Evidence record / openclaw
A dated record of one change, kept so the writing that cites it can be checked. Compiled from the sources listed below by the research run, not written for reading. The judgment lives in the issues that cite it, below.
2026-08-20-openclaw-codex-sandbox-stop-can-return-success-while-children-run
PR #125908 merge fd8326c5 (2026-08-18T17:48:11Z) is on main. Terminating a Codex sandbox command could acknowledge a single SIGTERM while a TERM-resistant local wrapper, remote sandbox descendant, or backend lease stayed alive. The fix completes descendant cleanup, TERM-to-KILL, reaping, and backend finalization before termination is acknowledged, and fails closed when a process survives SIGKILL. compare vs v2026.8.1-beta.2 is diverged, ahead_by=1180. Not in v2026.7.1-2. No new GitHub Release or npm publish in this window.
Channel: main-unreleased. The hole is the standing behavior on released Codex sandbox installs. Half: defect | security-relevant.
Operator consequence: on latest, beta, and extended-stable, a successful stop or timeout is not proof the process tree is gone. Check the host after cancelled sandbox execs. The containment is main-only.
Receipt
Finding metadata
Run: 2026-08-20-brief-2026-08-17_2026-08-20-frontier-v0
Finding ID: 2026-08-20-openclaw-codex-sandbox-stop-can-return-success-while-children-run
Profile citations
- OpenClaw / claim / sandbox-stop-success-while-children-run
Source links
Primary links, including exact changelog lines when available.
Versioned source: run artifact