safe fixes, rerun or await CI, and post concise status updates. Prioritize deterministic and test-backed fixes.",
  "cooldown_secs": 20
 }
 ```
 ## 5) Staging Batch Review (Every 8h)
 ```json
 {
  "name": "wf-staging-batch-review",
  "description": "Batch correctness review through staging, then merge to main",
  "trigger_type": "cron",
  "schedule": "0 0 */{{batch_interval_hours}} * * *",
  "action_type": "full_job",
  "prompt": "Every cycle: list ready PRs, merge ready ones into {{staging_branch}}, run deep correctness analysis in batch, fix discovered issues on affected branches, ensure CI green, then merge {{staging_branch}} into {{main_branch}} if clean.",
  "cooldown_secs": 120
 }
 ```
 ## 6) Post-Merge Learning -> Common Memory
 ```json
 {
  "name": "wf-learning-memory",
  "description": "Capture merge learnings into shared memory",
  "trigger_type": "system_event",
  "event_source": "github",
  "event_type": "pr.closed",
  "event_filters": {
    "repository": "{{repository}}",
    "pr_merged": "true"
  },
  "action_type": "full_job",
  "prompt": "From merged PR #{{pr_number}}, extract preventable mistakes, reviewer themes, CI failure causes, and successful patterns. Write/update a shared memory doc with actionable rules to reduce cycle time and regressions.",
  "cooldown_secs": 30
 }
 ```
 ## Optional: Synthetic Event Test
 ```json
 {
  "source": "github",
  "event_type": "issue.opened",
  "payload": {
    "repository": "{{repository}}",
    "issue_number": 99999,
    "sender": "test-bot"
  }
 }
 ```
 Use with `event_emit` after routine install.
@@ -14,15 +14,14 @@ Core agent logic. This is the most complex subsystem — read this before workin
 | `session_manager.rs` | Lifecycle: create/lookup sessions, map external thread IDs to internal UUIDs, prune stale sessions, manage undo managers. |
 | `router.rs` | Routes explicit `/commands` to `MessageIntent`. Natural language bypasses the router entirely. |
 | `scheduler.rs` | Parallel job scheduling. Maintains `jobs` map (full LLM-driven) and `subtasks` map (tool-exec/background). |
-| *(moved to `src/worker/job.rs`)* | Per-job execution now lives in `src/worker/job.rs` as `JobDelegate`, using the shared `run_agentic_loop()` engine. |
+| `worker.rs` | Per-job execution for background scheduler jobs: calls LLM, runs tools, handles the reasoning loop. Distinct from `dispatcher.rs`. |
 | `agentic_loop.rs` | Shared agentic loop engine: `run_agentic_loop()`, `LoopDelegate` trait, `LoopOutcome`, `LoopSignal`, `TextAction`. All three execution paths (chat, job, container) delegate to this. |
 | `compaction.rs` | Context window management: summarize old turns, write to workspace daily log, trim context. Three strategies. |
 | `context_monitor.rs` | Detects memory pressure. Suggests `CompactionStrategy` based on usage level. |
 | `self_repair.rs` | Detects stuck jobs and broken tools, attempts recovery. |
 | `heartbeat.rs` | Proactive periodic execution. Reads `HEARTBEAT.md`, notifies via channel if findings. |
 | `submission.rs` | Parses all user submissions into typed variants before routing. |
 | `undo.rs` | Turn-based undo/redo with checkpoints. Checkpoints store message lists (max 20 by default). |
-| `routine.rs` | `Routine` types: `Trigger` (cron/event/system_event/manual) + `RoutineAction` (lightweight/full_job) + `RoutineGuardrails`. |
+| `routine.rs` | `Routine` types: `Trigger` (cron/event/webhook/manual) + `RoutineAction` (lightweight/full_job) + `RoutineGuardrails`. |
 | `routine_engine.rs` | Cron ticker and event matcher. Fires routines when triggers match. Lightweight runs inline; full_job dispatches to `Scheduler`. |
 | `task.rs` | Task types for the scheduler: `Job`, `ToolExec`, `Background`. Used by `spawn_subtask` and `spawn_batch`. |
 | `cost_guard.rs` | LLM spend and action-rate enforcement. Tracks daily budget (cents) and hourly call rate. Lives in `AgentDeps`. |
@@ -50,28 +49,26 @@ Session (per user)
 ## Agentic Loop (dispatcher.rs)
-All three execution paths (chat, job, container) now use the shared `run_agentic_loop()` engine in `agentic_loop.rs`, each providing their own `LoopDelegate` implementation:
+The `dispatcher.rs` module handles **direct conversational turns** (user messages processed inline by the main agent). Background scheduler jobs use `worker.rs` instead — these are two separate execution paths.
 - **`ChatDelegate`** (`dispatcher.rs`) — conversational turns, tool approval, skill context injection
 - **`JobDelegate`** (`src/worker/job.rs`) — background scheduler jobs, planning support, completion detection
 - **`ContainerDelegate`** (`src/worker/container.rs`) — Docker container worker, sequential tool exec, HTTP event streaming
 ```
-run_agentic_loop(delegate, reasoning, reason_ctx, config)
+run_agentic_loop()  [dispatcher.rs — conversational turns]
-  1. Check signals (stop/cancel) via delegate.check_signals()
+  1. Load workspace system prompt (identity files: AGENTS.md, SOUL.md, etc.)
-  2. Pre-LLM hook via delegate.before_llm_call()
+  2. Detect group chat from metadata; exclude MEMORY.md if group chat
-  3. LLM call via delegate.call_llm()
+  3. Select active skills (keyword/pattern scoring against message content)
-  4. If text response → delegate.handle_text_response() → Continue or Return
+  4. Build skill context block (injected before user message)
-  5. If tool calls → delegate.execute_tool_calls() → Continue or Return
+  5. LLM call → text response OR tool calls
-  6. Post-iteration hook via delegate.after_iteration()
+  6. If tool calls:
-  7. Repeat until LoopOutcome returned or max_iterations reached
+     a. Check tool approval (session auto-approvals, pending approval queue)
     b. Execute tools (parallel via JoinSet)
     c. Sanitize results through SafetyLayer
     d. Feed results back → goto 5
  7. Return AgenticLoopResult::Response or NeedApproval
 ```
-**Tool approval:** Tools flagged `requires_approval` pause the loop — `ChatDelegate` returns `LoopOutcome::NeedApproval(pending)`. The web gateway stores the `PendingApproval` in session state and sends an `approval_needed` SSE event. The user's approval/deny resumes the loop.
+**Tool approval:** Tools flagged `requires_approval` pause the loop and return `NeedApproval`. The web gateway stores the `PendingApproval` in session state and sends an `approval_needed` SSE event. The user's approval/deny resumes the loop.
-**Shared tool execution:** `tools/execute.rs` provides `execute_tool_with_safety()` (validate → timeout → execute → serialize) and `process_tool_result()` (sanitize → wrap → ChatMessage), used by all three delegates.
+**worker.rs vs dispatcher.rs:** `dispatcher.rs` runs the agentic loop for user-initiated conversational turns (holds session lock, tracks turns). `worker.rs` is spawned by the `Scheduler` for background jobs created via `CreateJob` / `/job` — it runs independently of the session and has its own LLM reasoning loop with planning support (`use_planning` flag).
 **ChatDelegate vs JobDelegate:** `ChatDelegate` runs for user-initiated conversational turns (holds session lock, tracks turns). `JobDelegate` is spawned by the `Scheduler` for background jobs created via `CreateJob` / `/job` — it runs independently of the session and has planning support (`use_planning` flag).
 ## Command Routing (router.rs)
@@ -738,18 +738,6 @@ impl Agent {
    }
    async fn handle_message(&self, message: &IncomingMessage) -> Result<Option<String>, Error> {
        // Log at info level only for tracking without exposing PII (user_id can be a phone number)
        tracing::info!(message_id = %message.id, "Processing message");
        // Log sensitive details at debug level for troubleshooting
        tracing::debug!(
            message_id = %message.id,
            user_id = %message.user_id,
            channel = %message.channel,
            thread_id = ?message.thread_id,
            "Message details"
        );
        // Set message tool context for this turn (current channel and target)
        // For Signal, use signal_target from metadata (group:ID or phone number),
        // otherwise fall back to user_id
@@ -765,7 +753,7 @@ impl Agent {
        // Parse submission type first
        let mut submission = SubmissionParser::parse(&message.content);
-        tracing::trace!(
+        tracing::debug!(
            "[agent_loop] Parsed submission: {:?}",
            std::any::type_name_of_val(&submission)
        );
@@ -798,19 +786,10 @@ impl Agent {
        // Hydrate thread from DB if it's a historical thread not in memory
        if let Some(ref external_thread_id) = message.thread_id {
            tracing::trace!(
                message_id = %message.id,
                thread_id = %external_thread_id,
                "Hydrating thread from DB"
            );
            self.maybe_hydrate_thread(message, external_thread_id).await;
        }
        // Resolve session and thread
        tracing::debug!(
            message_id = %message.id,
            "Resolving session and thread"
        );
        let (session, thread_id) = self
            .session_manager
            .resolve_thread(
@@ -819,11 +798,6 @@ impl Agent {
                message.thread_id.as_deref(),
            )
            .await;
        tracing::debug!(
            message_id = %message.id,
            thread_id = %thread_id,
            "Resolved session and thread"
        );
        // Auth mode interception: if the thread is awaiting a token, route
        // the message directly to the credential store. Nothing touches
@@ -853,7 +827,7 @@ impl Agent {
            }
        }
-        tracing::trace!(
+        tracing::debug!(
            "Received message from {} on {} ({} chars)",
            message.user_id,
            message.channel,
@@ -1,587 +0,0 @@
 //! Unified agentic loop engine.
 //!
 //! Provides a single implementation of the core LLM call → tool execution →
 //! result processing → context update → repeat cycle. Three consumers
 //! (chat dispatcher, job worker, container runtime) customize behavior
 //! via the `LoopDelegate` trait.
 use async_trait::async_trait;
 use crate::agent::session::PendingApproval;
 use crate::error::Error;
 use crate::llm::{ChatMessage, Reasoning, ReasoningContext, RespondResult};
 /// Signal from the delegate indicating how the loop should proceed.
 pub enum LoopSignal {
    /// Continue normally.
    Continue,
    /// Stop the loop gracefully.
    Stop,
    /// Inject a user message into context and continue.
    InjectMessage(String),
 }
 /// Outcome of a text response from the LLM.
 pub enum TextAction {
    /// Return this as the final loop result.
    Return(LoopOutcome),
    /// Continue the loop (text was handled but loop should proceed).
    Continue,
 }
 /// Final outcome of the agentic loop.
 pub enum LoopOutcome {
    /// Completed with a text response.
    Response(String),
    /// Loop was stopped by a signal.
    Stopped,
    /// Max iterations exceeded.
    MaxIterations,
    /// A tool requires user approval before continuing (chat delegate only).
    NeedApproval(Box<PendingApproval>),
 }
 /// Configuration for the agentic loop.
 pub struct AgenticLoopConfig {
    pub max_iterations: usize,
    pub enable_tool_intent_nudge: bool,
    pub max_tool_intent_nudges: u32,
 }
 impl Default for AgenticLoopConfig {
    fn default() -> Self {
        Self {
            max_iterations: 50,
            enable_tool_intent_nudge: true,
            max_tool_intent_nudges: 2,
        }
    }
 }
 /// Strategy trait — each consumer implements this to customize I/O and lifecycle.
 ///
 /// The shared loop calls these methods at well-defined points. Consumers
 /// implement only the behavior that differs between chat, job, and container
 /// contexts. The loop itself handles the common logic: tool intent nudge,
 /// iteration counting, tool definition refresh, and the respond → execute → process cycle.
 ///
 /// # `Send + Sync` requirement
 ///
 /// This trait requires `Send + Sync` because the loop accepts `&dyn LoopDelegate`.
 /// Delegates using borrowed references (e.g. `ChatDelegate<'a>`) must ensure all
 /// borrowed fields are `Send + Sync`. This is a load-bearing constraint: if a
 /// delegate needs to be spawned into a detached task, it must use `Arc`-based
 /// ownership instead of borrows (as `JobDelegate` and `ContainerDelegate` do).
 #[async_trait]
 pub trait LoopDelegate: Send + Sync {
    /// Called at the start of each iteration. Check for external signals
    /// (cancellation, user messages, stop requests).
    async fn check_signals(&self) -> LoopSignal;
    /// Called before the LLM call. Allows the delegate to refresh tool
    /// definitions, enforce cost guards, or inject messages.
    /// Return `Some(outcome)` to break the loop early.
    async fn before_llm_call(
        &self,
        reason_ctx: &mut ReasoningContext,
        iteration: usize,
    ) -> Option<LoopOutcome>;
    /// Call the LLM and return the result. Delegates own the LLM call
    /// to handle consumer-specific concerns (rate limiting, auto-compaction,
    /// cost tracking, force_text mode).
    async fn call_llm(
        &self,
        reasoning: &Reasoning,
        reason_ctx: &mut ReasoningContext,
        iteration: usize,
    ) -> Result<crate::llm::RespondOutput, Error>;
    /// Handle a text-only response from the LLM.
    /// Return `TextAction::Return` to exit the loop, `TextAction::Continue` to proceed.
    async fn handle_text_response(
        &self,
        text: &str,
        reason_ctx: &mut ReasoningContext,
    ) -> TextAction;
    /// Execute tool calls and add results to context.
    /// Return `Some(outcome)` to break the loop (e.g. approval needed).
    async fn execute_tool_calls(
        &self,
        tool_calls: Vec<crate::llm::ToolCall>,
        content: Option<String>,
        reason_ctx: &mut ReasoningContext,
    ) -> Result<Option<LoopOutcome>, Error>;
    /// Called when the LLM expresses tool intent without actually calling a tool.
    /// Delegates can use this to emit events or log the nudge for observability.
    async fn on_tool_intent_nudge(&self, _text: &str, _reason_ctx: &mut ReasoningContext) {}
    /// Called after each successful iteration (no error, no early return).
    async fn after_iteration(&self, _iteration: usize) {}
 }
 /// Run the unified agentic loop.
 ///
 /// This is the single implementation used by all three consumers (chat, job, container).
 /// The `delegate` provides consumer-specific behavior via the `LoopDelegate` trait.
 pub async fn run_agentic_loop(
    delegate: &dyn LoopDelegate,
    reasoning: &Reasoning,
    reason_ctx: &mut ReasoningContext,
    config: &AgenticLoopConfig,
 ) -> Result<LoopOutcome, Error> {
    let mut consecutive_tool_intent_nudges: u32 = 0;
    for iteration in 1..=config.max_iterations {
        // Check for external signals (stop, cancellation, user messages)
        match delegate.check_signals().await {
            LoopSignal::Continue => {}
            LoopSignal::Stop => return Ok(LoopOutcome::Stopped),
            LoopSignal::InjectMessage(msg) => {
                reason_ctx.messages.push(ChatMessage::user(&msg));
            }
        }
        // Pre-LLM call hook (cost guard, tool refresh, iteration limit nudge)
        if let Some(outcome) = delegate.before_llm_call(reason_ctx, iteration).await {
            return Ok(outcome);
        }
        // Call LLM
        let output = delegate.call_llm(reasoning, reason_ctx, iteration).await?;
        match output.result {
            RespondResult::Text(text) => {
                // Tool intent nudge: if the LLM says "let me search..." without
                // actually calling a tool, inject a nudge message.
                if config.enable_tool_intent_nudge
                    && !reason_ctx.available_tools.is_empty()
                    && !reason_ctx.force_text
                    && consecutive_tool_intent_nudges < config.max_tool_intent_nudges
                    && crate::llm::llm_signals_tool_intent(&text)
                {
                    consecutive_tool_intent_nudges += 1;
                    tracing::info!(
                        iteration,
                        "LLM expressed tool intent without calling a tool, nudging"
                    );
                    delegate.on_tool_intent_nudge(&text, reason_ctx).await;
                    reason_ctx.messages.push(ChatMessage::assistant(&text));
                    reason_ctx
                        .messages
                        .push(ChatMessage::user(crate::llm::TOOL_INTENT_NUDGE));
                    delegate.after_iteration(iteration).await;
                    continue;
                }
                // Reset nudge counter since we got a non-intent text response
                if !crate::llm::llm_signals_tool_intent(&text) {
                    consecutive_tool_intent_nudges = 0;
                }
                match delegate.handle_text_response(&text, reason_ctx).await {
                    TextAction::Return(outcome) => return Ok(outcome),
                    TextAction::Continue => {}
                }
            }
            RespondResult::ToolCalls {
                tool_calls,
                content,
            } => {
                consecutive_tool_intent_nudges = 0;
                if let Some(outcome) = delegate
                    .execute_tool_calls(tool_calls, content, reason_ctx)
                    .await?
                {
                    return Ok(outcome);
                }
            }
        }
        delegate.after_iteration(iteration).await;
    }
    Ok(LoopOutcome::MaxIterations)
 }
 /// Truncate a string for log/status previews.
 ///
 /// `max` is a byte budget. The result is truncated at the last valid char
 /// boundary at or before `max` bytes, so it is always valid UTF-8.
 pub fn truncate_for_preview(s: &str, max: usize) -> String {
    if s.len() <= max {
        s.to_string()
    } else {
        let end = crate::util::floor_char_boundary(s, max);
        format!("{}...", &s[..end])
    }
 }
 #[cfg(test)]
 mod tests {
    use super::*;
    use crate::llm::{RespondOutput, TokenUsage, ToolCall};
    use crate::testing::StubLlm;
    use std::sync::Arc;
    use std::sync::atomic::{AtomicUsize, Ordering};
    use tokio::sync::Mutex;
    fn stub_reasoning() -> Reasoning {
        Reasoning::new(Arc::new(StubLlm::default()))
    }
    fn zero_usage() -> TokenUsage {
        TokenUsage {
            input_tokens: 0,
            output_tokens: 0,
            cache_read_input_tokens: 0,
            cache_creation_input_tokens: 0,
        }
    }
    fn text_output(text: &str) -> RespondOutput {
        RespondOutput {
            result: RespondResult::Text(text.to_string()),
            usage: zero_usage(),
        }
    }
    fn tool_calls_output(calls: Vec<ToolCall>) -> RespondOutput {
        RespondOutput {
            result: RespondResult::ToolCalls {
                tool_calls: calls,
                content: None,
            },
            usage: zero_usage(),
        }
    }
    /// Configurable mock delegate for testing run_agentic_loop.
    struct MockDelegate {
        signal: Mutex<LoopSignal>,
        llm_responses: Mutex<Vec<RespondOutput>>,
        tool_exec_count: AtomicUsize,
        tool_exec_outcome: Mutex<Option<LoopOutcome>>,
        iterations_seen: Mutex<Vec<usize>>,
        early_exit: Mutex<Option<(usize, LoopOutcome)>>,
        nudge_count: AtomicUsize,
    }
    impl MockDelegate {
        fn new(responses: Vec<RespondOutput>) -> Self {
            Self {
                signal: Mutex::new(LoopSignal::Continue),
                llm_responses: Mutex::new(responses),
                tool_exec_count: AtomicUsize::new(0),
                tool_exec_outcome: Mutex::new(None),
                iterations_seen: Mutex::new(Vec::new()),
                early_exit: Mutex::new(None),
                nudge_count: AtomicUsize::new(0),
            }
        }
        fn with_signal(mut self, signal: LoopSignal) -> Self {
            self.signal = Mutex::new(signal);
            self
        }
        fn with_early_exit(mut self, iteration: usize, outcome: LoopOutcome) -> Self {
            self.early_exit = Mutex::new(Some((iteration, outcome)));
            self
        }
    }
    #[async_trait]
    impl LoopDelegate for MockDelegate {
        async fn check_signals(&self) -> LoopSignal {
            let mut sig = self.signal.lock().await;
            std::mem::replace(&mut *sig, LoopSignal::Continue)
        }
        async fn before_llm_call(
            &self,
            _reason_ctx: &mut ReasoningContext,
            iteration: usize,
        ) -> Option<LoopOutcome> {
            let mut guard = self.early_exit.lock().await;
            let should_take = guard
                .as_ref()
                .is_some_and(|(target, _)| *target == iteration);
            if should_take {
                guard.take().map(|(_, o)| o)
            } else {
                None
            }
        }
        async fn call_llm(
            &self,
            _reasoning: &Reasoning,
            _reason_ctx: &mut ReasoningContext,
            _iteration: usize,
        ) -> Result<crate::llm::RespondOutput, crate::error::Error> {
            let mut responses = self.llm_responses.lock().await;
            if responses.is_empty() {
                panic!("MockDelegate: no more LLM responses queued");
            }
            Ok(responses.remove(0))
        }
        async fn handle_text_response(
            &self,
            text: &str,
            _reason_ctx: &mut ReasoningContext,
        ) -> TextAction {
            TextAction::Return(LoopOutcome::Response(text.to_string()))
        }
        async fn execute_tool_calls(
            &self,
            _tool_calls: Vec<ToolCall>,
            _content: Option<String>,
            reason_ctx: &mut ReasoningContext,
        ) -> Result<Option<LoopOutcome>, crate::error::Error> {
            self.tool_exec_count.fetch_add(1, Ordering::SeqCst);
            reason_ctx
                .messages
                .push(ChatMessage::user("tool result stub"));
            let outcome = self.tool_exec_outcome.lock().await.take();
            Ok(outcome)
        }
        async fn on_tool_intent_nudge(&self, _text: &str, _reason_ctx: &mut ReasoningContext) {
            self.nudge_count.fetch_add(1, Ordering::SeqCst);
        }
        async fn after_iteration(&self, iteration: usize) {
            self.iterations_seen.lock().await.push(iteration);
        }
    }
    // --- Tests ---
    #[tokio::test]
    async fn test_text_response_returns_immediately() {
        let delegate = MockDelegate::new(vec![text_output("Hello, world!")]);
        let reasoning = stub_reasoning();
        let mut ctx = ReasoningContext::new();
        let config = AgenticLoopConfig::default();
        let outcome = run_agentic_loop(&delegate, &reasoning, &mut ctx, &config)
            .await
            .unwrap();
        match outcome {
            LoopOutcome::Response(text) => assert_eq!(text, "Hello, world!"),
            _ => panic!("Expected LoopOutcome::Response"),
        }
        // after_iteration is NOT called when handle_text_response returns Return
        // (the loop exits before reaching after_iteration).
        assert!(delegate.iterations_seen.lock().await.is_empty());
    }
    #[tokio::test]
    async fn test_tool_call_then_text_response() {
        let tool_call = ToolCall {
            id: "call_1".to_string(),
            name: "echo".to_string(),
            arguments: serde_json::json!({}),
        };
        let delegate = MockDelegate::new(vec![
            tool_calls_output(vec![tool_call]),
            text_output("Done!"),
        ]);
        let reasoning = stub_reasoning();
        let mut ctx = ReasoningContext::new();
        let config = AgenticLoopConfig::default();
        let outcome = run_agentic_loop(&delegate, &reasoning, &mut ctx, &config)
            .await
            .unwrap();
        match outcome {
            LoopOutcome::Response(text) => assert_eq!(text, "Done!"),
            _ => panic!("Expected LoopOutcome::Response"),
        }
        assert_eq!(delegate.tool_exec_count.load(Ordering::SeqCst), 1);
        // after_iteration called for iteration 1 (tool call), but not 2
        // (text response exits before after_iteration).
        assert_eq!(*delegate.iterations_seen.lock().await, vec![1]);
    }
    #[tokio::test]
    async fn test_stop_signal_exits_immediately() {
        let delegate =
            MockDelegate::new(vec![text_output("unreachable")]).with_signal(LoopSignal::Stop);
        let reasoning = stub_reasoning();
        let mut ctx = ReasoningContext::new();
        let config = AgenticLoopConfig::default();
        let outcome = run_agentic_loop(&delegate, &reasoning, &mut ctx, &config)
            .await
            .unwrap();
        assert!(matches!(outcome, LoopOutcome::Stopped));
        assert!(delegate.iterations_seen.lock().await.is_empty());
    }
    #[tokio::test]
    async fn test_inject_message_adds_user_message() {
        let delegate = MockDelegate::new(vec![text_output("Got it")])
            .with_signal(LoopSignal::InjectMessage("injected prompt".to_string()));
        let reasoning = stub_reasoning();
        let mut ctx = ReasoningContext::new();
        let config = AgenticLoopConfig::default();
        let outcome = run_agentic_loop(&delegate, &reasoning, &mut ctx, &config)
            .await
            .unwrap();
        assert!(matches!(outcome, LoopOutcome::Response(_)));
        assert!(
            ctx.messages
                .iter()
                .any(|m| m.role == crate::llm::Role::User && m.content.contains("injected prompt")),
            "Injected message should appear in context"
        );
    }
    #[tokio::test]
    async fn test_max_iterations_reached() {
        struct ContinueDelegate;
        #[async_trait]
        impl LoopDelegate for ContinueDelegate {
            async fn check_signals(&self) -> LoopSignal {
                LoopSignal::Continue
            }
            async fn before_llm_call(
                &self,
                _: &mut ReasoningContext,
                _: usize,
            ) -> Option<LoopOutcome> {
                None
            }
            async fn call_llm(
                &self,
                _: &Reasoning,
                _: &mut ReasoningContext,
                _: usize,
            ) -> Result<crate::llm::RespondOutput, crate::error::Error> {
                Ok(text_output("still working"))
            }
            async fn handle_text_response(
                &self,
                _: &str,
                ctx: &mut ReasoningContext,
            ) -> TextAction {
                ctx.messages.push(ChatMessage::assistant("still working"));
                TextAction::Continue
            }
            async fn execute_tool_calls(
                &self,
                _: Vec<ToolCall>,
                _: Option<String>,
                _: &mut ReasoningContext,
            ) -> Result<Option<LoopOutcome>, crate::error::Error> {
                Ok(None)
            }
        }
        let delegate = ContinueDelegate;
        let reasoning = stub_reasoning();
        let mut ctx = ReasoningContext::new();
        let config = AgenticLoopConfig {
            max_iterations: 3,
            ..Default::default()
        };
        let outcome = run_agentic_loop(&delegate, &reasoning, &mut ctx, &config)
            .await
            .unwrap();
        assert!(matches!(outcome, LoopOutcome::MaxIterations));
        let assistant_count = ctx
            .messages
            .iter()
            .filter(|m| m.role == crate::llm::Role::Assistant)
            .count();
        assert_eq!(assistant_count, 3);
    }
    #[tokio::test]
    async fn test_tool_intent_nudge_fires_and_caps() {
        let delegate = MockDelegate::new(vec![
            text_output("Let me search for that file"),
            text_output("Let me search for that file"),
            text_output("Let me search for that file"),
        ]);
        let reasoning = stub_reasoning();
        let mut ctx = ReasoningContext::new();
        ctx.available_tools.push(crate::llm::ToolDefinition {
            name: "search".to_string(),
            description: "Search files".to_string(),
            parameters: serde_json::json!({"type": "object"}),
        });
        let config = AgenticLoopConfig {
            max_iterations: 10,
            enable_tool_intent_nudge: true,
            max_tool_intent_nudges: 2,
        };
        let outcome = run_agentic_loop(&delegate, &reasoning, &mut ctx, &config)
            .await
            .unwrap();
        assert!(matches!(outcome, LoopOutcome::Response(_)));
        assert_eq!(delegate.nudge_count.load(Ordering::SeqCst), 2);
        let nudge_messages = ctx
            .messages
            .iter()
            .filter(|m| {
                m.role == crate::llm::Role::User
                    && m.content.contains("you did not include any tool calls")
            })
            .count();
        assert_eq!(
            nudge_messages, 2,
            "Should have exactly 2 nudge messages in context"
        );
    }
    #[tokio::test]
    async fn test_before_llm_call_early_exit() {
        let delegate = MockDelegate::new(vec![text_output("unreachable")])
            .with_early_exit(1, LoopOutcome::Stopped);
        let reasoning = stub_reasoning();
        let mut ctx = ReasoningContext::new();
        let config = AgenticLoopConfig::default();
        let outcome = run_agentic_loop(&delegate, &reasoning, &mut ctx, &config)
            .await
            .unwrap();
        assert!(matches!(outcome, LoopOutcome::Stopped));
        assert!(delegate.iterations_seen.lock().await.is_empty());
    }
    #[test]
    fn test_truncate_short_string_unchanged() {
        assert_eq!(truncate_for_preview("hello", 10), "hello");
    }
    #[test]
    fn test_truncate_long_string_adds_ellipsis() {
        let result = truncate_for_preview("hello world", 5);
        assert_eq!(result, "hello...");
    }
    #[test]
    fn test_truncate_multibyte_safe() {
        let result = truncate_for_preview("café", 4);
        assert_eq!(result, "caf...");
    }
 }
@@ -405,8 +405,7 @@ impl Agent {
            .with_max_tokens(512)
            .with_temperature(0.3);
-        let reasoning =
+        let reasoning = Reasoning::new(self.llm().clone());
            Reasoning::new(self.llm().clone()).with_model_name(self.llm().active_model_name());
        match reasoning.complete(request).await {
            Ok((text, _usage)) => Ok(SubmissionResult::response(format!(
                "Thread Summary:\n\n{}",
@@ -454,8 +453,7 @@ impl Agent {
            .with_max_tokens(512)
            .with_temperature(0.5);
-        let reasoning =
+        let reasoning = Reasoning::new(self.llm().clone());
            Reasoning::new(self.llm().clone()).with_model_name(self.llm().active_model_name());
        match reasoning.complete(request).await {
            Ok((text, _usage)) => Ok(SubmissionResult::response(format!(
                "Suggested Next Steps:\n\n{}",
@@ -227,8 +227,7 @@ Be brief but capture all important details. Use bullet points."#,
            .with_max_tokens(1024)
            .with_temperature(0.3);
-        let reasoning =
+        let reasoning = Reasoning::new(self.llm.clone());
            Reasoning::new(self.llm.clone()).with_model_name(self.llm.active_model_name());
        let (text, _) = reasoning.complete(request).await?;
        Ok(text)
    }
@@ -14,12 +14,7 @@ use crate::agent::session::{PendingApproval, Session, ThreadState};
 use crate::channels::{IncomingMessage, StatusUpdate};
 use crate::context::JobContext;
 use crate::error::Error;
-use async_trait::async_trait;
+use crate::llm::{ChatMessage, Reasoning, ReasoningContext, RespondResult};
 use crate::agent::agentic_loop::{
    AgenticLoopConfig, LoopDelegate, LoopOutcome, LoopSignal, TextAction,
 };
 use crate::llm::{ChatMessage, Reasoning, ReasoningContext};
 use crate::tools::redact_params;
 /// Result of the agentic loop execution.
@@ -90,7 +85,7 @@ impl Agent {
                    crate::skills::SkillTrust::Installed => "INSTALLED",
                };
-                tracing::debug!(
+                tracing::info!(
                    skill_name = skill.name(),
                    skill_version = skill.version(),
                    trust = %skill.trust,
@@ -138,6 +133,9 @@ impl Agent {
            reasoning = reasoning.with_skill_context(ctx);
        }
        // Build context with messages that we'll mutate during the loop
        let mut context_messages = initial_messages;
        // Create a JobContext for tool execution (chat doesn't have a real job)
        let mut job_ctx =
            JobContext::with_user(&message.user_id, "chat", "Interactive chat session");
@@ -156,118 +154,53 @@ impl Agent {
        let cached_prompt_no_tools = reasoning.build_system_prompt_with_tools(&[]);
        let max_tool_iterations = self.config.max_tool_iterations;
        // Force a text-only response on the last iteration to guarantee termination
        // instead of hard-erroring. The penultimate iteration also gets a nudge
        // message so the LLM knows it should wrap up.
        let force_text_at = max_tool_iterations;
        let nudge_at = max_tool_iterations.saturating_sub(1);
-
+        let mut iteration = 0;
-        let delegate = ChatDelegate {
+        const MAX_TOOL_INTENT_NUDGES: u32 = 2;
-            agent: self,
+        let mut consecutive_tool_intent_nudges: u32 = 0;
-            session: session.clone(),
+        loop {
-            thread_id,
+            iteration += 1;
-            message,
+            // Hard ceiling one past the forced-text iteration (should never be reached
-            job_ctx,
+            // since force_text_at guarantees a text response, but kept as a safety net).
-            active_skills,
+            if iteration > max_tool_iterations + 1 {
-            cached_prompt,
+                return Err(crate::error::LlmError::InvalidResponse {
            cached_prompt_no_tools,
            nudge_at,
            force_text_at,
            user_tz,
        };
        let mut reason_ctx = ReasoningContext::new()
            .with_messages(initial_messages)
            .with_tools(initial_tool_defs)
            .with_system_prompt(delegate.cached_prompt.clone())
            .with_metadata({
                let mut m = std::collections::HashMap::new();
                m.insert("thread_id".to_string(), thread_id.to_string());
                m
            });
        let loop_config = AgenticLoopConfig {
            // Hard ceiling: one past force_text_at (safety net).
            max_iterations: max_tool_iterations + 1,
            enable_tool_intent_nudge: true,
            max_tool_intent_nudges: 2,
        };
        let outcome = crate::agent::agentic_loop::run_agentic_loop(
            &delegate,
            &reasoning,
            &mut reason_ctx,
            &loop_config,
        )
        .await?;
        match outcome {
            LoopOutcome::Response(text) => Ok(AgenticLoopResult::Response(text)),
            LoopOutcome::Stopped => Err(crate::error::JobError::ContextError {
                id: thread_id,
                reason: "Interrupted".to_string(),
            }
            .into()),
            LoopOutcome::MaxIterations => Err(crate::error::LlmError::InvalidResponse {
                    provider: "agent".to_string(),
                    reason: format!("Exceeded maximum tool iterations ({max_tool_iterations})"),
                }
-            .into()),
+                .into());
            LoopOutcome::NeedApproval(pending) => {
                Ok(AgenticLoopResult::NeedApproval { pending: *pending })
            }
        }
            }
-    /// Execute a tool for chat (without full job context).
+            // Check if interrupted
-    pub(super) async fn execute_chat_tool(
+            {
-        &self,
+                let sess = session.lock().await;
-        tool_name: &str,
+                if let Some(thread) = sess.threads.get(&thread_id)
        params: &serde_json::Value,
        job_ctx: &JobContext,
    ) -> Result<String, Error> {
        execute_chat_tool_standalone(self.tools(), self.safety(), tool_name, params, job_ctx).await
    }
 }
 /// Delegate for the chat (dispatcher) context.
 ///
 /// Implements `LoopDelegate` to customize the shared agentic loop for
 /// interactive chat sessions with the full 3-phase tool execution
 /// (preflight → parallel exec → post-flight), approval flow, hooks,
 /// auth intercept, and cost tracking.
 struct ChatDelegate<'a> {
    agent: &'a Agent,
    session: Arc<Mutex<Session>>,
    thread_id: Uuid,
    message: &'a IncomingMessage,
    job_ctx: JobContext,
    active_skills: Vec<crate::skills::LoadedSkill>,
    cached_prompt: String,
    cached_prompt_no_tools: String,
    nudge_at: usize,
    force_text_at: usize,
    user_tz: chrono_tz::Tz,
 }
 #[async_trait]
 impl<'a> LoopDelegate for ChatDelegate<'a> {
    async fn check_signals(&self) -> LoopSignal {
        let sess = self.session.lock().await;
        if let Some(thread) = sess.threads.get(&self.thread_id)
                    && thread.state == ThreadState::Interrupted
                {
-            return LoopSignal::Stop;
+                    return Err(crate::error::JobError::ContextError {
                        id: thread_id,
                        reason: "Interrupted".to_string(),
                    }
                    .into());
                }
        LoopSignal::Continue
            }
-    async fn before_llm_call(
+            // Enforce cost guardrails before the LLM call
-        &self,
+            if let Err(limit) = self.cost_guard().check_allowed().await {
-        reason_ctx: &mut ReasoningContext,
+                return Err(crate::error::LlmError::InvalidResponse {
-        iteration: usize,
+                    provider: "agent".to_string(),
-    ) -> Option<LoopOutcome> {
+                    reason: limit.to_string(),
                }
                .into());
            }
            // Inject a nudge message when approaching the iteration limit so the
            // LLM is aware it should produce a final answer on the next turn.
-        if iteration == self.nudge_at {
+            if iteration == nudge_at {
-            reason_ctx.messages.push(ChatMessage::system(
+                context_messages.push(ChatMessage::system(
                    "You are approaching the tool call limit. \
                     Provide your best final answer on the next response \
                     using the information you have gathered so far. \
@@ -275,15 +208,15 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                ));
            }
-        let force_text = iteration >= self.force_text_at;
+            let force_text = iteration >= force_text_at;
            // Refresh tool definitions each iteration so newly built tools become visible
-        let tool_defs = self.agent.tools().tool_definitions().await;
+            let tool_defs = self.tools().tool_definitions().await;
            // Apply trust-based tool attenuation if skills are active.
-        let tool_defs = if !self.active_skills.is_empty() {
+            let tool_defs = if !active_skills.is_empty() {
-            let result = crate::skills::attenuate_tools(&tool_defs, &self.active_skills);
+                let result = crate::skills::attenuate_tools(&tool_defs, &active_skills);
-            tracing::debug!(
+                tracing::info!(
                    min_trust = %result.min_trust,
                    tools_available = result.tools.len(),
                    tools_removed = result.removed_tools.len(),
@@ -296,14 +229,23 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                tool_defs
            };
-        // Update context for this iteration
+            // Call LLM with current context; force_text drops tools to guarantee a
-        reason_ctx.available_tools = tool_defs;
+            // text response on the final iteration. The pre-built system prompt
-        reason_ctx.system_prompt = Some(if force_text {
+            // avoids rebuilding the same ~1,500-token string each iteration.
-            self.cached_prompt_no_tools.clone()
+            let mut context = ReasoningContext::new()
                .with_messages(context_messages.clone())
                .with_tools(tool_defs)
                .with_system_prompt(if force_text {
                    cached_prompt_no_tools.clone()
                } else {
-            self.cached_prompt.clone()
+                    cached_prompt.clone()
                })
                .with_metadata({
                    let mut m = std::collections::HashMap::new();
                    m.insert("thread_id".to_string(), thread_id.to_string());
                    m
                });
-        reason_ctx.force_text = force_text;
+            context.force_text = force_text;
            if force_text {
                tracing::info!(
@@ -313,34 +255,15 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
            }
            let _ = self
            .agent
                .channels
                .send_status(
-                &self.message.channel,
+                    &message.channel,
                    StatusUpdate::Thinking("Calling LLM...".into()),
-                &self.message.metadata,
+                    &message.metadata,
                )
                .await;
-        None
+            let output = match reasoning.respond_with_tools(&context).await {
    }
    async fn call_llm(
        &self,
        reasoning: &Reasoning,
        reason_ctx: &mut ReasoningContext,
        iteration: usize,
    ) -> Result<crate::llm::RespondOutput, Error> {
        // Enforce cost guardrails before the LLM call
        if let Err(limit) = self.agent.cost_guard().check_allowed().await {
            return Err(crate::error::LlmError::InvalidResponse {
                provider: "agent".to_string(),
                reason: limit.to_string(),
            }
            .into());
        }
        let output = match reasoning.respond_with_tools(reason_ctx).await {
                Ok(output) => output,
                Err(crate::error::LlmError::ContextLengthExceeded { used, limit }) => {
                    tracing::warn!(
@@ -350,16 +273,23 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                        "Context length exceeded, compacting messages and retrying"
                    );
-                // Compact messages in place and retry
+                    // Compact: keep system messages + last user message + current turn
-                reason_ctx.messages = compact_messages_for_retry(&reason_ctx.messages);
+                    context_messages = compact_messages_for_retry(&context_messages);
-                // When force_text, clear tools to further reduce token count
+                    // Rebuild context with compacted messages, reusing cached prompt
-                if reason_ctx.force_text {
+                    let mut retry_context = ReasoningContext::new()
-                    reason_ctx.available_tools.clear();
+                        .with_messages(context_messages.clone())
-                }
+                        .with_tools(if force_text {
                            Vec::new()
                        } else {
                            context.available_tools.clone()
                        })
                        .with_metadata(context.metadata.clone());
                    retry_context.force_text = force_text;
                    retry_context.system_prompt = context.system_prompt.clone();
                    reasoning
-                    .respond_with_tools(reason_ctx)
+                        .respond_with_tools(&retry_context)
                        .await
                        .map_err(|retry_err| {
                            tracing::error!(
@@ -368,6 +298,7 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                                retry_error = %retry_err,
                                "Retry after auto-compaction also failed"
                            );
                            // Propagate the actual retry error so callers see the real failure
                            crate::error::Error::from(retry_err)
                        })?
                }
@@ -375,11 +306,10 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
            };
            // Record cost and track token usage
-        let model_name = self.agent.llm().active_model_name();
+            let model_name = self.llm().active_model_name();
-        let read_discount = self.agent.llm().cache_read_discount();
+            let read_discount = self.llm().cache_read_discount();
-        let write_multiplier = self.agent.llm().cache_write_multiplier();
+            let write_multiplier = self.llm().cache_write_multiplier();
            let call_cost = self
            .agent
                .cost_guard()
                .record_llm_call(
                    &model_name,
@@ -389,7 +319,7 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                    output.usage.cache_creation_input_tokens,
                    read_discount,
                    write_multiplier,
-                Some(self.agent.llm().cost_per_token()),
+                    Some(self.llm().cost_per_token()),
                )
                .await;
            tracing::debug!(
@@ -399,60 +329,72 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                call_cost,
            );
-        Ok(output)
+            match output.result {
                RespondResult::Text(text) => {
                    // Nudge the LLM if it expressed tool intent without calling tools.
                    // This is common with non-Anthropic models (e.g. GLM-5 via NEAR AI)
                    // that output "Let me search…" but don't issue tool_calls.
                    if !force_text
                        && !context.available_tools.is_empty()
                        && consecutive_tool_intent_nudges < MAX_TOOL_INTENT_NUDGES
                        && crate::llm::llm_signals_tool_intent(&text)
                    {
                        consecutive_tool_intent_nudges += 1;
                        tracing::info!(
                            iteration,
                            "LLM expressed tool intent without calling a tool, nudging"
                        );
                        context_messages.push(ChatMessage::assistant(&text));
                        context_messages.push(ChatMessage::user(crate::llm::TOOL_INTENT_NUDGE));
                        continue;
                    }
    async fn handle_text_response(
        &self,
        text: &str,
        _reason_ctx: &mut ReasoningContext,
    ) -> TextAction {
                    // Strip internal "[Called tool ...]" text that can leak when
                    // provider flattening (e.g. NEAR AI) converts tool_calls to
                    // plain text and the LLM echoes it back.
