These tests guard against catastrophic regex backtracking (seconds/minutes),
not 12ms differences. CI runners with coverage instrumentation (cargo-llvm-cov)
consistently exceed the 100ms threshold due to overhead, causing flaky failures.
500ms still catches real regressions while tolerating CI variability.
[skip-regression-check]
Co-authored-by: Claude Opus 4.6 (1M context) <[email protected]>
* fix: Rate limiter returns retry after None instead of a duration
linter fix
* review fixes
* fix: rate limiter returns None for retry_after duration
Add regression test to src/llm/retry.rs that verifies RateLimited errors
always have a fallback duration (never None) due to the 60-second fallback
applied in all rate limit error creation sites (nearai_chat.rs,
anthropic_oauth.rs, embeddings.rs).
The production code fix adds `.or(Some(Duration::from_secs(60)))` to ensure
the error message never displays "retry after None" to the user.
[skip-regression-check]
Co-Authored-By: Claude Haiku 4.5 <[email protected]>
---------
Co-authored-by: Claude Haiku 4.5 <[email protected]>
- Remove duplicate build_nearai_model_fetch_config() definition from setup/wizard.rs
(function already exists in llm/models.rs and is imported)
- Add missing cheap_model and smart_routing_cascade fields to LlmConfig
initializer in build_nearai_model_fetch_config() (llm/models.rs)
- Pass request_timeout_secs to create_registry_provider() call
(llm/mod.rs:432)
All clippy checks pass with zero warnings (--no-default-features --features libsql).
Co-Authored-By: Claude Haiku 4.5 <[email protected]>
Resolved merge conflicts in 5 files:
1. src/agent/job_monitor.rs - Used is_internal flag approach (HEAD) for safe internal message marking. Removed metadata-based approach which could be spoofed by external channels.
2. src/agent/agent_loop.rs - Used is_internal check (HEAD) for routing internal messages, consistent with security model where is_internal field cannot be spoofed.
3. src/agent/dispatcher.rs - Included notify_metadata in job context (main), needed for job routing through JobMonitorRoute.
4. src/setup/wizard.rs - Added build_nearai_model_fetch_config() function (main) for model selection during setup.
5. src/tools/builtin/job.rs - Used both comments from HEAD (clarifying notify_channel and notify_user logic) while removing metadata field from JobMonitorRoute (consistent with job_monitor.rs).
All conflicts resolved with security-first approach: use is_internal boolean field for internal message marking (cannot be spoofed), while passing routing metadata through context.
Co-Authored-By: Claude Haiku 4.5 <[email protected]>
* feat: add LLM_CHEAP_MODEL for generic smart routing across all backends
Add generic cheap model support that works with any LLM backend, not just
NearAI. New env vars: LLM_CHEAP_MODEL (cheap model for any backend) and
SMART_ROUTING_CASCADE (top-level cascade flag).
Resolution order: LLM_CHEAP_MODEL > NEARAI_CHEAP_MODEL (backward compat).
Registry-based providers (OpenAI, Anthropic, Groq, etc.) clone their
RegistryProviderConfig with the cheap model swapped in. Bedrock returns
an explicit error (not yet supported). All error paths use ok_or_else
with proper LlmError variants -- no unwrap/expect in production code.
* refactor: address Gemini review — remove unnecessary async, extract cheap_model_name()
- Remove async from create_cheap_provider_for_backend() and
create_cheap_llm_provider() — neither contains .await calls
- Extract duplicated cheap model resolution logic into
LlmConfig::cheap_model_name() helper method (DRY)
- Revert tests from tokio::test async back to sync #[test]
- Add test_cheap_model_name_resolution() unit test for the helper
---------
Co-authored-by: SMKRV <[email protected]>
Route messages and replies to the correct Telegram forum topic via
message_thread_id. Key behaviors:
- Parse message_thread_id, is_topic_message, is_forum from incoming updates
- Thread agent sessions by "chat_id:topic_id" for forum groups only
(non-forum reply threads are excluded via is_forum guard)
- Pass message_thread_id through all send methods (text, photo, document)
- Normalize thread_id=1 (General topic) to None for sendMessage/sendPhoto/
sendDocument since Telegram rejects it, but preserve it for sendChatAction
where Telegram requires it for typing indicators
- Hoist bot_username workspace read to avoid duplicate WASM host call per
group message
Co-authored-by: Claude Opus 4.6 (1M context) <[email protected]>
* feat(orchestrator): read ORCHESTRATOR_PORT env var for configurable API port
The orchestrator internal API port was hardcoded to 50051 in two places
(ContainerJobConfig and OrchestratorApi::start call), making it impossible
to run multiple IronClaw instances on the same host — the second instance
fails with "Address already in use".
