fix(workflow-executor): add per-invocation AI timeout to surface hanging provider errors [PRD-409]

[matthv]

Summary

When the AI provider hangs (no response, internal retries, or holds the connection open), the previous code relied on the global STEP_TIMEOUT_MS (default 5 min) to fail the step. From the user's perspective this looks like an infinite spinner.

This PR adds a dedicated timeout on each AI invocation (default 30s, configurable via AI_INVOKE_TIMEOUT_MS) by passing LangChain's native timeout call option to model.invoke. LangChain converts it into an AbortSignal.timeout(ms) and forwards it to the underlying HTTP request, so a hanging provider is actually cancelled — not merely raced.

On timeout, the abort surfaces as a TimeoutError/AbortError, which invokeWithTools maps to the new AiInvokeTimeoutError. BaseStepExecutor.execute() then converts it to an error outcome with a user-friendly message — the orchestrator sets context.error on the step and the frontend exits its isLoading state immediately.

Why delegate to LangChain instead of a manual AbortController

An earlier version wired up AbortController + setTimeout by hand. LangChain already does exactly this internally when given a timeout call option (verified in @langchain/core ensureConfig → AbortSignal.timeout → forwarded as signal to the request). Delegating removes the manual timer plumbing and lowers invokeWithTools complexity, while still producing a real request cancellation. The timeout call option is in milliseconds.

Why not just lower STEP_TIMEOUT_MS globally

STEP_TIMEOUT_MS covers more than the AI call (it also covers slow agent fetches, DB lookups, etc.). Lowering it globally would kill legitimately slow non-AI work. A dedicated AI timeout is more surgical.

Changes

• defaults.ts: new DEFAULT_AI_INVOKE_TIMEOUT_MS = 30_000
• errors.ts: new AiInvokeTimeoutError extends WorkflowExecutorError with provider-specific user message
• base-step-executor.ts: invokeWithTools passes { timeout: aiInvokeTimeoutMs } to model.invoke, and maps the resulting TimeoutError/AbortError to AiInvokeTimeoutError
• Config plumbing through RunnerConfig → StepContextConfig → ExecutionContext
• cli-core.ts: parse AI_INVOKE_TIMEOUT_MS env var
• 6 unit tests: TimeoutError/AbortError mapped to AiInvokeTimeoutError, { timeout } passed as the 2nd arg, disabled when unset/<=0 (abort not mapped), non-abort errors rethrown as-is

fixes PRD-409

Test plan

• workflow-executor test suite passes (base-step-executor.test.ts: 45 tests, incl. the 6 above)
• Lint clean on changed files; tsc --noEmit clean
• Live test: with SIMULATE_AI_HANG=1 AI_INVOKE_TIMEOUT_MS=10000, the frontend shows the new user message after 10s instead of spinning for 5min
• Default set to 30s

🤖 Generated with Claude Code

Note

Add per-invocation AI timeout to surface hanging provider errors in workflow executor

• Adds aiInvokeTimeoutMs (default 30s) to RunnerConfig, ExecutionContext, and ExecutorOptions, configurable via the AI_INVOKE_TIMEOUT_MS env var in the CLI.
• Wraps each AI model invocation in BaseStepExecutor.invokeWithTools with AbortSignal.timeout, throwing a distinct AiInvokeTimeoutError with a user-facing message when the provider does not respond in time.
• Introduces positiveOrDefault in build-workflow-executor.ts so non-positive or non-finite timeout values always fall back to their defaults rather than disabling the timeout.
• Behavioral Change: stepTimeoutMs values that are non-positive or non-finite now fall back to DEFAULT_STEP_TIMEOUT_MS instead of disabling the step timeout.

^{Macroscope summarized 7e7b418.}

[linear]

PRD-409

[qltysh]

All good ✅

[qltysh]

Coverage Impact

⬆️ Merging this pull request will increase total coverage on feat/prd-214-server-step-mapper by 0.03%.

Modified Files with Diff Coverage (4)

Rating	File	% Diff	Uncovered Line #s
	packages/workflow-executor/src/build-workflow-executor.ts	100.0%
	packages/workflow-executor/src/executors/base-step-executor.ts	100.0%
	packages/workflow-executor/src/errors.ts	100.0%
	packages/workflow-executor/src/defaults.ts	100.0%
	Total	100.0%

🚦 See full report on Qlty Cloud »

🛟 Help

• Diff Coverage: Coverage for added or modified lines of code (excludes deleted files). Learn more.

• Total Coverage: Coverage for the whole repository, calculated as the sum of all File Coverage. Learn more.

• File Coverage: Covered Lines divided by Covered Lines plus Missed Lines. (Excludes non-executable lines including blank lines and comments.)

  • Indirect Changes: Changes to File Coverage for files that were not modified in this PR. Learn more.