-        let sanitized = strip_internal_tool_call_text(text);
+                    let sanitized = strip_internal_tool_call_text(&text);
-        TextAction::Return(LoopOutcome::Response(sanitized))
+                    return Ok(AgenticLoopResult::Response(sanitized));
                }
-
+                RespondResult::ToolCalls {
-    async fn execute_tool_calls(
+                    tool_calls,
-        &self,
+                    content,
-        tool_calls: Vec<crate::llm::ToolCall>,
+                } => {
-        content: Option<String>,
+                    consecutive_tool_intent_nudges = 0;
        reason_ctx: &mut ReasoningContext,
    ) -> Result<Option<LoopOutcome>, Error> {
                    // Add the assistant message with tool_calls to context.
                    // OpenAI protocol requires this before tool-result messages.
-        reason_ctx
+                    context_messages.push(ChatMessage::assistant_with_tool_calls(
            .messages
            .push(ChatMessage::assistant_with_tool_calls(
                        content,
                        tool_calls.clone(),
                    ));
                    // Execute tools and add results to context
                    let _ = self
            .agent
                        .channels
                        .send_status(
-                &self.message.channel,
+                            &message.channel,
-                StatusUpdate::Thinking(format!("Executing {} tool(s)...", tool_calls.len())),
+                            StatusUpdate::Thinking(format!(
-                &self.message.metadata,
+                                "Executing {} tool(s)...",
                                tool_calls.len()
                            )),
                            &message.metadata,
                        )
                        .await;
                    // Record tool calls in the thread with sensitive params redacted.
                    // Look up each tool's sensitive_params before acquiring the session lock.
                    {
-            let mut redacted_args: Vec<serde_json::Value> = Vec::with_capacity(tool_calls.len());
+                        let mut redacted_args: Vec<serde_json::Value> =
                            Vec::with_capacity(tool_calls.len());
                        for tc in &tool_calls {
-                let safe = if let Some(tool) = self.agent.tools().get(&tc.name).await {
+                            let safe = if let Some(tool) = self.tools().get(&tc.name).await {
                                redact_params(&tc.arguments, tool.sensitive_params())
                            } else {
                                tc.arguments.clone()
                            };
                            redacted_args.push(safe);
                        }
-            let mut sess = self.session.lock().await;
+                        let mut sess = session.lock().await;
-            if let Some(thread) = sess.threads.get_mut(&self.thread_id)
+                        if let Some(thread) = sess.threads.get_mut(&thread_id)
                            && let Some(turn) = thread.last_turn_mut()
                        {
                            for (tc, safe_args) in tool_calls.iter().zip(redacted_args) {
@@ -465,8 +407,13 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                    // Walk tool_calls checking approval and hooks. Classify
                    // each tool as Rejected (by hook) or Runnable. Stop at the
                    // first tool that needs approval.
                    //
                    // Outcomes are indexed by original tool_calls position so
                    // Phase 3 can emit results in the correct order.
                    enum PreflightOutcome {
                        /// Hook rejected/blocked this tool; contains the error message.
                        Rejected(String),
                        /// Tool passed preflight and will be executed.
                        Runnable,
                    }
                    let mut preflight: Vec<(crate::llm::ToolCall, PreflightOutcome)> = Vec::new();
@@ -480,21 +427,26 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                    for (idx, original_tc) in tool_calls.iter().enumerate() {
                        let mut tc = original_tc.clone();
-            let tool_opt = self.agent.tools().get(&tc.name).await;
+                        // Fetch the tool upfront so we can redact sensitive params
                        // before they touch hooks or approval display.
                        let tool_opt = self.tools().get(&tc.name).await;
                        let sensitive = tool_opt
                            .as_ref()
                            .map(|t| t.sensitive_params())
                            .unwrap_or(&[]);
-            // Hook: BeforeToolCall
+                        // Hook: BeforeToolCall (runs before approval so hooks can
                        // modify parameters — approval is checked on final params).
                        // Hooks receive redacted params so sensitive values are not
                        // exposed to hook handlers or their logs.
                        let hook_params = redact_params(&tc.arguments, sensitive);
                        let event = crate::hooks::HookEvent::ToolCall {
                            tool_name: tc.name.clone(),
                            parameters: hook_params,
-                user_id: self.message.user_id.clone(),
+                            user_id: message.user_id.clone(),
                            context: "chat".to_string(),
                        };
-            match self.agent.hooks().run(&event).await {
+                        match self.hooks().run(&event).await {
                            Err(crate::hooks::HookError::Rejected { reason }) => {
                                preflight.push((
                                    tc,
@@ -503,7 +455,7 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                                        reason
                                    )),
                                ));
-                    continue;
+                                continue; // skip to next tool (not infinite: using for loop)
                            }
                            Err(err) => {
                                preflight.push((
@@ -519,9 +471,12 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                                modified: Some(new_params),
                            }) => match serde_json::from_str::<serde_json::Value>(&new_params) {
                                Ok(mut parsed) => {
                                    // Restore original sensitive param values so a hook
                                    // cannot overwrite them (they were sent as [REDACTED]).
                                    if let Some(obj) = parsed.as_object_mut() {
                                        for key in sensitive {
-                                if let Some(orig_val) = original_tc.arguments.get(*key) {
+                                            if let Some(orig_val) = original_tc.arguments.get(*key)
                                            {
                                                obj.insert((*key).to_string(), orig_val.clone());
                                            }
                                        }
@@ -539,15 +494,16 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                            _ => {}
                        }
-            // Check if tool requires approval
+                        // Check if tool requires approval on the final (post-hook)
-            if !self.agent.config.auto_approve_tools
+                        // parameters. Skipped when auto_approve_tools is set.
                        if !self.config.auto_approve_tools
                            && let Some(tool) = tool_opt
                        {
                            use crate::tools::ApprovalRequirement;
                            let needs_approval = match tool.requires_approval(&tc.arguments) {
                                ApprovalRequirement::Never => false,
                                ApprovalRequirement::UnlessAutoApproved => {
-                        let sess = self.session.lock().await;
+                                    let sess = session.lock().await;
                                    !sess.is_tool_auto_approved(&tc.name)
                                }
                                ApprovalRequirement::Always => true,
@@ -555,7 +511,7 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                            if needs_approval {
                                approval_needed = Some((idx, tc, tool));
-                    break;
+                                break; // remaining tools are deferred
                            }
                        }
@@ -565,58 +521,59 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                    }
                    // === Phase 2: Parallel execution ===
                    // Execute runnable tools and slot results back by preflight
                    // index so Phase 3 can iterate in original order.
                    let mut exec_results: Vec<Option<Result<String, Error>>> =
                        (0..preflight.len()).map(|_| None).collect();
                    if runnable.len() <= 1 {
                        // Single tool (or none): execute inline
                        for (pf_idx, tc) in &runnable {
                            let _ = self
                    .agent
                                .channels
                                .send_status(
-                        &self.message.channel,
+                                    &message.channel,
                                    StatusUpdate::ToolStarted {
                                        name: tc.name.clone(),
                                    },
-                        &self.message.metadata,
+                                    &message.metadata,
                                )
                                .await;
                            let result = self
-                    .agent
+                                .execute_chat_tool(&tc.name, &tc.arguments, &job_ctx)
                    .execute_chat_tool(&tc.name, &tc.arguments, &self.job_ctx)
                                .await;
-                let disp_tool = self.agent.tools().get(&tc.name).await;
+                            let disp_tool = self.tools().get(&tc.name).await;
                            let _ = self
                    .agent
                                .channels
                                .send_status(
-                        &self.message.channel,
+                                    &message.channel,
                                    StatusUpdate::tool_completed(
                                        tc.name.clone(),
                                        &result,
                                        &tc.arguments,
                                        disp_tool.as_deref(),
                                    ),
-                        &self.message.metadata,
+                                    &message.metadata,
                                )
                                .await;
                            exec_results[*pf_idx] = Some(result);
                        }
                    } else {
                        // Multiple tools: execute in parallel via JoinSet
                        let mut join_set = JoinSet::new();
                        for (pf_idx, tc) in &runnable {
                            let pf_idx = *pf_idx;
-                let tools = self.agent.tools().clone();
+                            let tools = self.tools().clone();
-                let safety = self.agent.safety().clone();
+                            let safety = self.safety().clone();
-                let channels = self.agent.channels.clone();
+                            let channels = self.channels.clone();
-                let job_ctx = self.job_ctx.clone();
+                            let job_ctx = job_ctx.clone();
                            let tc = tc.clone();
-                let channel = self.message.channel.clone();
+                            let channel = message.channel.clone();
-                let metadata = self.message.metadata.clone();
+                            let metadata = message.metadata.clone();
                            join_set.spawn(async move {
                                let _ = channels
@@ -665,20 +622,25 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                                    if e.is_panic() {
                                        tracing::error!("Chat tool execution task panicked: {}", e);
                                    } else {
-                            tracing::error!("Chat tool execution task cancelled: {}", e);
+                                        tracing::error!(
                                            "Chat tool execution task cancelled: {}",
                                            e
                                        );
                                    }
                                }
                            }
                        }
                        // Fill panicked slots with error results
-            for (pf_idx, tc) in runnable.iter() {
+                        for (runnable_idx, (pf_idx, tc)) in runnable.iter().enumerate() {
                            if exec_results[*pf_idx].is_none() {
                                tracing::error!(
                                    tool = %tc.name,
                                    runnable_idx,
                                    "Filling failed task slot with error"
                                );
-                    exec_results[*pf_idx] = Some(Err(crate::error::ToolError::ExecutionFailed {
+                                exec_results[*pf_idx] =
                                    Some(Err(crate::error::ToolError::ExecutionFailed {
                                        name: tc.name.clone(),
                                        reason: "Task failed during execution".to_string(),
                                    }
@@ -688,25 +650,30 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                    }
                    // === Phase 3: Post-flight (sequential, in original order) ===
                    // Process all results — both hook rejections and execution
                    // results — in the original tool_calls order. Auth intercept
                    // is deferred until after every result is recorded.
                    let mut deferred_auth: Option<String> = None;
                    for (pf_idx, (tc, outcome)) in preflight.into_iter().enumerate() {
                        match outcome {
                            PreflightOutcome::Rejected(error_msg) => {
                                // Record hook rejection in thread
                                {
-                        let mut sess = self.session.lock().await;
+                                    let mut sess = session.lock().await;
-                        if let Some(thread) = sess.threads.get_mut(&self.thread_id)
+                                    if let Some(thread) = sess.threads.get_mut(&thread_id)
                                        && let Some(turn) = thread.last_turn_mut()
                                    {
                                        turn.record_tool_error(error_msg.clone());
                                    }
                                }
-                    reason_ctx
+                                context_messages
                        .messages
                                    .push(ChatMessage::tool_result(&tc.id, &tc.name, error_msg));
                            }
                            PreflightOutcome::Runnable => {
-                    let tool_result = exec_results[pf_idx].take().unwrap_or_else(|| {
+                                // Retrieve the execution result for this slot
                                let tool_result =
                                    exec_results[pf_idx].take().unwrap_or_else(|| {
                                        Err(crate::error::ToolError::ExecutionFailed {
                                            name: tc.name.clone(),
                                            reason: "No result available".to_string(),
@@ -714,11 +681,13 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                                        .into())
                                    });
-                    // Detect image generation sentinel
+                                // Detect image generation sentinel in tool output
                                // (only from image tools — avoids parsing all tool outputs)
                                let is_image_sentinel = if let Ok(ref output) = tool_result
                                    && matches!(tc.name.as_str(), "image_generate" | "image_edit")
                                {
-                        if let Ok(sentinel) = serde_json::from_str::<serde_json::Value>(output)
+                                    if let Ok(sentinel) =
                                        serde_json::from_str::<serde_json::Value>(output)
                                        && sentinel.get("type").and_then(|v| v.as_str())
                                            == Some("image_generated")
                                    {
@@ -731,18 +700,19 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                                            .get("path")
                                            .and_then(|v| v.as_str())
                                            .map(String::from);
                                        // Skip broadcasting if data_url is empty to avoid
                                        // sending a broken ImageGenerated SSE event.
                                        if data_url.is_empty() {
                                            tracing::warn!(
                                                "Image generation sentinel has empty data URL, skipping broadcast"
                                            );
                                        } else {
                                            let _ = self
                                    .agent
                                                .channels
                                                .send_status(
-                                        &self.message.channel,
+                                                    &message.channel,
                                                    StatusUpdate::ImageGenerated { data_url, path },
-                                        &self.message.metadata,
+                                                    &message.metadata,
                                                )
                                                .await;
                                        }
@@ -754,49 +724,49 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                                    false
                                };
-                    // Send ToolResult preview
+                                // Send ToolResult preview (skip for image sentinels to avoid
                                // broadcasting multi-MB base64 data as a preview)
                                if !is_image_sentinel
                                    && let Ok(ref output) = tool_result
                                    && !output.is_empty()
                                {
                                    let _ = self
                            .agent
                                        .channels
                                        .send_status(
-                                &self.message.channel,
+                                            &message.channel,
                                            StatusUpdate::ToolResult {
                                                name: tc.name.clone(),
                                                preview: output.clone(),
                                            },
-                                &self.message.metadata,
+                                            &message.metadata,
                                        )
                                        .await;
                                }
-                    // Check for auth awaiting
+                                // Check for auth awaiting — defer the return
                                // until all results are recorded.
                                if deferred_auth.is_none()
                                    && let Some((ext_name, instructions)) =
                                        check_auth_required(&tc.name, &tool_result)
                                {
                                    let auth_data = parse_auth_result(&tool_result);
                                    {
-                            let mut sess = self.session.lock().await;
+                                        let mut sess = session.lock().await;
-                            if let Some(thread) = sess.threads.get_mut(&self.thread_id) {
+                                        if let Some(thread) = sess.threads.get_mut(&thread_id) {
                                            thread.enter_auth_mode(ext_name.clone());
                                        }
                                    }
                                    let _ = self
                            .agent
                                        .channels
                                        .send_status(
-                                &self.message.channel,
+                                            &message.channel,
                                            StatusUpdate::AuthRequired {
                                                extension_name: ext_name,
                                                instructions: Some(instructions.clone()),
                                                auth_url: auth_data.auth_url,
                                                setup_url: auth_data.setup_url,
                                            },
-                                &self.message.metadata,
+                                            &message.metadata,
                                        )
                                        .await;
                                    deferred_auth = Some(instructions);
@@ -804,7 +774,7 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                                // Stash full output so subsequent tools can reference it
                                if let Ok(ref output) = tool_result {
-                        self.job_ctx
+                                    job_ctx
                                        .tool_output_stash
                                        .write()
                                        .await
@@ -816,8 +786,8 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                                let result_content = match tool_result {
                                    Ok(output) => {
                                        let sanitized =
-                                self.agent.safety().sanitize_tool_output(&tc.name, &output);
+                                            self.safety().sanitize_tool_output(&tc.name, &output);
-                            self.agent.safety().wrap_for_llm(
+                                        self.safety().wrap_for_llm(
                                            &tc.name,
                                            &sanitized.content,
                                            sanitized.was_modified,
@@ -826,21 +796,24 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                                    Err(e) => format!("Tool '{}' failed: {}", tc.name, e),
                                };
-                    // Record sanitized result in thread
+                                // Record sanitized result in thread so messages()
                                // and persist_tool_calls() use cleaned content.
                                {
-                        let mut sess = self.session.lock().await;
+                                    let mut sess = session.lock().await;
-                        if let Some(thread) = sess.threads.get_mut(&self.thread_id)
+                                    if let Some(thread) = sess.threads.get_mut(&thread_id)
                                        && let Some(turn) = thread.last_turn_mut()
                                    {
                                        if is_tool_error {
                                            turn.record_tool_error(result_content.clone());
                                        } else {
-                                turn.record_tool_result(serde_json::json!(result_content));
+                                            turn.record_tool_result(serde_json::json!(
                                                result_content
                                            ));
                                        }
                                    }
                                }
-                    reason_ctx.messages.push(ChatMessage::tool_result(
+                                context_messages.push(ChatMessage::tool_result(
                                    &tc.id,
                                    &tc.name,
                                    result_content,
@@ -851,11 +824,14 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                    // Return auth response after all results are recorded
                    if let Some(instructions) = deferred_auth {
-            return Ok(Some(LoopOutcome::Response(instructions)));
+                        return Ok(AgenticLoopResult::Response(instructions));
                    }
                    // Handle approval if a tool needed it
                    if let Some((approval_idx, tc, tool)) = approval_needed {
                        // Show redacted params in the approval UI — the user already knows
                        // the sensitive value (they provided it); showing it again is
                        // unnecessary and creates a leakage path through channel logs.
                        let display_params = redact_params(&tc.arguments, tool.sensitive_params());
                        let pending = PendingApproval {
                            request_id: Uuid::new_v4(),
@@ -864,23 +840,34 @@ impl<'a> LoopDelegate for ChatDelegate<'a> {
                            display_parameters: display_params,
                            description: tool.description().to_string(),
                            tool_call_id: tc.id.clone(),
-                context_messages: reason_ctx.messages.clone(),
+                            context_messages: context_messages.clone(),
                            deferred_tool_calls: tool_calls[approval_idx + 1..].to_vec(),
-                user_timezone: Some(self.user_tz.name().to_string()),
+                            user_timezone: Some(user_tz.name().to_string()),
                        };
-            return Ok(Some(LoopOutcome::NeedApproval(Box::new(pending))));
+                        return Ok(AgenticLoopResult::NeedApproval { pending });
                    }
                }
            }
        }
    }
-        Ok(None)
+    /// Execute a tool for chat (without full job context).
    pub(super) async fn execute_chat_tool(
        &self,
        tool_name: &str,
        params: &serde_json::Value,
        job_ctx: &JobContext,
    ) -> Result<String, Error> {
        execute_chat_tool_standalone(self.tools(), self.safety(), tool_name, params, job_ctx).await
    }
 }
 /// Execute a chat tool without requiring `&Agent`.
 ///
 /// This standalone function enables parallel invocation from spawned JoinSet
-/// tasks, which cannot borrow `&self`. Delegates to the shared
+/// tasks, which cannot borrow `&self`. It replicates the logic from
-/// `execute_tool_with_safety` pipeline.
+/// `Agent::execute_chat_tool`.
 pub(super) async fn execute_chat_tool_standalone(
    tools: &crate::tools::ToolRegistry,
    safety: &crate::safety::SafetyLayer,
@@ -888,7 +875,91 @@ pub(super) async fn execute_chat_tool_standalone(
    params: &serde_json::Value,
    job_ctx: &crate::context::JobContext,
 ) -> Result<String, Error> {
-    crate::tools::execute::execute_tool_with_safety(tools, safety, tool_name, params, job_ctx).await
+    let tool = tools
        .get(tool_name)
        .await
        .ok_or_else(|| crate::error::ToolError::NotFound {
            name: tool_name.to_string(),
        })?;
    // Validate tool parameters
    let validation = safety.validator().validate_tool_params(params);
    if !validation.is_valid {
        let details = validation
            .errors
            .iter()
            .map(|e| format!("{}: {}", e.field, e.message))
            .collect::<Vec<_>>()
            .join("; ");
        return Err(crate::error::ToolError::InvalidParameters {
            name: tool_name.to_string(),
            reason: format!("Invalid tool parameters: {}", details),
        }
        .into());
    }
    let safe_params = redact_params(params, tool.sensitive_params());
    tracing::debug!(
        tool = %tool_name,
        params = %safe_params,
        "Tool call started"
    );
    // Execute with per-tool timeout
    let timeout = tool.execution_timeout();
    let start = std::time::Instant::now();
    let result = tokio::time::timeout(timeout, async {
        tool.execute(params.clone(), job_ctx).await
    })
    .await;
    let elapsed = start.elapsed();
    match &result {
        Ok(Ok(output)) => {
            let result_str = serde_json::to_string(&output.result)
                .unwrap_or_else(|_| "<serialize error>".to_string());
            tracing::debug!(
                tool = %tool_name,
                elapsed_ms = elapsed.as_millis() as u64,
                result = %result_str,
                "Tool call succeeded"
            );
        }
        Ok(Err(e)) => {
            tracing::debug!(
                tool = %tool_name,
                elapsed_ms = elapsed.as_millis() as u64,
                error = %e,
                "Tool call failed"
            );
        }
        Err(_) => {
            tracing::debug!(
                tool = %tool_name,
                elapsed_ms = elapsed.as_millis() as u64,
                timeout_secs = timeout.as_secs(),
                "Tool call timed out"
            );
        }
    }
    let result = result
        .map_err(|_| crate::error::ToolError::Timeout {
            name: tool_name.to_string(),
            timeout,
        })?
        .map_err(|e| crate::error::ToolError::ExecutionFailed {
            name: tool_name.to_string(),
            reason: e.to_string(),
        })?;
    serde_json::to_string_pretty(&result.result).map_err(|e| {
        crate::error::ToolError::ExecutionFailed {
            name: tool_name.to_string(),
            reason: format!("Failed to serialize result: {}", e),
        }
        .into()
    })
 }
 /// Parsed auth result fields for emitting StatusUpdate::AuthRequired.
@@ -189,7 +189,7 @@ impl HeartbeatRunner {
            // Skip during quiet hours
            if self.config.is_quiet_hours() {
-                tracing::trace!("Heartbeat skipped: quiet hours");
+                tracing::debug!("Heartbeat skipped: quiet hours");
                continue;
            }
@@ -212,7 +212,7 @@ impl HeartbeatRunner {
            match self.check_heartbeat().await {
                HeartbeatResult::Ok => {
-                    tracing::trace!("Heartbeat OK");
+                    tracing::debug!("Heartbeat OK");
                    self.consecutive_failures = 0;
                }
                HeartbeatResult::NeedsAttention(message) => {
@@ -221,7 +221,7 @@ impl HeartbeatRunner {
                    self.send_notification(&message).await;
                }
                HeartbeatResult::Skipped => {
-                    tracing::trace!("Heartbeat skipped");
+                    tracing::debug!("Heartbeat skipped");
                }
                HeartbeatResult::Failed(error) => {
                    tracing::error!("Heartbeat failed: {}", error);
@@ -303,8 +303,7 @@ impl HeartbeatRunner {
            .with_max_tokens(max_tokens)
            .with_temperature(0.3);
-        let reasoning =
+        let reasoning = Reasoning::new(self.llm.clone());
            Reasoning::new(self.llm.clone()).with_model_name(self.llm.active_model_name());
        let (content, _usage) = match reasoning.complete(request).await {
            Ok(r) => r,
            Err(e) => return HeartbeatResult::Failed(format!("LLM call failed: {}", e)),
@@ -11,7 +11,6 @@
 //! - Context compaction for long conversations
 mod agent_loop;
 pub mod agentic_loop;
 mod attachments;
 mod commands;
 pub mod compaction;
@@ -23,7 +22,7 @@ pub mod job_monitor;
 mod router;
 pub mod routine;
 pub mod routine_engine;
-pub(crate) mod scheduler;
+mod scheduler;
 mod self_repair;
 pub mod session;
 mod session_manager;
@@ -31,8 +30,8 @@ pub mod submission;
 pub mod task;
 mod thread_ops;
 pub mod undo;
 pub mod worker;
 pub use crate::worker::{Worker, WorkerDeps};
 pub(crate) use agent_loop::truncate_for_preview;
 pub use agent_loop::{Agent, AgentDeps};
 pub use compaction::{CompactionResult, ContextCompactor};
@@ -48,3 +47,4 @@ pub use session_manager::SessionManager;
 pub use submission::{Submission, SubmissionParser, SubmissionResult};
 pub use task::{Task, TaskContext, TaskHandler, TaskOutput};
 pub use undo::{Checkpoint, UndoManager};
 pub use worker::{Worker, WorkerDeps};
@@ -8,7 +8,7 @@
 //! ┌──────────┐     ┌─────────┐     ┌──────────────────┐
 //! │  Trigger  │────▶│ Engine  │────▶│  Execution Mode  │
 //! │ cron/event│     │guardrail│     │lightweight│full_job│
-//! │ system    │     │ check   │     └──────────────────┘
+//! │ webhook   │     │ check   │     └──────────────────┘
 //! │ manual    │     └─────────┘              │
 //! └──────────┘                               ▼
 //!                                     ┌──────────────┐
@@ -69,15 +69,12 @@ pub enum Trigger {
        /// Regex pattern to match against message content.
        pattern: String,
    },
-    /// Fire when a structured system event is emitted.
+    /// Fire on incoming webhook POST to /hooks/routine/{id}.
-    SystemEvent {
+    Webhook {
-        /// Event source namespace (e.g. "github", "workflow", "tool").
+        /// Optional webhook path suffix (defaults to routine id).
-        source: String,
+        path: Option<String>,
-        /// Event type within the source (e.g. "issue.opened").
+        /// Optional shared secret for HMAC validation.
-        event_type: String,
+        secret: Option<String>,
        /// Optional exact-match filters against payload top-level fields.
        #[serde(default)]
        filters: std::collections::HashMap<String, String>,
    },
    /// Only fires via tool call or CLI.
    Manual,
@@ -89,7 +86,7 @@ impl Trigger {
        match self {
            Trigger::Cron { .. } => "cron",
            Trigger::Event { .. } => "event",
-            Trigger::SystemEvent { .. } => "system_event",
+            Trigger::Webhook { .. } => "webhook",
            Trigger::Manual => "manual",
        }
    }
@@ -137,39 +134,16 @@ impl Trigger {
                    .map(String::from);
                Ok(Trigger::Event { channel, pattern })
            }
-            "system_event" => {
+            "webhook" => {
-                let source = config
+                let path = config
-                    .get("source")
+                    .get("path")
                    .and_then(|v| v.as_str())
-                    .ok_or_else(|| RoutineError::MissingField {
+                    .map(String::from);
-                        context: "system_event trigger".into(),
+                let secret = config
-                        field: "source".into(),
+                    .get("secret")
                    })?
                    .to_string();
                let event_type = config
                    .get("event_type")
                    .and_then(|v| v.as_str())
-                    .ok_or_else(|| RoutineError::MissingField {
+                    .map(String::from);
-                        context: "system_event trigger".into(),
+                Ok(Trigger::Webhook { path, secret })
                        field: "event_type".into(),
                    })?
                    .to_string();
                let filters = config
                    .get("filters")
                    .and_then(|v| v.as_object())
                    .map(|m| {
                        m.iter()
                            .filter_map(|(k, v)| {
                                json_value_as_filter_string(v).map(|s| (k.clone(), s))
                            })
                            .collect()
                    })
                    .unwrap_or_default();
                Ok(Trigger::SystemEvent {
                    source,
                    event_type,
                    filters,
                })
            }
            "manual" => Ok(Trigger::Manual),
            other => Err(RoutineError::UnknownTriggerType {
@@ -189,14 +163,9 @@ impl Trigger {
                "pattern": pattern,
                "channel": channel,
            }),
-            Trigger::SystemEvent {
+            Trigger::Webhook { path, secret } => serde_json::json!({
-                source,
+                "path": path,
-                event_type,
+                "secret": secret,
                filters,
            } => serde_json::json!({
                "source": source,
                "event_type": event_type,
                "filters": filters,
            }),
            Trigger::Manual => serde_json::json!({}),
        }
@@ -459,19 +428,6 @@ pub struct RoutineRun {
    pub created_at: DateTime<Utc>,
 }
 /// Convert a JSON value to a string for filter storage.
 ///
 /// Handles strings, numbers, and booleans — consistent with the matching
 /// logic in `routine_engine::json_value_as_string`.
 pub fn json_value_as_filter_string(v: &serde_json::Value) -> Option<String> {
    match v {
        serde_json::Value::String(s) => Some(s.clone()),
        serde_json::Value::Number(n) => Some(n.to_string()),
        serde_json::Value::Bool(b) => Some(b.to_string()),
        _ => None,
    }
 }
 /// Compute a content hash for event dedup.
 pub fn content_hash(content: &str) -> u64 {
    let mut hasher = DefaultHasher::new();
@@ -530,24 +486,6 @@ mod tests {
            if channel == Some("telegram".to_string()) && pattern == r"deploy\s+\w+"));
    }
    #[test]
    fn test_system_event_trigger_roundtrip() {
        let mut filters = std::collections::HashMap::new();
        filters.insert("repo".to_string(), "nearai/ironclaw".to_string());
        filters.insert("action".to_string(), "opened".to_string());
        let trigger = Trigger::SystemEvent {
            source: "github".to_string(),
            event_type: "issue".to_string(),
            filters: filters.clone(),
        };
        let json = trigger.to_config_json();
        let parsed = Trigger::from_db("system_event", json).expect("parse system_event");
        assert!(
            matches!(parsed, Trigger::SystemEvent { source, event_type, filters: f }
            if source == "github" && event_type == "issue" && f == filters)
        );
    }
    #[test]
    fn test_action_lightweight_roundtrip() {
        let action = RoutineAction::Lightweight {
@@ -685,13 +623,12 @@ mod tests {
            "event"
        );
        assert_eq!(
-            Trigger::SystemEvent {
+            Trigger::Webhook {
-                source: String::new(),
+                path: None,
-                event_type: String::new(),
+                secret: None
                filters: std::collections::HashMap::new(),
            }
            .type_tag(),
-            "system_event"
+            "webhook"
        );
        assert_eq!(Trigger::Manual.type_tag(), "manual");
    }
@@ -32,14 +32,9 @@ use crate::llm::{
    ChatMessage, CompletionRequest, FinishReason, LlmProvider, ToolCall, ToolCompletionRequest,
 };
 use crate::safety::SafetyLayer;
-use crate::tools::{ApprovalContext, ApprovalRequirement, ToolError, ToolRegistry};
+use crate::tools::{ApprovalContext, ApprovalRequirement, ToolError, ToolRegistry, redact_params};
 use crate::workspace::Workspace;
 enum EventMatcher {
    Message { routine: Routine, regex: Regex },
    System { routine: Routine },
 }
 /// The routine execution engine.
 pub struct RoutineEngine {
    config: RoutineConfig,
@@ -50,8 +45,8 @@ pub struct RoutineEngine {
    notify_tx: mpsc::Sender<OutgoingResponse>,
    /// Currently running routine count (across all routines).
    running_count: Arc<AtomicUsize>,
-    /// Cached matchers for all event-driven routines.
+    /// Compiled event regex cache: routine_id -> compiled regex.
-    event_cache: Arc<RwLock<Vec<EventMatcher>>>,
+    event_cache: Arc<RwLock<Vec<(Uuid, Routine, Regex)>>>,
    /// Scheduler for dispatching jobs (FullJob mode).
    scheduler: Option<Arc<Scheduler>>,
    /// Tool registry for lightweight routine tool execution.
@@ -92,12 +87,9 @@ impl RoutineEngine {
            Ok(routines) => {
                let mut cache = Vec::new();
                for routine in routines {
-                    match &routine.trigger {
+                    if let Trigger::Event { ref pattern, .. } = routine.trigger {
-                        Trigger::Event { pattern, .. } => match Regex::new(pattern) {
+                        match Regex::new(pattern) {
-                            Ok(re) => cache.push(EventMatcher::Message {
+                            Ok(re) => cache.push((routine.id, routine.clone(), re)),
                                routine: routine.clone(),
                                regex: re,
                            }),
                            Err(e) => {
                                tracing::warn!(
                                    routine = %routine.name,
@@ -105,18 +97,12 @@ impl RoutineEngine {
                                    pattern, e
                                );
                            }
                        },
                        Trigger::SystemEvent { .. } => {
                            cache.push(EventMatcher::System {
                                routine: routine.clone(),
                            });
                        }
                        _ => {}
                    }
                }
                let count = cache.len();
                *self.event_cache.write().await = cache;
-                tracing::trace!("Refreshed event cache: {} routines", count);
+                tracing::debug!("Refreshed event cache: {} routines", count);
            }
            Err(e) => {
                tracing::error!("Failed to refresh event cache: {}", e);
@@ -132,11 +118,7 @@ impl RoutineEngine {
        let cache = self.event_cache.read().await;
        let mut fired = 0;
-        for matcher in cache.iter() {
+        for (_, routine, re) in cache.iter() {
            let (routine, re) = match matcher {
                EventMatcher::Message { routine, regex } => (routine, regex),
                EventMatcher::System { .. } => continue,
            };
            // Channel filter
            if let Trigger::Event {
                channel: Some(ch), ..
@@ -153,13 +135,13 @@ impl RoutineEngine {
            // Cooldown check
            if !self.check_cooldown(routine) {
-                tracing::trace!(routine = %routine.name, "Skipped: cooldown active");
+                tracing::debug!(routine = %routine.name, "Skipped: cooldown active");
                continue;
            }
            // Concurrent run check
            if !self.check_concurrent(routine).await {
-                tracing::trace!(routine = %routine.name, "Skipped: max concurrent reached");
+                tracing::debug!(routine = %routine.name, "Skipped: max concurrent reached");
                continue;
            }
@@ -177,88 +159,6 @@ impl RoutineEngine {
        fired
    }
    /// Emit a structured event to system-event routines.
    ///
    /// Returns the number of routines that were fired.
    pub async fn emit_system_event(
        &self,
        source: &str,
        event_type: &str,
        payload: &serde_json::Value,
        user_id: Option<&str>,
    ) -> usize {
        let cache = self.event_cache.read().await;
        let mut fired = 0;
        for matcher in cache.iter() {
            let routine = match matcher {
                EventMatcher::System { routine } => routine,
                EventMatcher::Message { .. } => continue,
            };
            let Trigger::SystemEvent {
                source: expected_source,
                event_type: expected_event,
                filters,
            } = &routine.trigger
            else {
                continue;
            };
            if !expected_source.eq_ignore_ascii_case(source)
                || !expected_event.eq_ignore_ascii_case(event_type)
            {
                continue;
            }
            if let Some(uid) = user_id
                && routine.user_id != uid
            {
                continue;
            }
            let mut matched = true;
            for (key, expected) in filters {
                let Some(actual) = payload
                    .get(key)
                    .and_then(crate::agent::routine::json_value_as_filter_string)
                else {
                    tracing::debug!(routine = %routine.name, filter_key = %key, "Filter key not found in payload");
                    matched = false;
                    break;
                };
                if !actual.eq_ignore_ascii_case(expected) {
                    matched = false;
                    break;
                }
            }
            if !matched {
                continue;
            }
            if !self.check_cooldown(routine) {
                tracing::debug!(routine = %routine.name, "Skipped: cooldown active");
                continue;
            }
            if !self.check_concurrent(routine).await {
                tracing::debug!(routine = %routine.name, "Skipped: max concurrent reached");
                continue;
            }
            if self.running_count.load(Ordering::Relaxed) >= self.config.max_concurrent_routines {
                tracing::warn!(routine = %routine.name, "Skipped: global max concurrent reached");
                continue;
            }
            let detail = truncate(&format!("{source}:{event_type}"), 200);
            self.spawn_fire(routine.clone(), "system_event", Some(detail));
            fired += 1;
        }
        fired
    }
    /// Check all due cron routines and fire them. Called by the cron ticker.
    pub async fn check_cron_triggers(&self) {
        let routines = match self.store.list_due_cron_routines().await {
@@ -1013,6 +913,13 @@ async fn execute_routine_tool(
        return Err(format!("Invalid tool parameters: {}", details).into());
    }
    let safe_params = redact_params(&tc.arguments, tool.sensitive_params());
    tracing::debug!(
        tool = %tc.name,
        params = %safe_params,
        "Lightweight routine tool call started"
    );
    // Execute with per-tool timeout
    let timeout = tool.execution_timeout();
    let start = std::time::Instant::now();
@@ -1022,14 +929,12 @@ async fn execute_routine_tool(
    .await;
    let elapsed = start.elapsed();
    // Log tool execution result (single consolidated log)
    match &result {
        Ok(Ok(_)) => {
            tracing::debug!(
                tool = %tc.name,
                elapsed_ms = elapsed.as_millis() as u64,
-                status = "succeeded",
+                "Lightweight routine tool call succeeded"
                "Lightweight routine tool execution completed"
            );
        }
        Ok(Err(e)) => {
@@ -1037,8 +942,7 @@ async fn execute_routine_tool(
                tool = %tc.name,
                elapsed_ms = elapsed.as_millis() as u64,
                error = %e,
-                status = "failed",
+                "Lightweight routine tool call failed"
                "Lightweight routine tool execution completed"
            );
        }
        Err(_) => {
@@ -1046,8 +950,7 @@ async fn execute_routine_tool(
                tool = %tc.name,
                elapsed_ms = elapsed.as_millis() as u64,
                timeout_secs = timeout.as_secs(),
-                status = "timeout",
+                "Lightweight routine tool call timed out"
                "Lightweight routine tool execution completed"
            );
        }
    }
@@ -9,6 +9,7 @@ use tokio::task::JoinHandle;
 use uuid::Uuid;
 use crate::agent::task::{Task, TaskContext, TaskOutput};
 use crate::agent::worker::{Worker, WorkerDeps};
 use crate::channels::web::types::SseEvent;
 use crate::config::AgentConfig;
 use crate::context::{ContextManager, JobContext, JobState};
@@ -18,7 +19,6 @@ use crate::hooks::HookRegistry;
 use crate::llm::LlmProvider;
 use crate::safety::SafetyLayer;
 use crate::tools::{ApprovalContext, ToolRegistry};
 use crate::worker::job::{Worker, WorkerDeps};
 /// Message to send to a worker.
 #[derive(Debug)]
@@ -160,36 +160,24 @@ impl Scheduler {
            .create_job_for_user(user_id, title, description)
            .await?;
-        // Apply metadata and token budget in a single atomic update.
+        // Apply token budget from config, allowing per-job metadata override.
-        // This prevents concurrent workers from observing partial state.
+        let max_tokens = metadata
        // Cap user-supplied max_tokens at the configured limit (Issue #815).
        let user_max_tokens = metadata
            .as_ref()
            .and_then(|m| m.get("max_tokens"))
-            .and_then(|v| v.as_u64());
+            .and_then(|v| v.as_u64())
        let max_tokens = user_max_tokens
            .map(|user_val| {
                if self.config.max_tokens_per_job == 0 {
                    // Config is "unlimited": use the user-supplied value directly.
                    user_val
                } else {
                    std::cmp::min(user_val, self.config.max_tokens_per_job)
                }
            })
            .unwrap_or(self.config.max_tokens_per_job);
-        // Apply both metadata and token budget in one closure (Issue #813: atomic update)
+        // Apply metadata if provided
        if let Some(meta) = metadata {
            self.context_manager
                .update_context(job_id, |ctx| {
                    ctx.metadata = meta;
                    if max_tokens > 0 {
                        ctx.max_tokens = max_tokens;
                    }
                })
                .await?;
-        } else if max_tokens > 0 {
+        }
        // Set token budget (separate update to avoid overwriting metadata)
        if max_tokens > 0 {
            self.context_manager
                .update_context(job_id, |ctx| {
                    ctx.max_tokens = max_tokens;
@@ -474,9 +462,6 @@ impl Scheduler {
    }
    /// Execute a single tool as a subtask.
    ///
    /// Performs scheduler-specific checks (approval, cancellation) then
    /// delegates to the shared `execute_tool_with_safety` pipeline.
    async fn execute_tool_task(
        tools: Arc<ToolRegistry>,
        context_manager: Arc<ContextManager>,
@@ -488,7 +473,7 @@ impl Scheduler {
    ) -> Result<TaskOutput, Error> {
        let start = std::time::Instant::now();
-        // Get the tool for approval check
+        // Get the tool
        let tool = tools.get(tool_name).await.ok_or_else(|| {
            Error::Tool(crate::error::ToolError::NotFound {
                name: tool_name.to_string(),
@@ -505,7 +490,6 @@ impl Scheduler {
            .into());
        }
        // Scheduler-specific approval check
        let requirement = tool.requires_approval(&params);
        let blocked =
            ApprovalContext::is_blocked_or_default(&approval_context, tool_name, requirement);
@@ -516,23 +500,41 @@ impl Scheduler {
            .into());
        }
-        // Delegate to shared tool execution pipeline
+        // Validate tool parameters
-        let output_str = crate::tools::execute::execute_tool_with_safety(
+        let validation = safety.validator().validate_tool_params(&params);
-            &tools, &safety, tool_name, &params, &job_ctx,
+        if !validation.is_valid {
-        )
+            let details = validation
-        .await?;
+                .errors
                .iter()
                .map(|e| format!("{}: {}", e.field, e.message))
                .collect::<Vec<_>>()
                .join("; ");
            return Err(crate::error::ToolError::InvalidParameters {
                name: tool_name.to_string(),
                reason: format!("Invalid tool parameters: {}", details),
            }
            .into());
        }
-        // Parse back to Value for TaskOutput; this should be infallible given
+        // Execute with per-tool timeout
-        // `execute_tool_with_safety` uses `serde_json::to_string_pretty`, but if it
+        let tool_timeout = tool.execution_timeout();
-        // ever fails we surface a clear error instead of silently changing types.
+        let result =
-        let result_value: serde_json::Value = serde_json::from_str(&output_str).map_err(|e| {
+            tokio::time::timeout(tool_timeout, async { tool.execute(params, &job_ctx).await })
                .await
                .map_err(|_| {
                    Error::Tool(crate::error::ToolError::Timeout {
                        name: tool_name.to_string(),
                        timeout: tool_timeout,
                    })
                })?
                .map_err(|e| {
                    Error::Tool(crate::error::ToolError::ExecutionFailed {
                        name: tool_name.to_string(),
-                reason: format!("Failed to parse tool output as JSON: {}", e),
+                        reason: e.to_string(),
                    })
                })?;
-        Ok(TaskOutput::new(result_value, start.elapsed()))
+        Ok(TaskOutput::new(result.result, start.elapsed()))
    }
    /// Stop a running job.
