Summary

Three related bugs in the runner/task execution path:

• execute() could resolve with the wrong run's result. It matched on the generic execution:finished/execution:failed events without checking which execution actually settled, so a scheduled fire completing while a manual execute() call was still in flight could resolve the manual call with the scheduled run's result. Both InlineScheduledTask.execute() and BackgroundScheduledTask.execute() now correlate by the execution id they invoked. For background tasks, the id is generated in the parent, sent through the task:execute IPC command, threaded into the daemon's InlineScheduledTask.execute(id) call, and echoed back on the forwarded events.

• A throwing/rejecting user hook could reclassify a successful run as failed. onFinished was called inside the task's own try block, so a synchronously-throwing execution:finished listener made the runner treat a successful execution as failed (emitting execution:failed after execution:finished, and overwriting lastRun()). An async listener that rejected became an unhandled rejection. The runner now calls onFinished outside the task's try block, in its own try/catch routed to the internal error fallback logger, and the inline task's event emission catches both a throwing and a rejecting listener without affecting the task's own outcome.

• Manual execute() silently drifted toward maxExecutions. It incremented runCount on every call but never checked maxExecutions, so repeated manual calls could push the count past the limit. Per decision: maxExecutions (and runsLeft()) now only counts scheduled fires. Manual execute() never counts toward it and is never blocked by it, even after the task has auto-stopped from maxExecutions.

Test plan

• npm run build
• npx tsc --noEmit
• npm run lint
• npm test (741 tests, 100% statements/branches/functions/lines)
• TDD: each new test observed failing for the right reason before the fix (race resolves with the wrong result, hook throw reclassifies success as failure, unhandled rejection from async hook, runCount incrementing/gating on manual execute()), then passing after