* feat: add customTools support to AgentConfig for orchestrator-level tool injection
Users can now pass custom ToolDefinition objects via AgentConfig.customTools,
which are registered alongside built-in tools in all orchestrator paths
(runAgent, runTeam, runTasks). Custom tools bypass allowlist/preset filtering
but can still be blocked by disallowedTools.
Ref #108
* test: add disallowedTools blocking custom tool test
* fix: apply disallowedTools filtering to runtime-added custom tools
Previously runtime-added tools bypassed all filtering including
disallowedTools, contradicting the documented behavior. Now custom
tools still bypass preset/allowlist but respect the denylist.
- #99: pass per-call effectiveAbortSignal to buildToolContext() so tools
receive the correct signal instead of the static runner-level one
- #100: replace manual pending-task loop with queue.skipRemaining() on
abort, fixing blocked tasks left non-terminal and missing events
- #101: forward abortSignal in Gemini adapter's buildConfig() so the
SDK can cancel in-flight API calls
- Add 8 targeted tests for all three fixes
run() only handled 'done' events from stream(), silently dropping
'error' events. This caused failed LLM calls to return an empty
RunResult that the caller treated as successful.
AgentPool now maintains a per-agent Semaphore(1) that serializes
concurrent run() calls targeting the same Agent. This prevents
shared-state races on Agent.state (status, messages, tokenUsage)
when multiple independent tasks are assigned to the same agent.
Lock acquisition order: per-agent lock first, then pool semaphore,
so queued tasks don't waste pool slots while waiting.
* feat(agent): add smart loop detection for stuck agents (#16)
Detect when agents repeat the same tool calls or text outputs in a
sliding window. Three modes: warn (inject nudge, terminate on 2nd hit),
terminate (immediate stop), or custom callback. Fully opt-in via
`loopDetection` on AgentConfig — zero overhead when unconfigured.
* fix(agent): support async onLoopDetected callbacks and prevent orphaned tool_use events
- Await onLoopDetected callback result so async functions work correctly
instead of silently falling through to 'continue'
- Move loop detection before yielding tool_use events so terminate mode
never emits tool_use without a matching tool_result
* fix(agent): reset loopWarned on recovery and rename maxRepeatedToolCalls to maxRepetitions
- Reset loopWarned flag when the agent stops repeating, so a future
loop gets a fresh warning cycle instead of immediate termination
- Rename maxRepeatedToolCalls → maxRepetitions since the threshold
applies to both tool call and text output repetition detection
* test(agent): add tests for async callback, warn recovery, and injected warning text
- Verify async onLoopDetected callback is awaited correctly
- Verify loopWarned resets after recovery, giving fresh warning cycle
- Verify WARNING TextBlock is injected into user message content
When both timeoutMs and a caller-provided abortSignal were set, the
timeout signal silently replaced the caller's signal. Now they are
combined via mergeAbortSignals() so either source can cancel the run.
Also removes dead array-handling branch in text-tool-extractor.ts
(extractJSONObjects only returns objects, never arrays).
Local models (Ollama, vLLM) sometimes return tool calls as text instead
of using the native tool_calls wire format. This adds a safety-net
extractor that parses tool calls from model text output when native
tool_calls is empty.
- Add text-tool-extractor with support for bare JSON, code fences,
and Hermes <tool_call> tags
- Wire fallback into OpenAI adapter chat() and stream() paths
- Add onWarning callback when model ignores configured tools
- Add timeoutMs on AgentConfig for per-run abort (local models can
be slow)
- Add 26 tests for extractor and fallback behavior
- Document local model compatibility in README
* feat(agent): add beforeRun / afterRun lifecycle hooks (#31)
Add optional hook callbacks to AgentConfig for cross-cutting concerns
(guardrails, logging, token budgets) without modifying framework internals.
- beforeRun: receives prompt + agent config, can modify or throw to abort
- afterRun: receives AgentRunResult, can modify or throw to fail
- Works with all three execution modes: run(), prompt(), stream()
- 15 test cases covering modify, throw, async, composition, and history integrity
* fix(agent): preserve non-text content blocks in beforeRun hook
- applyHookContext now replaces only text blocks, keeping images and
tool results intact (was silently stripping them)
- Use backward loop instead of reverse() + find() for efficiency
- Clarify JSDoc that only `prompt` is applied from hook return value
- Add test for mixed-content user messages
* fix(agent): address review feedback on beforeRun/afterRun hooks
- Normalize stream done event to always yield AgentRunResult
- Move transitionTo('completed') after afterRun to fix state ordering
- Strip hook functions from BeforeRunHookContext.agent to avoid self-references
- Pass originalPrompt to applyHookContext to avoid redundant message scan
- Clarify afterRun JSDoc: not called when the run throws
- Add tests: error-path skip, outputSchema+afterRun, ctx.agent shape, multi-turn hooks
Add lightweight onTrace callback to OrchestratorConfig that emits
structured span events (llm_call, tool_call, task, agent) with timing,
token usage, and runId correlation. Zero overhead when not subscribed.
Closes#18
- Include error feedback user turn in mergedMessages to maintain
alternating user/assistant roles required by Anthropic API
- Use explicit undefined check instead of ?? for structured merge
to preserve null as a valid structured output value
When `outputSchema` is set on AgentConfig, the agent's final text output
is parsed as JSON, validated against the Zod schema, and exposed via
`result.structured`. On validation failure a single retry with error
feedback is attempted automatically.
Closes#29
Enable connecting to any OpenAI-compatible API (Ollama, vLLM, LM Studio,
etc.) by adding baseURL and apiKey fields to AgentConfig and
OrchestratorConfig, threaded through to adapter constructors.
- OpenAIAdapter and AnthropicAdapter accept optional baseURL
- createAdapter() forwards baseURL to both adapters, warns if used with copilot
- All execution paths (runAgent, runTeam coordinator, buildPool) merge defaults
- Fully backward compatible — omitting new fields preserves existing behavior
JackChen
MrAvalonApple
Ibrahim Kazimov
JackChen