@@ -697,140 +699,8 @@ impl Scheduler {
 mod tests {
    use super::*;
    use crate::config::SafetyConfig;
    use crate::llm::{
        CompletionRequest, CompletionResponse, LlmError, LlmProvider, ToolCompletionRequest,
        ToolCompletionResponse,
    };
    use crate::safety::SafetyLayer;
    use crate::tools::{ApprovalRequirement, Tool, ToolError, ToolOutput};
    use rust_decimal_macros::dec;
    /// Minimal LLM provider stub for scheduler tests that don't exercise LLM calls.
    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) {
            (dec!(0), dec!(0))
        }
        async fn complete(&self, _req: CompletionRequest) -> Result<CompletionResponse, LlmError> {
            Err(LlmError::RequestFailed {
                provider: "stub".into(),
                reason: "not implemented".into(),
            })
        }
        async fn complete_with_tools(
            &self,
            _req: ToolCompletionRequest,
        ) -> Result<ToolCompletionResponse, LlmError> {
            Err(LlmError::RequestFailed {
                provider: "stub".into(),
                reason: "not implemented".into(),
            })
        }
    }
    /// Create a Scheduler for token-budget tests. The LLM stub will fail if a
    /// worker actually tries to call it, but `dispatch_job` sets the token
    /// budget *before* spawning the worker so we can inspect the context
    /// immediately after dispatch.
    fn make_test_scheduler(max_tokens_per_job: u64) -> Scheduler {
        let config = AgentConfig {
            name: "test".to_string(),
            max_parallel_jobs: 5,
            job_timeout: std::time::Duration::from_secs(30),
            stuck_threshold: std::time::Duration::from_secs(300),
            repair_check_interval: std::time::Duration::from_secs(3600),
            max_repair_attempts: 0,
            use_planning: false,
            session_idle_timeout: std::time::Duration::from_secs(3600),
            allow_local_tools: true,
            max_cost_per_day_cents: None,
            max_actions_per_hour: None,
            max_tool_iterations: 10,
            auto_approve_tools: true,
            default_timezone: "UTC".to_string(),
            max_tokens_per_job,
        };
        let cm = Arc::new(ContextManager::new(5));
        let llm: Arc<dyn LlmProvider> = Arc::new(StubLlm);
        let safety = Arc::new(SafetyLayer::new(&SafetyConfig {
            max_output_length: 100_000,
            injection_check_enabled: false,
        }));
        let tools = Arc::new(ToolRegistry::new());
        let hooks = Arc::new(HookRegistry::default());
        Scheduler::new(config, cm, llm, safety, tools, None, hooks)
    }
    #[tokio::test]
    async fn test_dispatch_job_caps_user_max_tokens() {
        let sched = make_test_scheduler(1000);
        let meta = serde_json::json!({ "max_tokens": 5000 });
        let job_id = sched
            .dispatch_job("user1", "test", "desc", Some(meta))
            .await
            .unwrap();
        let ctx = sched.context_manager.get_context(job_id).await.unwrap();
        assert_eq!(ctx.max_tokens, 1000, "should cap at configured limit");
    }
    #[tokio::test]
    async fn test_dispatch_job_unlimited_config_preserves_user_tokens() {
        let sched = make_test_scheduler(0); // 0 = unlimited
        let meta = serde_json::json!({ "max_tokens": 5000 });
        let job_id = sched
            .dispatch_job("user1", "test", "desc", Some(meta))
            .await
            .unwrap();
        let ctx = sched.context_manager.get_context(job_id).await.unwrap();
        assert_eq!(
            ctx.max_tokens, 5000,
            "unlimited config should preserve user value"
        );
    }
    #[tokio::test]
    async fn test_dispatch_job_no_user_tokens_uses_config() {
        let sched = make_test_scheduler(2000);
        let job_id = sched
            .dispatch_job("user1", "test", "desc", None)
            .await
            .unwrap();
        let ctx = sched.context_manager.get_context(job_id).await.unwrap();
        assert_eq!(
            ctx.max_tokens, 2000,
            "should use config default when no user value"
        );
    }
    #[tokio::test]
    async fn test_dispatch_job_atomic_metadata_and_tokens() {
        let sched = make_test_scheduler(10_000);
        let meta = serde_json::json!({
            "max_tokens": 3000,
            "custom_key": "custom_value"
        });
        let job_id = sched
            .dispatch_job("user1", "test", "desc", Some(meta))
            .await
            .unwrap();
        let ctx = sched.context_manager.get_context(job_id).await.unwrap();
        assert_eq!(ctx.max_tokens, 3000, "should use user value within limit");
        assert_eq!(
            ctx.metadata.get("custom_key").and_then(|v| v.as_str()),
            Some("custom_value"),
            "metadata should be set atomically with token budget"
        );
    }
    #[test]
    fn test_scheduler_creation() {
@@ -334,21 +334,22 @@ impl RepairTask {
            // Check for stuck jobs
            let stuck_jobs = self.repair.detect_stuck_jobs().await;
            for job in stuck_jobs {
                tracing::info!("Attempting to repair stuck job {}", job.job_id);
                match self.repair.repair_stuck_job(&job).await {
                    Ok(RepairResult::Success { message }) => {
-                        tracing::info!(job = %job.job_id, status = "success", "Stuck job repair completed: {}", message);
+                        tracing::info!("Repair succeeded: {}", message);
                    }
                    Ok(RepairResult::Retry { message }) => {
-                        tracing::debug!(job = %job.job_id, status = "retry", "Stuck job repair needs retry: {}", message);
+                        tracing::warn!("Repair needs retry: {}", message);
                    }
                    Ok(RepairResult::Failed { message }) => {
-                        tracing::error!(job = %job.job_id, status = "failed", "Stuck job repair failed: {}", message);
+                        tracing::error!("Repair failed: {}", message);
                    }
                    Ok(RepairResult::ManualRequired { message }) => {
-                        tracing::warn!(job = %job.job_id, status = "manual", "Stuck job repair requires manual intervention: {}", message);
+                        tracing::warn!("Manual intervention needed: {}", message);
                    }
                    Err(e) => {
-                        tracing::error!(job = %job.job_id, "Stuck job repair error: {}", e);
+                        tracing::error!("Repair error: {}", e);
                    }
                }
            }
@@ -356,12 +357,13 @@ impl RepairTask {
            // Check for broken tools
            let broken_tools = self.repair.detect_broken_tools().await;
            for tool in broken_tools {
                tracing::info!("Attempting to repair broken tool: {}", tool.name);
                match self.repair.repair_broken_tool(&tool).await {
                    Ok(result) => {
-                        tracing::debug!(tool = %tool.name, status = "completed", "Tool repair completed: {:?}", result);
+                        tracing::info!("Tool repair result: {:?}", result);
                    }
                    Err(e) => {
-                        tracing::error!(tool = %tool.name, "Tool repair error: {}", e);
+                        tracing::error!("Tool repair error: {}", e);
                    }
                }
            }
@@ -113,13 +113,6 @@ impl Agent {
        thread_id: Uuid,
        content: &str,
    ) -> Result<SubmissionResult, Error> {
        tracing::debug!(
            message_id = %message.id,
            thread_id = %thread_id,
            content_len = content.len(),
            "Processing user input"
        );
        // First check thread state without holding lock during I/O
        let thread_state = {
            let sess = session.lock().await;
@@ -130,41 +123,19 @@ impl Agent {
            thread.state
        };
        tracing::debug!(
            message_id = %message.id,
            thread_id = %thread_id,
            thread_state = ?thread_state,
            "Checked thread state"
        );
        // Check thread state
        match thread_state {
            ThreadState::Processing => {
                tracing::warn!(
                    message_id = %message.id,
                    thread_id = %thread_id,
                    "Thread is processing, rejecting new input"
                );
                return Ok(SubmissionResult::error(
                    "Turn in progress. Use /interrupt to cancel.",
                ));
            }
            ThreadState::AwaitingApproval => {
                tracing::warn!(
                    message_id = %message.id,
                    thread_id = %thread_id,
                    "Thread awaiting approval, rejecting new input"
                );
                return Ok(SubmissionResult::error(
                    "Waiting for approval. Use /interrupt to cancel.",
                ));
            }
            ThreadState::Completed => {
                tracing::warn!(
                    message_id = %message.id,
                    thread_id = %thread_id,
                    "Thread completed, rejecting new input"
                );
                return Ok(SubmissionResult::error(
                    "Thread completed. Use /thread new.",
                ));
@@ -298,20 +269,9 @@ impl Agent {
        };
        // Persist user message to DB immediately so it survives crashes
        tracing::debug!(
            message_id = %message.id,
            thread_id = %thread_id,
            "Persisting user message to DB"
        );
        self.persist_user_message(thread_id, &message.user_id, effective_content)
            .await;
        tracing::debug!(
            message_id = %message.id,
            thread_id = %thread_id,
            "User message persisted, starting agentic loop"
        );
        // Send thinking status
        let _ = self
            .channels
@@ -852,12 +812,19 @@ impl Agent {
            // Sanitize tool result, then record the cleaned version in the
            // thread. Must happen before auth intercept check which may return early.
            let is_tool_error = tool_result.is_err();
-            let (result_content, _) = crate::tools::execute::process_tool_result(
+            let result_content = match &tool_result {
-                self.safety(),
+                Ok(output) => {
                    let sanitized = self
                        .safety()
                        .sanitize_tool_output(&pending.tool_name, output);
                    self.safety().wrap_for_llm(
                        &pending.tool_name,
-                &pending.tool_call_id,
+                        &sanitized.content,
-                &tool_result,
+                        sanitized.was_modified,
-            );
+                    )
                }
                Err(e) => format!("Error: {}", e),
            };
            // Record sanitized result in thread
            {
@@ -1097,12 +1064,17 @@ impl Agent {
                // Sanitize first, then record the cleaned version in thread.
                // Must happen before auth detection which may set deferred_auth.
                let is_deferred_error = deferred_result.is_err();
-                let (deferred_content, _) = crate::tools::execute::process_tool_result(
+                let deferred_content = match &deferred_result {
-                    self.safety(),
+                    Ok(output) => {
                        let sanitized = self.safety().sanitize_tool_output(&tc.name, output);
                        self.safety().wrap_for_llm(
                            &tc.name,
-                    &tc.id,
+                            &sanitized.content,
-                    &deferred_result,
+                            sanitized.was_modified,
-                );
+                        )
                    }
                    Err(e) => format!("Error: {}", e),
                };
                // Record sanitized result in thread
                {
@@ -572,7 +572,7 @@ impl AppBuilder {
        let (dev_loaded_tool_names, _) = tokio::join!(wasm_tools_future, mcp_servers_future);
        // Load registry catalog entries for extension discovery
-        let mut catalog_entries = match crate::registry::RegistryCatalog::load_or_embedded() {
+        let catalog_entries = match crate::registry::RegistryCatalog::load_or_embedded() {
            Ok(catalog) => {
                let entries: Vec<_> = catalog
                    .all()
@@ -591,15 +591,6 @@ impl AppBuilder {
            }
        };
        // Append builtin entries (e.g. channel-relay integrations) so they appear
        // in the web UI's available extensions list.
        let builtin = crate::extensions::registry::builtin_entries();
        for entry in builtin {
            if !catalog_entries.iter().any(|e| e.name == entry.name) {
                catalog_entries.push(entry);
            }
        }
        // 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) =
@@ -344,24 +344,6 @@ pub trait Channel: Send + Sync {
    }
 }
 /// Trait for channels that support hot-secret-swapping during SIGHUP reload.
 ///
 /// This allows channels to update authentication credentials without restarting,
 /// enabling zero-downtime configuration reloads. Channels that don't support
 /// secret updates can simply not implement this trait.
 #[async_trait]
 pub trait ChannelSecretUpdater: Send + Sync {
    /// Update the secret for this channel.
    ///
    /// Called during SIGHUP configuration reload. Implementation should:
    /// - Apply the new secret atomically
    /// - Not fail the entire reload if secret update fails
    /// - Log appropriate errors/info messages
    ///
    /// The secret is optional (may be None if secret is no longer configured).
    async fn update_secret(&self, new_secret: Option<secrecy::SecretString>);
 }
 #[cfg(test)]
 mod tests {
    use super::*;
@@ -10,7 +10,7 @@ use axum::{
    response::IntoResponse,
    routing::{get, post},
 };
-use secrecy::{ExposeSecret, SecretString};
+use secrecy::ExposeSecret;
 use serde::{Deserialize, Serialize};
 use subtle::ConstantTimeEq;
 use tokio::sync::{RwLock, mpsc, oneshot};
@@ -18,8 +18,7 @@ use tokio_stream::wrappers::ReceiverStream;
 use uuid::Uuid;
 use crate::channels::{
-    AttachmentKind, Channel, ChannelSecretUpdater, IncomingAttachment, IncomingMessage,
+    AttachmentKind, Channel, IncomingAttachment, IncomingMessage, MessageStream, OutgoingResponse,
    MessageStream, OutgoingResponse,
 };
 use crate::config::HttpConfig;
 use crate::error::ChannelError;
@@ -30,16 +29,13 @@ pub struct HttpChannel {
    state: Arc<HttpChannelState>,
 }
-pub struct HttpChannelState {
+struct HttpChannelState {
    /// Sender for incoming messages.
    tx: RwLock<Option<mpsc::Sender<IncomingMessage>>>,
    /// Pending responses keyed by message ID.
    pending_responses: RwLock<std::collections::HashMap<Uuid, oneshot::Sender<String>>>,
    /// Expected webhook secret for authentication (if configured).
-    /// Stored in a separate Arc<RwLock<>> to avoid contending with other state operations.
+    webhook_secret: Option<String>,
    /// Rarely changes (only on SIGHUP), so isolated from hot-path state accesses.
    /// Uses SecretString to prevent accidental logging and memory dump exposure.
    webhook_secret: Arc<RwLock<Option<SecretString>>>,
    /// Fixed user ID for this HTTP channel.
    user_id: String,
    /// Rate limiting state.
@@ -52,14 +48,6 @@ struct RateLimitState {
    request_count: u32,
 }
 impl HttpChannelState {
    /// Update the webhook secret in-place without restarting the listener.
    /// Called during SIGHUP to hot-swap credentials.
    pub async fn update_secret(&self, new_secret: Option<SecretString>) {
        *self.webhook_secret.write().await = new_secret;
    }
 }
 /// Maximum JSON body size for webhook requests (15 MB, to support base64 image attachments
 /// with ~33% overhead from base64 encoding).
 const MAX_BODY_BYTES: usize = 15 * 1024 * 1024;
@@ -79,7 +67,7 @@ impl HttpChannel {
        let webhook_secret = config
            .webhook_secret
            .as_ref()
-            .map(|s| SecretString::from(s.expose_secret().to_string()));
+            .map(|s| s.expose_secret().to_string());
        let user_id = config.user_id.clone();
        Self {
@@ -87,7 +75,7 @@ impl HttpChannel {
            state: Arc::new(HttpChannelState {
                tx: RwLock::new(None),
                pending_responses: RwLock::new(std::collections::HashMap::new()),
-                webhook_secret: Arc::new(RwLock::new(webhook_secret)),
+                webhook_secret,
                user_id,
                rate_limit: tokio::sync::Mutex::new(RateLimitState {
                    window_start: std::time::Instant::now(),
@@ -114,16 +102,6 @@ impl HttpChannel {
    pub fn addr(&self) -> (&str, u16) {
        (&self.config.host, self.config.port)
    }
    /// Return a shared handle to the channel state for out-of-band updates.
    pub fn shared_state(&self) -> Arc<HttpChannelState> {
        Arc::clone(&self.state)
    }
    /// Update the webhook secret in-place without restarting the listener.
    pub async fn update_secret(&self, new_secret: Option<SecretString>) {
        self.state.update_secret(new_secret).await;
    }
 }
 #[derive(Debug, Deserialize)]
@@ -223,10 +201,9 @@ async fn webhook_handler(
    });
    // Validate secret if configured
-    if let Some(ref expected_secret) = *state.webhook_secret.read().await {
+    if let Some(ref expected_secret) = state.webhook_secret {
        let expected_bytes = expected_secret.expose_secret().as_bytes();
        match &req.secret {
-            Some(provided) if bool::from(provided.as_bytes().ct_eq(expected_bytes)) => {
+            Some(provided) if bool::from(provided.as_bytes().ct_eq(expected_secret.as_bytes())) => {
                // Secret matches, continue
            }
            Some(_) => {
@@ -395,14 +372,9 @@ async fn process_message(
        None
    };
-    // Clone sender while holding read lock, then release lock before async send.
+    // Send message to the channel
-    // This prevents blocking other webhook handlers during the async I/O.
+    let tx_guard = state.tx.read().await;
-    let tx = {
+    if let Some(tx) = tx_guard.as_ref() {
        let guard = state.tx.read().await;
        guard.as_ref().cloned()
    };
    if let Some(tx) = tx {
        if tx.send(msg).await.is_err() {
            return (
                StatusCode::INTERNAL_SERVER_ERROR,
@@ -423,6 +395,7 @@ async fn process_message(
            }),
        );
    }
    drop(tx_guard);
    // Wait for response if requested
    let response = if let Some(rx) = response_rx {
@@ -455,7 +428,7 @@ impl Channel for HttpChannel {
    }
    async fn start(&self) -> Result<MessageStream, ChannelError> {
-        if self.state.webhook_secret.read().await.is_none() {
+        if self.state.webhook_secret.is_none() {
            return Err(ChannelError::StartupFailed {
                name: "http".to_string(),
                reason: "HTTP webhook secret is required (set HTTP_WEBHOOK_SECRET)".to_string(),
@@ -502,16 +475,6 @@ impl Channel for HttpChannel {
    }
 }
 /// Implement secret update for HTTP channel state.
 /// This allows SIGHUP handler to update secrets generically via the trait.
 #[async_trait]
 impl ChannelSecretUpdater for HttpChannelState {
    async fn update_secret(&self, new_secret: Option<SecretString>) {
        *self.webhook_secret.write().await = new_secret;
        tracing::info!("HTTP webhook secret updated");
    }
 }
 #[cfg(test)]
 mod tests {
    use axum::body::Body;
@@ -599,156 +562,4 @@ mod tests {
        let resp = app.oneshot(req).await.unwrap();
        assert_eq!(resp.status(), StatusCode::UNAUTHORIZED);
    }
    #[tokio::test]
    async fn test_update_secret_hot_swap() {
        let channel = test_channel(Some("old-secret"));
        let _stream = channel.start().await.unwrap();
        let app1 = channel.routes();
        // Request with old-secret should succeed
        let body_old = serde_json::json!({
            "content": "hello",
            "secret": "old-secret"
        });
        let req1 = Request::builder()
            .method("POST")
            .uri("/webhook")
            .header("content-type", "application/json")
            .body(Body::from(serde_json::to_vec(&body_old).unwrap()))
            .unwrap();
        let resp1 = app1.oneshot(req1).await.unwrap();
        assert_eq!(
            resp1.status(),
            StatusCode::OK,
            "old secret should work initially"
        );
        // Update secret to new-secret
        channel
            .update_secret(Some(SecretString::from("new-secret".to_string())))
            .await;
        let app2 = channel.routes();
        // Request with old-secret should fail
        let req2 = Request::builder()
            .method("POST")
            .uri("/webhook")
            .header("content-type", "application/json")
            .body(Body::from(serde_json::to_vec(&body_old).unwrap()))
            .unwrap();
        let resp2 = app2.oneshot(req2).await.unwrap();
        assert_eq!(
            resp2.status(),
            StatusCode::UNAUTHORIZED,
            "old secret should fail after update"
        );
        let app3 = channel.routes();
        // Request with new-secret should succeed
        let body_new = serde_json::json!({
            "content": "hello",
            "secret": "new-secret"
        });
        let req3 = Request::builder()
            .method("POST")
            .uri("/webhook")
            .header("content-type", "application/json")
            .body(Body::from(serde_json::to_vec(&body_new).unwrap()))
            .unwrap();
        let resp3 = app3.oneshot(req3).await.unwrap();
        assert_eq!(
            resp3.status(),
            StatusCode::OK,
            "new secret should work after update"
        );
    }
    #[tokio::test]
    async fn test_concurrent_requests_during_secret_update() {
        use std::sync::Arc as StdArc;
        use std::sync::atomic::{AtomicUsize, Ordering};
        use std::time::Duration;
        let channel = test_channel(Some("initial-secret"));
        let _stream = channel.start().await.unwrap();
        let app = channel.routes();
        // Counters for request outcomes
        let success_count = StdArc::new(AtomicUsize::new(0));
        let mut handles = vec![];
        // Spawn 5 concurrent tasks that keep making requests with the initial secret
        for i in 0..5 {
            let app = app.clone();
            let success = StdArc::clone(&success_count);
            let handle = tokio::spawn(async move {
                let body = serde_json::json!({
                    "content": format!("test-{}", i),
                    "secret": "initial-secret"
                });
                let req = Request::builder()
                    .method("POST")
                    .uri("/webhook")
                    .header("content-type", "application/json")
                    .body(Body::from(serde_json::to_vec(&body).unwrap()))
                    .unwrap();
                let resp = app.oneshot(req).await.unwrap();
                if resp.status() == StatusCode::OK {
                    success.fetch_add(1, Ordering::SeqCst);
                }
            });
            handles.push(handle);
        }
        // Update secret mid-flight (tests that RwLock allows readers while writer holds lock)
        tokio::time::sleep(Duration::from_millis(5)).await;
        channel
            .update_secret(Some(SecretString::from("updated-secret".to_string())))
            .await;
        // Spawn 5 more tasks that use the new secret
        for i in 5..10 {
            let app = app.clone();
            let success = StdArc::clone(&success_count);
            let handle = tokio::spawn(async move {
                let body = serde_json::json!({
                    "content": format!("test-{}", i),
                    "secret": "updated-secret"
                });
                let req = Request::builder()
                    .method("POST")
                    .uri("/webhook")
                    .header("content-type", "application/json")
                    .body(Body::from(serde_json::to_vec(&body).unwrap()))
                    .unwrap();
                let resp = app.oneshot(req).await.unwrap();
                if resp.status() == StatusCode::OK {
                    success.fetch_add(1, Ordering::SeqCst);
                }
            });
            handles.push(handle);
        }
        // Wait for all tasks to complete
        for handle in handles {
            let _ = handle.await;
        }
        // Verify all requests succeeded with their respective secrets
        assert_eq!(
            success_count.load(Ordering::SeqCst),
            10,
            "All concurrent requests should succeed with correct secrets after update"
        );
    }
 }
@@ -56,17 +56,6 @@ impl ChannelManager {
    /// the agent loop.
    pub async fn hot_add(&self, channel: Box<dyn Channel>) -> Result<(), ChannelError> {
        let name = channel.name().to_string();
        // Shut down any existing channel with the same name to avoid parallel consumers.
        // The old forwarding task will stop when the channel's stream ends after shutdown.
        {
            let channels = self.channels.read().await;
            if let Some(existing) = channels.get(&name) {
                tracing::debug!(channel = %name, "Shutting down existing channel before hot-add replacement");
                let _ = existing.shutdown().await;
            }
        }
        let stream = channel.start().await?;
        // Register for respond/broadcast/send_status
@@ -348,30 +337,4 @@ mod tests {
        let msg = stream.next().await.expect("stream ended");
        assert_eq!(msg.content, "background alert");
    }
    #[tokio::test]
    async fn test_hot_add_replaces_existing_channel() {
        // Regression: hot_add must shut down the existing channel before replacing it,
        // to prevent duplicate SSE consumers from running in parallel.
        let manager = ChannelManager::new();
        let (stub1, _tx1) = StubChannel::new("relay");
        manager.add(Box::new(stub1)).await;
        let mut stream = manager.start_all().await.expect("start_all");
        // Hot-add a replacement channel with the same name
        let (stub2, tx2) = StubChannel::new("relay");
        manager.hot_add(Box::new(stub2)).await.expect("hot_add");
        // Send through the new channel — should arrive in the merged stream
        tx2.send(IncomingMessage::new("relay", "u1", "from new"))
            .await
            .expect("send");
        let msg = stream.next().await.expect("stream");
        assert_eq!(msg.content, "from new");
        // Verify only one channel entry exists
        let channels = manager.channels.read().await;
        assert_eq!(channels.len(), 1);
        assert!(channels.contains_key("relay"));
    }
 }
@@ -30,7 +30,6 @@
 mod channel;
 mod http;
 mod manager;
 pub mod relay;
 mod repl;
 mod signal;
 pub mod wasm;
@@ -38,10 +37,10 @@ pub mod web;
 mod webhook_server;
 pub use channel::{
-    AttachmentKind, Channel, ChannelSecretUpdater, IncomingAttachment, IncomingMessage,
+    AttachmentKind, Channel, IncomingAttachment, IncomingMessage, MessageStream, OutgoingResponse,
-    MessageStream, OutgoingResponse, StatusUpdate,
+    StatusUpdate,
 };
-pub use http::{HttpChannel, HttpChannelState};
+pub use http::HttpChannel;
 pub use manager::ChannelManager;
 pub use repl::ReplChannel;
 pub use signal::SignalChannel;
@@ -1,642 +0,0 @@
 //! Channel trait implementation for channel-relay SSE streams.
 //!
 //! `RelayChannel` connects to a channel-relay service via SSE, converts
 //! incoming events to `IncomingMessage`s, and sends responses via the
 //! relay's provider-specific proxy API (Slack).
 use std::collections::HashMap;
 use std::sync::Arc;
 use async_trait::async_trait;
 use tokio::sync::{RwLock, mpsc};
 use crate::channels::relay::client::{RelayClient, RelayError};
 use crate::channels::{Channel, IncomingMessage, MessageStream, OutgoingResponse, StatusUpdate};
 use crate::error::ChannelError;
 /// Default channel name for the Slack relay integration.
 pub const DEFAULT_RELAY_NAME: &str = "slack-relay";
 /// The messaging provider backing a relay channel.
 #[derive(Debug, Clone, Copy, PartialEq, Eq)]
 pub enum RelayProvider {
    Slack,
 }
 impl RelayProvider {
    /// Provider string used in proxy API routes and metadata.
    pub fn as_str(&self) -> &'static str {
        match self {
            Self::Slack => "slack",
        }
    }
    /// The default channel name for this provider.
    pub fn channel_name(&self) -> &'static str {
        match self {
            Self::Slack => DEFAULT_RELAY_NAME,
        }
    }
 }
 /// Channel implementation that connects to a channel-relay SSE stream.
 pub struct RelayChannel {
    client: RelayClient,
    provider: RelayProvider,
    stream_token: Arc<RwLock<String>>,
    team_id: String,
    instance_id: String,
    user_id: String,
    /// SSE stream long-poll timeout in seconds.
    stream_timeout_secs: u64,
    /// Initial exponential backoff in milliseconds.
    backoff_initial_ms: u64,
    /// Maximum exponential backoff in milliseconds.
    backoff_max_ms: u64,
    /// Handle to the reconnect task for clean shutdown.
    reconnect_handle: RwLock<Option<tokio::task::JoinHandle<()>>>,
    /// Handle to the SSE parser task for clean shutdown.
    parser_handle: Arc<RwLock<Option<tokio::task::JoinHandle<()>>>>,
    /// Maximum consecutive reconnect failures before giving up.
    max_consecutive_failures: u64,
 }
 impl RelayChannel {
    /// Create a new relay channel for Slack (default provider).
    pub fn new(
        client: RelayClient,
        stream_token: String,
        team_id: String,
        instance_id: String,
        user_id: String,
    ) -> Self {
        Self::new_with_provider(
            client,
            RelayProvider::Slack,
            stream_token,
            team_id,
            instance_id,
            user_id,
        )
    }
    /// Create a new relay channel with a specific provider.
    pub fn new_with_provider(
        client: RelayClient,
        provider: RelayProvider,
        stream_token: String,
        team_id: String,
        instance_id: String,
        user_id: String,
    ) -> Self {
        Self {
            client,
            provider,
            stream_token: Arc::new(RwLock::new(stream_token)),
            team_id,
            instance_id,
            user_id,
            stream_timeout_secs: 86400,
            backoff_initial_ms: 1000,
            backoff_max_ms: 60000,
            reconnect_handle: RwLock::new(None),
            parser_handle: Arc::new(RwLock::new(None)),
            max_consecutive_failures: 50,
        }
    }
    /// Set backoff/timeout parameters from relay config values.
    pub fn with_timeouts(
        mut self,
        stream_timeout_secs: u64,
        backoff_initial_ms: u64,
        backoff_max_ms: u64,
    ) -> Self {
        self.stream_timeout_secs = stream_timeout_secs;
        self.backoff_initial_ms = backoff_initial_ms;
        self.backoff_max_ms = backoff_max_ms;
        self
    }
    /// Set the maximum number of consecutive reconnect failures before giving up.
    pub fn with_max_failures(mut self, max: u64) -> Self {
        self.max_consecutive_failures = max;
        self
    }
    /// Build a provider-appropriate proxy body for sending a message.
    fn build_send_body(
        &self,
        channel_id: &str,
        text: &str,
        thread_id: Option<&str>,
    ) -> (String, serde_json::Value) {
        match self.provider {
            RelayProvider::Slack => {
                let mut body = serde_json::json!({
                    "channel": channel_id,
                    "text": text,
                });
                if let Some(tid) = thread_id {
                    body["thread_ts"] = serde_json::Value::String(tid.to_string());
                }
                ("chat.postMessage".to_string(), body)
            }
        }
    }
    /// Send a message via the provider proxy.
    async fn proxy_send(
        &self,
        team_id: &str,
        method: &str,
        body: serde_json::Value,
    ) -> Result<serde_json::Value, RelayError> {
        self.client
            .proxy_provider(
                self.provider.as_str(),
                team_id,
                method,
                body,
                Some(&self.instance_id),
            )
            .await
    }
 }
 #[async_trait]
 impl Channel for RelayChannel {
    fn name(&self) -> &str {
        self.provider.channel_name()
    }
    async fn start(&self) -> Result<MessageStream, ChannelError> {
        let channel_name = self.name().to_string();
        let token = self.stream_token.read().await.clone();
        let (stream, initial_parser_handle) = self
            .client
            .connect_stream(&token, self.stream_timeout_secs)
            .await
            .map_err(|e| ChannelError::StartupFailed {
                name: channel_name.clone(),
                reason: e.to_string(),
            })?;
        *self.parser_handle.write().await = Some(initial_parser_handle);
        let (tx, rx) = mpsc::channel(64);
        // Spawn the stream reader + reconnect task
        let client = self.client.clone();
        let stream_token = Arc::clone(&self.stream_token);
        let instance_id = self.instance_id.clone();
        let user_id = self.user_id.clone();
        let team_id = self.team_id.clone();
        let stream_timeout_secs = self.stream_timeout_secs;
        let backoff_initial_ms = self.backoff_initial_ms;
        let backoff_max_ms = self.backoff_max_ms;
        let max_consecutive_failures = self.max_consecutive_failures;
        let parser_handle = Arc::clone(&self.parser_handle);
        let provider_str = self.provider.as_str().to_string();
        let relay_name = channel_name.clone();
        let handle = tokio::spawn(async move {
            use futures::StreamExt;
            let mut current_stream = stream;
            let mut backoff_ms = backoff_initial_ms;
            let mut consecutive_failures: u64 = 0;
            loop {
                // Read events from the current stream
                while let Some(event) = current_stream.next().await {
                    // Reset backoff and failure count on successful event
                    backoff_ms = backoff_initial_ms;
                    consecutive_failures = 0;
                    // Validate required fields
                    if event.sender_id.is_empty()
                        || event.channel_id.is_empty()
                        || event.provider_scope.is_empty()
                    {
                        tracing::debug!(
                            event_type = %event.event_type,
                            sender_id = %event.sender_id,
                            channel_id = %event.channel_id,
                            "Relay: skipping event with missing required fields"
                        );
                        continue;
                    }
                    // Skip non-message events
                    if !event.is_message() {
                        tracing::debug!(
                            event_type = %event.event_type,
                            "Relay: skipping non-message event"
                        );
                        continue;
                    }
                    tracing::info!(
                        event_type = %event.event_type,
                        sender = %event.sender_id,
                        channel = %event.channel_id,
                        provider = %provider_str,
                        "Relay: received message from {}", provider_str
                    );
                    let msg = IncomingMessage::new(&relay_name, &event.sender_id, event.text())
                        .with_user_name(event.display_name())
                        .with_metadata(serde_json::json!({
                            "team_id": event.team_id(),
                            "channel_id": event.channel_id,
                            "sender_id": event.sender_id,
                            "sender_name": event.display_name(),
                            "event_type": event.event_type,
                            "thread_id": event.thread_id,
                            "provider": event.provider,
                        }));
                    let msg = if let Some(ref thread_id) = event.thread_id {
                        msg.with_thread(thread_id)
                    } else {
                        msg.with_thread(&event.channel_id)
                    };
                    if tx.send(msg).await.is_err() {
                        tracing::info!("Relay channel receiver dropped, stopping");
                        return;
                    }
                }
                // Stream ended, attempt reconnect with backoff
                consecutive_failures += 1;
                if consecutive_failures >= max_consecutive_failures {
                    tracing::error!(
                        channel = %relay_name,
                        failures = consecutive_failures,
                        "Relay channel giving up after {} consecutive failures",
                        consecutive_failures
                    );
                    break;
                }
                tracing::warn!(
                    backoff_ms = backoff_ms,
                    failures = consecutive_failures,
                    "Relay SSE stream ended, reconnecting..."
                );
                tokio::time::sleep(std::time::Duration::from_millis(backoff_ms)).await;
                backoff_ms = (backoff_ms * 2).min(backoff_max_ms);
                // Try to reconnect
                let token = stream_token.read().await.clone();
                match client.connect_stream(&token, stream_timeout_secs).await {
                    Ok((new_stream, new_parser)) => {
                        tracing::info!("Relay SSE stream reconnected");
                        current_stream = new_stream;
                        // Abort old parser before replacing
                        if let Some(old) = parser_handle.write().await.take() {
                            old.abort();
                        }
                        *parser_handle.write().await = Some(new_parser);
                    }
                    Err(RelayError::TokenExpired) => {
                        // Attempt token renewal
                        tracing::info!("Relay stream token expired, renewing...");
                        match client.renew_token(&instance_id, &user_id).await {
                            Ok(new_token) => {
                                *stream_token.write().await = new_token.clone();
                                match client.connect_stream(&new_token, stream_timeout_secs).await {
                                    Ok((new_stream, new_parser)) => {
                                        tracing::info!(
                                            "Relay SSE stream reconnected with new token"
                                        );
                                        current_stream = new_stream;
                                        if let Some(old) = parser_handle.write().await.take() {
                                            old.abort();
                                        }
                                        *parser_handle.write().await = Some(new_parser);
                                    }
                                    Err(e) => {
                                        tracing::error!(
                                            error = %e,
                                            "Failed to reconnect after token renewal"
                                        );
                                    }
                                }
                            }
                            Err(e) => {
                                tracing::error!(
                                    error = %e,
                                    "Failed to renew relay stream token"
                                );
                            }
                        }
                    }
                    Err(e) => {
                        tracing::error!(error = %e, "Failed to reconnect relay SSE stream");
                    }
                }
                // Check if the team is still valid (skip when team_id is unknown,
                // e.g. when no DB store was available at activation time)
                if !team_id.is_empty() {
                    match client.list_connections(&instance_id).await {
                        Ok(conns) => {
                            let has_team =
                                conns.iter().any(|c| c.team_id == team_id && c.connected);
                            if !has_team {
                                tracing::warn!(
                                    team_id = %team_id,
                                    "Team no longer connected, stopping relay channel"
                                );
                                return;
                            }
                        }
                        Err(e) => {
                            tracing::warn!(
                                error = %e,
                                "Could not verify team connection, will retry next iteration"
                            );
                        }
                    }
                }
            }
        });
        *self.reconnect_handle.write().await = Some(handle);
        let stream = tokio_stream::wrappers::ReceiverStream::new(rx);
        Ok(Box::pin(stream))
    }
    async fn respond(
        &self,
        msg: &IncomingMessage,
        response: OutgoingResponse,
    ) -> Result<(), ChannelError> {
        let channel_name = self.name().to_string();
        let metadata = &msg.metadata;
        let team_id = metadata
            .get("team_id")
            .and_then(|v| v.as_str())
            .unwrap_or(&self.team_id);
        let channel_id = metadata
            .get("channel_id")
            .and_then(|v| v.as_str())
            .ok_or_else(|| ChannelError::SendFailed {
                name: channel_name.clone(),
                reason: "Missing channel_id in message metadata".to_string(),
            })?;
        // Determine thread_id from response or metadata
        let thread_id = response
            .thread_id
            .as_deref()
            .or_else(|| metadata.get("thread_id").and_then(|v| v.as_str()));
        let (method, body) = self.build_send_body(channel_id, &response.content, thread_id);
        self.proxy_send(team_id, &method, body)
            .await
            .map_err(|e| ChannelError::SendFailed {
                name: channel_name,
                reason: e.to_string(),
            })?;
        Ok(())
    }
    /// Status updates are not forwarded to messaging providers to avoid noise.
    async fn send_status(
        &self,
        _status: StatusUpdate,
        _metadata: &serde_json::Value,
    ) -> Result<(), ChannelError> {
        Ok(())
    }
    async fn broadcast(
        &self,
        target: &str,
        response: OutgoingResponse,
    ) -> Result<(), ChannelError> {
        let channel_name = self.name().to_string();
        // Determine thread_id from response or metadata
        let thread_id = response
            .thread_id
            .as_deref()
            .or_else(|| response.metadata.get("thread_ts").and_then(|v| v.as_str()));
        let (method, body) = self.build_send_body(target, &response.content, thread_id);
        self.proxy_send(&self.team_id, &method, body)
            .await
            .map_err(|e| ChannelError::SendFailed {
                name: channel_name,
                reason: e.to_string(),
            })?;
        Ok(())
    }
    async fn health_check(&self) -> Result<(), ChannelError> {
        self.client
            .list_connections(&self.instance_id)
            .await
            .map_err(|_| ChannelError::HealthCheckFailed {
                name: self.name().to_string(),
            })?;
        Ok(())
    }
    fn conversation_context(&self, metadata: &serde_json::Value) -> HashMap<String, String> {
        let mut ctx = HashMap::new();
        if let Some(sender) = metadata.get("sender_name").and_then(|v| v.as_str()) {
            ctx.insert("sender".to_string(), sender.to_string());
        }
        if let Some(sender_id) = metadata.get("sender_id").and_then(|v| v.as_str()) {
            ctx.insert("sender_uuid".to_string(), sender_id.to_string());
        }
        if let Some(channel_id) = metadata.get("channel_id").and_then(|v| v.as_str()) {
            ctx.insert("group".to_string(), channel_id.to_string());
        }
        ctx.insert("platform".to_string(), self.provider.as_str().to_string());
        ctx
    }
    async fn shutdown(&self) -> Result<(), ChannelError> {
        if let Some(handle) = self.reconnect_handle.write().await.take() {
            handle.abort();
        }
        if let Some(handle) = self.parser_handle.write().await.take() {
            handle.abort();
        }
        Ok(())
    }
 }
 #[cfg(test)]
 mod tests {
    use super::*;
    fn test_client() -> RelayClient {
        RelayClient::new(
            "http://localhost:3001".into(),
            secrecy::SecretString::from("key".to_string()),
            30,
        )
        .expect("client")
    }
    #[test]
    fn relay_channel_name() {
        let channel = RelayChannel::new(
            test_client(),
            "token".into(),
            "T123".into(),
            "inst1".into(),
            "user1".into(),
        );
        assert_eq!(channel.name(), DEFAULT_RELAY_NAME);
    }
    #[test]
    fn conversation_context_extracts_metadata() {
        let channel = RelayChannel::new(
            test_client(),
            "token".into(),
            "T123".into(),
            "inst1".into(),
            "user1".into(),
        );
        let metadata = serde_json::json!({
            "sender_name": "bob",
            "sender_id": "U123",
            "channel_id": "C456",
        });
        let ctx = channel.conversation_context(&metadata);
        assert_eq!(ctx.get("sender"), Some(&"bob".to_string()));
        assert_eq!(ctx.get("sender_uuid"), Some(&"U123".to_string()));
        assert_eq!(ctx.get("platform"), Some(&"slack".to_string()));
    }
    #[test]
    fn metadata_shape_includes_event_type_and_sender_name() {
        // Regression: metadata JSON must include event_type and sender_name
        // for downstream routing (DM vs channel) and conversation_context().
        let metadata = serde_json::json!({
            "team_id": "T123",
            "channel_id": "C456",
            "sender_id": "U789",
            "sender_name": "alice",
            "event_type": "direct_message",
            "thread_id": null,
            "provider": "slack",
        });
        // event_type must be present for DM-vs-channel routing
        assert_eq!(
            metadata.get("event_type").and_then(|v| v.as_str()),
            Some("direct_message")
        );
        // sender_name must be present for conversation_context
        assert_eq!(
            metadata.get("sender_name").and_then(|v| v.as_str()),
            Some("alice")
        );
    }
    #[test]
    fn with_timeouts_sets_values() {
        let channel = RelayChannel::new(
            test_client(),
            "token".into(),
            "T123".into(),
            "inst1".into(),
            "user1".into(),
        )
        .with_timeouts(43200, 2000, 120000);
        assert_eq!(channel.stream_timeout_secs, 43200);
        assert_eq!(channel.backoff_initial_ms, 2000);
        assert_eq!(channel.backoff_max_ms, 120000);
    }
    #[test]
    fn build_send_body_slack() {
        let channel = RelayChannel::new(
            test_client(),
            "token".into(),
            "T123".into(),
            "inst1".into(),
            "user1".into(),
        );
        let (method, body) = channel.build_send_body("C456", "hello", Some("1234567.890"));
        assert_eq!(method, "chat.postMessage");
        assert_eq!(body["channel"], "C456");
        assert_eq!(body["text"], "hello");
        assert_eq!(body["thread_ts"], "1234567.890");
    }
    #[test]
    fn parser_handle_is_shared_arc() {
        let channel = RelayChannel::new(
            test_client(),
            "token".into(),
            "T123".into(),
            "inst1".into(),
            "user1".into(),
        );
        // parser_handle should be an Arc — cloning should give a second reference
        let handle_clone = Arc::clone(&channel.parser_handle);
        // Both point to the same allocation
        assert!(Arc::ptr_eq(&channel.parser_handle, &handle_clone));
    }
    #[test]
    fn with_max_failures_sets_value() {
        let channel = RelayChannel::new(
            test_client(),
            "token".into(),
            "T123".into(),
            "inst1".into(),
            "user1".into(),
        )
        .with_max_failures(10);
        assert_eq!(channel.max_consecutive_failures, 10);
    }
    #[test]
    fn default_max_failures_is_50() {
        let channel = RelayChannel::new(
            test_client(),
            "token".into(),
            "T123".into(),
            "inst1".into(),
            "user1".into(),
        );
        assert_eq!(channel.max_consecutive_failures, 50);
    }
    #[test]
    fn empty_team_id_accepted_at_construction() {
        // Regression: empty team_id (when no DB store is available) must not
        // prevent channel construction or cause immediate shutdown.
        let channel = RelayChannel::new(
            test_client(),
            "token".into(),
            String::new(), // empty team_id
            "inst1".into(),
            "user1".into(),
        );
        assert_eq!(channel.team_id, "");
        // The reconnect loop now skips team validation when team_id is empty,
        // so the channel remains alive.
