fix(pi-claude-cli): defer to engine StuckTaskDetector instead of 300s subprocess kill

The engine already runs a StuckTaskDetector with a configurable per-task
heartbeat (default 1 hour) and aborts triage/executor sessions through
AbortSignal when it judges them quiet. pi-claude-cli forwards that signal to
the subprocess. The 300s subprocess-level inactivity timeout was a redundant,
much-tighter shadow that kept killing Sonnet 4.6 mid-thinking on the 40k-char
triage prompt — exactly what the engine-level detector is designed not to do.

Move the local timeout to 30 minutes purely as a last-resort guard for embeds
that don't pass an abort signal. Stuck-detection responsibility now lives in
the caller, where it can be configured per task.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
gsxdsm
2026-04-25 17:49:09 -07:00
parent 833b9dac61
commit f0e3413b69
2 changed files with 22 additions and 13 deletions

View File

@@ -45,13 +45,22 @@ import { handleControlRequest } from "./control-handler.js";
import { mapThinkingEffort } from "./thinking-config.js";
import { isPiKnownClaudeTool } from "./tool-mapping.js";
/**
* Inactivity timeout: kill the Claude CLI subprocess if no stdout arrives
* for this long. Sonnet 4.6 with extended thinking on large system prompts
* (e.g. the triage prompt + AGENTS.md + skill blocks ~ 40k chars) can take
* minutes between thinking deltas; a 3-minute timeout was killing those
* sessions before they could call fn_review_spec. Bump to 5 minutes.
* Inactivity safety net for the Claude CLI subprocess.
*
* Set very high (30 minutes) because the caller is the authoritative source of
* truth for "this session is stuck": Fusion's engine runs a `StuckTaskDetector`
* with a configurable heartbeat (default 1 hour) and aborts the session via
* `AbortSignal` when it decides the agent has gone quiet. pi-claude-cli already
* forwards that signal to the subprocess (`forceKillProcess` on `signal.abort`).
*
* A short timeout here was racing the engine: Sonnet 4.6 with extended thinking
* on the triage prompt (~40k chars) routinely goes >3 minutes between thinking
* deltas, and we were killing those subprocesses before they could write
* PROMPT.md and call `fn_review_spec`. The half-hour ceiling is just a
* last-resort guard for catastrophically hung processes when no abort signal
* arrives (e.g. someone embeds pi-claude-cli without a stuck detector).
*/
const INACTIVITY_TIMEOUT_MS = 300_000;
const INACTIVITY_TIMEOUT_MS = 30 * 60_000;
/** Extended stream options: pi's SimpleStreamOptions plus optional cwd and mcpConfigPath */
type StreamViaCLiOptions = SimpleStreamOptions & {