NETWORK_SECURITY.md already documents ORCHESTRATOR_PORT as configurable,
and ContainerJobConfig.orchestrator_port is propagated to worker containers
via IRONCLAW_ORCHESTRATOR_URL, but the env var was never actually read.
Extract resolve_orchestrator_port() that reads ORCHESTRATOR_PORT and falls
back to 50051. Includes tests for valid, invalid, and out-of-range values.
* test: add ENV_LOCK mutex for env-var test serialization
Address Gemini review: add std::sync::Mutex to serialize env var access
across test threads. Keep unsafe blocks — required in Rust edition 2024
where std::env::set_var/remove_var are unsafe functions.
---------
Co-authored-by: SMKRV <[email protected]>
* feat(transcription): add Chat Completions API provider for audio transcription
The existing transcription pipeline only supports the OpenAI Whisper API
(/v1/audio/transcriptions with multipart upload). Providers like OpenRouter
expose audio transcription through the Chat Completions API instead, using
base64-encoded audio in the `input_audio` content type.
Add `ChatCompletionsTranscriptionProvider` that sends audio as base64 in
a chat completion request and extracts the transcript from the response.
Compatible with OpenRouter, OpenAI GPT-4o-audio, and any provider that
supports audio input via Chat Completions.
Config changes:
- TRANSCRIPTION_PROVIDER=chat_completions selects the new provider
- TRANSCRIPTION_API_KEY overrides provider-specific keys
- LLM_API_KEY used as fallback for chat_completions provider
- Default model per provider (whisper-1 for openai, gemini-2.0-flash for
chat_completions)
* style: address review feedback — formatting, idiomatic patterns
- Fix rustfmt formatting for provider constructor chain
- Use or_else for resolve_api_key priority chain (Gemini review)
- Use trim_end_matches('/') instead of while loop (Gemini review)
---------
Co-authored-by: SMKRV <[email protected]>
* fix(jobs): make completed->completed transition idempotent to prevent race errors
Both execution_loop and the worker wrapper in execute() can race to call
mark_completed(). Previously the second call hit "Cannot transition from
completed to completed" and errored the job despite successful completion.
This narrowly allows only the Completed->Completed self-transition as
idempotent (early return with debug log, no duplicate history entry).
All other self-transitions remain rejected to preserve state machine
strictness.
Co-Authored-By: Claude Opus 4.6 <[email protected]>
* style: fix assert! formatting in idempotent completion test
Co-Authored-By: Claude Opus 4.6 <[email protected]>
---------
Co-authored-by: Claude Opus 4.6 <[email protected]>
* fix(llm): persist refreshed Anthropic OAuth token after Keychain re-read (#1136)
The Anthropic OAuth provider stored its token as an immutable SecretString.
When a 401 triggered a Keychain re-read, the fresh token was used for a
single retry but never persisted — every subsequent request reused the
expired original token, causing repeated auth failures.
Changes:
- Wrap token in RwLock<SecretString> so it can be updated after refresh
- Persist refreshed token via update_token() on successful retry
- Add 500ms delay before Keychain re-read to give Claude Code time to
complete its async token refresh write (reduces race window)
- Add regression test verifying token updates persist across reads
Closes#1136
Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
* style: fix formatting
Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <[email protected]>
* fix(worker): prevent orphaned tool_results and fix parallel merging
Two fixes for tool result handling in the Worker:
1. Preserve reasoning text from select_tools() in the RespondResult
content field so it appears in the assistant_with_tool_calls message
pushed by execute_tool_calls. Without this, the LLM's reasoning
context was lost when using the select_tools path.
2. Merge consecutive tool_result messages into a single User message
in rig_adapter's convert_messages(). When parallel tools execute,
each produces a separate ChatMessage with role: Tool. Without
merging, these become consecutive User messages which Anthropic
rejects. Now consecutive tool results are merged into one User
message with multiple ToolResult content items.
Includes regression tests for both fixes.