    }
 }
@@ -1,549 +0,0 @@
 //! HTTP client for the channel-relay service.
 //!
 //! Wraps reqwest for all channel-relay API calls: OAuth initiation,
 //! SSE streaming, token renewal, and Slack API proxy.
 use std::pin::Pin;
 use std::task::{Context, Poll};
 use futures::Stream;
 use secrecy::{ExposeSecret, SecretString};
 use serde::{Deserialize, Serialize};
 use tokio::sync::mpsc;
 /// Known relay event types.
 pub mod event_types {
    pub const MESSAGE: &str = "message";
    pub const DIRECT_MESSAGE: &str = "direct_message";
    pub const MENTION: &str = "mention";
 }
 /// A parsed SSE event from the channel-relay stream.
 ///
 /// Field names match the channel-relay `ChannelEvent` struct exactly.
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct ChannelEvent {
    /// Unique event ID.
    #[serde(default)]
    pub id: String,
    /// Event type enum from channel-relay (e.g., "direct_message", "message", "mention").
    pub event_type: String,
    /// Provider (e.g., "slack").
    #[serde(default)]
    pub provider: String,
    /// Team/workspace ID (called `provider_scope` in channel-relay).
    #[serde(alias = "team_id", default)]
    pub provider_scope: String,
    /// Channel or DM conversation ID.
    #[serde(default)]
    pub channel_id: String,
    /// Sender user ID.
    #[serde(default)]
    pub sender_id: String,
    /// Sender display name.
    #[serde(default)]
    pub sender_name: Option<String>,
    /// Message text content (called `content` in channel-relay).
    #[serde(alias = "text", default)]
    pub content: Option<String>,
    /// Thread ID (for threaded replies, called `thread_id` in channel-relay).
    #[serde(alias = "thread_ts", default)]
    pub thread_id: Option<String>,
    /// Full raw event data.
    #[serde(default)]
    pub raw: serde_json::Value,
    /// Event timestamp (ISO 8601 from channel-relay).
    #[serde(default)]
    pub timestamp: Option<String>,
 }
 impl ChannelEvent {
    /// Get the team_id (provider_scope).
    pub fn team_id(&self) -> &str {
        &self.provider_scope
    }
    /// Get the message text content.
    pub fn text(&self) -> &str {
        self.content.as_deref().unwrap_or("")
    }
    /// Get the sender name or fallback to sender_id.
    pub fn display_name(&self) -> &str {
        self.sender_name.as_deref().unwrap_or(&self.sender_id)
    }
    /// Check if this is a message-like event that should be forwarded to the agent.
    pub fn is_message(&self) -> bool {
        matches!(
            self.event_type.as_str(),
            event_types::MESSAGE | event_types::DIRECT_MESSAGE | event_types::MENTION
        )
    }
 }
 /// Connection info returned by list_connections.
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct Connection {
    pub provider: String,
    pub team_id: String,
    pub team_name: Option<String>,
    pub connected: bool,
 }
 /// HTTP client for the channel-relay service.
 #[derive(Clone)]
 pub struct RelayClient {
    http: reqwest::Client,
    base_url: String,
    api_key: SecretString,
 }
 impl RelayClient {
    /// Create a new relay client.
    pub fn new(
        base_url: String,
        api_key: SecretString,
        request_timeout_secs: u64,
    ) -> Result<Self, RelayError> {
        let http = reqwest::Client::builder()
            .timeout(std::time::Duration::from_secs(request_timeout_secs))
            .redirect(reqwest::redirect::Policy::none())
            .build()
            .map_err(|e| RelayError::Network(format!("Failed to build HTTP client: {e}")))?;
        Ok(Self {
            http,
            base_url: base_url.trim_end_matches('/').to_string(),
            api_key,
        })
    }
    /// Initiate Slack OAuth flow via channel-relay.
    ///
    /// Calls `GET /oauth/slack/auth` with `redirect(Policy::none())` and
    /// returns the `Location` header (Slack OAuth URL) without following it.
    pub async fn initiate_oauth(
        &self,
        instance_id: &str,
        user_id: &str,
        callback_url: &str,
    ) -> Result<String, RelayError> {
        let resp = self
            .http
            .get(format!("{}/oauth/slack/auth", self.base_url))
            .header("X-API-Key", self.api_key.expose_secret())
            .query(&[
                ("instance_id", instance_id),
                ("user_id", user_id),
                ("callback", callback_url),
            ])
            .send()
            .await
            .map_err(|e| RelayError::Network(e.to_string()))?;
        let status = resp.status();
        if status.is_redirection() {
            let location = resp
                .headers()
                .get(reqwest::header::LOCATION)
                .and_then(|v| v.to_str().ok())
                .map(|s| s.to_string())
                .ok_or_else(|| {
                    RelayError::Protocol("Redirect response missing Location header".to_string())
                })?;
            Ok(location)
        } else if status.is_success() {
            // Some relay implementations return the URL in JSON body instead
            let body: serde_json::Value = resp
                .json()
                .await
                .map_err(|e| RelayError::Protocol(e.to_string()))?;
            body.get("auth_url")
                .or_else(|| body.get("url"))
                .and_then(|v| v.as_str())
                .map(|s| s.to_string())
                .ok_or_else(|| RelayError::Protocol("Response missing auth_url field".to_string()))
        } else {
            let body = resp.text().await.unwrap_or_default();
            Err(RelayError::Api {
                status: status.as_u16(),
                message: body,
            })
        }
    }
    /// Connect to the SSE event stream.
    ///
    /// Returns a stream of parsed `ChannelEvent`s and the `JoinHandle` of the
    /// background SSE parser task. The caller is responsible for reconnection
    /// logic on stream end/error and for aborting the handle on shutdown.
    pub async fn connect_stream(
        &self,
        stream_token: &str,
        stream_timeout_secs: u64,
    ) -> Result<(ChannelEventStream, tokio::task::JoinHandle<()>), RelayError> {
        let resp = self
            .http
            .get(format!("{}/stream", self.base_url))
            .query(&[("token", stream_token)])
            .timeout(std::time::Duration::from_secs(stream_timeout_secs))
            .send()
            .await
            .map_err(|e| RelayError::Network(e.to_string()))?;
        let status = resp.status();
        if status == reqwest::StatusCode::UNAUTHORIZED {
            return Err(RelayError::TokenExpired);
        }
        if !status.is_success() {
            let body = resp.text().await.unwrap_or_default();
            return Err(RelayError::Api {
                status: status.as_u16(),
                message: body,
            });
        }
        // Spawn a background task that reads the SSE stream and sends parsed events
        let (tx, rx) = mpsc::channel(64);
        let byte_stream = resp.bytes_stream();
        let handle = tokio::spawn(parse_sse_stream(byte_stream, tx));
        Ok((ChannelEventStream { rx }, handle))
    }
    /// Renew an expired stream token.
    ///
    /// Calls `POST /stream/renew` with API key auth, returns a new stream token.
    pub async fn renew_token(
        &self,
        instance_id: &str,
        user_id: &str,
    ) -> Result<String, RelayError> {
        let resp = self
            .http
            .post(format!("{}/stream/renew", self.base_url))
            .header("X-API-Key", self.api_key.expose_secret())
            .json(&serde_json::json!({
                "instance_id": instance_id,
                "user_id": user_id,
            }))
            .send()
            .await
            .map_err(|e| RelayError::Network(e.to_string()))?;
        let status = resp.status();
        if !status.is_success() {
            let body = resp.text().await.unwrap_or_default();
            return Err(RelayError::Api {
                status: status.as_u16(),
                message: body,
            });
        }
        let body: serde_json::Value = resp
            .json()
            .await
            .map_err(|e| RelayError::Protocol(e.to_string()))?;
        body.get("stream_token")
            .or_else(|| body.get("token"))
            .and_then(|v| v.as_str())
            .map(|s| s.to_string())
            .ok_or_else(|| RelayError::Protocol("Response missing stream_token field".to_string()))
    }
    /// Proxy an API call through channel-relay for any provider.
    ///
    /// Calls `POST /proxy/{provider}/{method}?team_id=X&instance_id=Y` with the given JSON body.
    pub async fn proxy_provider(
        &self,
        provider: &str,
        team_id: &str,
        method: &str,
        body: serde_json::Value,
        instance_id: Option<&str>,
    ) -> Result<serde_json::Value, RelayError> {
        let mut query: Vec<(&str, &str)> = vec![("team_id", team_id)];
        if let Some(iid) = instance_id {
            query.push(("instance_id", iid));
        }
        let resp = self
            .http
            .post(format!("{}/proxy/{}/{}", self.base_url, provider, method))
            .header("X-API-Key", self.api_key.expose_secret())
            .query(&query)
            .json(&body)
            .send()
            .await
            .map_err(|e| RelayError::Network(e.to_string()))?;
        if !resp.status().is_success() {
            let status = resp.status().as_u16();
            let body = resp.text().await.unwrap_or_default();
            return Err(RelayError::Api {
                status,
                message: body,
            });
        }
        resp.json()
            .await
            .map_err(|e| RelayError::Protocol(e.to_string()))
    }
    /// List active connections for an instance.
    pub async fn list_connections(&self, instance_id: &str) -> Result<Vec<Connection>, RelayError> {
        let resp = self
            .http
            .get(format!("{}/connections", self.base_url))
            .header("X-API-Key", self.api_key.expose_secret())
            .query(&[("instance_id", instance_id)])
            .send()
            .await
            .map_err(|e| RelayError::Network(e.to_string()))?;
        if !resp.status().is_success() {
            let status = resp.status().as_u16();
            let body = resp.text().await.unwrap_or_default();
            return Err(RelayError::Api {
                status,
                message: body,
            });
        }
        resp.json()
            .await
            .map_err(|e| RelayError::Protocol(e.to_string()))
    }
 }
 /// Async stream of parsed channel events from SSE.
 pub struct ChannelEventStream {
    rx: mpsc::Receiver<ChannelEvent>,
 }
 impl Stream for ChannelEventStream {
    type Item = ChannelEvent;
    fn poll_next(mut self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Option<Self::Item>> {
        self.rx.poll_recv(cx)
    }
 }
 /// Parse SSE format from a reqwest bytes stream.
 ///
 /// SSE format:
 /// ```text
 /// event: message
 /// data: {"key": "value"}
 ///
 /// ```
 /// Blank line terminates an event.
 async fn parse_sse_stream(
    byte_stream: impl futures::Stream<Item = Result<bytes::Bytes, reqwest::Error>> + Send + 'static,
    tx: mpsc::Sender<ChannelEvent>,
 ) {
    use futures::StreamExt;
    let mut buffer = Vec::<u8>::new();
    let mut event_type = String::new();
    let mut data_lines = Vec::new();
    let mut byte_stream = std::pin::pin!(byte_stream);
    while let Some(chunk_result) = byte_stream.next().await {
        let chunk = match chunk_result {
            Ok(c) => c,
            Err(e) => {
                tracing::debug!(error = %e, "SSE stream chunk error");
                break;
            }
        };
        buffer.extend_from_slice(&chunk);
        // Process complete lines (decode UTF-8 only on full lines to avoid
        // corruption when multi-byte characters span chunk boundaries)
        while let Some(newline_pos) = buffer.iter().position(|&b| b == b'\n') {
            let line = String::from_utf8_lossy(&buffer[..newline_pos])
                .trim_end_matches('\r')
                .to_string();
            buffer.drain(..=newline_pos);
            if line.is_empty() {
                // Blank line = end of event
                if !data_lines.is_empty() {
                    let data = data_lines.join("\n");
                    if let Ok(mut event) = serde_json::from_str::<ChannelEvent>(&data) {
                        if event.event_type.is_empty() && !event_type.is_empty() {
                            event.event_type = event_type.clone();
                        }
                        if tx.send(event).await.is_err() {
                            return; // receiver dropped
                        }
                    } else {
                        tracing::debug!(
                            event_type = %event_type,
                            data_len = data.len(),
                            "Failed to parse SSE event data as ChannelEvent"
                        );
                    }
                }
                event_type.clear();
                data_lines.clear();
            } else if let Some(value) = line.strip_prefix("event:") {
                event_type = value.trim().to_string();
            } else if let Some(value) = line.strip_prefix("data:") {
                data_lines.push(value.trim().to_string());
            }
            // Ignore other fields (id:, retry:, comments)
        }
    }
    tracing::debug!("SSE stream ended");
 }
 /// Errors from relay client operations.
 #[derive(Debug, thiserror::Error)]
 pub enum RelayError {
    #[error("Network error: {0}")]
    Network(String),
    #[error("API error (HTTP {status}): {message}")]
    Api { status: u16, message: String },
    #[error("Protocol error: {0}")]
    Protocol(String),
    #[error("Stream token expired")]
    TokenExpired,
 }
 #[cfg(test)]
 mod tests {
    use super::*;
    #[test]
    fn channel_event_deserialize_minimal() {
        let json = r#"{"event_type": "message", "content": "hello"}"#;
        let event: ChannelEvent = serde_json::from_str(json).expect("parse failed");
        assert_eq!(event.event_type, "message");
        assert_eq!(event.text(), "hello");
        assert!(event.provider_scope.is_empty());
    }
    #[test]
    fn channel_event_deserialize_relay_format() {
        // Matches the actual channel-relay ChannelEvent serialization format.
        let json = r#"{
            "id": "evt_123",
            "event_type": "direct_message",
            "provider": "slack",
            "provider_scope": "T123",
            "channel_id": "D456",
            "sender_id": "U789",
            "sender_name": "bob",
            "content": "hi there",
            "thread_id": "1234567890.123456",
            "raw": {},
            "timestamp": "2026-03-09T21:00:00Z"
        }"#;
        let event: ChannelEvent = serde_json::from_str(json).expect("parse failed");
        assert_eq!(event.provider, "slack");
        assert_eq!(event.team_id(), "T123");
        assert_eq!(event.display_name(), "bob");
        assert_eq!(event.thread_id, Some("1234567890.123456".to_string()));
        assert!(event.is_message());
    }
    #[test]
    fn channel_event_is_message() {
        let make = |et: &str| ChannelEvent {
            id: String::new(),
            event_type: et.to_string(),
            provider: String::new(),
            provider_scope: String::new(),
            channel_id: String::new(),
            sender_id: String::new(),
            sender_name: None,
            content: None,
            thread_id: None,
            raw: serde_json::Value::Null,
            timestamp: None,
        };
        assert!(make("message").is_message());
        assert!(make("direct_message").is_message());
        assert!(make("mention").is_message());
        assert!(!make("reaction").is_message());
    }
    #[test]
    fn connection_deserialize() {
        let json = r#"{"provider": "slack", "team_id": "T123", "team_name": "My Team", "connected": true}"#;
        let conn: Connection = serde_json::from_str(json).expect("parse failed");
        assert_eq!(conn.provider, "slack");
        assert!(conn.connected);
    }
    #[test]
    fn relay_error_display() {
        let err = RelayError::Network("timeout".into());
        assert_eq!(err.to_string(), "Network error: timeout");
        let err = RelayError::Api {
            status: 401,
            message: "unauthorized".into(),
        };
        assert_eq!(err.to_string(), "API error (HTTP 401): unauthorized");
        let err = RelayError::TokenExpired;
        assert_eq!(err.to_string(), "Stream token expired");
    }
    #[test]
    fn event_type_constants_match_is_message() {
        let make = |et: &str| ChannelEvent {
            id: String::new(),
            event_type: et.to_string(),
            provider: String::new(),
            provider_scope: String::new(),
            channel_id: String::new(),
            sender_id: String::new(),
            sender_name: None,
            content: None,
            thread_id: None,
            raw: serde_json::Value::Null,
            timestamp: None,
        };
        assert!(make(event_types::MESSAGE).is_message());
        assert!(make(event_types::DIRECT_MESSAGE).is_message());
        assert!(make(event_types::MENTION).is_message());
    }
    #[tokio::test]
    async fn parse_sse_handles_multibyte_utf8_across_chunks() {
        // The crab emoji (🦀) is 4 bytes: [0xF0, 0x9F, 0xA6, 0x80].
        // Split it across two chunks to verify no U+FFFD corruption.
        let event_json = r#"{"event_type":"message","content":"hello 🦀 world","provider_scope":"T1","channel_id":"C1","sender_id":"U1"}"#;
        let full = format!("event: message\ndata: {}\n\n", event_json);
        let bytes = full.as_bytes();
        // Find the crab emoji and split mid-character
        let crab_pos = bytes
            .windows(4)
            .position(|w| w == [0xF0, 0x9F, 0xA6, 0x80])
            .expect("crab emoji not found");
        let split_at = crab_pos + 2; // split in the middle of the 4-byte emoji
        let chunk1 = bytes::Bytes::copy_from_slice(&bytes[..split_at]);
        let chunk2 = bytes::Bytes::copy_from_slice(&bytes[split_at..]);
        let chunks: Vec<Result<bytes::Bytes, reqwest::Error>> = vec![Ok(chunk1), Ok(chunk2)];
        let stream = futures::stream::iter(chunks);
        let (tx, mut rx) = mpsc::channel(8);
        parse_sse_stream(stream, tx).await;
        let event = rx.recv().await.expect("should receive event");
        assert_eq!(event.text(), "hello 🦀 world");
    }
 }
@@ -1,12 +0,0 @@
 //! Channel-relay integration for connecting to external messaging platforms
 //! (Slack) via the channel-relay service.
 //!
 //! The relay service handles OAuth, credential storage, webhook ingestion,
 //! and SSE event streaming. IronClaw consumes the SSE stream and sends
 //! messages via the relay's proxy API.
 pub mod channel;
 pub mod client;
 pub use channel::{DEFAULT_RELAY_NAME, RelayChannel};
 pub use client::RelayClient;
@@ -218,15 +218,8 @@ async fn register_channel(
    }
    // Inject credentials from secrets store / environment.
-    match inject_channel_credentials(
+    if let Some(secrets) = secrets_store {
-        &channel_arc,
+        match inject_channel_credentials(&channel_arc, secrets.as_ref(), &channel_name).await {
        secrets_store
            .as_ref()
            .map(|s| s.as_ref() as &dyn SecretsStore),
        &channel_name,
    )
    .await
    {
            Ok(count) => {
                if count > 0 {
                    tracing::info!(
@@ -244,6 +237,7 @@ async fn register_channel(
                );
            }
        }
    }
    (channel_name, Box::new(SharedWasmChannel::new(channel_arc)))
 }
@@ -253,33 +247,24 @@ async fn register_channel(
 /// Looks for secrets matching the pattern `{channel_name}_*` and injects them
 /// as credential placeholders (e.g., `telegram_bot_token` -> `{TELEGRAM_BOT_TOKEN}`).
 ///
-/// Falls back to environment variables starting with the uppercase channel name
+/// Falls back to environment variables with the uppercase name if not found
-/// prefix (e.g., `TELEGRAM_` for channel `telegram`) for missing credentials.
+/// in the secrets store (e.g., `TELEGRAM_BOT_TOKEN`).
 ///
 /// Returns the number of credentials injected.
 pub async fn inject_channel_credentials(
    channel: &Arc<WasmChannel>,
-    secrets: Option<&dyn SecretsStore>,
+    secrets: &dyn SecretsStore,
    channel_name: &str,
 ) -> anyhow::Result<usize> {
    if channel_name.trim().is_empty() {
        return Ok(0);
    }
    let mut count = 0;
    let mut injected_placeholders = HashSet::new();
    // 1. Try injecting from persistent secrets store if available
    if let Some(secrets) = secrets {
    let all_secrets = secrets
        .list("default")
        .await
        .map_err(|e| anyhow::anyhow!("Failed to list secrets: {}", e))?;
-        let prefix = format!("{}_", channel_name.to_ascii_lowercase());
+    let prefix = format!("{}_", channel_name);
    let mut count = 0;
    let mut injected_placeholders = HashSet::new();
    for secret_meta in all_secrets {
-            if !secret_meta.name.to_ascii_lowercase().starts_with(&prefix) {
+        if !secret_meta.name.starts_with(&prefix) {
            continue;
        }
@@ -310,13 +295,10 @@ pub async fn inject_channel_credentials(
        injected_placeholders.insert(placeholder);
        count += 1;
    }
    }
-    // 2. Fall back to environment variables for credentials not in the secrets store.
+    // Fall back to environment variables for required secrets not found in the store.
-    // Only env vars starting with the channel's uppercase prefix are allowed
+    // This allows channels to work when configured via env vars (e.g., TELEGRAM_BOT_TOKEN)
-    // (e.g., TELEGRAM_ for channel "telegram") to prevent reading unrelated host
+    // without requiring the setup wizard to have run.
    // credentials like AWS_SECRET_ACCESS_KEY.
    let prefix = format!("{}_", channel_name.to_ascii_uppercase());
    let caps = channel.capabilities();
    if let Some(ref http_cap) = caps.tool_capabilities.http {
        for cred_mapping in http_cap.credentials.values() {
@@ -324,14 +306,6 @@ pub async fn inject_channel_credentials(
            if injected_placeholders.contains(&placeholder) {
                continue;
            }
            if !placeholder.starts_with(&prefix) {
                tracing::warn!(
                    channel = %channel_name,
                    placeholder = %placeholder,
                    "Ignoring non-prefixed credential placeholder in environment fallback"
                );
                continue;
            }
            if let Ok(env_value) = std::env::var(&placeholder)
                && !env_value.is_empty()
            {
@@ -35,7 +35,6 @@ pub async fn chat_send_handler(
    }
    let msg_id = msg.id;
    let thread_id = msg.thread_id.clone();
    let tx_guard = state.msg_tx.read().await;
    let tx = tx_guard.as_ref().ok_or((
@@ -50,13 +49,6 @@ pub async fn chat_send_handler(
        )
    })?;
    tracing::debug!(
        message_id = %msg_id,
        thread_id = ?thread_id,
        content_len = req.content.len(),
        "Message queued to agent loop"
    );
    Ok((
        StatusCode::ACCEPTED,
        Json(SendMessageResponse {
@@ -271,6 +263,7 @@ pub async fn chat_history_handler(
    ))?;
    let session = session_manager.get_or_create_session(&state.user_id).await;
    let sess = session.lock().await;
    let limit = query.limit.unwrap_or(50);
    let before_cursor = query
@@ -288,12 +281,11 @@ pub async fn chat_history_handler(
        })
        .transpose()?;
-    // Find the thread (lock only briefly to get active_thread if needed)
+    // Find the thread
    let thread_id = if let Some(ref tid) = query.thread_id {
        Uuid::parse_str(tid)
            .map_err(|_| (StatusCode::BAD_REQUEST, "Invalid thread_id".to_string()))?
    } else {
        let sess = session.lock().await;
        sess.active_thread
            .ok_or((StatusCode::NOT_FOUND, "No active thread".to_string()))?
    };
@@ -306,13 +298,10 @@ pub async fn chat_history_handler(
            .conversation_belongs_to_user(thread_id, &state.user_id)
            .await
            .unwrap_or(false);
-        if !owned {
+        if !owned && !sess.threads.contains_key(&thread_id) {
            let sess = session.lock().await;
            if !sess.threads.contains_key(&thread_id) {
            return Err((StatusCode::NOT_FOUND, "Thread not found".to_string()));
        }
    }
    }
    // For paginated requests (before cursor set), always go to DB
    if before_cursor.is_some()
@@ -335,9 +324,6 @@ pub async fn chat_history_handler(
    }
    // Try in-memory first (freshest data for active threads)
    // Lock only when checking in-memory state
    {
        let sess = session.lock().await;
    if let Some(thread) = sess.threads.get(&thread_id)
        && (!thread.turns.is_empty() || thread.pending_approval.is_some())
    {
@@ -389,7 +375,6 @@ pub async fn chat_history_handler(
            pending_approval,
        }));
    }
    }
    // Fall back to DB for historical threads not in memory (paginated)
    if let Some(ref store) = state.store {
@@ -430,6 +415,7 @@ pub async fn chat_threads_handler(
    ))?;
    let session = session_manager.get_or_create_session(&state.user_id).await;
    let sess = session.lock().await;
    // Try DB first for persistent thread list
    if let Some(ref store) = state.store {
@@ -479,22 +465,15 @@ pub async fn chat_threads_handler(
                });
            }
            // Read active thread while holding minimal lock (just before return)
            let active_thread = {
                let sess = session.lock().await;
                sess.active_thread
            };
            return Ok(Json(ThreadListResponse {
                assistant_thread,
                threads,
-                active_thread,
+                active_thread: sess.active_thread,
            }));
        }
    }
    // Fallback: in-memory only (no assistant thread without DB)
    let sess = session.lock().await;
    let mut sorted_threads: Vec<_> = sess.threads.values().collect();
    sorted_threads.sort_by(|a, b| b.updated_at.cmp(&a.updated_at));
    let threads: Vec<ThreadInfo> = sorted_threads
@@ -511,13 +490,10 @@ pub async fn chat_threads_handler(
        })
        .collect();
    let active_thread = sess.active_thread;
    drop(sess); // Explicit drop to release lock
    Ok(Json(ThreadListResponse {
        assistant_thread: None,
        threads,
-        active_thread,
+        active_thread: sess.active_thread,
    }))
 }
@@ -46,14 +46,6 @@ pub async fn extensions_list_handler(
                } else {
                    "configured".to_string()
                })
            } else if ext.kind == crate::extensions::ExtensionKind::ChannelRelay {
                Some(if ext.active {
                    "active".to_string()
                } else if ext.authenticated {
                    "configured".to_string()
                } else {
                    "installed".to_string()
                })
            } else {
                None
            };
@@ -111,7 +103,6 @@ pub async fn extensions_install_handler(
        "mcp_server" => Some(crate::extensions::ExtensionKind::McpServer),
        "wasm_tool" => Some(crate::extensions::ExtensionKind::WasmTool),
        "wasm_channel" => Some(crate::extensions::ExtensionKind::WasmChannel),
        "channel_relay" => Some(crate::extensions::ExtensionKind::ChannelRelay),
        _ => None,
    });
@@ -124,6 +115,62 @@ pub async fn extensions_install_handler(
    }
 }
 pub async fn extensions_activate_handler(
    State(state): State<Arc<GatewayState>>,
    Path(name): Path<String>,
 ) -> 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.activate(&name).await {
        Ok(result) => {
            // Activation just loads the WASM module. Auth (OAuth/manual) is
            // triggered separately via save_setup_secrets or the auth endpoint.
            Ok(Json(ActionResponse::ok(result.message)))
        }
        Err(activate_err) => {
            let err_str = activate_err.to_string();
            let needs_auth = err_str.contains("authentication")
                || err_str.contains("401")
                || err_str.contains("Unauthorized");
            if !needs_auth {
                return Ok(Json(ActionResponse::fail(err_str)));
            }
            // Activation failed due to auth; try authenticating first.
            match ext_mgr.auth(&name, None).await {
                Ok(auth_result) if auth_result.is_authenticated() => {
                    // Auth succeeded, retry activation.
                    match ext_mgr.activate(&name).await {
                        Ok(result) => Ok(Json(ActionResponse::ok(result.message))),
                        Err(e) => Ok(Json(ActionResponse::fail(e.to_string()))),
                    }
                }
                Ok(auth_result) => {
                    // Auth in progress (OAuth URL or awaiting manual token).
                    let mut resp = ActionResponse::fail(
                        auth_result
                            .instructions()
                            .map(String::from)
                            .unwrap_or_else(|| format!("'{}' requires authentication.", name)),
                    );
                    resp.auth_url = auth_result.auth_url().map(String::from);
                    resp.awaiting_token = Some(auth_result.is_awaiting_token());
                    resp.instructions = auth_result.instructions().map(String::from);
                    Ok(Json(resp))
                }
                Err(auth_err) => Ok(Json(ActionResponse::fail(format!(
                    "Authentication failed: {}",
                    auth_err
                )))),
            }
        }
    }
 }
 pub async fn extensions_remove_handler(
    State(state): State<Arc<GatewayState>>,
    Path(name): Path<String>,
@@ -27,7 +27,7 @@ pub async fn routines_list_handler(
        .await
        .map_err(|e| (StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))?;
-    let items: Vec<RoutineInfo> = routines.iter().map(RoutineInfo::from_routine).collect();
+    let items: Vec<RoutineInfo> = routines.iter().map(routine_to_info).collect();
    Ok(Json(RoutineListResponse { routines: items }))
 }
@@ -263,6 +263,54 @@ pub async fn routines_runs_handler(
    })))
 }
 /// Convert a Routine to the trimmed RoutineInfo for list display.
 fn routine_to_info(r: &crate::agent::routine::Routine) -> RoutineInfo {
    let (trigger_type, trigger_summary) = match &r.trigger {
        crate::agent::routine::Trigger::Cron { schedule, .. } => {
            ("cron".to_string(), format!("cron: {}", schedule))
        }
        crate::agent::routine::Trigger::Event {
            pattern, channel, ..
        } => {
            let ch = channel.as_deref().unwrap_or("any");
            ("event".to_string(), format!("on {} /{}/", ch, pattern))
        }
        crate::agent::routine::Trigger::Webhook { path, .. } => {
            let p = path.as_deref().unwrap_or("/");
            ("webhook".to_string(), format!("webhook: {}", p))
        }
        crate::agent::routine::Trigger::Manual => ("manual".to_string(), "manual only".to_string()),
    };
    let action_type = match &r.action {
        crate::agent::routine::RoutineAction::Lightweight { .. } => "lightweight",
        crate::agent::routine::RoutineAction::FullJob { .. } => "full_job",
    };
    let status = if !r.enabled {
        "disabled"
    } else if r.consecutive_failures > 0 {
        "failing"
    } else {
        "active"
    };
    RoutineInfo {
        id: r.id,
        name: r.name.clone(),
        description: r.description.clone(),
        enabled: r.enabled,
        trigger_type,
        trigger_summary,
        action_type: action_type.to_string(),
        last_run_at: r.last_run_at.map(|dt| dt.to_rfc3339()),
        next_fire_at: r.next_fire_at.map(|dt| dt.to_rfc3339()),
        run_count: r.run_count,
        consecutive_failures: r.consecutive_failures,
        status: status.to_string(),
    }
 }
 /// Map `RoutineError` variants to appropriate HTTP status codes.
 fn routine_error_status(err: &RoutineError) -> StatusCode {
    match err {
@@ -97,7 +97,6 @@ impl GatewayChannel {
            skill_registry: None,
            skill_catalog: None,
            chat_rate_limiter: server::RateLimiter::new(30, 60),
            oauth_rate_limiter: server::RateLimiter::new(10, 60),
            registry_entries: Vec::new(),
            cost_guard: None,
            routine_engine: Arc::new(tokio::sync::RwLock::new(None)),
@@ -134,7 +133,6 @@ impl GatewayChannel {
            skill_registry: self.state.skill_registry.clone(),
            skill_catalog: self.state.skill_catalog.clone(),
            chat_rate_limiter: server::RateLimiter::new(30, 60),
            oauth_rate_limiter: server::RateLimiter::new(10, 60),
            registry_entries: self.state.registry_entries.clone(),
            cost_guard: self.state.cost_guard.clone(),
            routine_engine: Arc::clone(&self.state.routine_engine),
@@ -28,7 +28,6 @@ use uuid::Uuid;
 use crate::agent::SessionManager;
 use crate::bootstrap::ironclaw_base_dir;
 use crate::channels::IncomingMessage;
 use crate::channels::relay::DEFAULT_RELAY_NAME;
 use crate::channels::web::auth::{AuthState, auth_middleware};
 use crate::channels::web::handlers::jobs::{
    job_files_list_handler, job_files_read_handler, jobs_cancel_handler, jobs_detail_handler,
@@ -165,8 +164,6 @@ pub struct GatewayState {
    pub scheduler: Option<crate::tools::builtin::SchedulerSlot>,
    /// Rate limiter for chat endpoints (30 messages per 60 seconds).
    pub chat_rate_limiter: RateLimiter,
    /// Rate limiter for OAuth callback endpoints (10 requests per 60 seconds).
    pub oauth_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>,
@@ -203,11 +200,7 @@ pub async fn start_server(
    // Public routes (no auth)
    let public = Router::new()
        .route("/api/health", get(health_handler))
-        .route("/oauth/callback", get(oauth_callback_handler))
+        .route("/oauth/callback", get(oauth_callback_handler));
        .route(
            "/oauth/slack/callback",
            get(slack_relay_oauth_callback_handler),
        );
    // Protected routes (require auth)
    let auth_state = AuthState { token: auth_token };
@@ -613,208 +606,6 @@ async fn oauth_callback_handler(
    axum::response::Html(html).into_response()
 }
 /// OAuth callback for Slack via channel-relay.
 ///
 /// This is a PUBLIC route (no Bearer token required) because channel-relay
 /// redirects the user's browser here after Slack OAuth completes.
 /// Query params: `stream_token`, `provider`, `team_id`.
 async fn slack_relay_oauth_callback_handler(
    State(state): State<Arc<GatewayState>>,
    Query(params): Query<std::collections::HashMap<String, String>>,
 ) -> impl IntoResponse {
    // Rate limit
    if !state.oauth_rate_limiter.check() {
        return axum::response::Html(
            "<html><body style='font-family: system-ui; text-align: center; padding: 60px;'>\
             <h2>Too Many Requests</h2>\
             <p>Please try again later.</p>\
             </body></html>"
                .to_string(),
        )
        .into_response();
    }
    // Validate stream_token: required, non-empty, max 2048 bytes
    let stream_token = match params.get("stream_token") {
        Some(t) if !t.is_empty() && t.len() <= 2048 => t.clone(),
        Some(t) if t.len() > 2048 => {
            return axum::response::Html(
                "<html><body style='font-family: system-ui; text-align: center; padding: 60px;'>\
                 <h2>Error</h2><p>Invalid callback parameters.</p></body></html>"
                    .to_string(),
            )
            .into_response();
        }
        _ => {
            return axum::response::Html(
                "<html><body style='font-family: system-ui; text-align: center; padding: 60px;'>\
                 <h2>Error</h2><p>Invalid callback parameters.</p></body></html>"
                    .to_string(),
            )
            .into_response();
        }
    };
    // Validate team_id format: empty or T followed by alphanumeric (max 20 chars)
    let team_id = params.get("team_id").cloned().unwrap_or_default();
    if !team_id.is_empty() {
        let valid_team_id = team_id.len() <= 21
            && team_id.starts_with('T')
            && team_id[1..].chars().all(|c| c.is_ascii_alphanumeric());
        if !valid_team_id {
            return axum::response::Html(
                "<html><body style='font-family: system-ui; text-align: center; padding: 60px;'>\
                 <h2>Error</h2><p>Invalid callback parameters.</p></body></html>"
                    .to_string(),
            )
            .into_response();
        }
    }
    // Validate provider: must be "slack" (only supported provider)
    let provider = params
        .get("provider")
        .cloned()
        .unwrap_or_else(|| "slack".into());
    if provider != "slack" {
        return axum::response::Html(
            "<html><body style='font-family: system-ui; text-align: center; padding: 60px;'>\
             <h2>Error</h2><p>Invalid callback parameters.</p></body></html>"
                .to_string(),
        )
        .into_response();
    }
    let ext_mgr = match state.extension_manager.as_ref() {
        Some(mgr) => mgr,
        None => {
            return axum::response::Html(
                "<html><body style='font-family: system-ui; text-align: center; padding: 60px;'>\
                 <h2>Error</h2><p>Extension manager not available.</p></body></html>"
                    .to_string(),
            )
            .into_response();
        }
    };
    // Validate CSRF state parameter
    let state_param = match params.get("state") {
        Some(s) if !s.is_empty() && s.len() <= 128 => s.clone(),
        _ => {
            return axum::response::Html(
                "<html><body style='font-family: system-ui; text-align: center; padding: 60px;'>\
                 <h2>Error</h2><p>Invalid or expired authorization.</p></body></html>"
                    .to_string(),
            )
            .into_response();
        }
    };
    let state_key = format!("relay:{}:oauth_state", DEFAULT_RELAY_NAME);
    let stored_state = match ext_mgr
        .secrets()
        .get_decrypted(&state.user_id, &state_key)
        .await
    {
        Ok(secret) => secret.expose().to_string(),
        Err(_) => {
            return axum::response::Html(
                "<html><body style='font-family: system-ui; text-align: center; padding: 60px;'>\
                 <h2>Error</h2><p>Invalid or expired authorization.</p></body></html>"
                    .to_string(),
            )
            .into_response();
        }
    };
    if state_param != stored_state {
        return axum::response::Html(
            "<html><body style='font-family: system-ui; text-align: center; padding: 60px;'>\
             <h2>Error</h2><p>Invalid or expired authorization.</p></body></html>"
                .to_string(),
        )
        .into_response();
    }
    // Delete the nonce (one-time use)
    let _ = ext_mgr.secrets().delete(&state.user_id, &state_key).await;
    let result: Result<(), String> = async {
        // Store the stream token as a secret
        let token_key = format!("relay:{}:stream_token", DEFAULT_RELAY_NAME);
        let _ = ext_mgr.secrets().delete(&state.user_id, &token_key).await;
        ext_mgr
            .secrets()
            .create(
                &state.user_id,
                crate::secrets::CreateSecretParams {
                    name: token_key,
                    value: secrecy::SecretString::from(stream_token),
                    provider: Some(provider.clone()),
                    expires_at: None,
                },
            )
            .await
            .map_err(|e| format!("Failed to store stream token: {}", e))?;
        // Store team_id in settings
        if let Some(ref store) = state.store {
            let team_id_key = format!("relay:{}:team_id", DEFAULT_RELAY_NAME);
            let _ = store
                .set_setting(&state.user_id, &team_id_key, &serde_json::json!(team_id))
                .await;
        }
        // Activate the relay channel
        ext_mgr
            .activate_stored_relay(DEFAULT_RELAY_NAME)
            .await
            .map_err(|e| format!("Failed to activate relay channel: {}", e))?;
        Ok(())
    }
    .await;
    let (success, message) = match &result {
        Ok(()) => (true, "Slack connected successfully!".to_string()),
        Err(e) => {
            tracing::error!(error = %e, "Slack relay OAuth callback failed");
            (
                false,
                "Connection failed. Check server logs for details.".to_string(),
            )
        }
    };
    // Broadcast SSE event to notify the web UI
    state.sse.broadcast(SseEvent::AuthCompleted {
        extension_name: DEFAULT_RELAY_NAME.to_string(),
        success,
        message: message.clone(),
    });
    if success {
        axum::response::Html(
            "<html><body style='font-family: system-ui; text-align: center; padding: 60px;'>\
             <h2>Slack Connected!</h2>\
             <p>You can close this tab and return to IronClaw.</p>\
             <script>window.close()</script>\
             </body></html>"
                .to_string(),
        )
        .into_response()
    } else {
        axum::response::Html(format!(
            "<html><body style='font-family: system-ui; text-align: center; padding: 60px;'>\
             <h2>Connection Failed</h2>\
             <p>{}</p>\
             </body></html>",
            message
        ))
        .into_response()
    }
 }
 // --- Chat handlers ---
 /// Convert web gateway `ImageData` to `IncomingAttachment` objects.
@@ -872,9 +663,9 @@ async fn chat_send_handler(
    headers: axum::http::HeaderMap,
    Json(req): Json<SendMessageRequest>,
 ) -> Result<(StatusCode, Json<SendMessageResponse>), (StatusCode, String)> {
-    tracing::trace!(
+    tracing::debug!(
-        "[chat_send_handler] Received message: content_len={}, thread_id={:?}",
+        "[chat_send_handler] Received message: content={:?}, thread_id={:?}",
-        req.content.len(),
+        req.content,
        req.thread_id
    );
@@ -907,10 +698,10 @@ async fn chat_send_handler(
    }
    let msg_id = msg.id;
-    tracing::trace!(
+    tracing::debug!(
-        "[chat_send_handler] Created message id={}, content_len={}, images={}",
+        "[chat_send_handler] Created message id={}, content={:?}, images={}",
        msg_id,
-        req.content.len(),
+        req.content,
        req.images.len()
    );
@@ -1848,13 +1639,13 @@ async fn extensions_activate_handler(
            Ok(Json(resp))
        }
        Err(activate_err) => {
-            let needs_auth = matches!(
+            let err_str = activate_err.to_string();
-                &activate_err,
+            let needs_auth = err_str.contains("authentication")
-                crate::extensions::ExtensionError::AuthRequired
+                || err_str.contains("401")
-            );
+                || err_str.contains("Unauthorized");
            if !needs_auth {
-                return Ok(Json(ActionResponse::fail(activate_err.to_string())));
+                return Ok(Json(ActionResponse::fail(err_str)));
            }
            // Activation failed due to auth; try authenticating first.
