You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
timeout_ms defaults to 30000 (exec_command input schema) and the watchdog kills the process group at the deadline. On small sandboxes (2C/4G) install/build/test reliably exceed 30 s, so the default turns routine builds into kills. yield_time_ms (foreground window, default 10 s) is already a separate knob — the gap is the lifetime default and the watchdog's treatment of deliberately backgrounded work.
Raise the default total lifetime to 120–300 s and make it deployable via env var; document "foreground window vs process lifetime" in the schema descriptions.
Watchdog should not pre-empt commands the model explicitly backgrounded (yield_time_ms: 0, or subsequently polled) before the hard timeout_ms cap.
Status must distinguish timeout | killed | exited, with the real exit code (pairs with the telemetry outcome split).
Targeted: 0.5.0 (reliability program). Baseline:
release/0.3.0.timeout_msdefaults to 30000 (exec_command input schema) and the watchdog kills the process group at the deadline. On small sandboxes (2C/4G)install/build/testreliably exceed 30 s, so the default turns routine builds into kills.yield_time_ms(foreground window, default 10 s) is already a separate knob — the gap is the lifetime default and the watchdog's treatment of deliberately backgrounded work.yield_time_ms: 0, or subsequently polled) before the hardtimeout_mscap.timeout|killed|exited, with the real exit code (pairs with the telemetry outcome split).MAX_ACTIVE_COMMANDS = 16changes the resource envelope — document, and leave fairness to Per-client resource quotas for the shared workspace runtime #46.