Co-Authored-By: Claude Opus 4.6 <[email protected]>
* fix(worker): use find_map for first non-empty reasoning extraction
The previous code only checked the first ToolSelection's reasoning,
missing cases where the first selection has empty reasoning but
subsequent ones do not. Switch to find_map to get the first non-empty
reasoning across all selections.
Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
---------
Co-authored-by: Claude Opus 4.6 <[email protected]>
* feat(heartbeat): fire_at time-of-day scheduling with IANA timezone support
- HEARTBEAT_FIRE_AT=HH:MM — fire heartbeat at a specific time of day instead
of on a rolling interval; format is 24h HH:MM (e.g. "14:00")
- HEARTBEAT_TIMEZONE=Region/City — IANA timezone name for fire_at (e.g.
"Pacific/Auckland", "America/New_York"). Defaults to UTC.
- When fire_at is set, interval_secs is ignored
- Config also readable from settings.toml [heartbeat] section
Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
* feat(heartbeat): wire fire_at + timezone into HeartbeatConfig runner
Missed file from heartbeat scheduling commit. HeartbeatConfig struct in
agent/heartbeat.rs now carries fire_at: Option<NaiveTime> and timezone: Tz
so the runner can schedule against a fixed time of day.
Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
* fix: add chrono-tz dependency for heartbeat fire_at timezone support
The chrono-tz crate was used in the heartbeat fire_at commits but
its Cargo.toml entry was lost during rebase conflict resolution.
Co-Authored-By: Claude Opus 4.6 <[email protected]>
* style: rustfmt fix for chained method call
Co-Authored-By: Claude Opus 4.6 <[email protected]>
* test(heartbeat): add fire_at scheduling and DST safety tests
- test_default_config_has_no_fire_at: interval-based default unchanged
- test_with_fire_at_builder: builder sets time and timezone
- test_duration_until_next_fire_is_bounded: result always 1s–24h
- test_duration_until_next_fire_dst_timezone_no_panic: US Eastern DST
- test_resolved_tz_defaults_to_utc: missing timezone falls back to UTC
- test_resolved_tz_parses_iana: IANA string resolves correctly
Co-Authored-By: Claude Opus 4.6 <[email protected]>
* fix(heartbeat): restore drift-free interval, add settings.json fallback for fire_at
- Interval path: restore tokio::time::interval (drift-free) instead of
tokio::time::sleep which drifts by loop body execution time
- fire_at config: fall back to settings.heartbeat.fire_at when
HEARTBEAT_FIRE_AT env var is not set, consistent with other settings
Addresses Gemini Code Assist review feedback.
Co-Authored-By: Claude Opus 4.6 <[email protected]>
---------
Co-authored-by: IronClaw <[email protected]>
Co-authored-by: Claude Sonnet 4.6 <[email protected]>
* feat: add Codex auth.json token reuse for LLM authentication
When LLM_USE_CODEX_AUTH=true, IronClaw reads the Codex CLI's auth.json
(default ~/.codex/auth.json) and extracts the API key or OAuth access
token. This lets IronClaw piggyback on a Codex login without
implementing its own OAuth flow.
New env vars:
- LLM_USE_CODEX_AUTH: enable Codex auth fallback (default: false)
- CODEX_AUTH_PATH: override path to auth.json
* fix: handle ChatGPT auth mode correctly
Switch base_url to chatgpt.com/backend-api/codex when auth.json
contains ChatGPT OAuth tokens. The access_token is a JWT that only
works against the private ChatGPT backend, not the public OpenAI API.
Refactored codex_auth.rs to return CodexCredentials (token +
is_chatgpt_mode) instead of just a string key.
* fix: Codex auth takes highest priority over secrets store
When LLM_USE_CODEX_AUTH=true, Codex credentials are now loaded before
checking env vars or the secrets store overlay. Previously the secrets
store key (injected during onboarding) would shadow the Codex token.
* feat: Responses API provider for ChatGPT backend
- New CodexChatGptProvider speaks the Responses API protocol
- Auto-detects model from /models endpoint (gpt-4o -> gpt-5.2-codex)
- Adds store=false (required by ChatGPT backend)
- Error handling with timeout for HTTP 400 responses
- Message format translation: Chat Completions -> Responses API
- SSE response parsing for text, tool calls, and usage stats
- 7 unit tests for message conversion and SSE parsing
* fix: SSE parser uses item_id instead of call_id for tool call deltas
The Responses API sends function_call_arguments.delta events with
item_id (e.g. fc_...) not call_id (e.g. call_...). The parser now
keys pending tool calls by item_id from output_item.added and
tracks call_id separately for result matching.