@@ -2145,7 +1936,7 @@ async fn routines_list_handler(
        .await
        .map_err(|e| (StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))?;
-    let items: Vec<RoutineInfo> = routines.iter().map(RoutineInfo::from_routine).collect();
+    let items: Vec<RoutineInfo> = routines.iter().map(routine_to_info).collect();
    Ok(Json(RoutineListResponse { routines: items }))
 }
@@ -2389,6 +2180,54 @@ async fn routines_runs_handler(
    })))
 }
 /// Convert a Routine to the trimmed RoutineInfo for list display.
 fn routine_to_info(r: &crate::agent::routine::Routine) -> RoutineInfo {
    let (trigger_type, trigger_summary) = match &r.trigger {
        crate::agent::routine::Trigger::Cron { schedule, .. } => {
            ("cron".to_string(), format!("cron: {}", schedule))
        }
        crate::agent::routine::Trigger::Event {
            pattern, channel, ..
        } => {
            let ch = channel.as_deref().unwrap_or("any");
            ("event".to_string(), format!("on {} /{}/", ch, pattern))
        }
        crate::agent::routine::Trigger::Webhook { path, .. } => {
            let p = path.as_deref().unwrap_or("/");
            ("webhook".to_string(), format!("webhook: {}", p))
        }
        crate::agent::routine::Trigger::Manual => ("manual".to_string(), "manual only".to_string()),
    };
    let action_type = match &r.action {
        crate::agent::routine::RoutineAction::Lightweight { .. } => "lightweight",
        crate::agent::routine::RoutineAction::FullJob { .. } => "full_job",
    };
    let status = if !r.enabled {
        "disabled"
    } else if r.consecutive_failures > 0 {
        "failing"
    } else {
        "active"
    };
    RoutineInfo {
        id: r.id,
        name: r.name.clone(),
        description: r.description.clone(),
        enabled: r.enabled,
        trigger_type,
        trigger_summary,
        action_type: action_type.to_string(),
        last_run_at: r.last_run_at.map(|dt| dt.to_rfc3339()),
        next_fire_at: r.next_fire_at.map(|dt| dt.to_rfc3339()),
        run_count: r.run_count,
        consecutive_failures: r.consecutive_failures,
        status: status.to_string(),
    }
 }
 // --- Settings handlers ---
 async fn settings_list_handler(
@@ -2690,7 +2529,6 @@ mod tests {
            skill_catalog: None,
            scheduler: None,
            chat_rate_limiter: RateLimiter::new(30, 60),
            oauth_rate_limiter: RateLimiter::new(10, 60),
            registry_entries: vec![],
            cost_guard: None,
            routine_engine: Arc::new(tokio::sync::RwLock::new(None)),
@@ -3013,177 +2851,4 @@ mod tests {
                .is_none()
        );
    }
    // --- Slack relay OAuth CSRF tests ---
    fn test_relay_oauth_router(state: Arc<GatewayState>) -> Router {
        Router::new()
            .route(
                "/oauth/slack/callback",
                get(slack_relay_oauth_callback_handler),
            )
            .with_state(state)
    }
    fn test_secrets_store() -> Arc<dyn crate::secrets::SecretsStore + Send + Sync> {
        Arc::new(crate::secrets::InMemorySecretsStore::new(Arc::new(
            crate::secrets::SecretsCrypto::new(secrecy::SecretString::from(
                "test-key-at-least-32-chars-long!!".to_string(),
            ))
            .expect("crypto"),
        )))
    }
    fn test_ext_mgr(
        secrets: Arc<dyn crate::secrets::SecretsStore + Send + Sync>,
    ) -> Arc<ExtensionManager> {
        let tool_registry = Arc::new(ToolRegistry::new());
        let mcp_sm = Arc::new(crate::tools::mcp::session::McpSessionManager::new());
        let mcp_pm = Arc::new(crate::tools::mcp::process::McpProcessManager::new());
        Arc::new(ExtensionManager::new(
            mcp_sm,
            mcp_pm,
            secrets,
            tool_registry,
            None,
            None,
            std::path::PathBuf::from("/tmp/wasm_tools"),
            std::path::PathBuf::from("/tmp/wasm_channels"),
            None,
            "test".to_string(),
            None,
            vec![],
        ))
    }
    #[tokio::test]
    async fn test_relay_oauth_callback_missing_state_param() {
        use axum::body::Body;
        use tower::ServiceExt;
        let secrets = test_secrets_store();
        let ext_mgr = test_ext_mgr(secrets);
        let state = test_gateway_state(Some(ext_mgr));
        let app = test_relay_oauth_router(state);
        // Callback without state param should be rejected
        let req = axum::http::Request::builder()
            .uri("/oauth/slack/callback?stream_token=tok123&team_id=T123&provider=slack")
            .body(Body::empty())
            .expect("request");
        let resp = ServiceExt::<axum::http::Request<Body>>::oneshot(app, req)
            .await
            .expect("response");
        let body = axum::body::to_bytes(resp.into_body(), 1024 * 64)
            .await
            .expect("body");
        let html = String::from_utf8_lossy(&body);
        assert!(
            html.contains("Invalid or expired authorization"),
            "Expected CSRF error, got: {}",
            &html[..html.len().min(300)]
        );
    }
    #[tokio::test]
    async fn test_relay_oauth_callback_wrong_state_param() {
        use axum::body::Body;
        use tower::ServiceExt;
        let secrets = test_secrets_store();
        // Store a valid nonce
        secrets
            .create(
                "test",
                crate::secrets::CreateSecretParams::new(
                    format!("relay:{}:oauth_state", DEFAULT_RELAY_NAME),
                    "correct-nonce-value",
                ),
            )
            .await
            .expect("store nonce");
        let ext_mgr = test_ext_mgr(secrets);
        let state = test_gateway_state(Some(ext_mgr));
        let app = test_relay_oauth_router(state);
        // Callback with wrong state param
        let req = axum::http::Request::builder()
            .uri("/oauth/slack/callback?stream_token=tok123&team_id=T123&provider=slack&state=wrong-nonce")
            .body(Body::empty())
            .expect("request");
        let resp = ServiceExt::<axum::http::Request<Body>>::oneshot(app, req)
            .await
            .expect("response");
        let body = axum::body::to_bytes(resp.into_body(), 1024 * 64)
            .await
            .expect("body");
        let html = String::from_utf8_lossy(&body);
        assert!(
            html.contains("Invalid or expired authorization"),
            "Expected CSRF error for wrong nonce, got: {}",
            &html[..html.len().min(300)]
        );
    }
    #[tokio::test]
    async fn test_relay_oauth_callback_correct_state_proceeds() {
        use axum::body::Body;
        use tower::ServiceExt;
        let secrets = test_secrets_store();
        let nonce = "valid-test-nonce-12345";
        // Store the correct nonce
        secrets
            .create(
                "test",
                crate::secrets::CreateSecretParams::new(
                    format!("relay:{}:oauth_state", DEFAULT_RELAY_NAME),
                    nonce,
                ),
            )
            .await
            .expect("store nonce");
        let ext_mgr = test_ext_mgr(secrets.clone());
        let state = test_gateway_state(Some(ext_mgr));
        let app = test_relay_oauth_router(state);
        // Callback with correct state param — will pass CSRF check
        // but may fail downstream (no real relay service) — that's OK,
        // we just verify it doesn't return a CSRF error.
        let req = axum::http::Request::builder()
            .uri(format!(
                "/oauth/slack/callback?stream_token=tok123&team_id=T123&provider=slack&state={}",
                nonce
            ))
            .body(Body::empty())
            .expect("request");
        let resp = ServiceExt::<axum::http::Request<Body>>::oneshot(app, req)
            .await
            .expect("response");
        let body = axum::body::to_bytes(resp.into_body(), 1024 * 64)
            .await
            .expect("body");
        let html = String::from_utf8_lossy(&body);
        // Should NOT contain the CSRF error message
        assert!(
            !html.contains("Invalid or expired authorization"),
            "Should have passed CSRF check, got: {}",
            &html[..html.len().min(300)]
        );
        // Verify the nonce was consumed (deleted)
        let state_key = format!("relay:{}:oauth_state", DEFAULT_RELAY_NAME);
        let exists = secrets.exists("test", &state_key).await.unwrap_or(true);
        assert!(!exists, "CSRF nonce should be deleted after use");
    }
 }
@@ -2350,8 +2350,8 @@ function renderExtensionCard(ext) {
    activeLabel.textContent = ext.active ? 'Active' : 'Installed';
    actions.appendChild(activeLabel);
-    // MCP servers and channel-relay extensions may be installed but inactive — show Activate button
+    // MCP servers may be installed but inactive — show Activate button
-    if ((ext.kind === 'mcp_server' || ext.kind === 'channel_relay') && !ext.active) {
+    if (ext.kind === 'mcp_server' && !ext.active) {
      const activateBtn = document.createElement('button');
      activateBtn.className = 'btn-ext activate';
      activateBtn.textContent = 'Activate';
@@ -1277,8 +1277,6 @@ body {
  gap: 8px;
  background: var(--bg-secondary);
  border-top: 1px solid var(--border);
  flex-shrink: 0;
  min-height: 56px;
 }
 .chat-input textarea {
@@ -3722,21 +3720,6 @@ mark {
  .ext-install-form input {
    width: 100%;
  }
  /* Chat input: ensure visibility on mobile */
  .chat-input {
    min-height: 52px;
  }
  .chat-input textarea {
    min-height: 36px;
    max-height: 100px;
  }
  .chat-input button {
    padding: 6px 16px;
    font-size: 14px;
  }
 }
 /* Slash command autocomplete dropdown */
@@ -82,7 +82,6 @@ impl TestGatewayBuilder {
            skill_catalog: None,
            scheduler: None,
            chat_rate_limiter: RateLimiter::new(30, 60),
            oauth_rate_limiter: RateLimiter::new(10, 60),
            registry_entries: Vec::new(),
            cost_guard: None,
            routine_engine: Arc::new(tokio::sync::RwLock::new(None)),
@@ -735,60 +735,6 @@ pub struct RoutineInfo {
    pub status: String,
 }
 impl RoutineInfo {
    /// Convert a `Routine` to the trimmed `RoutineInfo` for list display.
    pub fn from_routine(r: &crate::agent::routine::Routine) -> Self {
        let (trigger_type, trigger_summary) = match &r.trigger {
            crate::agent::routine::Trigger::Cron { schedule, .. } => {
                ("cron".to_string(), format!("cron: {}", schedule))
            }
            crate::agent::routine::Trigger::Event {
                pattern, channel, ..
            } => {
                let ch = channel.as_deref().unwrap_or("any");
                ("event".to_string(), format!("on {} /{}/", ch, pattern))
            }
            crate::agent::routine::Trigger::SystemEvent {
                source, event_type, ..
            } => (
                "system_event".to_string(),
                format!("event: {}.{}", source, event_type),
            ),
            crate::agent::routine::Trigger::Manual => {
                ("manual".to_string(), "manual only".to_string())
            }
        };
        let action_type = match &r.action {
            crate::agent::routine::RoutineAction::Lightweight { .. } => "lightweight",
            crate::agent::routine::RoutineAction::FullJob { .. } => "full_job",
        };
        let status = if !r.enabled {
            "disabled"
        } else if r.consecutive_failures > 0 {
            "failing"
        } else {
            "active"
        };
        RoutineInfo {
            id: r.id,
            name: r.name.clone(),
            description: r.description.clone(),
            enabled: r.enabled,
            trigger_type,
            trigger_summary,
            action_type: action_type.to_string(),
            last_run_at: r.last_run_at.map(|dt| dt.to_rfc3339()),
            next_fire_at: r.next_fire_at.map(|dt| dt.to_rfc3339()),
            run_count: r.run_count,
            consecutive_failures: r.consecutive_failures,
            status: status.to_string(),
        }
    }
 }
 #[derive(Debug, Serialize)]
 pub struct RoutineListResponse {
    pub routines: Vec<RoutineInfo>,
@@ -509,7 +509,6 @@ mod tests {
            skill_registry: None,
            skill_catalog: None,
            chat_rate_limiter: crate::channels::web::server::RateLimiter::new(30, 60),
            oauth_rate_limiter: crate::channels::web::server::RateLimiter::new(10, 60),
            registry_entries: Vec::new(),
            cost_guard: None,
            routine_engine: Arc::new(tokio::sync::RwLock::new(None)),
@@ -24,8 +24,6 @@ pub struct WebhookServerConfig {
 pub struct WebhookServer {
    config: WebhookServerConfig,
    routes: Vec<Router>,
    /// Merged router saved after start() for restart_with_addr().
    merged_router: Option<Router>,
    shutdown_tx: Option<oneshot::Sender<()>>,
    handle: Option<JoinHandle<()>>,
 }
@@ -36,7 +34,6 @@ impl WebhookServer {
        Self {
            config,
            routes: Vec::new(),
            merged_router: None,
            shutdown_tx: None,
            handle: None,
        }
@@ -54,13 +51,7 @@ impl WebhookServer {
        for fragment in self.routes.drain(..) {
            app = app.merge(fragment);
        }
        self.merged_router = Some(app.clone());
        self.bind_and_spawn(app).await
    }
    /// Bind a listener to the configured address and spawn the server task.
    /// Private helper used by both start() and restart_with_addr().
    async fn bind_and_spawn(&mut self, app: Router) -> Result<(), ChannelError> {
        let listener = tokio::net::TcpListener::bind(self.config.addr)
            .await
            .map_err(|e| ChannelError::StartupFailed {
@@ -89,54 +80,6 @@ impl WebhookServer {
        Ok(())
    }
    /// Gracefully shut down the current listener and rebind to a new address.
    /// The merged router from the original `start()` call is reused.
    ///
    /// If binding to the new address fails, the old listener remains active and
    /// state is restored. This prevents a denial-of-service if the new address
    /// is invalid or already in use.
    pub async fn restart_with_addr(&mut self, new_addr: SocketAddr) -> Result<(), ChannelError> {
        let app = self
            .merged_router
            .clone()
            .ok_or_else(|| ChannelError::StartupFailed {
                name: "webhook_server".to_string(),
                reason: "restart_with_addr called before start()".to_string(),
            })?;
        // Save old state for rollback if new bind fails
        let old_addr = self.config.addr;
        let old_shutdown_tx = self.shutdown_tx.take();
        let old_handle = self.handle.take();
        // Update config to new address and try to bind
        self.config.addr = new_addr;
        match self.bind_and_spawn(app).await {
            Ok(()) => {
                // New listener is running, gracefully shut down the old one
                if let Some(tx) = old_shutdown_tx {
                    let _ = tx.send(());
                }
                if let Some(handle) = old_handle {
                    let _ = handle.await;
                }
                Ok(())
            }
            Err(e) => {
                // Restore old state; old listener remains active
                self.config.addr = old_addr;
                self.shutdown_tx = old_shutdown_tx;
                self.handle = old_handle;
                Err(e)
            }
        }
    }
    /// Return the current bind address.
    pub fn current_addr(&self) -> SocketAddr {
        self.config.addr
    }
    /// Signal graceful shutdown and wait for the server task to finish.
    pub async fn shutdown(&mut self) {
        if let Some(tx) = self.shutdown_tx.take() {
@@ -147,182 +90,3 @@ impl WebhookServer {
        }
    }
 }
 #[cfg(test)]
 mod tests {
    use super::*;
    use axum::Json;
    use serde_json::json;
    #[tokio::test]
    async fn test_restart_with_addr_rebinds_listener() {
        use std::net::TcpListener as StdTcpListener;
        // Find two available ports by binding and immediately closing
        let port1 = {
            let listener =
                StdTcpListener::bind("127.0.0.1:0").expect("Failed to find available port 1");
            listener
                .local_addr()
                .expect("Failed to get local addr")
                .port()
        };
        let port2 = {
            let listener =
                StdTcpListener::bind("127.0.0.1:0").expect("Failed to find available port 2");
            listener
                .local_addr()
                .expect("Failed to get local addr")
                .port()
        };
        assert_ne!(port1, port2, "Should have different ports");
        assert_ne!(port1, 0, "Port 1 should be non-zero");
        assert_ne!(port2, 0, "Port 2 should be non-zero");
        // Start server on first port
        let addr1 = format!("127.0.0.1:{}", port1).parse().unwrap();
        let mut server = WebhookServer::new(WebhookServerConfig { addr: addr1 });
        // Create a test router that responds to health checks
        let test_router = axum::Router::new().route(
            "/health",
            axum::routing::get(|| async { Json(json!({"status": "ok"})) }),
        );
        server.add_routes(test_router);
        // Start the server on first port
        server.start().await.expect("Failed to start server");
        assert_eq!(
            server.current_addr(),
            addr1,
            "Server should be bound to initial address"
        );
        // Verify the first server is actually listening
        let client = reqwest::Client::new();
        let response = client
            .get(format!("http://{}/health", addr1))
            .send()
            .await
            .expect("Failed to send request to first server");
        assert_eq!(
            response.status(),
            200,
            "First server should respond to health check"
        );
        // Restart on second port
        let addr2 = format!("127.0.0.1:{}", port2).parse().unwrap();
        server
            .restart_with_addr(addr2)
            .await
            .expect("Failed to restart with new addr");
        // Assert the address changed
        assert_eq!(
            server.current_addr(),
            addr2,
            "Server address should be updated after restart"
        );
        assert_ne!(
            addr1, addr2,
            "Address should change after restart_with_addr"
        );
        // Verify the new server is actually listening on the new address
        let response = client
            .get(format!("http://{}/health", addr2))
            .send()
            .await
            .expect("Failed to send request to restarted server");
        assert_eq!(
            response.status(),
            200,
            "Restarted server should respond to health check on new address"
        );
        // Verify the old address is no longer responding
        let old_result = tokio::time::timeout(
            std::time::Duration::from_millis(200),
            client.get(format!("http://{}/health", addr1)).send(),
        )
        .await;
        assert!(
            old_result.is_err() || old_result.as_ref().unwrap().is_err(),
            "Old address should not respond after server restarts"
        );
        // Clean up
        server.shutdown().await;
    }
    #[tokio::test]
    async fn test_restart_with_addr_rollback_on_bind_failure() {
        use std::net::TcpListener as StdTcpListener;
        // Find an available port
        let port1 = {
            let listener =
                StdTcpListener::bind("127.0.0.1:0").expect("Failed to find available port");
            listener
                .local_addr()
                .expect("Failed to get local addr")
                .port()
        };
        // Start server on first port
        let addr1 = format!("127.0.0.1:{}", port1).parse().unwrap();
        let mut server = WebhookServer::new(WebhookServerConfig { addr: addr1 });
        // Create a test router
        let test_router = axum::Router::new().route(
            "/health",
            axum::routing::get(|| async { Json(json!({"status": "ok"})) }),
        );
        server.add_routes(test_router);
        // Start the server on first port
        server.start().await.expect("Failed to start server");
        // Verify the server is listening
        let client = reqwest::Client::new();
        let response = client
            .get(format!("http://{}/health", addr1))
            .send()
            .await
            .expect("Failed to send request");
        assert_eq!(response.status(), 200, "Server should be listening");
        // Try to restart on an invalid address (port 0 is reserved, won't bind)
        // Use port 1 which typically requires elevated privileges
        let invalid_addr: SocketAddr = "127.0.0.1:1".parse().unwrap();
        // Attempt restart (should fail)
        let result = server.restart_with_addr(invalid_addr).await;
        assert!(result.is_err(), "Restart with invalid address should fail");
        // Verify the old address is still responding (rollback succeeded)
        let response = client
            .get(format!("http://{}/health", addr1))
            .send()
            .await
            .expect("Failed to send request to old address");
        assert_eq!(
            response.status(),
            200,
            "Old listener should still be running after failed restart"
        );
        // Verify the server address is unchanged
        assert_eq!(
            server.current_addr(),
            addr1,
            "Server address should be restored after failed restart"
        );
        // Clean up
        server.shutdown().await;
    }
 }
@@ -7,7 +7,6 @@
 use std::path::PathBuf;
 use crate::bootstrap::ironclaw_base_dir;
 use crate::settings::Settings;
 /// Run all diagnostic checks and print results.
 pub async fn run_doctor_command() -> anyhow::Result<()> {
@@ -16,35 +15,14 @@ pub async fn run_doctor_command() -> anyhow::Result<()> {
    let mut passed = 0u32;
    let mut failed = 0u32;
    let mut skipped = 0u32;
-    // Load settings once for checks that need them.
+    // ── Configuration checks ──────────────────────────────────
    let settings = Settings::load();
    // ── Settings & core config ─────────────────────────────────
    check(
        "Settings file",
        check_settings_file(),
        &mut passed,
        &mut failed,
        &mut skipped,
    );
    check(
        "NEAR AI session",
        check_nearai_session().await,
        &mut passed,
        &mut failed,
        &mut skipped,
    );
    check(
        "LLM configuration",
        check_llm_config(&settings),
        &mut passed,
        &mut failed,
        &mut skipped,
    );
    check(
@@ -52,7 +30,6 @@ pub async fn run_doctor_command() -> anyhow::Result<()> {
        check_database().await,
        &mut passed,
        &mut failed,
        &mut skipped,
    );
    check(
@@ -60,75 +37,15 @@ pub async fn run_doctor_command() -> anyhow::Result<()> {
        check_workspace_dir(),
        &mut passed,
        &mut failed,
        &mut skipped,
    );
    // ── Subsystem configuration checks ─────────────────────────
    check(
        "Embeddings",
        check_embeddings(&settings),
        &mut passed,
        &mut failed,
        &mut skipped,
    );
    check(
        "Routines config",
        check_routines_config(),
        &mut passed,
        &mut failed,
        &mut skipped,
    );
    check(
        "Gateway config",
        check_gateway_config(&settings),
        &mut passed,
        &mut failed,
        &mut skipped,
    );
    check(
        "MCP servers",
        check_mcp_config().await,
        &mut passed,
        &mut failed,
        &mut skipped,
    );
    check(
        "Skills",
        check_skills().await,
        &mut passed,
        &mut failed,
        &mut skipped,
    );
    check(
        "Secrets",
        check_secrets(&settings),
        &mut passed,
        &mut failed,
        &mut skipped,
    );
    check(
        "Service",
        check_service_installed(),
        &mut passed,
        &mut failed,
        &mut skipped,
    );
    // ── External binary checks ────────────────────────────────
    check(
-        "Docker daemon",
+        "Docker",
-        check_docker_daemon().await,
+        check_binary("docker", &["--version"]),
        &mut passed,
        &mut failed,
        &mut skipped,
    );
    check(
@@ -136,7 +53,6 @@ pub async fn run_doctor_command() -> anyhow::Result<()> {
        check_binary("cloudflared", &["--version"]),
        &mut passed,
        &mut failed,
        &mut skipped,
    );
    check(
@@ -144,7 +60,6 @@ pub async fn run_doctor_command() -> anyhow::Result<()> {
        check_binary("ngrok", &["version"]),
        &mut passed,
        &mut failed,
        &mut skipped,
    );
    check(
@@ -152,13 +67,12 @@ pub async fn run_doctor_command() -> anyhow::Result<()> {
        check_binary("tailscale", &["version"]),
        &mut passed,
        &mut failed,
        &mut skipped,
    );
    // ── Summary ───────────────────────────────────────────────
    println!();
-    println!("  {passed} passed, {failed} failed, {skipped} skipped");
+    println!("  {passed} passed, {failed} failed");
    if failed > 0 {
        println!("\n  Some checks failed. This is normal if you don't use those features.");
@@ -169,7 +83,7 @@ pub async fn run_doctor_command() -> anyhow::Result<()> {
 // ── Individual checks ───────────────────────────────────────
-fn check(name: &str, result: CheckResult, passed: &mut u32, failed: &mut u32, skipped: &mut u32) {
+fn check(name: &str, result: CheckResult, passed: &mut u32, failed: &mut u32) {
    match result {
        CheckResult::Pass(detail) => {
            *passed += 1;
@@ -180,7 +94,6 @@ fn check(name: &str, result: CheckResult, passed: &mut u32, failed: &mut u32, sk
            println!("  [FAIL] {name}: {detail}");
        }
        CheckResult::Skip(reason) => {
            *skipped += 1;
            println!("  [skip] {name}: {reason}");
        }
    }
@@ -192,29 +105,6 @@ enum CheckResult {
    Skip(String),
 }
 // ── Settings file ───────────────────────────────────────────
 fn check_settings_file() -> CheckResult {
    let path = Settings::default_path();
    if !path.exists() {
        return CheckResult::Pass("no settings file (defaults will be used)".into());
    }
    match std::fs::read_to_string(&path) {
        Ok(data) => match serde_json::from_str::<serde_json::Value>(&data) {
            Ok(_) => CheckResult::Pass(format!("valid ({})", path.display())),
            Err(e) => CheckResult::Fail(format!(
                "settings.json is malformed: {}. Fix or delete {}",
                e,
                path.display()
            )),
        },
        Err(e) => CheckResult::Fail(format!("cannot read {}: {}", path.display(), e)),
    }
 }
 // ── NEAR AI session ─────────────────────────────────────────
 async fn check_nearai_session() -> CheckResult {
    // Check if session file exists
    let session_path = crate::config::llm::default_session_path();
@@ -239,27 +129,6 @@ async fn check_nearai_session() -> CheckResult {
    }
 }
 // ── LLM configuration ──────────────────────────────────────
 fn check_llm_config(settings: &Settings) -> CheckResult {
    match crate::llm::LlmConfig::resolve(settings) {
        Ok(config) => {
            // Show the model for the active backend, not always nearai.model.
            let model = if let Some(ref bedrock) = config.bedrock {
                &bedrock.model
            } else if let Some(ref provider) = config.provider {
                &provider.model
            } else {
                &config.nearai.model
            };
            CheckResult::Pass(format!("backend={}, model={}", config.backend, model))
        }
        Err(e) => CheckResult::Fail(format!("LLM config error: {e}")),
    }
 }
 // ── Database ────────────────────────────────────────────────
 async fn check_database() -> CheckResult {
    let backend = std::env::var("DATABASE_BACKEND")
        .ok()
@@ -323,8 +192,6 @@ async fn try_pg_connect() -> Result<(), String> {
    Err("postgres feature not compiled in".into())
 }
 // ── Workspace directory ─────────────────────────────────────
 fn check_workspace_dir() -> CheckResult {
    let dir = ironclaw_base_dir();
@@ -339,222 +206,6 @@ fn check_workspace_dir() -> CheckResult {
    }
 }
 // ── Embeddings ──────────────────────────────────────────────
 fn check_embeddings(settings: &Settings) -> CheckResult {
    match crate::config::EmbeddingsConfig::resolve(settings) {
        Ok(config) => {
            if !config.enabled {
                return CheckResult::Skip("disabled (set EMBEDDING_ENABLED=true)".into());
            }
            let has_creds = match config.provider.as_str() {
                "openai" => config.openai_api_key().is_some(),
                "nearai" => {
                    // NearAiEmbeddings uses SessionManager::get_token() which
                    // only returns session tokens, NOT NEARAI_API_KEY
                    // (src/workspace/embeddings.rs:309, src/llm/session.rs:132).
                    let session_path = crate::config::llm::default_session_path();
                    session_path.exists()
                        && std::fs::read_to_string(&session_path)
                            .map(|s| !s.trim().is_empty())
                            .unwrap_or(false)
                }
                "ollama" => true, // local, no creds needed
                _ => config.openai_api_key().is_some(),
            };
            if has_creds {
                CheckResult::Pass(format!(
                    "provider={}, model={}",
                    config.provider, config.model
                ))
            } else {
                let hint = match config.provider.as_str() {
                    "nearai" => "run `ironclaw onboard` to create a session",
                    _ => "set OPENAI_API_KEY",
                };
                CheckResult::Fail(format!(
                    "provider={} but credentials missing ({})",
                    config.provider, hint
                ))
            }
        }
        Err(e) => CheckResult::Fail(format!("config error: {e}")),
    }
 }
 // ── Routines config ─────────────────────────────────────────
 fn check_routines_config() -> CheckResult {
    match crate::config::RoutineConfig::resolve() {
        Ok(config) => {
            if config.enabled {
                CheckResult::Pass(format!(
                    "enabled (interval={}s, max_concurrent={})",
                    config.cron_check_interval_secs, config.max_concurrent_routines
                ))
            } else {
                CheckResult::Skip("disabled".into())
            }
        }
        Err(e) => CheckResult::Fail(format!("config error: {e}")),
    }
 }
 // ── Gateway config ──────────────────────────────────────────
 fn check_gateway_config(settings: &Settings) -> CheckResult {
    // Use the same resolve() path as runtime so invalid env values
    // (e.g. GATEWAY_PORT=abc) are caught here too.
    match crate::config::ChannelsConfig::resolve(settings) {
        Ok(channels) => match channels.gateway {
            Some(gw) => {
                if gw.auth_token.is_some() {
                    CheckResult::Pass(format!(
                        "enabled at {}:{} (auth token set)",
                        gw.host, gw.port
                    ))
                } else {
                    CheckResult::Pass(format!(
                        "enabled at {}:{} (no auth token — random token will be generated)",
                        gw.host, gw.port
                    ))
                }
            }
            None => CheckResult::Skip("disabled (GATEWAY_ENABLED=false)".into()),
        },
        Err(e) => CheckResult::Fail(format!("config error: {e}")),
    }
 }
 // ── MCP servers ─────────────────────────────────────────────
 async fn check_mcp_config() -> CheckResult {
    match crate::tools::mcp::config::load_mcp_servers().await {
        Ok(file) => {
            let servers: Vec<_> = file.enabled_servers().collect();
            if servers.is_empty() {
                return CheckResult::Skip("no MCP servers configured".into());
            }
            let mut invalid = Vec::new();
            for server in &servers {
                if let Err(e) = server.validate() {
                    invalid.push(format!("{}: {}", server.name, e));
                }
            }
            if invalid.is_empty() {
                CheckResult::Pass(format!("{} server(s) configured, all valid", servers.len()))
            } else {
                CheckResult::Fail(format!(
                    "{} server(s), {} invalid: {}",
                    servers.len(),
                    invalid.len(),
                    invalid.join("; ")
                ))
            }
        }
        Err(e) => {
            // Distinguish no config from corrupted config
            let msg = e.to_string();
            if msg.contains("not found") || msg.contains("No such file") {
                CheckResult::Skip("no MCP config file".into())
            } else {
                CheckResult::Fail(format!("config error: {e}"))
            }
        }
    }
 }
 // ── Skills ──────────────────────────────────────────────────
 async fn check_skills() -> CheckResult {
    let user_dir = ironclaw_base_dir().join("skills");
    let installed_dir = ironclaw_base_dir().join("installed_skills");
    let mut registry = crate::skills::SkillRegistry::new(user_dir.clone());
    registry = registry.with_installed_dir(installed_dir);
    // discover_all() returns loaded skill names (not warnings).
    let _loaded_names = registry.discover_all().await;
    let count = registry.count();
    if count == 0 {
        return CheckResult::Skip("no skills discovered".into());
    }
    CheckResult::Pass(format!("{count} skill(s) loaded"))
 }
 // ── Secrets ─────────────────────────────────────────────────
 fn check_secrets(settings: &Settings) -> CheckResult {
    match settings.secrets_master_key_source {
        crate::settings::KeySource::Keychain => {
            CheckResult::Pass("master key source: OS keychain".into())
        }
        crate::settings::KeySource::Env => {
            if std::env::var("SECRETS_MASTER_KEY").is_ok() {
                CheckResult::Pass("master key source: env var (set)".into())
            } else {
                CheckResult::Fail(
                    "master key source: env var but SECRETS_MASTER_KEY not set".into(),
                )
            }
        }
        crate::settings::KeySource::None => {
            CheckResult::Skip("secrets not configured (run `ironclaw onboard`)".into())
        }
    }
 }
 // ── Service ─────────────────────────────────────────────────
 fn check_service_installed() -> CheckResult {
    if cfg!(target_os = "macos") {
        let plist =
            dirs::home_dir().map(|h| h.join("Library/LaunchAgents/com.ironclaw.daemon.plist"));
        match plist {
            Some(path) if path.exists() => {
                CheckResult::Pass(format!("launchd plist installed ({})", path.display()))
            }
            Some(_) => CheckResult::Skip("not installed (run `ironclaw service install`)".into()),
            None => CheckResult::Skip("cannot determine home directory".into()),
        }
    } else if cfg!(target_os = "linux") {
        let unit = dirs::home_dir().map(|h| h.join(".config/systemd/user/ironclaw.service"));
        match unit {
            Some(path) if path.exists() => {
                CheckResult::Pass(format!("systemd unit installed ({})", path.display()))
            }
            Some(_) => CheckResult::Skip("not installed (run `ironclaw service install`)".into()),
            None => CheckResult::Skip("cannot determine home directory".into()),
        }
    } else {
        CheckResult::Skip("service management not supported on this platform".into())
    }
 }
 // ── Docker daemon ───────────────────────────────────────────
 async fn check_docker_daemon() -> CheckResult {
    let detection = crate::sandbox::check_docker().await;
    match detection.status {
        crate::sandbox::DockerStatus::Available => CheckResult::Pass("running".into()),
        crate::sandbox::DockerStatus::NotInstalled => CheckResult::Skip(format!(
            "not installed. {}",
            detection.platform.install_hint()
        )),
        crate::sandbox::DockerStatus::NotRunning => CheckResult::Fail(format!(
            "installed but not running. {}",
            detection.platform.start_hint()
        )),
        crate::sandbox::DockerStatus::Disabled => CheckResult::Skip("sandbox disabled".into()),
    }
 }
 // ── External binary ─────────────────────────────────────────
 fn check_binary(name: &str, args: &[&str]) -> CheckResult {
    match std::process::Command::new(name)
        .args(args)
@@ -622,193 +273,6 @@ mod tests {
        }
    }
    #[test]
    fn check_settings_file_handles_missing() {
        // Settings::default_path() might or might not exist, but must not panic
        let result = check_settings_file();
        match result {
            CheckResult::Pass(_) | CheckResult::Fail(_) | CheckResult::Skip(_) => {}
        }
    }
    #[test]
    fn check_llm_config_does_not_panic() {
        let settings = Settings::default();
        let result = check_llm_config(&settings);
        match result {
            CheckResult::Pass(_) | CheckResult::Fail(_) | CheckResult::Skip(_) => {}
        }
    }
    #[test]
    fn check_routines_config_does_not_panic() {
        let result = check_routines_config();
        match result {
            CheckResult::Pass(_) | CheckResult::Fail(_) | CheckResult::Skip(_) => {}
        }
    }
    #[test]
    fn check_gateway_config_does_not_panic() {
        let settings = Settings::default();
        let result = check_gateway_config(&settings);
        match result {
            CheckResult::Pass(_) | CheckResult::Fail(_) | CheckResult::Skip(_) => {}
        }
    }
    #[test]
    fn check_embeddings_does_not_panic() {
        let settings = Settings::default();
        let result = check_embeddings(&settings);
        match result {
            CheckResult::Pass(_) | CheckResult::Fail(_) | CheckResult::Skip(_) => {}
        }
    }
    #[test]
    fn check_secrets_none_returns_skip() {
        let settings = Settings::default();
        match check_secrets(&settings) {
            CheckResult::Skip(msg) => {
                assert!(
                    msg.contains("not configured"),
                    "expected 'not configured' in skip message, got: {msg}"
                );
            }
            other => panic!(
                "expected Skip for default settings, got: {}",
                format_result(&other)
            ),
        }
    }
    #[test]
    fn check_service_installed_does_not_panic() {
        let result = check_service_installed();
        match result {
            CheckResult::Pass(_) | CheckResult::Fail(_) | CheckResult::Skip(_) => {}
        }
    }
    #[tokio::test]
    async fn check_docker_daemon_does_not_panic() {
        let result = check_docker_daemon().await;
        match result {
            CheckResult::Pass(_) | CheckResult::Fail(_) | CheckResult::Skip(_) => {}
        }
    }
    #[tokio::test]
    async fn check_mcp_config_does_not_panic() {
        let result = check_mcp_config().await;
        match result {
            CheckResult::Pass(_) | CheckResult::Fail(_) | CheckResult::Skip(_) => {}
        }
    }
    #[tokio::test]
    async fn check_skills_does_not_panic() {
        let result = check_skills().await;
        match result {
            CheckResult::Pass(_) | CheckResult::Fail(_) | CheckResult::Skip(_) => {}
        }
    }
    #[test]
    fn check_llm_config_shows_nearai_model_for_nearai_backend() {
        let _guard = crate::config::helpers::ENV_MUTEX.lock().expect("env mutex");
        // SAFETY: Under ENV_MUTEX, no concurrent env access.
        unsafe {
            std::env::remove_var("LLM_BACKEND");
        }
        let settings = Settings::default();
        match check_llm_config(&settings) {
            CheckResult::Pass(msg) => {
                assert!(
                    msg.contains("backend=nearai"),
                    "expected nearai backend, got: {msg}"
                );
                // Must NOT show a bedrock or registry model when backend is nearai
                assert!(
                    !msg.contains("anthropic.claude"),
                    "should not show bedrock model for nearai backend: {msg}"
                );
            }
            other => panic!(
                "expected Pass for default LLM config, got: {}",
                format_result(&other)
            ),
        }
    }
    #[test]
    fn check_embeddings_disabled_by_default_returns_skip() {
        let _guard = crate::config::helpers::ENV_MUTEX.lock().expect("env mutex");
        // SAFETY: Under ENV_MUTEX.
        unsafe {
            std::env::remove_var("EMBEDDING_ENABLED");
        }
        let settings = Settings::default();
        match check_embeddings(&settings) {
            CheckResult::Skip(msg) => {
                assert!(
                    msg.contains("disabled"),
                    "expected 'disabled' in skip message, got: {msg}"
                );
            }
            other => panic!(
                "expected Skip for disabled embeddings, got: {}",
                format_result(&other)
            ),
        }
    }
    #[test]
    fn check_routines_enabled_by_default() {
        let _guard = crate::config::helpers::ENV_MUTEX.lock().expect("env mutex");
        // SAFETY: Under ENV_MUTEX.
        unsafe {
            std::env::remove_var("ROUTINES_ENABLED");
        }
        match check_routines_config() {
            CheckResult::Pass(msg) => {
                assert!(
                    msg.contains("enabled"),
                    "routines should be enabled by default, got: {msg}"
                );
            }
            other => panic!(
                "expected Pass for default routines, got: {}",
                format_result(&other)
            ),
        }
    }
    #[test]
    fn check_secrets_env_without_var_returns_fail() {
        let settings = Settings {
            secrets_master_key_source: crate::settings::KeySource::Env,
            ..Default::default()
        };
        match check_secrets(&settings) {
            CheckResult::Fail(msg) => {
                assert!(
                    msg.contains("SECRETS_MASTER_KEY not set"),
                    "expected mention of missing env var, got: {msg}"
                );
            }
            CheckResult::Pass(_) => {
                // If SECRETS_MASTER_KEY happens to be set in the environment,
                // Pass is correct — don't fail the test.
            }
            other => panic!(
                "expected Fail or Pass for env key source, got: {}",
                format_result(&other)
            ),
        }
    }
    fn format_result(r: &CheckResult) -> String {
        match r {
            CheckResult::Pass(s) => format!("Pass({s})"),
@@ -209,7 +209,6 @@ impl LlmConfig {
            extra_headers_env,
            api_key_required,
            base_url_required,
            unsupported_params,
        ) = if let Some(def) = def {
            (
                def.id.as_str(),
@@ -222,7 +221,6 @@ impl LlmConfig {
                def.extra_headers_env.as_deref(),
                def.api_key_required,
                def.base_url_required,
                def.unsupported_params.clone(),
            )
        } else {
            // Absolute fallback: treat as generic openai_completions
@@ -237,7 +235,6 @@ impl LlmConfig {
                Some("LLM_EXTRA_HEADERS"),
                false,
                true,
                Vec::new(),
            )
        };
@@ -341,7 +338,6 @@ impl LlmConfig {
            extra_headers,
            oauth_token,
            cache_retention,
            unsupported_params,
        })
    }
 }
@@ -629,12 +625,6 @@ mod tests {
        let provider = cfg.provider.expect("provider config should be present");
        assert_eq!(provider.base_url, "https://inference.tinfoil.sh/v1");
        assert_eq!(provider.model, "kimi-k2-5");
        assert!(
            provider
                .unsupported_params
                .contains(&"temperature".to_string()),
            "tinfoil should propagate unsupported_params from registry"
        );
    }
    #[test]
@@ -14,7 +14,6 @@ mod heartbeat;
 pub(crate) mod helpers;
 mod hygiene;
 pub(crate) mod llm;
 pub mod relay;
 mod routines;
 mod safety;
 mod sandbox;
@@ -39,7 +38,6 @@ pub use self::embeddings::EmbeddingsConfig;
 pub use self::heartbeat::HeartbeatConfig;
 pub use self::hygiene::HygieneConfig;
 pub use self::llm::default_session_path;
 pub use self::relay::RelayConfig;
 pub use self::routines::RoutineConfig;
 pub use self::safety::SafetyConfig;
 pub use self::sandbox::{ClaudeCodeConfig, SandboxModeConfig};
@@ -87,9 +85,6 @@ pub struct Config {
    pub skills: SkillsConfig,
    pub transcription: TranscriptionConfig,
    pub observability: crate::observability::ObservabilityConfig,
    /// Channel-relay integration (Slack via external relay service).
    /// Present only when both `CHANNEL_RELAY_URL` and `CHANNEL_RELAY_API_KEY` are set.
    pub relay: Option<RelayConfig>,
 }
 impl Config {
@@ -162,7 +157,6 @@ impl Config {
            },
            transcription: TranscriptionConfig::default(),
            observability: crate::observability::ObservabilityConfig::default(),
            relay: None,
        }
    }
@@ -316,7 +310,6 @@ impl Config {
            observability: crate::observability::ObservabilityConfig {
                backend: std::env::var("OBSERVABILITY_BACKEND").unwrap_or_else(|_| "none".into()),
            },
            relay: RelayConfig::from_env(),
        })
    }
 }
@@ -1,157 +0,0 @@
 //! Channel-relay service configuration.
 use secrecy::SecretString;
 /// Configuration for connecting to a channel-relay service.
