feat(tool): configurable timeout protection for tool and task execution

[ESRE-dev]

Issue for this PR

Closes #15080

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

Adds configurable timeout protection for both normal tool execution and subagent task execution to prevent indefinite hangs.

Key changes:

1. Shared abort utility

   • Adds a raceSignal(...) helper that cleanly races async work against an abort signal.

2. Global non-task tool timeout

   • Adds a default timeout for non-task tools in the tool execution wrapper.
   • Supports config override via experimental.tool_timeout.
   • Keeps task excluded from this outer timeout so it can manage its own deadline.

3. Task tool deadline management

   • Adds task-level timeout behavior (default deadline + optional override).
   • Supports config via experimental.task_timeout.
   • On timeout, cancels child session execution and returns structured output with task_id.
   • Preserves partial-output recovery behavior where available.

4. Config schema updates

   • Adds timeout fields used by the above behavior.

Related overlap (not duplicates):

• Tasks/Subagents with Codex / OpenAI are frequently getting stuck with no timeout/retry, which then hangs the session forever #11865 (subagent sessions hanging forever)
• Explore subagent hangs indefinitely with Anthropic Claude Opus 4.6 -- no timeout or recovery #13841 (explore subagent hang without timeout/recovery)
• Subagent tasks hang indefinitely in high-concurrency environment #18378 (subagent hangs in high-concurrency paths)

How did you verify your code works?

• Added and ran packages/opencode/test/tool/timeout.test.ts.
• Ran targeted tool/session tests to confirm timeout behavior and cancellation flow.
• Verified that task timeout behavior remains isolated from non-task global tool timeout.

Screenshots / recordings

Not a UI change.

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

If you do not follow this template your PR will be automatically rejected.

[github-actions]

Thanks for updating your PR! It now meets our contributing guidelines. 👍

[georgeglarson]

Hi @ESRE-dev — I opened #25141 yesterday for the same issue (#15080) without noticing this PR. Apologies for the noise; you got there first and your scope is broader (this PR covers non-task tool timeouts and adds a shared raceSignal helper, mine only races the Task tool prompt against Effect.sleep).

I've drafted #25141. If you'd like a hand getting the failing checks (unit linux/windows + e2e linux/windows) over the line, happy to chip in — let me know. If maintainers prefer the broader patch this PR offers, mine can close.

One thing worth surfacing on this thread regardless: @maiterth left a comment on #25141 suggesting that some "subagent hangs" may actually be a deeper-nesting permission-prompt-bubbling failure (ask prompts not propagating past the first subagent layer), which a timeout would mask rather than fix. Worth verifying our shared assumption that Task-tool-level timeouts address the actual bug shape, not just the visible symptom — applies equally to both PRs.