* fix: strip empty string values from tool call arguments
gpt-5.2-codex fills optional tool parameters with empty strings
(e.g. timestamp: ""), which IronClaw's tool validation rejects.
Strip them before passing to tool execution.
* fix: prevent apiKey mode fallback to ChatGPT token
When auth_mode is explicitly 'apiKey' but the key is missing/empty,
do not fall through to check for a ChatGPT access_token. This prevents
returning credentials with is_chatgpt_mode: true and routing to the
wrong LLM provider.
* refactor: reuse single reqwest::Client across model discovery and LLM calls
Create Client once in with_auto_model, pass &Client to
fetch_default_model, and move it into the provider struct.
Eliminates the redundant Client::new() that wasted a connection pool.
* fix: bump client_version to 1.0.0 to unlock gpt-5.3-codex and gpt-5.4
The /models endpoint gates newer models behind client_version.
Version 0.1.0 only returns up to gpt-5.2-codex, while 1.0.0+
also returns gpt-5.3-codex and gpt-5.4.
* feat: user-configured LLM_MODEL takes priority over auto-detection
Fetch the full model list from /models endpoint. If LLM_MODEL is set,
validate it against the supported list and warn with available models
if not found. If LLM_MODEL is not set, auto-detect the highest-priority
model. Also bumps client_version to 1.0.0 to unlock gpt-5.3/5.4.
* fix: add 10s timeout to model discovery HTTP request
Prevents startup from blocking indefinitely if chatgpt.com
is slow or unreachable. Uses reqwest per-request timeout.
* docs: add private API warning for ChatGPT backend endpoint
The chatgpt.com/backend-api/codex endpoint is private and
undocumented. Add warning in module docs and a runtime log
on first use to inform users of potential ToS implications.
* feat: implement OAuth 401 token refresh for Codex ChatGPT provider
On HTTP 401, if a refresh_token is available, the provider now
automatically refreshes the access token via auth.openai.com/oauth/token
(same protocol as Codex CLI) and retries the request once. Refreshed
tokens are persisted back to auth.json.
Changes:
- codex_auth: read refresh_token, add refresh_access_token() and
persist_refreshed_tokens()
- codex_chatgpt: RwLock for api_key, 401 detection + retry in
send_request, send_http_request helper
- config/llm: thread refresh_token/auth_path through RegistryProviderConfig
- llm/mod: pass refresh params to with_auto_model
* refactor: lazy model detection via OnceCell, remove block_in_place
Model is no longer resolved during provider construction. Instead,
resolve_model() uses tokio::sync::OnceCell to lazily fetch from
/models on the first LLM call. This eliminates the block_in_place
+ block_on workaround in create_codex_chatgpt_from_registry.
- with_auto_model (async) -> with_lazy_model (sync constructor)
- resolve_model() added with OnceCell-based lazy init
- build_request_body takes model as parameter
- model_name() returns resolved or configured_model as fallback
* feat: support multimodal content (images) in Codex ChatGPT provider
message_to_input_items now checks content_parts for user messages.
ContentPart::Text maps to input_text and ContentPart::ImageUrl maps
to input_image, matching the Responses API format used by Codex CLI.
Falls back to plain text when content_parts is empty.
Also updates client_version to 0.111.0 for /models endpoint.
Adds test: test_message_conversion_user_with_image
* refactor: move codex_auth module from src/ to src/llm/
codex_auth is only used by the LLM layer (codex_chatgpt provider
and config/llm). Moving it under src/llm/ reflects its actual scope.
- Remove pub mod codex_auth from lib.rs
- Add pub mod codex_auth to llm/mod.rs
- Update imports: super::codex_auth, crate::llm::codex_auth
* Fix codex provider style issues
* Use SecretString throughout codex auth refresh flow
* Use SecretString for codex access tokens
* Reuse provider client for codex token refresh
* Stream Codex SSE responses incrementally
* Fix Windows clippy and SQLite test linkage
* Trigger checks after regression skip label
* Tighten codex auth module handling
Coffee