 #[derive(Clone)]
 pub struct RelayConfig {
    /// Base URL of the channel-relay service (e.g., `http://localhost:3001`).
    pub url: String,
    /// API key for authenticated channel-relay endpoints.
    pub api_key: SecretString,
    /// Override for the OAuth callback URL (e.g., a tunnel URL).
    pub callback_url: Option<String>,
    /// Override for the instance identifier.
    pub instance_id: Option<String>,
    /// HTTP request timeout in seconds (default: 30).
    pub request_timeout_secs: u64,
    /// SSE stream long-poll timeout in seconds (default: 86400 = 24 h).
    pub stream_timeout_secs: u64,
    /// Initial exponential backoff in milliseconds (default: 1000).
    pub backoff_initial_ms: u64,
    /// Maximum exponential backoff in milliseconds (default: 60000).
    pub backoff_max_ms: u64,
 }
 impl std::fmt::Debug for RelayConfig {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("RelayConfig")
            .field("url", &self.url)
            .field("api_key", &"[REDACTED]")
            .field("callback_url", &self.callback_url)
            .field("instance_id", &self.instance_id)
            .field("request_timeout_secs", &self.request_timeout_secs)
            .field("stream_timeout_secs", &self.stream_timeout_secs)
            .field("backoff_initial_ms", &self.backoff_initial_ms)
            .field("backoff_max_ms", &self.backoff_max_ms)
            .finish()
    }
 }
 impl RelayConfig {
    /// Load relay config from environment variables.
    ///
    /// Returns `None` if either `CHANNEL_RELAY_URL` or `CHANNEL_RELAY_API_KEY`
    /// is not set, making the relay integration opt-in.
    pub fn from_env() -> Option<Self> {
        Self::from_env_reader(|key| std::env::var(key).ok())
    }
    /// Build a config for tests without touching the process environment.
    pub fn from_values(url: impl Into<String>, api_key: impl Into<String>) -> Self {
        Self {
            url: url.into(),
            api_key: SecretString::from(api_key.into()),
            callback_url: None,
            instance_id: None,
            request_timeout_secs: 30,
            stream_timeout_secs: 86400,
            backoff_initial_ms: 1000,
            backoff_max_ms: 60000,
        }
    }
    /// Internal constructor that reads values through a closure, enabling safe testing.
    fn from_env_reader(env: impl Fn(&str) -> Option<String>) -> Option<Self> {
        let url = env("CHANNEL_RELAY_URL")?;
        let api_key = SecretString::from(env("CHANNEL_RELAY_API_KEY")?);
        Some(Self {
            url,
            api_key,
            callback_url: env("IRONCLAW_OAUTH_CALLBACK_URL"),
            instance_id: env("IRONCLAW_INSTANCE_ID"),
            request_timeout_secs: env("RELAY_REQUEST_TIMEOUT_SECS")
                .and_then(|v| v.parse().ok())
                .unwrap_or(30),
            stream_timeout_secs: env("RELAY_STREAM_TIMEOUT_SECS")
                .and_then(|v| v.parse().ok())
                .unwrap_or(86400),
            backoff_initial_ms: env("RELAY_BACKOFF_INITIAL_MS")
                .and_then(|v| v.parse().ok())
                .unwrap_or(1000),
            backoff_max_ms: env("RELAY_BACKOFF_MAX_MS")
                .and_then(|v| v.parse().ok())
                .unwrap_or(60000),
        })
    }
 }
 #[cfg(test)]
 mod tests {
    use super::*;
    #[test]
    fn from_env_reader_returns_none_when_unset() {
        let config = RelayConfig::from_env_reader(|_| None);
        assert!(config.is_none());
    }
    #[test]
    fn from_env_reader_loads_defaults() {
        let config = RelayConfig::from_env_reader(|key| match key {
            "CHANNEL_RELAY_URL" => Some("http://localhost:3001".into()),
            "CHANNEL_RELAY_API_KEY" => Some("test-key".into()),
            _ => None,
        })
        .expect("config should be Some");
        assert_eq!(config.url, "http://localhost:3001");
        assert_eq!(config.request_timeout_secs, 30);
        assert_eq!(config.stream_timeout_secs, 86400);
        assert_eq!(config.backoff_initial_ms, 1000);
        assert_eq!(config.backoff_max_ms, 60000);
        assert!(config.callback_url.is_none());
        assert!(config.instance_id.is_none());
    }
    #[test]
    fn from_env_reader_loads_overrides() {
        let config = RelayConfig::from_env_reader(|key| match key {
            "CHANNEL_RELAY_URL" => Some("http://relay:3001".into()),
            "CHANNEL_RELAY_API_KEY" => Some("secret".into()),
            "IRONCLAW_OAUTH_CALLBACK_URL" => Some("https://tunnel.example.com".into()),
            "IRONCLAW_INSTANCE_ID" => Some("my-instance".into()),
            "RELAY_REQUEST_TIMEOUT_SECS" => Some("60".into()),
            "RELAY_STREAM_TIMEOUT_SECS" => Some("43200".into()),
            "RELAY_BACKOFF_INITIAL_MS" => Some("2000".into()),
            "RELAY_BACKOFF_MAX_MS" => Some("120000".into()),
            _ => None,
        })
        .expect("config should be Some");
        assert_eq!(
            config.callback_url.as_deref(),
            Some("https://tunnel.example.com")
        );
        assert_eq!(config.instance_id.as_deref(), Some("my-instance"));
        assert_eq!(config.request_timeout_secs, 60);
        assert_eq!(config.stream_timeout_secs, 43200);
        assert_eq!(config.backoff_initial_ms, 2000);
        assert_eq!(config.backoff_max_ms, 120000);
    }
    #[test]
    fn from_values_builds_with_defaults() {
        let config = RelayConfig::from_values("http://localhost:3001", "key");
        assert_eq!(config.url, "http://localhost:3001");
        assert_eq!(config.request_timeout_secs, 30);
    }
    #[test]
    fn debug_redacts_api_key() {
        let config = RelayConfig::from_values("http://localhost:3001", "super-secret");
        let debug = format!("{:?}", config);
        assert!(debug.contains("[REDACTED]"));
        assert!(!debug.contains("super-secret"));
    }
 }
@@ -30,9 +30,8 @@ impl JobStore for LibSqlBackend {
                    id, conversation_id, title, description, category, status, source,
                    user_id,
                    budget_amount, budget_token, bid_amount, estimated_cost, estimated_time_secs,
-                    actual_cost, repair_attempts, max_tokens, total_tokens_used,
+                    actual_cost, repair_attempts, created_at, started_at, completed_at
-                    created_at, started_at, completed_at
+                ) VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10, ?11, ?12, ?13, ?14, ?15, ?16, ?17, ?18)
                ) VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10, ?11, ?12, ?13, ?14, ?15, ?16, ?17, ?18, ?19, ?20)
                ON CONFLICT (id) DO UPDATE SET
                    title = excluded.title,
                    description = excluded.description,
@@ -43,8 +42,6 @@ impl JobStore for LibSqlBackend {
                    estimated_time_secs = excluded.estimated_time_secs,
                    actual_cost = excluded.actual_cost,
                    repair_attempts = excluded.repair_attempts,
                    max_tokens = excluded.max_tokens,
                    total_tokens_used = excluded.total_tokens_used,
                    started_at = excluded.started_at,
                    completed_at = excluded.completed_at
                "#,
@@ -64,8 +61,6 @@ impl JobStore for LibSqlBackend {
                    estimated_time_secs,
                    ctx.actual_cost.to_string(),
                    ctx.repair_attempts as i64,
                    ctx.max_tokens as i64,
                    ctx.total_tokens_used as i64,
                    fmt_ts(&ctx.created_at),
                    fmt_opt_ts(&ctx.started_at),
                    fmt_opt_ts(&ctx.completed_at),
@@ -83,8 +78,7 @@ impl JobStore for LibSqlBackend {
                r#"
                SELECT id, conversation_id, title, description, category, status, user_id,
                       budget_amount, budget_token, bid_amount, estimated_cost, estimated_time_secs,
-                       actual_cost, repair_attempts, max_tokens, total_tokens_used,
+                       actual_cost, repair_attempts, created_at, started_at, completed_at
                       created_at, started_at, completed_at
                FROM agent_jobs WHERE id = ?1
                "#,
                params![id.to_string()],
@@ -117,12 +111,12 @@ impl JobStore for LibSqlBackend {
                    estimated_duration: estimated_time_secs
                        .map(|s| std::time::Duration::from_secs(s as u64)),
                    actual_cost: get_decimal(&row, 12),
-                    max_tokens: get_i64(&row, 14) as u64,
+                    total_tokens_used: 0,
-                    total_tokens_used: get_i64(&row, 15) as u64,
+                    max_tokens: 0,
                    repair_attempts: get_i64(&row, 13) as u32,
-                    created_at: get_ts(&row, 16),
+                    created_at: get_ts(&row, 14),
-                    started_at: get_opt_ts(&row, 17),
+                    started_at: get_opt_ts(&row, 15),
-                    completed_at: get_opt_ts(&row, 18),
+                    completed_at: get_opt_ts(&row, 16),
                    transitions: Vec::new(),
                    metadata: serde_json::Value::Null,
                    extra_env: std::sync::Arc::new(std::collections::HashMap::new()),
@@ -167,7 +167,7 @@ impl RoutineStore for LibSqlBackend {
        let mut rows = conn
            .query(
                &format!(
-                    "SELECT {} FROM routines WHERE enabled = 1 AND trigger_type IN ('event', 'system_event')",
+                    "SELECT {} FROM routines WHERE enabled = 1 AND trigger_type = 'event'",
                    ROUTINE_COLUMNS
                ),
                (),
@@ -583,8 +583,7 @@ INSERT OR IGNORE INTO leak_detection_patterns (id, name, pattern, severity, acti
 ///
 /// Each entry is `(version, name, sql)`. Migrations are idempotent: the
 /// `_migrations` table tracks which versions have been applied.
-pub const INCREMENTAL_MIGRATIONS: &[(i64, &str, &str)] = &[
+pub const INCREMENTAL_MIGRATIONS: &[(i64, &str, &str)] = &[(
    (
    9,
    "flexible_embedding_dimension",
    // Rebuild memory_chunks to remove the fixed F32_BLOB(1536) type
@@ -645,18 +644,7 @@ CREATE TRIGGER IF NOT EXISTS memory_chunks_fts_update AFTER UPDATE ON memory_chu
    INSERT INTO memory_chunks_fts(rowid, content) VALUES (new._rowid, new.content);
 END;
 "#,
-    ),
+)];
    (
        12,
        "job_token_budget",
        // Add token budget tracking columns to agent_jobs.
        // SQLite supports ALTER TABLE ADD COLUMN, so no table rebuild needed.
        r#"
 ALTER TABLE agent_jobs ADD COLUMN max_tokens INTEGER NOT NULL DEFAULT 0;
 ALTER TABLE agent_jobs ADD COLUMN total_tokens_used INTEGER NOT NULL DEFAULT 0;
 "#,
    ),
 ];
 /// Run incremental migrations that haven't been applied yet.
 ///
@@ -665,7 +653,6 @@ ALTER TABLE agent_jobs ADD COLUMN total_tokens_used INTEGER NOT NULL DEFAULT 0;
 pub async fn run_incremental(conn: &libsql::Connection) -> Result<(), crate::error::DatabaseError> {
    use crate::error::DatabaseError;
    let mut applied_count = 0;
    for &(version, name, sql) in INCREMENTAL_MIGRATIONS {
        // Check if already applied
        let mut rows = conn
@@ -682,6 +669,8 @@ pub async fn run_incremental(conn: &libsql::Connection) -> Result<(), crate::err
            continue; // Already applied
        }
        tracing::info!(version, name, "libSQL: applying incremental migration");
        // Wrap migration + recording in a transaction for atomicity.
        // If the process crashes mid-migration, the transaction rolls back
        // and the migration will be retried on next startup.
@@ -713,12 +702,7 @@ pub async fn run_incremental(conn: &libsql::Connection) -> Result<(), crate::err
            ))
        })?;
-        applied_count += 1;
+        tracing::info!(version, name, "libSQL: migration applied successfully");
        tracing::debug!(version, name, "libSQL: migration applied");
    }
    if applied_count > 0 {
        tracing::info!("libSQL: applied {} incremental migrations", applied_count);
    }
    Ok(())
@@ -250,7 +250,6 @@ fn extract_source(source: &ExtensionSource) -> String {
        ExtensionSource::Discovered { url } => url.clone(),
        ExtensionSource::WasmDownload { wasm_url, .. } => wasm_url.clone(),
        ExtensionSource::WasmBuildable { source_dir, .. } => source_dir.clone(),
        ExtensionSource::ChannelRelay { relay_url } => relay_url.clone(),
    }
 }
@@ -84,8 +84,6 @@ pub struct ExtensionManager {
    // WASM channel hot-activation infrastructure (set post-construction)
    channel_runtime: RwLock<Option<ChannelRuntimeState>>,
    /// Channel manager for hot-adding relay channels (set independently of WASM runtime).
    relay_channel_manager: RwLock<Option<Arc<ChannelManager>>>,
    // Shared
    secrets: Arc<dyn SecretsStore + Send + Sync>,
@@ -99,8 +97,6 @@ pub struct ExtensionManager {
    store: Option<Arc<dyn crate::db::Database>>,
    /// Names of WASM channels that were successfully loaded at startup.
    active_channel_names: RwLock<HashSet<String>>,
    /// Installed channel-relay extensions (no on-disk artifact, tracked in memory).
    installed_relay_extensions: RwLock<HashSet<String>>,
    /// Last activation error for each WASM channel (ephemeral, cleared on success).
    activation_errors: RwLock<HashMap<String, String>>,
    /// SSE broadcast sender (set post-construction via `set_sse_sender()`).
@@ -115,34 +111,6 @@ pub struct ExtensionManager {
    /// Gateway auth token for authenticating with the platform token exchange proxy.
    /// Read once at construction from `GATEWAY_AUTH_TOKEN` env var.
    gateway_token: Option<String>,
    /// Relay config captured at startup. Used by `auth_channel_relay` and
    /// `activate_channel_relay` instead of re-reading env vars.
    relay_config: Option<crate::config::RelayConfig>,
 }
 /// Sanitize a URL for logging by removing query parameters and credentials.
 /// Prevents accidental logging of API keys, OAuth tokens, or other sensitive data in URLs.
 fn sanitize_url_for_logging(url: &str) -> String {
    // If URL is very short or doesn't look like a URL, just use as-is
    if url.len() < 10 || !url.contains("://") {
        return url.to_string();
    }
    // Try to parse and remove sensitive components
    if let Ok(mut parsed) = url::Url::parse(url) {
        // Remove query string and fragment
        parsed.set_query(None);
        parsed.set_fragment(None);
        // Remove userinfo (username and password) if present
        let _ = parsed.set_username("");
        let _ = parsed.set_password(None);
        parsed.to_string()
    } else {
        // Fallback: strip after ? or #
        url.split(['?', '#']).next().unwrap_or(url).to_string()
    }
 }
 impl ExtensionManager {
@@ -176,7 +144,6 @@ impl ExtensionManager {
            wasm_tools_dir,
            wasm_channels_dir,
            channel_runtime: RwLock::new(None),
            relay_channel_manager: RwLock::new(None),
            secrets,
            tool_registry,
            hooks,
@@ -185,24 +152,13 @@ impl ExtensionManager {
            user_id,
            store,
            active_channel_names: RwLock::new(HashSet::new()),
            installed_relay_extensions: RwLock::new(HashSet::new()),
            activation_errors: RwLock::new(HashMap::new()),
            sse_sender: RwLock::new(None),
            pending_oauth_flows: crate::cli::oauth_defaults::new_pending_oauth_registry(),
            gateway_token: std::env::var("GATEWAY_AUTH_TOKEN").ok(),
            relay_config: crate::config::RelayConfig::from_env(),
        }
    }
    /// Get the relay config stored at startup.
    fn relay_config(&self) -> Result<&crate::config::RelayConfig, ExtensionError> {
        self.relay_config.as_ref().ok_or_else(|| {
            ExtensionError::Config(
                "CHANNEL_RELAY_URL and CHANNEL_RELAY_API_KEY must be set".to_string(),
            )
        })
    }
    /// Configure the channel runtime infrastructure for hot-activating WASM channels.
    ///
    /// Call after construction (and after wrapping in `Arc`) once the channel
@@ -216,8 +172,6 @@ impl ExtensionManager {
        wasm_channel_router: Arc<WasmChannelRouter>,
        wasm_channel_owner_ids: std::collections::HashMap<String, i64>,
    ) {
        // Also store the channel manager for relay channel activation.
        *self.relay_channel_manager.write().await = Some(Arc::clone(&channel_manager));
        *self.channel_runtime.write().await = Some(ChannelRuntimeState {
            channel_manager,
            wasm_channel_runtime,
@@ -227,58 +181,6 @@ impl ExtensionManager {
        });
    }
    /// Set just the channel manager for relay channel hot-activation.
    ///
    /// Call this when WASM channel runtime is not available but relay channels
    /// still need to be hot-added.
    pub async fn set_relay_channel_manager(&self, channel_manager: Arc<ChannelManager>) {
        *self.relay_channel_manager.write().await = Some(channel_manager);
    }
    /// Check if a channel name corresponds to a relay extension (has stored stream token).
    pub async fn is_relay_channel(&self, name: &str) -> bool {
        self.secrets
            .exists(&self.user_id, &format!("relay:{}:stream_token", name))
            .await
            .unwrap_or(false)
    }
    /// Restore persisted relay channels after startup.
    ///
    /// Loads the persisted active channel list, filters to relay types (those with
    /// a stored stream token), and activates each via `activate_stored_relay()`.
    /// Skips channels that are already active. Call this after `set_relay_channel_manager()`.
    pub async fn restore_relay_channels(&self) {
        let persisted = self.load_persisted_active_channels().await;
        let already_active = self.active_channel_names.read().await.clone();
        for name in &persisted {
            if already_active.contains(name) {
                continue;
            }
            if !self.is_relay_channel(name).await {
                continue;
            }
            match self.activate_stored_relay(name).await {
                Ok(_) => {
                    tracing::debug!(channel = %name, "Restored persisted relay channel");
                }
                Err(e) => {
                    tracing::warn!(
                        channel = %name,
                        error = %e,
                        "Failed to restore persisted relay channel"
                    );
                }
            }
        }
    }
    /// Access the secrets store (used by OAuth callback handlers).
    pub fn secrets(&self) -> &Arc<dyn SecretsStore + Send + Sync> {
        &self.secrets
    }
    /// Register channel names that were loaded at startup.
    /// Called after WASM channels are loaded so `list()` reports accurate active status.
    pub async fn set_active_channels(&self, names: Vec<String>) {
@@ -397,8 +299,7 @@ impl ExtensionManager {
        url: Option<&str>,
        kind_hint: Option<ExtensionKind>,
    ) -> Result<InstallResult, ExtensionError> {
-        let sanitized_url = url.map(sanitize_url_for_logging);
+        tracing::info!(extension = %name, url = ?url, kind = ?kind_hint, "Installing extension");
        tracing::info!(extension = %name, url = ?sanitized_url, kind = ?kind_hint, "Installing extension");
        Self::validate_extension_name(name)?;
        // If we have a registry entry, use it (prefer kind_hint to resolve collisions)
@@ -418,16 +319,9 @@ impl ExtensionManager {
                ExtensionKind::WasmChannel => {
                    self.install_wasm_channel_from_url(name, url, None).await
                }
                ExtensionKind::ChannelRelay => {
                    // ChannelRelay extensions are installed from registry, not by URL
                    Err(ExtensionError::InstallFailed(
                        "Channel relay extensions cannot be installed by URL".to_string(),
                    ))
                }
            }
            .map_err(|e| {
-                let sanitized = sanitize_url_for_logging(url);
+                tracing::error!(extension = %name, url = %url, error = %e, "Extension install from URL failed");
                tracing::error!(extension = %name, url = %sanitized, error = %e, "Extension install from URL failed");
                e
            });
        }
@@ -456,7 +350,6 @@ impl ExtensionManager {
            ExtensionKind::McpServer => self.auth_mcp(name, token).await,
            ExtensionKind::WasmTool => self.auth_wasm_tool(name, token).await,
            ExtensionKind::WasmChannel => self.auth_wasm_channel(name, token).await,
            ExtensionKind::ChannelRelay => self.auth_channel_relay(name, token).await,
        }
    }
@@ -469,7 +362,6 @@ impl ExtensionManager {
            ExtensionKind::McpServer => self.activate_mcp(name).await,
            ExtensionKind::WasmTool => self.activate_wasm_tool(name).await,
            ExtensionKind::WasmChannel => self.activate_wasm_channel(name).await,
            ExtensionKind::ChannelRelay => self.activate_channel_relay(name).await,
        }
    }
@@ -641,41 +533,6 @@ impl ExtensionManager {
            }
        }
        // List channel-relay extensions
        if kind_filter.is_none() || kind_filter == Some(ExtensionKind::ChannelRelay) {
            let installed = self.installed_relay_extensions.read().await;
            let active_names = self.active_channel_names.read().await;
            for name in installed.iter() {
                let active = active_names.contains(name);
                let has_token = self
                    .secrets
                    .exists(&self.user_id, &format!("relay:{}:stream_token", name))
                    .await
                    .unwrap_or(false);
                let registry_entry = self
                    .registry
                    .get_with_kind(name, Some(ExtensionKind::ChannelRelay))
                    .await;
                let display_name = registry_entry.as_ref().map(|e| e.display_name.clone());
                let description = registry_entry.as_ref().map(|e| e.description.clone());
                extensions.push(InstalledExtension {
                    name: name.clone(),
                    kind: ExtensionKind::ChannelRelay,
                    display_name,
                    description,
                    url: None,
                    authenticated: has_token,
                    active,
                    tools: Vec::new(),
                    needs_setup: false,
                    has_auth: true,
                    installed: true,
                    activation_error: None,
                    version: None,
                });
            }
        }
        // Append available-but-not-installed registry entries
        if include_available {
            let installed_names: std::collections::HashSet<(String, ExtensionKind)> = extensions
@@ -814,37 +671,6 @@ impl ExtensionManager {
                    name
                ))
            }
            ExtensionKind::ChannelRelay => {
                // Remove from installed set
                self.installed_relay_extensions.write().await.remove(name);
                // Remove from active channels
                self.active_channel_names.write().await.remove(name);
                self.persist_active_channels().await;
                // Remove stored stream token
                let _ = self
                    .secrets
                    .delete(&self.user_id, &format!("relay:{}:stream_token", name))
                    .await;
                // Shut down the channel (check both runtime paths for WASM+relay and relay-only modes)
                let mut shut_down = false;
                if let Some(ref rt) = *self.channel_runtime.read().await
                    && let Some(channel) = rt.channel_manager.get_channel(name).await
                {
                    let _ = channel.shutdown().await;
                    shut_down = true;
                }
                if !shut_down
                    && let Some(ref cm) = *self.relay_channel_manager.read().await
                    && let Some(channel) = cm.get_channel(name).await
                {
                    let _ = channel.shutdown().await;
                }
                Ok(format!("Removed channel relay '{}'", name))
            }
        }
    }
@@ -932,12 +758,12 @@ impl ExtensionManager {
                &self.wasm_channels_dir,
                crate::tools::wasm::WIT_CHANNEL_VERSION,
            ),
-            ExtensionKind::McpServer | ExtensionKind::ChannelRelay => {
+            ExtensionKind::McpServer => {
                return UpgradeOutcome {
                    name: name.to_string(),
                    kind,
                    status: "failed".to_string(),
-                    detail: "This extension type cannot be upgraded this way".to_string(),
+                    detail: "MCP servers cannot be upgraded this way".to_string(),
                };
            }
        };
@@ -958,7 +784,7 @@ impl ExtensionManager {
                                .ok()
                                .and_then(|c| c.wit_version)
                        }
-                        ExtensionKind::McpServer | ExtensionKind::ChannelRelay => None,
+                        ExtensionKind::McpServer => None,
                    };
                    wit
                }
@@ -1118,14 +944,6 @@ impl ExtensionManager {
                });
                Ok(info)
            }
            ExtensionKind::ChannelRelay => {
                let info = serde_json::json!({
                    "name": name,
                    "kind": "channel_relay",
                    "active": self.active_channel_names.read().await.contains(name),
                });
                Ok(info)
            }
        }
    }
@@ -1290,21 +1108,6 @@ impl ExtensionManager {
                    "WASM channel entry has no download URL or build info".to_string(),
                )),
            },
            ExtensionKind::ChannelRelay => {
                // No download needed — just mark as installed.
                self.installed_relay_extensions
                    .write()
                    .await
                    .insert(entry.name.clone());
                Ok(InstallResult {
                    name: entry.name.clone(),
                    kind: ExtensionKind::ChannelRelay,
                    message: format!(
                        "'{}' installed. Click Activate to connect your workspace.",
                        entry.display_name
                    ),
                })
            }
        }
    }
@@ -1409,11 +1212,10 @@ impl ExtensionManager {
            .build()
            .map_err(|e| ExtensionError::DownloadFailed(e.to_string()))?;
-        let sanitized_url = sanitize_url_for_logging(url);
+        tracing::debug!(extension = %name, url = %url, "Downloading WASM extension");
        tracing::debug!(extension = %name, url = %sanitized_url, "Downloading WASM extension");
        let response = client.get(url).send().await.map_err(|e| {
-            tracing::error!(extension = %name, url = %sanitized_url, error = %e, "Download request failed");
+            tracing::error!(extension = %name, url = %url, error = %e, "Download request failed");
            ExtensionError::DownloadFailed(e.to_string())
        })?;
@@ -1421,7 +1223,7 @@ impl ExtensionManager {
            let status = response.status();
            tracing::error!(
                extension = %name,
-                url = %sanitized_url,
+                url = %url,
                status = %status,
                "Download returned non-success HTTP status"
            );
@@ -1664,7 +1466,6 @@ impl ExtensionManager {
            ExtensionKind::WasmTool => "WASM tool",
            ExtensionKind::WasmChannel => "WASM channel",
            ExtensionKind::McpServer => "MCP server",
            ExtensionKind::ChannelRelay => "channel relay",
        };
        tracing::info!(
@@ -2974,9 +2775,9 @@ impl ExtensionManager {
        }
        // Inject credentials
-        match inject_channel_credentials_from_secrets(
+        match crate::extensions::manager::inject_channel_credentials_from_secrets(
            &channel_arc,
-            Some(self.secrets.as_ref()),
+            self.secrets.as_ref(),
            &channel_name,
            &self.user_id,
        )
@@ -3061,7 +2862,7 @@ impl ExtensionManager {
        // Re-inject credentials from secrets store into the running channel
        let cred_count = match inject_channel_credentials_from_secrets(
            &existing_channel,
-            Some(self.secrets.as_ref()),
+            self.secrets.as_ref(),
            name,
            &self.user_id,
        )
@@ -3204,192 +3005,7 @@ impl ExtensionManager {
        })
    }
    // ── Channel-relay extension methods ──────────────────────────────────
    /// Derive a stable instance ID from the relay config and user_id.
    fn relay_instance_id(&self, config: &crate::config::RelayConfig) -> String {
        config.instance_id.clone().unwrap_or_else(|| {
            uuid::Uuid::new_v5(&uuid::Uuid::NAMESPACE_DNS, self.user_id.as_bytes()).to_string()
        })
    }
    /// Authenticate a channel-relay extension.
    ///
    /// For Slack: initiates OAuth flow (redirect-based).
    /// For Telegram: accepts a bot token, registers it with channel-relay,
    /// and stores the returned stream token.
    async fn auth_channel_relay(
        &self,
        name: &str,
        _token: Option<&str>,
    ) -> Result<AuthResult, ExtensionError> {
        // Check if already authenticated (stream token exists)
        let token_key = format!("relay:{}:stream_token", name);
        if self
            .secrets
            .exists(&self.user_id, &token_key)
            .await
            .unwrap_or(false)
        {
            return Ok(AuthResult::authenticated(name, ExtensionKind::ChannelRelay));
        }
        // Use relay config captured at startup
        let relay_config = self.relay_config()?;
        let instance_id = self.relay_instance_id(relay_config);
        let user_id_uuid = std::env::var("IRONCLAW_USER_ID").unwrap_or_else(|_| {
            uuid::Uuid::new_v5(&uuid::Uuid::NAMESPACE_DNS, self.user_id.as_bytes()).to_string()
        });
        let client = crate::channels::relay::RelayClient::new(
            relay_config.url.clone(),
            relay_config.api_key.clone(),
            relay_config.request_timeout_secs,
        )
        .map_err(|e| ExtensionError::Config(e.to_string()))?;
        // OAuth redirect flow
        let callback_base = self
            .tunnel_url
            .clone()
            .or_else(|| relay_config.callback_url.clone())
            .unwrap_or_else(|| {
                let host = std::env::var("GATEWAY_HOST").unwrap_or_else(|_| "127.0.0.1".into());
                let port = std::env::var("GATEWAY_PORT").unwrap_or_else(|_| "3001".into());
                format!("http://{}:{}", host, port)
            });
        // Generate CSRF nonce for OAuth state parameter
        let state_nonce = uuid::Uuid::new_v4().to_string();
        let state_key = format!("relay:{}:oauth_state", name);
        // Delete any stale nonce before storing the new one
        let _ = self.secrets.delete(&self.user_id, &state_key).await;
        self.secrets
            .create(
                &self.user_id,
                CreateSecretParams::new(&state_key, &state_nonce),
            )
            .await
            .map_err(|e| ExtensionError::AuthFailed(format!("Failed to store OAuth state: {e}")))?;
        let callback_url = format!(
            "{}/oauth/slack/callback?state={}",
            callback_base, state_nonce
        );
        match client
            .initiate_oauth(&instance_id, &user_id_uuid, &callback_url)
            .await
        {
            Ok(auth_url) => Ok(AuthResult::awaiting_authorization(
                name,
                ExtensionKind::ChannelRelay,
                auth_url,
                "redirect".to_string(),
            )),
            Err(e) => Err(ExtensionError::AuthFailed(e.to_string())),
        }
    }
    /// Activate a channel-relay extension.
    async fn activate_channel_relay(&self, name: &str) -> Result<ActivateResult, ExtensionError> {
        let token_key = format!("relay:{}:stream_token", name);
        let team_id_key = format!("relay:{}:team_id", name);
        // Check if we have a stream token
        let stream_token = match self.secrets.get_decrypted(&self.user_id, &token_key).await {
            Ok(secret) => secret.expose().to_string(),
            Err(_) => {
                return Err(ExtensionError::AuthRequired);
            }
        };
        // Get team_id from settings
        let team_id = if let Some(ref store) = self.store {
            store
                .get_setting(&self.user_id, &team_id_key)
                .await
                .ok()
                .flatten()
                .and_then(|v| v.as_str().map(|s| s.to_string()))
                .unwrap_or_default()
        } else {
            String::new()
        };
        // Use relay config captured at startup
        let relay_config = self.relay_config()?;
        let instance_id = self.relay_instance_id(relay_config);
        let client = crate::channels::relay::RelayClient::new(
            relay_config.url.clone(),
            relay_config.api_key.clone(),
            relay_config.request_timeout_secs,
        )
        .map_err(|e| ExtensionError::ActivationFailed(e.to_string()))?;
        let channel = crate::channels::relay::RelayChannel::new_with_provider(
            client,
            crate::channels::relay::channel::RelayProvider::Slack,
            stream_token,
            team_id,
            instance_id,
            self.user_id.clone(),
        )
        .with_timeouts(
            relay_config.stream_timeout_secs,
            relay_config.backoff_initial_ms,
            relay_config.backoff_max_ms,
        );
        // Hot-add to channel manager
        let cm_guard = self.relay_channel_manager.read().await;
        let channel_mgr = cm_guard.as_ref().ok_or_else(|| {
            ExtensionError::ActivationFailed("Channel manager not initialized".to_string())
        })?;
        channel_mgr
            .hot_add(Box::new(channel))
            .await
            .map_err(|e| ExtensionError::ActivationFailed(e.to_string()))?;
        // Mark as active
        self.active_channel_names
            .write()
            .await
            .insert(name.to_string());
        self.persist_active_channels().await;
        // Broadcast status
        let status_msg = "Slack connected via channel relay".to_string();
        self.broadcast_extension_status(name, "active", Some(&status_msg))
            .await;
        Ok(ActivateResult {
            name: name.to_string(),
            kind: ExtensionKind::ChannelRelay,
            tools_loaded: Vec::new(),
            message: status_msg,
        })
    }
    /// Activate a channel-relay extension from stored credentials (for startup reconnect).
    pub async fn activate_stored_relay(&self, name: &str) -> Result<(), ExtensionError> {
        self.installed_relay_extensions
            .write()
            .await
            .insert(name.to_string());
        self.activate_channel_relay(name).await?;
        Ok(())
    }
    /// Determine what kind of installed extension this is.
    ///
    /// This is a read-only check — it never modifies `installed_relay_extensions`.
    /// To mark a relay extension as installed, use `activate_stored_relay()` or
    /// the explicit install flow.
    async fn determine_installed_kind(&self, name: &str) -> Result<ExtensionKind, ExtensionError> {
        // Check MCP servers first
        if self.get_mcp_server(name).await.is_ok() {
@@ -3408,22 +3024,8 @@ impl ExtensionManager {
            return Ok(ExtensionKind::WasmChannel);
        }
        // Check channel-relay extensions (installed in memory or has stored token)
        if self.installed_relay_extensions.read().await.contains(name) {
            return Ok(ExtensionKind::ChannelRelay);
        }
        // Also check if there's a stored stream token (persisted across restarts)
        if self
            .secrets
            .exists(&self.user_id, &format!("relay:{}:stream_token", name))
            .await
            .unwrap_or(false)
        {
            return Ok(ExtensionKind::ChannelRelay);
        }
        Err(ExtensionError::NotInstalled(format!(
-            "'{}' is not installed as an MCP server, WASM tool, WASM channel, or channel relay",
+            "'{}' is not installed as an MCP server, WASM tool, or WASM channel",
            name
        )))
    }
@@ -3839,30 +3441,23 @@ impl ExtensionManager {
 /// Looks for secrets matching the pattern `{channel_name}_*` and injects them
 /// as credential placeholders (e.g., `telegram_bot_token` -> `{TELEGRAM_BOT_TOKEN}`).
 ///
 /// Falls back to environment variables starting with the uppercase channel name
 /// prefix (e.g., `TELEGRAM_` for channel `telegram`) for missing credentials.
 ///
 /// Returns the number of credentials injected.
 async fn inject_channel_credentials_from_secrets(
    channel: &Arc<crate::channels::wasm::WasmChannel>,
-    secrets: Option<&dyn SecretsStore>,
+    secrets: &dyn SecretsStore,
    channel_name: &str,
    user_id: &str,
 ) -> Result<usize, String> {
    let mut count = 0;
    let mut injected_placeholders = std::collections::HashSet::new();
    // 1. Try injecting from persistent secrets store if available
    if let Some(secrets) = secrets {
    let all_secrets = secrets
        .list(user_id)
        .await
        .map_err(|e| format!("Failed to list secrets: {}", e))?;
-        let prefix = format!("{}_", channel_name.to_ascii_lowercase());
+    let prefix = format!("{}_", channel_name);
    let mut count = 0;
    for secret_meta in all_secrets {
-            if !secret_meta.name.to_ascii_lowercase().starts_with(&prefix) {
+        if !secret_meta.name.starts_with(&prefix) {
            continue;
        }
@@ -3882,88 +3477,12 @@ async fn inject_channel_credentials_from_secrets(
        channel
            .set_credential(&placeholder, decrypted.expose().to_string())
            .await;
            injected_placeholders.insert(placeholder);
        count += 1;
    }
    }
    // 2. Fallback to environment variables for missing credentials
    count += inject_env_credentials(channel, channel_name, &injected_placeholders).await;
    Ok(count)
 }
 /// Inject missing credentials from environment variables.
 ///
 /// Only environment variables starting with the uppercase channel name prefix
 /// (e.g., `TELEGRAM_` for channel `telegram`) are considered for security.
 async fn inject_env_credentials(
    channel: &Arc<crate::channels::wasm::WasmChannel>,
    channel_name: &str,
    already_injected: &std::collections::HashSet<String>,
 ) -> usize {
    if channel_name.trim().is_empty() {
        return 0;
    }
    let caps = channel.capabilities();
    let Some(ref http_cap) = caps.tool_capabilities.http else {
        return 0;
    };
    let placeholders: Vec<String> = http_cap
        .credentials
        .values()
        .map(|m| m.secret_name.to_uppercase())
        .collect();
    let resolved = resolve_env_credentials(&placeholders, channel_name, already_injected);
    let count = resolved.len();
    for (placeholder, value) in resolved {
        channel.set_credential(&placeholder, value).await;
    }
    count
 }
 /// Pure helper: from a list of credential placeholder names, return those that
 /// pass the channel-prefix security check and have a non-empty env var value.
 ///
 /// Placeholders already covered by the secrets store (`already_injected`) are
 /// skipped. Only names starting with `{CHANNEL_NAME}_` are allowed to prevent
 /// a WASM channel from reading unrelated host credentials (e.g. `AWS_SECRET_ACCESS_KEY`).
 pub(crate) fn resolve_env_credentials(
    placeholders: &[String],
    channel_name: &str,
    already_injected: &std::collections::HashSet<String>,
 ) -> Vec<(String, String)> {
    if channel_name.trim().is_empty() {
        return Vec::new();
    }
    let prefix = format!("{}_", channel_name.to_ascii_uppercase());
    let mut out = Vec::new();
    for placeholder in placeholders {
        if already_injected.contains(placeholder) {
            continue;
        }
        if !placeholder.starts_with(&prefix) {
            tracing::warn!(
                channel = %channel_name,
                placeholder = %placeholder,
                "Ignoring non-prefixed credential placeholder in environment fallback"
            );
            continue;
        }
        if let Ok(value) = std::env::var(placeholder)
            && !value.is_empty()
        {
            out.push((placeholder.clone(), value));
        }
    }
    out
 }
 /// Infer the extension kind from a URL.
 fn infer_kind_from_url(url: &str) -> ExtensionKind {
    if url.ends_with(".wasm") || url.ends_with(".tar.gz") {
@@ -4414,276 +3933,4 @@ mod tests {
            Vec::new(),
        )
    }
    // ── resolve_env_credentials tests ────────────────────────────────────
    #[test]
    fn test_security_prefix_check() {
        // Placeholders that don't start with the channel prefix must be rejected.
        // All env var names are prefixed with ICTEST1_ to avoid CI collisions.
        let placeholders = vec![
            "ICTEST1_BOT_TOKEN".to_string(), // valid: matches channel prefix
            "ICTEST2_TOKEN".to_string(),     // invalid: wrong channel prefix
            "ICTEST1_UNRELATED_OTHER".to_string(), // valid prefix, but env var not set — not injected
        ];
        let already_injected = std::collections::HashSet::new();
        unsafe { std::env::set_var("ICTEST1_BOT_TOKEN", "good-secret") };
        unsafe { std::env::set_var("ICTEST2_TOKEN", "bad-secret") };
        // ICTEST1_UNRELATED_OTHER intentionally not set — tests both prefix rejection and absence
        let resolved = super::resolve_env_credentials(&placeholders, "ictest1", &already_injected);
        // Only ICTEST1_BOT_TOKEN passes the prefix check for channel "ictest1"
        assert_eq!(resolved.len(), 1);
        assert_eq!(resolved[0].0, "ICTEST1_BOT_TOKEN");
        assert_eq!(resolved[0].1, "good-secret");
        unsafe { std::env::remove_var("ICTEST1_BOT_TOKEN") };
        unsafe { std::env::remove_var("ICTEST2_TOKEN") };
    }
    #[test]
    fn test_already_injected_skipped() {
        // Use unique env var names (ictest3_*) to avoid interference with other tests.
        let placeholders = vec!["ICTEST3_TOKEN".to_string()];
        let mut already_injected = std::collections::HashSet::new();
        already_injected.insert("ICTEST3_TOKEN".to_string());
        unsafe { std::env::set_var("ICTEST3_TOKEN", "secret") };
        let resolved = super::resolve_env_credentials(&placeholders, "ictest3", &already_injected);
        // Already covered by secrets store — env var must be skipped
        assert!(resolved.is_empty());
        unsafe { std::env::remove_var("ICTEST3_TOKEN") };
    }
    #[test]
    fn test_missing_env_var_not_injected() {
        // Use unique env var names (ictest4_*) to avoid interference with other tests.
        let placeholders = vec!["ICTEST4_TOKEN".to_string()];
        let already_injected = std::collections::HashSet::new();
        unsafe { std::env::remove_var("ICTEST4_TOKEN") };
        let resolved = super::resolve_env_credentials(&placeholders, "ictest4", &already_injected);
        assert!(resolved.is_empty());
    }
    #[test]
    fn test_empty_env_var_not_injected() {
        // An env var that exists but is empty must not be injected.
