OpenClaw has merged a broad repair to agent execution and cancellation, aimed at keeping active Codex turns alive, preserving partial output and making terminal failures tell the truth. The important catch is operational rather than ornamental: the commit verifies tests and isolated live behaviour, but says there is no deployment or release yet.
OpenClaw Watch analysis
What happened
The 2 September commit moves native Codex turn liveness away from inferred idle watches and gives the native lifecycle ownership of execution and completion. Its validation summary says an isolated check reproduced the old 300-second interruption, then verified a 320-second stream and a subsequent turn.
It also consolidates deadline and cancellation handling across agent paths. Approval pauses and compaction grace are preserved, unlimited execution no longer depends on a sentinel timer, and terminal timeout explanations are carried through settlement instead of being overwritten by an earlier tool error or a misleading success state.
The same repair keeps stopped sessions visibly stopped while retaining partial output, and reduces the risk that stale observations from an earlier execution terminate resumed or replacement work.
Why it matters
For people running long Codex turns, this is the difference between “the provider went quiet” and “the system knew the work was still alive”. For operators, clearer ownership should make approval pauses, compaction grace, manual Stop actions and retries easier to reason about.
But a green validation trail is not a production rollout. Until a released OpenClaw build contains the change, current timeout and recovery safeguards still earn their keep.
Our read
This is a meaningful reliability repair, not a cosmetic sweep of error messages. Check whether your running version includes the commit before altering timeout or recovery procedures, and treat the reported benefits as pending availability.
What to watch
- Whether a released OpenClaw build incorporates these lifecycle changes.
- Whether operators report fewer false interruptions and misleading success states after deployment.
- Whether resumed or replacement sessions retain partial output as intended in production.
Discussion spark: Which lifecycle failure should operators prioritise validating first: long-turn interruption, Stop-state accuracy or resumed-session recovery?
Sources and evidence
- fix(codex): stop interrupting active turns during quiet native work (… (2 September 2026, 03:15 UTC)
OpenClaw Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by the OpenClaw Foundation.