Repository navigation
PASS
Baseline. This re-reviews 4bc4644e against the CONDITIONS on f162bbe0 (this PR), on the same arch 7a86ec7b. I checked that since.diff equals git diff f162bbe0 4bc4644e, and that pr.diff equals git diff 3bcea66b 4bc4644e (index and hunk lines normalised). 3bcea66b (#3031) is the base and the merge base. The compare has two parts. The first is one branch commit, feaec74e (fix(minimald): keep a timeout's ending when its record is not saved). The second is merge 4bc4644e, which brings only #3031. It has no conflict hunks (git show --cc is empty), and it touches no file of this PR. The ledger cuts the f162bbe0 ruling at 400 characters. So I rebuilt its open items from the head's PR body and from the compare.
Governs: architecture.md, min box — every box, any type. Line 499 says a box whose timeout elapses ends exited, with reason = timeout and no exit code. The same line says stop is SIGTERM to every process, execs included, then grace, then SIGKILL. Line 497 puts an exec under the box's timeout ceiling. Line 1097 limits timeout to lifetime = "until_complete". Spec 29 BOX-037 states the same.
Open items on f162bbe0 and where they stand. (1) At the timeout, a failed record read logged the box is not ended and let the box run past its timeout. That fails open (rubric 7). end_on_timeout now ends the box when its own host is live. Without a live host, it arms the deadline again after EXIT_RECORD_RETRY_BACKOFF. (2) A failed exited write let a later min box stop record stopped. Line 499 says a timeout end is exited. The actor now keeps unrecorded_exit. launch_host refuses while it is set, as for unrecorded_stop (BOX-013). stop_box_inner writes the kept exited timeout record and event (write_kept_exit). The StopBox arm then ends the actor, and BoxManager::ended_while_stopping answers AlreadyExited(Timeout) from the record. Two new tests pin both paths: timeout_record_read_failure_still_ends_box_reason_timeout and timeout_record_write_failure_refuses_resume_and_stop_keeps_timeout. Everything the compare does not touch (the CLI 124 line, the docs, the deadline arm and disarm) keeps its status from f162bbe0.
Ruling. BOX-037 conforms. The timeout reuses stop_box_processes with BoxProcesses::of_container, the BOX-013 stop path, so it signals the same process set that line 499 gives a stop. No condition is open.
Not checked. I did not build or run tests. The PR's test table shows exit 101 for every leg because cargo nextest is not installed in the coder's sandbox. CI is the gate for those legs. The proof that a real exec, in the box's PID namespace, dies at the timeout rests on the BOX-013 stop path. The mock launcher cannot show it, and only the VM or e2e lane can. I did not see the full condition list of the f162bbe0 ruling.
Ruled at 4bc4644 against gominimal/arch@7a86ec7bb3c0 by claude-opus-5-5; arch_rulings 1063, job f56e66cc-666c-4cbc-b7d4-7a5ca4ab4540.