        // Use unique env var names (ictest5_*) to avoid interference with other tests.
        let placeholders = vec!["ICTEST5_TOKEN".to_string()];
        let already_injected = std::collections::HashSet::new();
        unsafe { std::env::set_var("ICTEST5_TOKEN", "") };
        let resolved = super::resolve_env_credentials(&placeholders, "ictest5", &already_injected);
        assert!(resolved.is_empty());
        unsafe { std::env::remove_var("ICTEST5_TOKEN") };
    }
    #[test]
    fn test_empty_channel_name_returns_nothing() {
        // An empty channel name must never match any env var (prefix would be "_").
        let placeholders = vec!["_TOKEN".to_string(), "ICTEST6_TOKEN".to_string()];
        let already_injected = std::collections::HashSet::new();
        unsafe { std::env::set_var("_TOKEN", "bad") };
        unsafe { std::env::set_var("ICTEST6_TOKEN", "bad") };
        let resolved = super::resolve_env_credentials(&placeholders, "", &already_injected);
        assert!(resolved.is_empty(), "empty channel name must match nothing");
        unsafe { std::env::remove_var("_TOKEN") };
        unsafe { std::env::remove_var("ICTEST6_TOKEN") };
    }
    #[tokio::test]
    async fn test_determine_installed_kind_does_not_auto_install_relay() {
        // Regression: determine_installed_kind used to auto-insert into
        // installed_relay_extensions when a ChannelRelay registry entry existed,
        // even though the user never installed it. It should be read-only.
        let dir = tempfile::tempdir().expect("temp dir");
        let mgr = make_test_manager(None, dir.path().to_path_buf());
        // The manager has no relay extensions installed
        assert!(
            mgr.installed_relay_extensions.read().await.is_empty(),
            "Should start with no installed relay extensions"
        );
        // Calling determine_installed_kind for a non-installed name returns NotInstalled
        let result = mgr.determine_installed_kind("slack-relay").await;
        assert!(result.is_err(), "Should return NotInstalled");
        // Crucially: installed_relay_extensions must still be empty
        assert!(
            mgr.installed_relay_extensions.read().await.is_empty(),
            "determine_installed_kind must not modify installed_relay_extensions"
        );
    }
    #[tokio::test]
    async fn test_is_relay_channel_detects_stored_token() {
        let dir = tempfile::tempdir().expect("temp dir");
        let mgr = make_test_manager(None, dir.path().to_path_buf());
        // No token stored → not a relay channel
        assert!(!mgr.is_relay_channel("slack-relay").await);
        // Store a stream token
        mgr.secrets
            .create(
                "test",
                crate::secrets::CreateSecretParams::new("relay:slack-relay:stream_token", "tok123"),
            )
            .await
            .expect("store token");
        // Now it's detected as a relay channel
        assert!(mgr.is_relay_channel("slack-relay").await);
    }
    #[tokio::test]
    async fn test_remove_relay_shuts_down_via_relay_channel_manager() {
        // Regression: remove() only checked channel_runtime for shutdown, missing
        // relay-only mode where only relay_channel_manager is set.
        let dir = tempfile::tempdir().expect("temp dir");
        let mgr = make_test_manager(None, dir.path().to_path_buf());
        // Set up relay channel manager with a stub channel
        let cm = Arc::new(crate::channels::ChannelManager::new());
        let (stub, _tx) = crate::testing::StubChannel::new("slack-relay");
        cm.add(Box::new(stub)).await;
        mgr.set_relay_channel_manager(Arc::clone(&cm)).await;
        // Mark as installed + store a token so determine_installed_kind finds it
        mgr.installed_relay_extensions
            .write()
            .await
            .insert("slack-relay".to_string());
        mgr.secrets
            .create(
                "test",
                crate::secrets::CreateSecretParams::new("relay:slack-relay:stream_token", "tok123"),
            )
            .await
            .expect("store token");
        // Verify channel exists before removal
        assert!(cm.get_channel("slack-relay").await.is_some());
        // Remove should succeed and shut down the channel
        let result = mgr.remove("slack-relay").await;
        assert!(result.is_ok(), "remove should succeed: {:?}", result.err());
        // installed_relay_extensions should be cleared
        assert!(
            !mgr.installed_relay_extensions
                .read()
                .await
                .contains("slack-relay"),
            "Should be removed from installed set"
        );
    }
    #[test]
    fn test_sanitize_url_with_query_params() {
        let url = "https://api.example.com/path?api_key=secret123&token=abc";
        let result = super::sanitize_url_for_logging(url);
        assert_eq!(result, "https://api.example.com/path");
        assert!(!result.contains("api_key"));
        assert!(!result.contains("secret123"));
        assert!(!result.contains("token"));
    }
    #[test]
    fn test_sanitize_url_with_credentials() {
        let url = "https://user:[email protected]:8080/path";
        let result = super::sanitize_url_for_logging(url);
        assert!(!result.contains("user"));
        assert!(!result.contains("password"));
        assert!(!result.contains("@"));
        assert!(result.contains("api.example.com"));
        assert!(result.contains(":8080"));
    }
    #[test]
    fn test_sanitize_url_with_fragment() {
        let url = "https://api.example.com/path#section";
        let result = super::sanitize_url_for_logging(url);
        assert_eq!(result, "https://api.example.com/path");
        assert!(!result.contains("#"));
        assert!(!result.contains("section"));
    }
    #[test]
    fn test_sanitize_url_with_port() {
        let url = "https://api.example.com:9443/path?key=value";
        let result = super::sanitize_url_for_logging(url);
        assert_eq!(result, "https://api.example.com:9443/path");
        assert!(result.contains(":9443"));
        assert!(!result.contains("key"));
    }
    #[test]
    fn test_sanitize_url_with_all_components() {
        let url = "https://admin:[email protected]:8080/v1/data?api_key=xyz#results";
        let result = super::sanitize_url_for_logging(url);
        assert!(!result.contains("admin"));
        assert!(!result.contains("secret"));
        assert!(!result.contains("@"));
        assert!(!result.contains("api_key"));
        assert!(!result.contains("xyz"));
        assert!(!result.contains("#"));
        assert!(!result.contains("results"));
        assert!(result.contains("api.example.com:8080"));
        assert!(result.contains("/v1/data"));
    }
    #[test]
    fn test_sanitize_url_malformed() {
        // Malformed URL should fallback to string splitting
        let url = "https://[invalid-url";
        let result = super::sanitize_url_for_logging(url);
        // Malformed URL without query should return as-is via fallback
        assert_eq!(result, url);
        // Should still strip query params via fallback
        let url_with_query = "https://[invalid-url?key=secret";
        let result_with_query = super::sanitize_url_for_logging(url_with_query);
        assert_eq!(result_with_query, "https://[invalid-url");
        assert!(!result_with_query.contains("?"));
        assert!(!result_with_query.contains("secret"));
    }
    #[test]
    fn test_sanitize_url_short_string() {
        let url = "short";
        let result = super::sanitize_url_for_logging(url);
        assert_eq!(result, "short");
    }
    #[test]
    fn test_sanitize_url_not_url_like() {
        let input = "this is not a url";
        let result = super::sanitize_url_for_logging(input);
        assert_eq!(result, input);
    }
    #[test]
    fn test_sanitize_url_preserves_path() {
        let url = "https://api.example.com/v1/users/123/profile";
        let result = super::sanitize_url_for_logging(url);
        assert_eq!(result, url);
        assert!(result.contains("/v1/users/123/profile"));
    }
 }
@@ -37,8 +37,6 @@ pub enum ExtensionKind {
    WasmTool,
    /// WASM channel module with hot-activation support.
    WasmChannel,
    /// External channel via channel-relay service (Slack, etc.).
    ChannelRelay,
 }
 impl std::fmt::Display for ExtensionKind {
@@ -47,7 +45,6 @@ impl std::fmt::Display for ExtensionKind {
            ExtensionKind::McpServer => write!(f, "mcp_server"),
            ExtensionKind::WasmTool => write!(f, "wasm_tool"),
            ExtensionKind::WasmChannel => write!(f, "wasm_channel"),
            ExtensionKind::ChannelRelay => write!(f, "channel_relay"),
        }
    }
 }
@@ -102,8 +99,6 @@ pub enum ExtensionSource {
    },
    /// Discovered online (not yet validated for a specific source type).
    Discovered { url: String },
    /// External channel via channel-relay service.
    ChannelRelay { relay_url: String },
 }
 /// Hint about what authentication method is needed.
@@ -121,8 +116,6 @@ pub enum AuthHint {
    CapabilitiesAuth,
    /// No authentication needed.
    None,
    /// OAuth via channel-relay service.
    ChannelRelayOAuth,
 }
 /// Where a search result came from.
@@ -506,9 +499,6 @@ pub enum ExtensionError {
    #[error("Activation failed: {0}")]
    ActivationFailed(String),
    #[error("Authentication required")]
    AuthRequired,
    #[error("Installation failed: {0}")]
    InstallFailed(String),
@@ -986,7 +976,6 @@ mod tests {
                ExtensionError::Config("missing key".into()),
                "Config error: missing key",
            ),
            (ExtensionError::AuthRequired, "Authentication required"),
            (
                ExtensionError::Other("something broke".into()),
                "something broke",
@@ -224,16 +224,8 @@ fn score_entry(entry: &RegistryEntry, tokens: &[String]) -> u32 {
 }
 /// Well-known extensions that ship with ironclaw.
-///
+fn builtin_entries() -> Vec<RegistryEntry> {
-/// If `relay_url` is provided, a channel-relay Slack entry is included in the list.
+    vec![
 /// Pass `None` when the relay is not configured.
 pub fn builtin_entries() -> Vec<RegistryEntry> {
    builtin_entries_with_relay(std::env::var("CHANNEL_RELAY_URL").ok())
 }
 /// Well-known extensions, with an optional relay URL for the channel-relay entry.
 pub fn builtin_entries_with_relay(relay_url: Option<String>) -> Vec<RegistryEntry> {
    let mut entries = vec![
        // -- MCP Servers --
        RegistryEntry {
            name: "notion".to_string(),
@@ -423,29 +415,7 @@ pub fn builtin_entries_with_relay(relay_url: Option<String>) -> Vec<RegistryEntr
        // WASM channels (telegram, slack, discord, whatsapp) come from the embedded
        // registry catalog (registry/channels/*.json) with WasmDownload URLs pointing
        // to GitHub release artifacts. See new_with_catalog() for merging.
-    ];
+    ]
    // Conditionally add channel-relay entries when relay URL is configured
    if let Some(relay_url) = relay_url {
        entries.push(RegistryEntry {
            name: crate::channels::relay::DEFAULT_RELAY_NAME.to_string(),
            display_name: "Slack".to_string(),
            kind: ExtensionKind::ChannelRelay,
            description: "Connect Slack workspace via channel relay".to_string(),
            keywords: vec![
                "slack".into(),
                "chat".into(),
                "messaging".into(),
                "relay".into(),
            ],
            source: ExtensionSource::ChannelRelay { relay_url },
            fallback_source: None,
            auth_hint: AuthHint::ChannelRelayOAuth,
            version: None,
        });
    }
    entries
 }
 #[cfg(test)]
@@ -965,30 +935,4 @@ mod tests {
        // The first catalog entry added is the channel.
        assert_eq!(entry.unwrap().kind, ExtensionKind::WasmChannel);
    }
    #[test]
    fn test_builtin_entries_with_relay_none_excludes_relay() {
        let entries = super::builtin_entries_with_relay(None);
        assert!(
            !entries
                .iter()
                .any(|e| e.kind == ExtensionKind::ChannelRelay),
            "No ChannelRelay entry when relay URL is None"
        );
    }
    #[test]
    fn test_builtin_entries_with_relay_some_includes_relay() {
        let entries =
            super::builtin_entries_with_relay(Some("http://relay.example.com".to_string()));
        let relay = entries
            .iter()
            .find(|e| e.kind == ExtensionKind::ChannelRelay);
        assert!(relay.is_some(), "ChannelRelay entry should be present");
        if let ExtensionSource::ChannelRelay { relay_url } = &relay.unwrap().source {
            assert_eq!(relay_url, "http://relay.example.com");
        } else {
            panic!("Expected ChannelRelay source");
        }
    }
 }
@@ -151,9 +151,8 @@ impl Store {
                id, conversation_id, title, description, category, status, source,
                user_id,
                budget_amount, budget_token, bid_amount, estimated_cost, estimated_time_secs,
-                actual_cost, repair_attempts, max_tokens, total_tokens_used,
+                actual_cost, repair_attempts, created_at, started_at, completed_at
-                created_at, started_at, completed_at
+            ) VALUES ($1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16, $17, $18)
            ) VALUES ($1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16, $17, $18, $19, $20)
            ON CONFLICT (id) DO UPDATE SET
                title = EXCLUDED.title,
                description = EXCLUDED.description,
@@ -164,8 +163,6 @@ impl Store {
                estimated_time_secs = EXCLUDED.estimated_time_secs,
                actual_cost = EXCLUDED.actual_cost,
                repair_attempts = EXCLUDED.repair_attempts,
                max_tokens = EXCLUDED.max_tokens,
                total_tokens_used = EXCLUDED.total_tokens_used,
                started_at = EXCLUDED.started_at,
                completed_at = EXCLUDED.completed_at
            "#,
@@ -185,8 +182,6 @@ impl Store {
                &estimated_time_secs,
                &ctx.actual_cost,
                &(ctx.repair_attempts as i32),
                &(ctx.max_tokens as i64),
                &(ctx.total_tokens_used as i64),
                &ctx.created_at,
                &ctx.started_at,
                &ctx.completed_at,
@@ -206,8 +201,7 @@ impl Store {
                r#"
                SELECT id, conversation_id, title, description, category, status, user_id,
                       budget_amount, budget_token, bid_amount, estimated_cost, estimated_time_secs,
-                       actual_cost, repair_attempts, max_tokens, total_tokens_used,
+                       actual_cost, repair_attempts, created_at, started_at, completed_at
                       created_at, started_at, completed_at
                FROM agent_jobs WHERE id = $1
                "#,
                &[&id],
@@ -243,9 +237,8 @@ impl Store {
                    completed_at: row.get("completed_at"),
                    transitions: Vec::new(), // Not loaded from DB for now
                    metadata: serde_json::Value::Null,
-                    max_tokens: row.get::<_, Option<i64>>("max_tokens").unwrap_or(0) as u64,
+                    total_tokens_used: 0,
-                    total_tokens_used: row.get::<_, Option<i64>>("total_tokens_used").unwrap_or(0)
+                    max_tokens: 0,
                        as u64,
                    extra_env: std::sync::Arc::new(std::collections::HashMap::new()),
                    http_interceptor: None,
                    tool_output_stash: std::sync::Arc::new(tokio::sync::RwLock::new(
@@ -1094,7 +1087,7 @@ impl Store {
        let conn = self.conn().await?;
        let rows = conn
            .query(
-                "SELECT * FROM routines WHERE enabled AND trigger_type IN ('event', 'system_event')",
+                "SELECT * FROM routines WHERE enabled AND trigger_type = 'event'",
                &[],
            )
            .await?;
@@ -6,8 +6,6 @@
 //!
 //! Pattern follows `nearai_chat.rs`: direct HTTP calls via `reqwest::Client`.
 use std::collections::HashSet;
 use async_trait::async_trait;
 use reqwest::Client;
 use rust_decimal::Decimal;
@@ -19,8 +17,7 @@ use crate::llm::costs;
 use crate::llm::error::LlmError;
 use crate::llm::provider::{
    ChatMessage, CompletionRequest, CompletionResponse, FinishReason, LlmProvider, Role, ToolCall,
-    ToolCompletionRequest, ToolCompletionResponse, strip_unsupported_completion_params,
+    ToolCompletionRequest, ToolCompletionResponse,
    strip_unsupported_tool_params,
 };
 const ANTHROPIC_API_URL: &str = "https://api.anthropic.com/v1/messages";
@@ -38,8 +35,6 @@ pub struct AnthropicOAuthProvider {
    model: String,
    base_url: Option<String>,
    active_model: std::sync::RwLock<String>,
    /// Parameter names that this provider does not support.
    unsupported_params: HashSet<String>,
 }
 impl AnthropicOAuthProvider {
@@ -66,29 +61,15 @@ impl AnthropicOAuthProvider {
            Some(config.base_url.clone())
        };
        let unsupported_params: HashSet<String> =
            config.unsupported_params.iter().cloned().collect();
        Ok(Self {
            client,
            token,
            model: config.model.clone(),
            base_url,
            active_model,
            unsupported_params,
        })
    }
    /// Strip unsupported fields from a `CompletionRequest` in place.
    fn strip_unsupported_completion_params(&self, req: &mut CompletionRequest) {
        strip_unsupported_completion_params(&self.unsupported_params, req);
    }
    /// Strip unsupported fields from a `ToolCompletionRequest` in place.
    fn strip_unsupported_tool_params(&self, req: &mut ToolCompletionRequest) {
        strip_unsupported_tool_params(&self.unsupported_params, req);
    }
    fn api_url(&self) -> String {
        if let Some(ref base) = self.base_url {
            let base = base.trim_end_matches('/');
@@ -216,9 +197,8 @@ impl AnthropicOAuthProvider {
 #[async_trait]
 impl LlmProvider for AnthropicOAuthProvider {
-    async fn complete(&self, mut req: CompletionRequest) -> Result<CompletionResponse, LlmError> {
+    async fn complete(&self, req: CompletionRequest) -> Result<CompletionResponse, LlmError> {
-        let model = req.model.take().unwrap_or_else(|| self.active_model_name());
+        let model = req.model.unwrap_or_else(|| self.active_model_name());
        self.strip_unsupported_completion_params(&mut req);
        let (system, messages) = convert_messages(req.messages);
        let request = AnthropicRequest {
@@ -253,10 +233,9 @@ impl LlmProvider for AnthropicOAuthProvider {
    async fn complete_with_tools(
        &self,
-        mut req: ToolCompletionRequest,
+        req: ToolCompletionRequest,
    ) -> Result<ToolCompletionResponse, LlmError> {
-        let model = req.model.take().unwrap_or_else(|| self.active_model_name());
+        let model = req.model.unwrap_or_else(|| self.active_model_name());
        self.strip_unsupported_tool_params(&mut req);
        let (system, messages) = convert_messages(req.messages);
        let tools: Vec<AnthropicTool> = req
@@ -87,10 +87,6 @@ pub struct RegistryProviderConfig {
    pub oauth_token: Option<SecretString>,
    /// Prompt cache retention (Anthropic-specific).
    pub cache_retention: CacheRetention,
    /// Parameter names that this provider does not support (e.g., `["temperature"]`).
    /// Supported keys: `"temperature"`, `"max_tokens"`, `"stop_sequences"`.
    /// Listed parameters are stripped from requests before sending to avoid 400 errors.
    pub unsupported_params: Vec<String>,
 }
 /// Configuration for AWS Bedrock (native Converse API).
@@ -29,7 +29,6 @@ pub mod session;
 pub mod smart_routing;
 pub mod image_models;
 pub mod reasoning_models;
 pub mod vision_models;
 pub use circuit_breaker::{CircuitBreakerConfig, CircuitBreakerProvider};
@@ -229,9 +228,7 @@ fn create_openai_compat_from_registry(
        "Using OpenAI-compatible provider"
    );
-    let adapter = RigAdapter::new(model, &config.model)
+    Ok(Arc::new(RigAdapter::new(model, &config.model)))
        .with_unsupported_params(config.unsupported_params.clone());
    Ok(Arc::new(adapter))
 }
 fn create_anthropic_from_registry(
@@ -299,9 +296,7 @@ fn create_anthropic_from_registry(
    );
    Ok(Arc::new(
-        RigAdapter::new(model, &config.model)
+        RigAdapter::new(model, &config.model).with_cache_retention(cache_retention),
            .with_cache_retention(cache_retention)
            .with_unsupported_params(config.unsupported_params.clone()),
    ))
 }
@@ -329,9 +324,7 @@ fn create_ollama_from_registry(
        "Using Ollama provider"
    );
-    let adapter = RigAdapter::new(model, &config.model)
+    Ok(Arc::new(RigAdapter::new(model, &config.model)))
        .with_unsupported_params(config.unsupported_params.clone());
    Ok(Arc::new(adapter))
 }
 /// Create a cheap/fast LLM provider for lightweight tasks (heartbeat, routing, evaluation).
@@ -270,15 +270,8 @@ impl NearAiChatProvider {
            reason: format!("Failed to read response body: {}", e),
        })?;
        if tracing::enabled!(tracing::Level::DEBUG) {
        tracing::debug!("NEAR AI Chat response status: {}", status);
-        }
+        tracing::debug!("NEAR AI Chat response body: {}", response_text);
        // Log response body only at TRACE level to avoid exposing sensitive content
        // (user-generated data, tool outputs, leaked secrets) in DEBUG logs
        if tracing::enabled!(tracing::Level::TRACE) {
            tracing::trace!("NEAR AI Chat response body: {}", response_text);
        }
        if !status.is_success() {
            let status_code = status.as_u16();
@@ -455,73 +455,6 @@ pub fn sanitize_tool_messages(messages: &mut [ChatMessage]) {
    }
 }
 /// Represents a request parameter that may not be supported by all LLM providers.
 ///
 /// This typed enum replaces stringly-typed parameter names across the codebase,
 /// providing type safety and single-point-of-maintenance for parameter handling.
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
 pub enum UnsupportedParam {
    Temperature,
    MaxTokens,
    StopSequences,
 }
 impl UnsupportedParam {
    /// Get the string name of this parameter for config/error messages.
    pub fn name(&self) -> &'static str {
        match self {
            UnsupportedParam::Temperature => "temperature",
            UnsupportedParam::MaxTokens => "max_tokens",
            UnsupportedParam::StopSequences => "stop_sequences",
        }
    }
 }
 /// Strip unsupported parameters from a `CompletionRequest` in place.
 ///
 /// This is the single helper function used by all providers to remove
 /// parameters they don't support, replacing duplicate stringly-typed logic.
 pub fn strip_unsupported_completion_params(
    unsupported: &std::collections::HashSet<String>,
    req: &mut CompletionRequest,
 ) {
    if unsupported.is_empty() {
        return;
    }
    if unsupported.contains(UnsupportedParam::Temperature.name()) {
        req.temperature = None;
    }
    if unsupported.contains(UnsupportedParam::MaxTokens.name()) {
        req.max_tokens = None;
    }
    if unsupported.contains(UnsupportedParam::StopSequences.name()) {
        req.stop_sequences = None;
    }
 }
 /// Strip unsupported parameters from a `ToolCompletionRequest` in place.
 ///
 /// This is the single helper function used by all providers to remove
 /// parameters they don't support from tool calls, replacing duplicate stringly-typed logic.
 ///
 /// Note: Only `Temperature` and `MaxTokens` are supported in `ToolCompletionRequest`.
 /// `StopSequences` is only available in `CompletionRequest` and is not applicable to tool calls.
 pub fn strip_unsupported_tool_params(
    unsupported: &std::collections::HashSet<String>,
    req: &mut ToolCompletionRequest,
 ) {
    if unsupported.is_empty() {
        return;
    }
    if unsupported.contains(UnsupportedParam::Temperature.name()) {
        req.temperature = None;
    }
    if unsupported.contains(UnsupportedParam::MaxTokens.name()) {
        req.max_tokens = None;
    }
    // Note: StopSequences is not a field in ToolCompletionRequest, so no action needed
 }
 #[cfg(test)]
 mod tests {
    use super::*;
@@ -450,8 +450,7 @@ impl Reasoning {
            cache_read_input_tokens: response.cache_read_input_tokens,
            cache_creation_input_tokens: response.cache_creation_input_tokens,
        };
-        let pre_truncated = truncate_at_tool_tags(&response.content);
+        Ok((clean_response(&response.content), usage))
        Ok((clean_response(&pre_truncated), usage))
    }
    /// Generate a plan for completing a goal.
@@ -481,11 +480,8 @@ impl Reasoning {
        let response = self.llm.complete(request).await?;
-        // Clean reasoning model artifacts before parsing JSON.
+        // Clean reasoning model artifacts before parsing JSON
-        // Pre-truncate at tool tags to avoid strip_xml_tag discarding
+        let cleaned = clean_response(&response.content);
        // content after unclosed tags (issue #789).
        let pre_truncated = truncate_at_tool_tags(&response.content);
        let cleaned = clean_response(&pre_truncated);
        self.parse_plan(&cleaned)
    }
@@ -579,11 +575,8 @@ Respond in JSON format:
        let response = self.llm.complete(request).await?;
-        // Clean reasoning model artifacts before parsing JSON.
+        // Clean reasoning model artifacts before parsing JSON
-        // Pre-truncate at tool tags to avoid strip_xml_tag discarding
+        let cleaned = clean_response(&response.content);
        // content after unclosed tags (issue #789).
        let pre_truncated = truncate_at_tool_tags(&response.content);
        let cleaned = clean_response(&pre_truncated);
        self.parse_evaluation(&cleaned)
    }
@@ -660,10 +653,7 @@ Respond in JSON format:
                return Ok(RespondOutput {
                    result: RespondResult::ToolCalls {
                        tool_calls: response.tool_calls,
-                        content: response.content.map(|c| {
+                        content: response.content.map(|c| clean_response(&c)),
                            let pre_truncated = truncate_at_tool_tags(&c);
                            clean_response(&pre_truncated)
                        }),
                    },
                    usage,
                });
@@ -676,13 +666,9 @@ Respond in JSON format:
            // Some models (e.g. GLM-4.7) emit tool calls as XML tags in content
            // instead of using the structured tool_calls field. Try to recover
            // them before giving up and returning plain text.
            // NOTE: Recovery runs on the raw content (before truncation) so it can
            // parse tool-call JSON from the XML tags. Truncation only applies to the
            // remaining *text* content returned alongside the recovered tool calls.
            let recovered = recover_tool_calls_from_content(&content, &context.available_tools);
            if !recovered.is_empty() {
-                let pre_truncated = truncate_at_tool_tags(&content);
+                let cleaned = clean_response(&content);
                let cleaned = clean_response(&pre_truncated);
                return Ok(RespondOutput {
                    result: RespondResult::ToolCalls {
                        tool_calls: recovered,
@@ -696,16 +682,12 @@ Respond in JSON format:
                });
            }
-            // Guard against empty text after cleaning. This can happen when:
+            // Guard against empty text after cleaning. This can happen
-            // 1. Reasoning models (e.g. GLM-5) return chain-of-thought in
+            // when reasoning models (e.g. GLM-5) return chain-of-thought
-            //    reasoning_content wrapped in <think> tags — clean_response
+            // in reasoning_content wrapped in <think> tags and content is
-            //    strips the think tags leaving an empty string.
+            // null — the .or(reasoning_content) fallback picks it up, then
-            // 2. Local models (Qwen3, DeepSeek) emit <tool_call> XML in text
+            // clean_response strips the think tags leaving an empty string.
-            //    responses even in force_text mode — strip_xml_tag discards
+            let cleaned = clean_response(&content);
            //    from unclosed opening tag onward (issue #789).
            // Pre-truncate at tool tags to preserve text before the tag.
            let pre_truncated = truncate_at_tool_tags(&content);
            let cleaned = clean_response(&pre_truncated);
            let final_text = if cleaned.trim().is_empty() {
                tracing::warn!(
                    "LLM response was empty after cleaning (original len={}), using fallback",
@@ -727,8 +709,7 @@ Respond in JSON format:
            request.metadata = context.metadata.clone();
            let response = self.llm.complete(request).await?;
-            let pre_truncated = truncate_at_tool_tags(&response.content);
+            let cleaned = clean_response(&response.content);
            let cleaned = clean_response(&pre_truncated);
            let final_text = if cleaned.trim().is_empty() {
                tracing::warn!(
                    "LLM response was empty after cleaning (original len={}), using fallback",
@@ -866,22 +847,10 @@ Respond with a JSON plan in this format:
                .to_string()
        };
-        // Models with native thinking (Qwen3, DeepSeek-R1, etc.) produce their
+        format!(
-        // own <think> tags or reasoning_content. Injecting our <think>/<final>
+            r#"You are IronClaw Agent, a secure autonomous assistant.
        // format collides with their native behavior, causing thinking-only
        // responses that clean to empty strings. See issue #789.
        let has_native_thinking = self
            .model_name
            .as_ref()
            .is_some_and(|n| crate::llm::reasoning_models::has_native_thinking(n));
-        let response_format = if has_native_thinking {
+## Response Format — CRITICAL
            r#"## Response Format
 Respond directly with your answer. Do not wrap your response in any special tags.
 Your reasoning process is handled natively — just provide the final user-facing answer."#
        } else {
            r#"## Response Format — CRITICAL
 ALL internal reasoning MUST be inside <think>...</think> tags.
 Do not output any analysis, planning, or self-talk outside <think>.
@@ -891,13 +860,7 @@ Only text inside <final> is shown to the user; everything else is discarded.
 Example:
 <think>The user is asking about X.</think>
-<final>Here is the answer about X.</final>"#
+<final>Here is the answer about X.</final>
        };
        format!(
            r#"You are IronClaw Agent, a secure autonomous assistant.
 {response_format}
 ## Guidelines
 - Be concise and direct
@@ -1479,99 +1442,6 @@ fn strip_bracket_tool_calls(text: &str) -> String {
 /// Tool-related tags stripped with simple string matching (no code-awareness needed).
 const TOOL_TAGS: &[&str] = &["tool_call", "function_call", "tool_calls"];
 /// Patterns that indicate tool-call XML in model output.
 const TOOL_TAG_PATTERNS: &[&str] = &[
    "<tool_call>",
    "<tool_call ",
    "<function_call>",
    "<function_call ",
    "<tool_calls>",
    "<tool_calls ",
    "<|tool_call|>",
    "<|function_call|>",
    "<|tool_calls|>",
 ];
 /// Truncate text at the first **unclosed** tool-call XML tag, preserving content
 /// before it.
 ///
 /// Local models (Qwen3, DeepSeek, etc.) often emit `<tool_call>` XML in text
 /// responses even when no tools are available. The downstream `clean_response()`
 /// → `strip_xml_tag()` pipeline discards everything from an unclosed opening
 /// tag onward, which can leave an empty string and trigger the fallback message.
 ///
 /// This function truncates at the first *unclosed* tool tag BEFORE
 /// `clean_response()` runs, so the useful text before the tag is preserved.
 /// Properly closed tags (e.g. `<tool_call>...</tool_call>`) are left intact for
 /// `clean_response()` to strip normally. Tags inside fenced markdown code blocks
 /// or inline code spans are ignored. See issue #789.
 fn truncate_at_tool_tags(text: &str) -> String {
    let code_regions = find_code_regions(text);
    // Use ASCII-only lowercasing so byte offsets stay valid for the original
    // string. Full `to_lowercase()` can change byte lengths for non-ASCII
    // chars (e.g. the Kelvin sign), making positions unreliable.
    let lower = text.to_ascii_lowercase();
    let first_unclosed = TOOL_TAG_PATTERNS
        .iter()
        .filter_map(|p| {
            let mut search_from = 0;
            loop {
                match lower[search_from..].find(p) {
                    Some(offset) => {
                        let pos = search_from + offset;
                        if is_inside_code(pos, &code_regions) {
                            search_from = pos + 1;
                            continue;
                        }
                        // Check if this tag has a matching closing tag after it.
                        // If so, clean_response() can handle it — skip to next.
                        let after_open = pos + p.len();
                        if closing_tag_for(p)
                            .is_some_and(|close| lower[after_open..].contains(close.as_str()))
                        {
                            search_from = after_open;
                            continue;
                        }
                        // Unclosed tag — truncate here
                        return Some(pos);
                    }
                    None => return None,
                }
            }
        })
        .min();
    match first_unclosed {
        Some(pos) => {
            tracing::debug!(
                original_len = text.len(),
                truncated_at = pos,
                "Truncated response at unclosed tool-call XML tag (issue #789)"
            );
            text[..pos].to_string()
        }
        None => text.to_string(),
    }
 }
 /// Derive the closing tag for a tool-call opening pattern.
 ///
 /// Examples: `<tool_call>` → `</tool_call>`, `<|tool_call|>` → `<|/tool_call|>`.
 fn closing_tag_for(open_pattern: &str) -> Option<String> {
    if let Some(name) = open_pattern
        .strip_prefix("<|")
        .and_then(|s| s.strip_suffix("|>"))
    {
        // Pipe-delimited: <|tool_call|> → <|/tool_call|>
        Some(format!("<|/{name}|>"))
    } else if let Some(rest) = open_pattern.strip_prefix('<') {
        // Standard XML: <tool_call> or <tool_call  → </tool_call>
        let name = rest.trim_end_matches('>').trim();
        Some(format!("</{name}>"))
    } else {
        None
    }
 }
 /// Strip thinking/reasoning tags using regex, respecting code regions.
 ///
 /// Strict mode: an unclosed opening tag discards all trailing text after it.
@@ -2544,588 +2414,4 @@ That's my plan."#;
        let text = "I said let me be clear, then let me fetch the data.";
        assert!(llm_signals_tool_intent(text));
    }
    // ---- Issue #789: truncate_at_tool_tags tests ----
    #[test]
    fn test_truncate_preserves_text_before_tool_tag() {
        let input = "Here is my answer about the topic.\n<tool_call>{\"name\": \"search\"}";
        assert_eq!(
            truncate_at_tool_tags(input),
            "Here is my answer about the topic.\n"
        );
    }
    #[test]
    fn test_truncate_no_tool_tags_unchanged() {
        let input = "Just a normal response with no tool tags.";
        assert_eq!(truncate_at_tool_tags(input), input);
    }
    #[test]
    fn test_truncate_empty_string() {
        assert_eq!(truncate_at_tool_tags(""), "");
    }
    #[test]
    fn test_truncate_tool_tag_at_start() {
        assert_eq!(
            truncate_at_tool_tags("<tool_call>{\"name\": \"search\"}"),
            ""
        );
    }
    #[test]
    fn test_truncate_picks_earliest_unclosed_tag() {
        // <function_call>...</function_call> is closed — skipped.
        // <tool_call>second is unclosed — truncated here.
        let input = "Text before <function_call>first</function_call> and <tool_call>second";
        assert_eq!(
            truncate_at_tool_tags(input),
            "Text before <function_call>first</function_call> and "
        );
    }
    #[test]
    fn test_truncate_pipe_delimited_tags() {
        let input = "Answer here\n<|tool_call|>{\"name\": \"fetch\"}";
        assert_eq!(truncate_at_tool_tags(input), "Answer here\n");
    }
    #[test]
    fn test_truncate_closed_tag_with_attributes_preserved() {
        // Closed tag (even with attributes) is left for clean_response()
        let input = "Some text <tool_call id=\"123\">{\"name\": \"test\"}</tool_call>";
        assert_eq!(truncate_at_tool_tags(input), input);
    }
    #[test]
    fn test_truncate_unclosed_tag_with_attributes() {
        let input = "Some text <tool_call id=\"123\">{\"name\": \"test\"}";
        assert_eq!(truncate_at_tool_tags(input), "Some text ");
    }
    #[test]
    fn test_truncate_whitespace_only_before_tag() {
        assert_eq!(truncate_at_tool_tags("   \n\n<tool_call>{}"), "   \n\n");
    }
    #[test]
    fn test_truncate_ignores_tags_inside_code_blocks() {
        let input = "Here's the XML format:\n\n```xml\n<tool_call>{\"name\": \"search\"}</tool_call>\n```\n\nYou can use this to call tools.";
        assert_eq!(truncate_at_tool_tags(input), input);
    }
    #[test]
    fn test_truncate_finds_tag_after_code_block() {
        let input = "Example:\n\n```\n<tool_call>example</tool_call>\n```\n\nReal output:\n<tool_call>{\"name\": \"x\"}";
        assert_eq!(
            truncate_at_tool_tags(input),
            "Example:\n\n```\n<tool_call>example</tool_call>\n```\n\nReal output:\n"
        );
    }
    // ---- Issue #789: full pipeline (truncate + clean_response) tests ----
    #[test]
    fn test_issue_789_force_text_unclosed_tool_tag() {
        let model_output = "The file contains a main function that initializes the server.\n<tool_call>{\"name\": \"read_file\", \"arguments\": {\"path\": \"src/main.rs\"}}";
        let pre_truncated = truncate_at_tool_tags(model_output);
        let cleaned = clean_response(&pre_truncated);
        assert_eq!(
            cleaned,
            "The file contains a main function that initializes the server."
        );
    }
    #[test]
    fn test_issue_789_only_tool_tag_produces_empty() {
        let model_output = "<tool_call>{\"name\": \"search\", \"arguments\": {\"q\": \"test\"}}";
        let pre_truncated = truncate_at_tool_tags(model_output);
        let cleaned = clean_response(&pre_truncated);
        assert!(cleaned.trim().is_empty());
    }
    #[test]
    fn test_issue_789_thinking_then_tool_tag() {
        let model_output =
            "<think>I should search for this</think>Let me help you.\n<tool_call>{\"name\": \"s\"}";
        let pre_truncated = truncate_at_tool_tags(model_output);
        let cleaned = clean_response(&pre_truncated);
        assert_eq!(cleaned, "Let me help you.");
    }
    #[test]
    fn test_issue_789_closed_tool_tag_preserved_for_clean_response() {
        // Closed tags are left intact — clean_response() strips them normally,
        // preserving any text after the tag.
        let model_output = "Info here.\n<tool_call>{\"name\": \"x\"}</tool_call>\nMore text.";
        let pre_truncated = truncate_at_tool_tags(model_output);
        assert_eq!(
            pre_truncated, model_output,
            "Closed tag should not be truncated"
        );
        let cleaned = clean_response(&pre_truncated);
        assert_eq!(cleaned, "Info here.\n\nMore text.");
    }
    // ---- Issue #789: conditional system prompt tests ----
    fn make_reasoning_with_model(model: &str) -> Reasoning {
        use crate::testing::StubLlm;
        Reasoning::new(Arc::new(StubLlm::new("test"))).with_model_name(model.to_string())
    }
    #[test]
    fn test_system_prompt_skips_think_final_for_native_thinking() {
        let reasoning = make_reasoning_with_model("qwen3-8b");
        let prompt = reasoning.build_system_prompt_with_tools(&[]);
        assert!(
            !prompt.contains("<think>"),
            "Native thinking model should NOT have <think> in system prompt"
        );
        assert!(prompt.contains("Respond directly with your answer"));
    }
    #[test]
    fn test_system_prompt_includes_think_final_for_regular_model() {
        let reasoning = make_reasoning_with_model("llama-3.1-70b");
        let prompt = reasoning.build_system_prompt_with_tools(&[]);
        assert!(prompt.contains("<think>"));
        assert!(prompt.contains("<final>"));
    }
    #[test]
    fn test_system_prompt_defaults_to_think_final_when_no_model() {
        use crate::testing::StubLlm;
        let reasoning = Reasoning::new(Arc::new(StubLlm::new("test")));
        let prompt = reasoning.build_system_prompt_with_tools(&[]);
        assert!(prompt.contains("<think>"));
        assert!(prompt.contains("<final>"));
    }
    #[test]
    fn test_system_prompt_deepseek_r1_skips_think_final() {
        let reasoning = make_reasoning_with_model("deepseek-r1-distill-qwen-32b");
        let prompt = reasoning.build_system_prompt_with_tools(&[]);
        assert!(!prompt.contains("CRITICAL"));
        assert!(prompt.contains("Respond directly"));
    }
    // ---- Issue #789: additional edge case tests for truncate_at_tool_tags ----
    #[test]
    fn test_truncate_unicode_content_before_tool_tag() {
        let input = "こんにちは世界！素晴らしい結果です。\n<tool_call>{\"name\": \"search\"}";
        assert_eq!(
            truncate_at_tool_tags(input),
            "こんにちは世界！素晴らしい結果です。\n"
        );
    }
    #[test]
    fn test_truncate_emoji_content_preserved() {
        let input = "The answer is 42 🎉🚀\n<function_call>{\"name\": \"x\"}";
        assert_eq!(truncate_at_tool_tags(input), "The answer is 42 🎉🚀\n");
    }
    #[test]
    fn test_truncate_very_long_text_before_tag() {
        let long_text = "A".repeat(10_000);
        let input = format!("{}\n<tool_call>{{\"name\": \"x\"}}", long_text);
        let result = truncate_at_tool_tags(&input);
        assert_eq!(result.len(), long_text.len() + 1); // +1 for \n
        assert!(result.starts_with("AAAA"));
    }
    #[test]
    fn test_truncate_multiple_code_blocks_with_tags() {
        let input = "Explanation:\n\n```python\n# <tool_call> in comment\nprint('hi')\n```\n\nAnd also:\n\n```xml\n<function_call>example</function_call>\n```\n\nFinal answer here.";
        // Both tags are inside code blocks, so nothing is truncated
        assert_eq!(truncate_at_tool_tags(input), input);
    }
    #[test]
    fn test_truncate_inline_code_with_tool_tag() {
        let input = "Use `<tool_call>` to invoke tools.\n<tool_call>{\"name\": \"real\"}";
        // First occurrence is in inline code, second is real
        assert_eq!(
            truncate_at_tool_tags(input),
            "Use `<tool_call>` to invoke tools.\n"
        );
    }
    #[test]
    fn test_truncate_tag_immediately_after_code_block() {
        let input = "```\nexample\n```\n<tool_call>{\"name\": \"x\"}";
        assert_eq!(truncate_at_tool_tags(input), "```\nexample\n```\n");
    }
    #[test]
    fn test_truncate_interleaved_thinking_and_tool_tags() {
        // Simulate: thinking tag + text + tool tag
        let input = "<think>reasoning</think>Here's the answer.\n<tool_call>{\"name\": \"y\"}";
        let truncated = truncate_at_tool_tags(input);
        let cleaned = clean_response(&truncated);
        assert_eq!(cleaned, "Here's the answer.");
    }
    #[test]
    fn test_truncate_closed_tool_calls_plural_preserved() {
        // Closed <tool_calls>...</tool_calls> left for clean_response()
        let input = "Answer.\n<tool_calls>[{\"name\": \"a\"}, {\"name\": \"b\"}]</tool_calls>";
        assert_eq!(truncate_at_tool_tags(input), input);
    }
    #[test]
    fn test_truncate_unclosed_tool_calls_plural() {
        let input = "Answer.\n<tool_calls>[{\"name\": \"a\"}, {\"name\": \"b\"}]";
        assert_eq!(truncate_at_tool_tags(input), "Answer.\n");
    }
    #[test]
    fn test_truncate_closed_pipe_function_call_preserved() {
        let input = "Done!\n<|function_call|>{\"name\": \"x\"}<|/function_call|>";
        assert_eq!(truncate_at_tool_tags(input), input);
    }
    #[test]
    fn test_truncate_unclosed_pipe_function_call() {
        let input = "Done!\n<|function_call|>{\"name\": \"x\"}";
        assert_eq!(truncate_at_tool_tags(input), "Done!\n");
    }
    #[test]
    fn test_truncate_adversarial_nested_code_blocks() {
        // Adversarial: code block inside another structure
        let input = "```\nouter\n```\n\nReal text.\n\n```\n<tool_call>inside</tool_call>\n```\n\n<tool_call>{\"name\": \"real\"}";
        let result = truncate_at_tool_tags(input);
        assert!(result.contains("Real text."));
        assert!(!result.contains("{\"name\": \"real\"}"));
    }
    // ---- Issue #789: StubLlm integration tests ----
    #[tokio::test]
    async fn test_complete_truncates_tool_tags_from_response() {
        use crate::testing::StubLlm;
        let response = "The server has 3 endpoints.\n<tool_call>{\"name\": \"read_file\"}";
        let llm = Arc::new(StubLlm::new(response));
        let reasoning = Reasoning::new(llm);
        let request = CompletionRequest::new(vec![ChatMessage::user("describe the server")]);
        let (result, _usage) = reasoning.complete(request).await.unwrap();
        assert_eq!(result, "The server has 3 endpoints.");
    }
    #[tokio::test]
    async fn test_complete_with_only_tool_tag_returns_empty() {
        use crate::testing::StubLlm;
        let response = "<tool_call>{\"name\": \"search\", \"arguments\": {}}";
        let llm = Arc::new(StubLlm::new(response));
        let reasoning = Reasoning::new(llm);
        let request = CompletionRequest::new(vec![ChatMessage::user("hello")]);
        let (result, _usage) = reasoning.complete(request).await.unwrap();
        assert!(result.trim().is_empty());
    }
    #[tokio::test]
    async fn test_respond_with_tools_force_text_truncates_tool_tags() {
        use crate::testing::StubLlm;
        let response = "Here is my analysis of the code.\n<tool_call>{\"name\": \"read_file\", \"arguments\": {\"path\": \"main.rs\"}}";
        let llm = Arc::new(StubLlm::new(response));
        let reasoning = Reasoning::new(llm);
        let mut context =
            ReasoningContext::new().with_message(ChatMessage::user("analyze the code"));
        context.force_text = true;
        let output = reasoning.respond_with_tools(&context).await.unwrap();
        match output.result {
            RespondResult::Text(text) => {
                assert_eq!(text, "Here is my analysis of the code.");
            }
            RespondResult::ToolCalls { .. } => {
                panic!("Expected text result in force_text mode");
            }
        }
    }
    #[tokio::test]
    async fn test_respond_with_tools_force_text_only_tag_uses_fallback() {
        use crate::testing::StubLlm;
        let response = "<tool_call>{\"name\": \"search\"}";
        let llm = Arc::new(StubLlm::new(response));
        let reasoning = Reasoning::new(llm);
        let mut context = ReasoningContext::new().with_message(ChatMessage::user("hi"));
        context.force_text = true;
        let output = reasoning.respond_with_tools(&context).await.unwrap();
        match output.result {
            RespondResult::Text(text) => {
                assert_eq!(text, "I'm not sure how to respond to that.");
            }
            RespondResult::ToolCalls { .. } => {
                panic!("Expected fallback text, not tool calls");
            }
        }
    }
    #[tokio::test]
    async fn test_plan_truncates_tool_tags_before_json() {
        use crate::testing::StubLlm;
        let response = r#"<think>Let me plan</think>{"goal": "Test goal", "actions": [{"tool_name": "search", "parameters": {}, "reasoning": "find files", "expected_outcome": "results"}], "confidence": 0.9}
 <tool_call>{"name": "search"}"#;
        let llm = Arc::new(StubLlm::new(response));
        let reasoning = Reasoning::new(llm);
        let context = ReasoningContext::new()
            .with_message(ChatMessage::user("plan a search"))
            .with_job("Search for relevant files");
        let plan = reasoning.plan(&context).await.unwrap();
        assert_eq!(plan.goal, "Test goal");
        assert!(!plan.actions.is_empty());
    }
    // ---- Issue #789: model name propagation test ----
    #[tokio::test]
    async fn test_with_model_name_affects_system_prompt() {
        use crate::testing::StubLlm;
        // StubLlm model_name is "stub-model" by default, but Reasoning.model_name
        // is what matters for system prompt building.
        let llm = Arc::new(StubLlm::new("test").with_model_name("qwen3-8b"));
        let reasoning = Reasoning::new(llm.clone()).with_model_name("qwen3-8b".to_string());
        let prompt = reasoning.build_system_prompt_with_tools(&[]);
        assert!(
            !prompt.contains("<think>"),
            "Qwen3 model should get native thinking system prompt"
        );
        assert!(prompt.contains("Respond directly"));
        // Now create reasoning WITHOUT with_model_name — should get default prompt
        let reasoning_no_model = Reasoning::new(llm);
        let prompt2 = reasoning_no_model.build_system_prompt_with_tools(&[]);
        assert!(
            prompt2.contains("<think>"),
            "Without model name, should get default think/final prompt"
        );
    }
    // ---- Issue #789: case-insensitive truncation ----
    #[test]
    fn test_truncate_case_insensitive_upper() {
        let input = "Some answer.\n<TOOL_CALL>{\"name\": \"search\"}";
        assert_eq!(truncate_at_tool_tags(input), "Some answer.\n");
    }
    #[test]
    fn test_truncate_case_insensitive_mixed() {
        let input = "Result here.\n<Tool_Call>{\"name\": \"x\"}";
        assert_eq!(truncate_at_tool_tags(input), "Result here.\n");
    }
    #[test]
    fn test_truncate_unicode_before_case_insensitive_tag_no_panic() {
        // Regression: to_lowercase() can change byte lengths for non-ASCII chars
        // (e.g. Kelvin sign U+212A is 3 bytes, lowercases to 'k' which is 1 byte).
        // Using to_ascii_lowercase() keeps byte offsets stable.
        let input = "Ответ: 42\n<TOOL_CALL>{\"name\": \"x\"}";
        assert_eq!(truncate_at_tool_tags(input), "Ответ: 42\n");
    }
    #[test]
    fn test_truncate_case_insensitive_function_call_closed() {
        // Closed tag (case-insensitive) preserved for clean_response()
        let input = "Done.\n<FUNCTION_CALL>{\"name\": \"y\"}</FUNCTION_CALL>";
        assert_eq!(truncate_at_tool_tags(input), input);
    }
    #[test]
    fn test_truncate_case_insensitive_function_call_unclosed() {
        let input = "Done.\n<FUNCTION_CALL>{\"name\": \"y\"}";
        assert_eq!(truncate_at_tool_tags(input), "Done.\n");
    }
    // ---- Issue #789: evaluate_success integration test ----
    #[tokio::test]
    async fn test_evaluate_success_truncates_tool_tags() {
        use crate::testing::StubLlm;
        let response = r#"<think>evaluating</think>{"success": true, "confidence": 0.85, "reasoning": "Task completed", "issues": [], "suggestions": []}
 <tool_call>{"name": "verify"}"#;
        let llm = Arc::new(StubLlm::new(response));
        let reasoning = Reasoning::new(llm);
        let context = ReasoningContext::new().with_job("Test task");
        let eval = reasoning
            .evaluate_success(&context, "The job is done")
            .await
            .unwrap();
        assert!(eval.success);
        assert_eq!(eval.confidence, 0.85);
    }
    // ---- Issue #789: respond_with_tools recovered tool calls path ----
    #[tokio::test]
    async fn test_respond_with_tools_recovered_tool_calls_preserves_text() {
        use crate::testing::StubLlm;
        // StubLlm returns empty tool_calls + content with XML tool tags.
        // The recovery path should parse the tool call AND preserve text before it.
        let response = "Let me search for that.\n<tool_call>{\"name\": \"tool_list\", \"arguments\": {}}</tool_call>";
        let llm = Arc::new(StubLlm::new(response));
        let reasoning = Reasoning::new(llm);
        let context = ReasoningContext::new()
            .with_message(ChatMessage::user("list tools"))
            .with_tools(vec![ToolDefinition {
                name: "tool_list".to_string(),
                description: "Lists tools".to_string(),
                parameters: serde_json::json!({}),
            }]);
        let output = reasoning.respond_with_tools(&context).await.unwrap();
        match output.result {
            RespondResult::ToolCalls {
                tool_calls,
                content,
            } => {
                assert_eq!(tool_calls.len(), 1);
                assert_eq!(tool_calls[0].name, "tool_list");
                // Text before the tag should be preserved
                assert_eq!(content.as_deref(), Some("Let me search for that."));
            }
            RespondResult::Text(_) => {
                panic!("Expected recovered tool calls, got text");
            }
        }
    }
    #[tokio::test]
    async fn test_respond_with_tools_recovered_only_tag_content_is_none() {
        use crate::testing::StubLlm;
        // Content is ONLY a tool call tag — after truncation+cleaning, content should be None
        let response = "<tool_call>{\"name\": \"tool_list\", \"arguments\": {}}</tool_call>";
        let llm = Arc::new(StubLlm::new(response));
        let reasoning = Reasoning::new(llm);
        let context = ReasoningContext::new()
            .with_message(ChatMessage::user("list tools"))
            .with_tools(vec![ToolDefinition {
                name: "tool_list".to_string(),
                description: "Lists tools".to_string(),
                parameters: serde_json::json!({}),
            }]);
        let output = reasoning.respond_with_tools(&context).await.unwrap();
        match output.result {
            RespondResult::ToolCalls {
                tool_calls,
                content,
            } => {
                assert_eq!(tool_calls.len(), 1);
                assert_eq!(tool_calls[0].name, "tool_list");
                assert!(
                    content.is_none(),
                    "Content should be None when only tool tags present"
                );
            }
            RespondResult::Text(_) => {
                panic!("Expected recovered tool calls, got text");
            }
        }
    }
    // ---- Issue #789: OpenAI reasoning models negative test ----
    #[test]
    fn test_openai_reasoning_models_not_detected() {
        use crate::llm::reasoning_models::has_native_thinking;
        assert!(!has_native_thinking("o1"));
        assert!(!has_native_thinking("o1-mini"));
        assert!(!has_native_thinking("o1-preview"));
        assert!(!has_native_thinking("o3-mini"));
        assert!(!has_native_thinking("o4-mini"));
    }
    // ---- closing_tag_for() unit tests ----
    #[test]
    fn test_closing_tag_for_standard_tags() {
        assert_eq!(
            closing_tag_for("<tool_call>").as_deref(),
            Some("</tool_call>")
        );
        assert_eq!(
            closing_tag_for("<function_call>").as_deref(),
            Some("</function_call>")
        );
        assert_eq!(
            closing_tag_for("<tool_calls>").as_deref(),
            Some("</tool_calls>")
        );
    }
    #[test]
    fn test_closing_tag_for_space_suffixed_patterns() {
        // Patterns with trailing space (for attribute matching)
        assert_eq!(
            closing_tag_for("<tool_call ").as_deref(),
            Some("</tool_call>")
        );
        assert_eq!(
            closing_tag_for("<function_call ").as_deref(),
            Some("</function_call>")
        );
        assert_eq!(
            closing_tag_for("<tool_calls ").as_deref(),
            Some("</tool_calls>")
        );
    }
    #[test]
    fn test_closing_tag_for_pipe_delimited() {
        assert_eq!(
            closing_tag_for("<|tool_call|>").as_deref(),
            Some("<|/tool_call|>")
        );
        assert_eq!(
            closing_tag_for("<|function_call|>").as_deref(),
            Some("<|/function_call|>")
        );
        assert_eq!(
            closing_tag_for("<|tool_calls|>").as_deref(),
            Some("<|/tool_calls|>")
        );
    }
    #[test]
    fn test_closing_tag_for_covers_all_patterns() {
        // Every entry in TOOL_TAG_PATTERNS must produce a closing tag
        for pattern in TOOL_TAG_PATTERNS {
            assert!(
                closing_tag_for(pattern).is_some(),
                "closing_tag_for({:?}) returned None",
                pattern
            );
        }
    }
    // ---- truncation with multiple tags: first closed, second unclosed ----
    #[test]
    fn test_truncate_mixed_closed_then_unclosed_different_types() {
        let input = "Text <function_call>{}</function_call> middle <tool_call>{\"name\": \"x\"}";
        // function_call is closed → skipped. tool_call is unclosed → truncated.
        assert_eq!(
            truncate_at_tool_tags(input),
            "Text <function_call>{}</function_call> middle "
        );
    }
 }
@@ -1,134 +0,0 @@
 //! Reasoning/thinking model detection utilities.
 //!
 //! Models with native thinking support produce structured chain-of-thought
 //! via `reasoning_content` fields or built-in `<think>` tags. Injecting
 //! IronClaw's own `<think>/<final>` format instructions into the system
 //! prompt collides with these models' native behavior, causing:
 //! - Thinking-only responses with no visible content
 //! - Double-wrapped thinking tags that confuse response cleaning
 //!
 //! When a model has native thinking, we skip the `<think>/<final>` prompt
 //! injection and let the model use its own format. The response cleaning
 //! pipeline already handles stripping all known thinking tag variants.
 //!
 //! ## Design note: why match broadly (e.g. all Qwen3)?
 //!
 //! Some families (Qwen3) have ALL variants trained with native `<think>` tags,
 //! even tiny models like 0.6B. Thinking can be disabled at inference time via
 //! `enable_thinking=false`, but we can't detect that from the model name alone.
 //! We err on the safe side: skip injection for all variants because:
 //! - False negative (inject when model thinks natively) = broken responses
 //! - False positive (skip injection for non-thinking model) = less structured
 //!   but working responses
 //!
 //! For families where only SOME variants reason (GLM-4), we match specific
 //! sub-families (glm-z1, glm-4-plus) to avoid false positives.
 /// Known model families with native thinking/reasoning support.
 ///
 /// These models produce chain-of-thought reasoning either via a dedicated
 /// `reasoning_content` response field or via built-in `<think>` tags that
 /// the model was trained to emit without prompt injection.
 const NATIVE_THINKING_PATTERNS: &[&str] = &[
    // Qwen3 family — ALL variants (0.6B through 235B) emit native <think> tags
    // by default. Thinking can be toggled via `enable_thinking` parameter or
    // `/think` `/no_think` soft switches, but the default is ON and we can't
    // detect the runtime setting from the model name.
    "qwen3",
    // QwQ is Qwen's dedicated reasoning model (based on Qwen2.5-32B + RL).
    // Always thinks, no disable toggle.
    "qwq",
    // DeepSeek reasoning models — native reasoning_content field
    "deepseek-r1",
    "deepseek-reasoner",
    // GLM reasoning variants only (glm-4-flash, glm-4-air, glm-4v do NOT reason)
    "glm-z1",
    "glm-4-plus",
    "glm-5",
    // Nanbeige reasoning models
    "nanbeige",
    // Step reasoning models (3.5+ have native thinking; step-3 base does not)
    "step-3.5",
    // MiniMax reasoning models
    "minimax-m2",
 ];
 /// Check if a model name indicates native thinking/reasoning support.
 ///
 /// Models that return `true` should NOT have IronClaw's `<think>/<final>`
 /// format instructions injected into their system prompt, as this collides
 /// with their built-in reasoning behavior.
 ///
 /// Note: this is a best-effort heuristic based on model name. Some models
 /// support toggling thinking at runtime (e.g. Qwen3's `enable_thinking`),
 /// which we cannot detect here. We default to assuming thinking is ON for
 /// models that have it, since that's the default behavior.
 pub fn has_native_thinking(model: &str) -> bool {
    let lower = model.to_ascii_lowercase();
    NATIVE_THINKING_PATTERNS.iter().any(|p| lower.contains(p))
 }
 #[cfg(test)]
 mod tests {
    use super::*;
    #[test]
    fn detects_qwen3_models() {
        // All Qwen3 variants have native thinking (even small ones)
        assert!(has_native_thinking("qwen3-coder-next-80b"));
        assert!(has_native_thinking("Qwen3.5-35B"));
        assert!(has_native_thinking("qwen3-0.6b"));
        assert!(has_native_thinking("qwen3:8b"));
        assert!(has_native_thinking("qwen3-30b-a3b"));
        // Ollama-style tag format
        assert!(has_native_thinking("qwen3-coder:latest"));
    }
    #[test]
    fn detects_qwq() {
        assert!(has_native_thinking("qwq-32b"));
        assert!(has_native_thinking("QwQ-32B-Preview"));
    }
    #[test]
    fn detects_deepseek_reasoning() {
        assert!(has_native_thinking("deepseek-r1-distill-qwen-32b"));
        assert!(has_native_thinking("deepseek-reasoner"));
    }
    #[test]
    fn detects_glm_reasoning_variants() {
        assert!(has_native_thinking("glm-z1-airx"));
        assert!(has_native_thinking("glm-4-plus"));
        assert!(has_native_thinking("GLM-5"));
    }
    #[test]
    fn detects_other_reasoning_models() {
        assert!(has_native_thinking("nanbeige-4.1-3b"));
        assert!(has_native_thinking("step-3.5-flash-197b"));
        assert!(has_native_thinking("minimax-m2.5-139b"));
    }
    #[test]
    fn rejects_non_reasoning_models() {
        assert!(!has_native_thinking("gpt-4o"));
        assert!(!has_native_thinking("claude-3-5-sonnet"));
        assert!(!has_native_thinking("llama-3.1-70b"));
        assert!(!has_native_thinking("mistral-7b"));
        assert!(!has_native_thinking("gemini-2.0-flash"));
    }
    #[test]
    fn rejects_non_reasoning_variants_in_same_family() {
        // Qwen2.5 does NOT have native thinking (only Qwen3/QwQ do)
        assert!(!has_native_thinking("qwen2.5:7b"));
        assert!(!has_native_thinking("qwen2.5-instruct"));
        // GLM-4 base variants do NOT have reasoning_content
        assert!(!has_native_thinking("glm-4-flash"));
        assert!(!has_native_thinking("glm-4-air"));
        assert!(!has_native_thinking("glm-4v"));
        // step-3 base does not reason (only 3.5+)
        assert!(!has_native_thinking("step-3-mini"));
    }
 }
@@ -113,33 +113,6 @@ impl SetupHint {
    }
 }
 /// Validates unsupported_params during deserialization.
 ///
 /// Only allows: "temperature", "max_tokens", "stop_sequences".
 /// Invalid parameter names cause a deserialization error.
 mod unsupported_params_de {
    use serde::{Deserialize, Deserializer};
    const VALID_PARAMS: &[&str] = &["temperature", "max_tokens", "stop_sequences"];
    pub fn deserialize<'de, D>(deserializer: D) -> Result<Vec<String>, D::Error>
    where
        D: Deserializer<'de>,
    {
        let params: Vec<String> = Deserialize::deserialize(deserializer)?;
        for param in &params {
            if !VALID_PARAMS.contains(&param.as_str()) {
                return Err(serde::de::Error::custom(format!(
                    "unsupported parameter name '{}': must be one of: {}",
                    param,
                    VALID_PARAMS.join(", ")
                )));
            }
        }
        Ok(params)
    }
 }
 /// Declarative definition of an LLM provider.
 ///
 /// One JSON object in `providers.json` maps to one `ProviderDefinition`.
@@ -179,12 +152,6 @@ pub struct ProviderDefinition {
    /// Setup wizard hints.
    #[serde(default)]
    pub setup: Option<SetupHint>,
    /// Parameter names that this provider does not support (e.g., `["temperature"]`).
    /// Supported keys: `"temperature"`, `"max_tokens"`, `"stop_sequences"`.
    /// Listed parameters are stripped from requests before sending to avoid 400 errors.
    /// Invalid parameter names cause a deserialization error.
    #[serde(default, deserialize_with = "unsupported_params_de::deserialize")]
    pub unsupported_params: Vec<String>,
 }
 /// Registry of known LLM providers.
@@ -411,7 +378,6 @@ mod tests {
            description: "Custom tinfoil".to_string(),
            extra_headers_env: None,
            setup: None,
            unsupported_params: vec![],
        });
        let registry = ProviderRegistry::new(all);
        let tf = registry.find("tinfoil").expect("tinfoil should exist");
@@ -551,7 +517,6 @@ mod tests {
            description: "No setup".to_string(),
            extra_headers_env: None,
            setup: None, // no setup hint
            unsupported_params: vec![],
        }];
        let registry = ProviderRegistry::new(providers.clone());
@@ -581,7 +546,6 @@ mod tests {
                can_list_models: false,
                models_filter: None,
            }),
            unsupported_params: vec![],
        });
        let registry = ProviderRegistry::new(providers);
@@ -623,7 +587,6 @@ mod tests {
                    can_list_models: false,
                    models_filter: None,
                }),
                unsupported_params: vec![],
            },
            // User override removes setup
            ProviderDefinition {
@@ -640,7 +603,6 @@ mod tests {
                description: "No setup now".to_string(),
                extra_headers_env: None,
                setup: None,
                unsupported_params: vec![],
            },
        ];
@@ -678,7 +640,6 @@ mod tests {
                    display_name: "A".to_string(),
                    can_list_models: false,
                }),
                unsupported_params: vec![],
            },
            ProviderDefinition {
                id: "bbb".to_string(),
@@ -697,7 +658,6 @@ mod tests {
                    display_name: "B".to_string(),
                    can_list_models: false,
                }),
                unsupported_params: vec![],
            },
            ProviderDefinition {
                id: "ccc".to_string(),
@@ -716,7 +676,6 @@ mod tests {
                    display_name: "C".to_string(),
                    can_list_models: false,
                }),
                unsupported_params: vec![],
            },
            // User override for B
            ProviderDefinition {
@@ -736,7 +695,6 @@ mod tests {
                    display_name: "B".to_string(),
                    can_list_models: false,
                }),
                unsupported_params: vec![],
            },
        ];
@@ -750,81 +708,6 @@ mod tests {
        );
    }
    #[test]
    fn test_unsupported_params_deserialized() {
        let providers: Vec<ProviderDefinition> =
            serde_json::from_str(include_str!("../../providers.json")).unwrap();
        // Tinfoil should have temperature in unsupported_params
        let tinfoil = providers.iter().find(|p| p.id == "tinfoil").unwrap();
        assert!(
            tinfoil
                .unsupported_params
                .contains(&"temperature".to_string()),
            "tinfoil should have 'temperature' in unsupported_params"
        );
        // OpenAI should also have temperature in unsupported_params
        let openai = providers.iter().find(|p| p.id == "openai").unwrap();
        assert!(
            openai
                .unsupported_params
                .contains(&"temperature".to_string()),
            "openai should have 'temperature' in unsupported_params"
        );
        // Providers without the field in JSON should deserialize to empty vec
        let groq = providers.iter().find(|p| p.id == "groq").unwrap();
        assert!(
            groq.unsupported_params.is_empty(),
            "groq should have empty unsupported_params (field absent in JSON)"
        );
        // All entries should only contain valid param names
        // (Invalid names should be rejected at deserialization time)
        for def in &providers {
            for param in &def.unsupported_params {
                assert!(
                    !param.is_empty(),
                    "{}: unsupported_params contains empty string",
                    def.id
                );
                assert!(
                    matches!(
                        param.as_str(),
                        "temperature" | "max_tokens" | "stop_sequences"
                    ),
                    "{}: unsupported_params contains invalid parameter '{}'",
                    def.id,
                    param
                );
            }
        }
    }
    #[test]
    fn test_unsupported_params_validation_rejects_invalid() {
        // Invalid parameter names should cause deserialization error
        let invalid_json = r#"[{
            "id": "test",
            "protocol": "open_ai_completions",
            "model_env": "TEST_MODEL",
            "default_model": "test-model",
            "description": "Test provider",
            "unsupported_params": ["temperrature"]
        }]"#;
        let result: Result<Vec<ProviderDefinition>, _> = serde_json::from_str(invalid_json);
        assert!(
            result.is_err(),
            "should reject invalid parameter name 'temperrature'"
        );
        assert!(
            result.err().unwrap().to_string().contains("temperrature"),
            "error message should mention the invalid parameter"
        );
    }
    #[test]
    fn test_all_builtin_api_key_providers_have_api_key_env() {
        // Every built-in provider with SetupHint::ApiKey must have api_key_env
@@ -205,7 +205,7 @@ impl LlmProvider for CachedProvider {
                    let hit_count = entry.hit_count;
                    // Clone now so we can release the mutable borrow before stats.
                    let cached_response = entry.response.clone();
-                    tracing::trace!(hits = hit_count, "response cache hit");
+                    tracing::debug!(hits = hit_count, "response cache hit");
                    // Drop the mutable borrow of `entry` before reading `guard` immutably.
                    let _ = entry;
                    let total_hits = self.total_hit_count.fetch_add(1, Ordering::Relaxed) + 1;
@@ -28,8 +28,7 @@ use crate::llm::error::LlmError;
 use crate::llm::provider::{
    ChatMessage, CompletionRequest, CompletionResponse, FinishReason, LlmProvider,
    ToolCall as IronToolCall, ToolCompletionRequest, ToolCompletionResponse,
-    ToolDefinition as IronToolDefinition, strip_unsupported_completion_params,
+    ToolDefinition as IronToolDefinition,
    strip_unsupported_tool_params,
 };
 /// Adapter that wraps a rig-core `CompletionModel` and implements `LlmProvider`.
@@ -43,9 +42,6 @@ pub struct RigAdapter<M: CompletionModel> {
    /// via `additional_params` for Anthropic automatic caching. Also controls
    /// the cost multiplier for cache-creation tokens.
    cache_retention: CacheRetention,
    /// Parameter names that this provider does not support (e.g., `"temperature"`).
    /// These are stripped from requests before sending to avoid 400 errors.
    unsupported_params: HashSet<String>,
 }
 impl<M: CompletionModel> RigAdapter<M> {
@@ -60,7 +56,6 @@ impl<M: CompletionModel> RigAdapter<M> {
            input_cost,
            output_cost,
            cache_retention: CacheRetention::None,
            unsupported_params: HashSet::new(),
        }
    }
@@ -89,25 +84,6 @@ impl<M: CompletionModel> RigAdapter<M> {
        }
        self
    }
    /// Set the list of unsupported parameter names for this provider.
    ///
    /// Parameters in this set are stripped from requests before sending.
    /// Supported parameter names: `"temperature"`, `"max_tokens"`, `"stop_sequences"`.
    pub fn with_unsupported_params(mut self, params: Vec<String>) -> Self {
        self.unsupported_params = params.into_iter().collect();
        self
    }
    /// Strip unsupported fields from a `CompletionRequest` in place.
    fn strip_unsupported_completion_params(&self, req: &mut CompletionRequest) {
        strip_unsupported_completion_params(&self.unsupported_params, req);
    }
    /// Strip unsupported fields from a `ToolCompletionRequest` in place.
    fn strip_unsupported_tool_params(&self, req: &mut ToolCompletionRequest) {
        strip_unsupported_tool_params(&self.unsupported_params, req);
    }
 }
 // -- Type conversion helpers --
@@ -563,10 +539,7 @@ where
        }
    }
-    async fn complete(
+    async fn complete(&self, request: CompletionRequest) -> Result<CompletionResponse, LlmError> {
        &self,
        mut request: CompletionRequest,
    ) -> Result<CompletionResponse, LlmError> {
        if let Some(requested_model) = request.model.as_deref()
            && requested_model != self.model_name.as_str()
        {
@@ -577,8 +550,6 @@ where
            );
        }
        self.strip_unsupported_completion_params(&mut request);
        let mut messages = request.messages;
        crate::llm::provider::sanitize_tool_messages(&mut messages);
        let (preamble, history) = convert_messages(&messages);
@@ -628,7 +599,7 @@ where
    async fn complete_with_tools(
        &self,
-        mut request: ToolCompletionRequest,
+        request: ToolCompletionRequest,
    ) -> Result<ToolCompletionResponse, LlmError> {
        if let Some(requested_model) = request.model.as_deref()
            && requested_model != self.model_name.as_str()
@@ -640,8 +611,6 @@ where
            );
        }
        self.strip_unsupported_tool_params(&mut request);
        let known_tool_names: HashSet<String> =
            request.tools.iter().map(|t| t.name.clone()).collect();
@@ -1187,97 +1156,4 @@ mod tests {
        assert!(!supports_prompt_cache("gpt-4o"));
        assert!(!supports_prompt_cache("llama3"));
    }
    #[test]
    fn test_with_unsupported_params_populates_set() {
        use rig::client::CompletionClient;
        use rig::providers::openai;
        let client: openai::Client = openai::Client::builder()
            .api_key("test-key")
            .base_url("http://localhost:0")
            .build()
            .unwrap();
        let client = client.completions_api();
        let model = client.completion_model("test-model");
        let adapter = RigAdapter::new(model, "test-model")
            .with_unsupported_params(vec!["temperature".to_string()]);
        assert!(adapter.unsupported_params.contains("temperature"));
        assert!(!adapter.unsupported_params.contains("max_tokens"));
    }
    #[test]
    fn test_strip_unsupported_completion_params() {
        use rig::client::CompletionClient;
        use rig::providers::openai;
        let client: openai::Client = openai::Client::builder()
            .api_key("test-key")
            .base_url("http://localhost:0")
            .build()
            .unwrap();
        let client = client.completions_api();
        let model = client.completion_model("test-model");
        let adapter = RigAdapter::new(model, "test-model").with_unsupported_params(vec![
            "temperature".to_string(),
            "stop_sequences".to_string(),
        ]);
        let mut req = CompletionRequest::new(vec![ChatMessage::user("hi")]);
        req.temperature = Some(0.7);
        req.max_tokens = Some(100);
        req.stop_sequences = Some(vec!["STOP".to_string()]);
        adapter.strip_unsupported_completion_params(&mut req);
        assert!(req.temperature.is_none(), "temperature should be stripped");
        assert_eq!(req.max_tokens, Some(100), "max_tokens should be preserved");
        assert!(
            req.stop_sequences.is_none(),
            "stop_sequences should be stripped"
        );
    }
    #[test]
    fn test_strip_unsupported_tool_params() {
        use rig::client::CompletionClient;
        use rig::providers::openai;
        let client: openai::Client = openai::Client::builder()
            .api_key("test-key")
            .base_url("http://localhost:0")
            .build()
            .unwrap();
        let client = client.completions_api();
        let model = client.completion_model("test-model");
        let adapter = RigAdapter::new(model, "test-model")
            .with_unsupported_params(vec!["temperature".to_string(), "max_tokens".to_string()]);
        let mut req = ToolCompletionRequest::new(vec![ChatMessage::user("hi")], vec![]);
        req.temperature = Some(0.5);
        req.max_tokens = Some(200);
        adapter.strip_unsupported_tool_params(&mut req);
        assert!(req.temperature.is_none(), "temperature should be stripped");
        assert!(req.max_tokens.is_none(), "max_tokens should be stripped");
    }
    #[test]
    fn test_unsupported_params_empty_by_default() {
        use rig::client::CompletionClient;
        use rig::providers::openai;
        let client: openai::Client = openai::Client::builder()
            .api_key("test-key")
            .base_url("http://localhost:0")
            .build()
            .unwrap();
        let client = client.completions_api();
        let model = client.completion_model("test-model");
        let adapter = RigAdapter::new(model, "test-model");
        assert!(adapter.unsupported_params.is_empty());
    }
 }
@@ -770,7 +770,7 @@ impl SmartRoutingProvider {
                }
            };
            let complexity = TaskComplexity::from(tier);
-            tracing::trace!(
+            tracing::debug!(
                %tier,
                ?complexity,
                "Smart routing: explicit tier hint"
@@ -782,7 +782,7 @@ impl SmartRoutingProvider {
        for po in DEFAULT_OVERRIDES.iter() {
            if po.regex.is_match(last_user_msg) {
                let complexity = TaskComplexity::from(po.tier);
-                tracing::trace!(
+                tracing::debug!(
                    tier = %po.tier,
                    ?complexity,
                    "Smart routing: pattern override matched"
@@ -798,7 +798,7 @@ impl SmartRoutingProvider {
            &self.domain_regex,
        );
        let complexity = TaskComplexity::from(breakdown.tier);
-        tracing::trace!(
+        tracing::debug!(
            score = breakdown.total,
            tier = %breakdown.tier,
            ?complexity,
@@ -872,7 +872,7 @@ impl LlmProvider for SmartRoutingProvider {
        match complexity {
            TaskComplexity::Simple => {
-                tracing::trace!(
+                tracing::debug!(
                    model = %self.cheap.model_name(),
                    "Smart routing: Simple task -> cheap model"
                );
@@ -880,7 +880,7 @@ impl LlmProvider for SmartRoutingProvider {
                self.cheap.complete(request).await
            }
            TaskComplexity::Complex => {
-                tracing::trace!(
+                tracing::debug!(
                    model = %self.primary.model_name(),
                    "Smart routing: Complex task -> primary model"
                );
@@ -889,7 +889,7 @@ impl LlmProvider for SmartRoutingProvider {
            }
            TaskComplexity::Moderate => {
                if self.config.cascade_enabled {
-                    tracing::trace!(
+                    tracing::debug!(
                        model = %self.cheap.model_name(),
                        "Smart routing: Moderate task -> cheap model (cascade enabled)"
                    );
@@ -913,7 +913,7 @@ impl LlmProvider for SmartRoutingProvider {
                    }
                } else {
                    // Without cascade, moderate tasks go to cheap model
-                    tracing::trace!(
+                    tracing::debug!(
                        model = %self.cheap.model_name(),
                        "Smart routing: Moderate task -> cheap model (cascade disabled)"
                    );
@@ -931,7 +931,7 @@ impl LlmProvider for SmartRoutingProvider {
    ) -> Result<ToolCompletionResponse, LlmError> {
        self.stats.total_requests.fetch_add(1, Ordering::Relaxed);
        self.stats.primary_requests.fetch_add(1, Ordering::Relaxed);
-        tracing::trace!(
+        tracing::debug!(
            model = %self.primary.model_name(),
            "Smart routing: Tool use -> primary model (always)"
        );
@@ -322,16 +322,10 @@ async fn async_main() -> anyhow::Result<()> {
    // Add HTTP channel if configured and not CLI-only mode.
    let mut webhook_server_addr: Option<std::net::SocketAddr> = None;
    #[cfg(unix)]
    let mut http_channel_state: Option<Arc<ironclaw::channels::HttpChannelState>> = None;
    if !cli.cli_only
        && let Some(ref http_config) = config.channels.http
    {
        let http_channel = HttpChannel::new(http_config.clone());
        #[cfg(unix)]
        {
            http_channel_state = Some(http_channel.shared_state());
        }
        webhook_routes.push(http_channel.routes());
        let (host, port) = http_channel.addr();
        webhook_server_addr = Some(
@@ -349,9 +343,7 @@ async fn async_main() -> anyhow::Result<()> {
    }
    // Start the unified webhook server if any routes were registered.
-    let webhook_server: Option<Arc<tokio::sync::Mutex<WebhookServer>>> = if !webhook_routes
+    let mut webhook_server = if !webhook_routes.is_empty() {
        .is_empty()
    {
        let addr =
            webhook_server_addr.unwrap_or_else(|| std::net::SocketAddr::from(([0, 0, 0, 0], 8080)));
        if addr.ip().is_unspecified() {
@@ -366,7 +358,7 @@ async fn async_main() -> anyhow::Result<()> {
            server.add_routes(routes);
        }
        server.start().await?;
-        Some(Arc::new(tokio::sync::Mutex::new(server)))
+        Some(server)
    } else {
        None
    };
@@ -564,39 +556,28 @@ async fn async_main() -> anyhow::Result<()> {
            .await;
        tracing::debug!("Channel runtime wired into extension manager for hot-activation");
-        // Auto-activate WASM channels that were active in a previous session.
+        // Auto-activate channels that were active in a previous session.
        // Relay channels are handled separately below via restore_relay_channels().
        let persisted = ext_mgr.load_persisted_active_channels().await;
        for name in &persisted {
-            if active_at_startup.contains(name) || ext_mgr.is_relay_channel(name).await {
+            if !active_at_startup.contains(name) {
                continue;
            }
                match ext_mgr.activate(name).await {
                    Ok(result) => {
                        tracing::debug!(
                            channel = %name,
                            message = %result.message,
-                        "Auto-activated persisted WASM channel"
+                            "Auto-activated persisted channel"
                        );
                    }
                    Err(e) => {
                        tracing::warn!(
                            channel = %name,
                            error = %e,
-                        "Failed to auto-activate persisted WASM channel"
+                            "Failed to auto-activate persisted channel"
                        );
                    }
                }
            }
        }
    // Ensure the relay channel manager is always set (even without WASM runtime),
    // then restore any persisted relay channels.
    if let Some(ref ext_mgr) = components.extension_manager {
        ext_mgr
            .set_relay_channel_manager(Arc::clone(&channels))
            .await;
        ext_mgr.restore_relay_channels().await;
    }
    // Wire SSE sender into extension manager for broadcasting status events.
@@ -620,13 +601,6 @@ async fn async_main() -> anyhow::Result<()> {
    // Clone context_manager for the reaper before it's moved into Agent::new()
    let reaper_context_manager = Arc::clone(&components.context_manager);
    // Capture db reference for SIGHUP handler before it's moved into AgentDeps (Unix only)
    #[cfg(unix)]
    let sighup_settings_store: Option<Arc<dyn ironclaw::db::SettingsStore>> = components
        .db
        .as_ref()
        .map(|db| Arc::clone(db) as Arc<dyn ironclaw::db::SettingsStore>);
    let deps = AgentDeps {
        store: components.db,
        llm: components.llm,
@@ -687,157 +661,10 @@ async fn async_main() -> anyhow::Result<()> {
        agent.set_routine_engine_slot(slot);
    }
    // Prepare SIGHUP handler for hot-reloading HTTP webhook config
    // Broadcast channel for clean shutdown of background tasks
    let (shutdown_tx, _) = tokio::sync::broadcast::channel::<()>(1);
    #[cfg(unix)]
    {
        use ironclaw::channels::ChannelSecretUpdater;
        // Collect all channels that support secret updates
        let mut secret_updaters: Vec<Arc<dyn ChannelSecretUpdater>> = Vec::new();
        if let Some(ref state) = http_channel_state {
            secret_updaters.push(Arc::clone(state) as Arc<dyn ChannelSecretUpdater>);
        }
        let sighup_webhook_server = webhook_server.clone();
        let sighup_settings_store_clone = sighup_settings_store.clone();
        let sighup_secrets_store = components.secrets_store.clone();
        let mut shutdown_rx = shutdown_tx.subscribe();
        tokio::spawn(async move {
            use tokio::signal::unix::{SignalKind, signal};
            let mut sighup = match signal(SignalKind::hangup()) {
                Ok(s) => s,
                Err(e) => {
                    tracing::warn!("Failed to register SIGHUP handler: {}", e);
                    return;
                }
            };
            loop {
                // Exit loop on shutdown signal or when SIGHUP is received
                tokio::select! {
                    _ = shutdown_rx.recv() => {
                        tracing::debug!("SIGHUP handler shutting down");
                        break;
                    }
                    _ = sighup.recv() => {
                        // Handle SIGHUP signal
                    }
                }
                tracing::info!("SIGHUP received — reloading HTTP webhook config");
                // Inject channel secrets from database into thread-safe overlay
                // (similar to inject_llm_keys_from_secrets for LLM providers)
                if let Some(ref secrets_store) = sighup_secrets_store {
                    // Inject HTTP webhook secret from encrypted store
                    if let Ok(webhook_secret) = secrets_store
                        .get_decrypted("default", "http_webhook_secret")
                        .await
                    {
                        // Thread-safe: Uses INJECTED_VARS mutex instead of unsafe std::env::set_var
                        // Config::from_env() will read from the overlay via optional_env()
                        ironclaw::config::inject_single_var(
                            "HTTP_WEBHOOK_SECRET",
                            webhook_secret.expose(),
                        );
                        tracing::debug!("Injected HTTP_WEBHOOK_SECRET from secrets store");
                    }
                }
                // Reload config (now with secrets injected into environment)
                let new_config = match &sighup_settings_store_clone {
                    Some(store) => {
                        ironclaw::config::Config::from_db(store.as_ref(), "default").await
                    }
                    None => ironclaw::config::Config::from_env().await,
                };
                let new_config = match new_config {
                    Ok(c) => c,
                    Err(e) => {
                        tracing::error!("SIGHUP config reload failed: {}", e);
                        continue;
                    }
                };
                let new_http = match new_config.channels.http {
                    Some(c) => c,
                    None => {
                        tracing::warn!("SIGHUP: HTTP channel no longer configured, skipping");
                        continue;
                    }
                };
                // Compute new socket addr
                let new_addr: std::net::SocketAddr =
                    match format!("{}:{}", new_http.host, new_http.port).parse() {
                        Ok(a) => a,
                        Err(e) => {
                            tracing::error!("SIGHUP: invalid addr in config: {}", e);
                            continue;
                        }
                    };
                // Restart listener if addr changed.
                // Minimize lock scope: acquire, read old addr, release, then restart.
                let mut restart_failed = false;
                if let Some(ref ws_arc) = sighup_webhook_server {
                    let old_addr = {
                        let ws = ws_arc.lock().await;
                        ws.current_addr()
                    }; // Lock released here
                    if old_addr != new_addr {
                        tracing::info!(
                            "SIGHUP: HTTP addr {} -> {}, restarting listener",
                            old_addr,
                            new_addr
                        );
                        // NOTE: Lock is held across restart_with_addr().await. This is
                        // acceptable because SIGHUP is infrequent and restart is fast. A full
                        // fix would require refactoring restart_with_addr to separate state
                        // mutation from async I/O.
                        let mut ws = ws_arc.lock().await;
                        match ws.restart_with_addr(new_addr).await {
                            Ok(()) => {
                                tracing::info!("SIGHUP: webhook server restarted on {}", new_addr);
                            }
                            Err(e) => {
                                tracing::error!("SIGHUP: listener restart failed: {}", e);
                                restart_failed = true;
                            }
                        }
                    } else {
                        tracing::debug!("SIGHUP: addr unchanged ({})", old_addr);
                    }
                }
                // Update secrets in all configured channels (if restart succeeded or wasn't needed)
                if !restart_failed {
                    use secrecy::{ExposeSecret, SecretString};
                    let new_secret = new_http
                        .webhook_secret
                        .as_ref()
                        .map(|s| SecretString::from(s.expose_secret().to_string()));
                    // Update all channels that support secret swapping
                    for updater in &secret_updaters {
                        updater.update_secret(new_secret.clone()).await;
                    }
                }
            }
        });
    }
    agent.run().await?;
    // ── Shutdown ────────────────────────────────────────────────────────
    // Signal background tasks (SIGHUP handler, etc.) to gracefully shut down
    let _ = shutdown_tx.send(());
    // Shut down all stdio MCP server child processes.
    components.mcp_process_manager.shutdown_all().await;
@@ -848,8 +675,8 @@ async fn async_main() -> anyhow::Result<()> {
        tracing::warn!("Failed to write LLM trace: {}", e);
    }
-    if let Some(ref ws_arc) = webhook_server {
+    if let Some(ref mut server) = webhook_server {
-        ws_arc.lock().await.shutdown().await;
+        server.shutdown().await;
    }
    if let Some(tunnel) = active_tunnel {
--- a/Show More
+++ b/Show More