Codex review: needs real behavior proof before merge. Reviewed July 9, 2026, 9:04 AM ET / 13:04 UTC.

Summary
The PR changes the isolated cron agent watchdog so phase progress after runner entry clears the pre-execution guard, while runner entry alone remains guarded, and expands timer regression coverage.

PR surface: Source -12, Tests +158. Total +146 across 2 files.

Reproducibility: yes. Source inspection shows current main still routes isolated cron start and phase updates into the watchdog, maps setup phases to pre_execution, and only clears the guard for execution-stage progress or firstModelCallStarted; I did not run a live scheduled cron repro in this read-only review.

Review metrics: 1 noteworthy metric.

• Pre-execution phases newly clearing guard: 5 newly clear, 1 excluded. workspace, runtime_plugins, model_resolution, auth, and context_engine would move under the full job timeout while runner_entered remains under the short guard, which is the key availability tradeoff.

Root-cause cluster
Relationship: fixed_by_candidate
Canonical: #93912
Summary: This PR is an open candidate fix for the narrowed setup-progress watchdog issue; sibling PRs overlap but none is a merged canonical replacement.

Members:

• canonical: cron isolated agent pre-execution watchdog aborts runs despite setup progress #93912 - The PR uses closing syntax for this open issue, and the issue narrows the bug to setup-like phases being treated as non-progress.
• same_root_cause: cron isolated-agent turn: pre-execution watchdog race with runEmbeddedAgent phase reporting #93530 - The issue reports the same isolated cron pre-execution watchdog phase gap and proposes the same watchdog-layer option.
• partial_overlap: fix(cron): emit before_agent_reply phase unconditionally for cron triggers #93535 - The open PR targets the narrower before_agent_reply phase-emission symptom rather than the broader setup-progress watchdog transition.
• partial_overlap: fix(cron): clear pre-execution watchdog on any phase notification (#93530) #93591 - The open PR overlaps the no-hook before_agent_reply path and remains unmerged.
• partial_overlap: [Bug]: cron agentTurn isolated-session "model-call-started" timeout blocks all isolated cron jobs intermittently #93867 - The closed report covers overlapping isolated cron startup timeout symptoms with broader operator observations.
• 2 more in the report.

Proposal only: this assessment does not dispatch repair, suppress jobs, mutate sibling items, close, or merge anything.

Merge readiness
Overall: 🧂 unranked krab
Proof: 🧂 unranked krab
Patch quality: 🦪 silver shellfish
Result: blocked until real behavior proof from a real setup is added.

Overall follows the weaker of proof and patch quality, so missing proof can cap an otherwise strong patch.

Rank-up moves:

• Use the existing stage lookup in the new guard so check-prod-types and check-lint can pass.
• [P1] Add redacted live scheduled-cron terminal output, copied live output, or runtime logs showing setup progress survives past the old 60-second guard.

Proof guidance:

• [P1] Needs real behavior proof before merge: The PR body and latest comment provide copied Vitest/fake-timer terminal output only; external real behavior proof still needs redacted live scheduled-cron output, logs, or an explicit maintainer override. After adding proof, update the PR body; ClawSweeper should re-review automatically. If it does not, the PR author or someone with repository write access can comment @clawsweeper re-review.

Risk before merge

• [P2] The PR intentionally moves workspace, runtime_plugins, model_resolution, auth, and context_engine stalls from the 60-second pre-execution guard to the configured job timeout; maintainers need to accept that availability tradeoff before merge.
• [P1] The supplied proof is copied Vitest/fake-timer output, not redacted live scheduled-cron output or runtime logs showing the after-fix path in a real setup.
• [P1] Several sibling PRs still propose narrower before_agent_reply-based fixes, so maintainers should keep one canonical cron watchdog contract before landing this cluster.

Maintainer options:

1. Fix Stage Use And Require Proof (recommended)
   Use the existing stage lookup in the new guard, keep the runner_entered regression, and require redacted real scheduled-cron proof before merge.
2. Accept Focused Proof Only
   Maintainers can choose to override the external proof gate if they consider the fake-timer regression sufficient for this watchdog timeout contract.
3. Defer To A Canonical Sibling
   Pause or close this PR if a narrower sibling watchdog or phase-emission PR becomes the chosen landing path.

Next step before merge

• [P1] Human follow-up is needed because the PR has a narrow type/lint blocker plus a contributor real-behavior-proof gate that ClawSweeper should not repair around automatically.

Maintainer decision needed

• Question: After the unused-stage blocker is fixed, should this external PR require redacted live scheduled-cron proof, or should maintainers override with focused source-level regression proof?
• Rationale: ClawSweeper policy treats tests and fake timers as supplemental only for external PRs, while this PR changes availability-sensitive cron timeout behavior that maintainers may decide to waive or require live proof for.
• Likely owner: steipete — The current watchdog module boundary dates to steipete's cron timer split, and the remaining choice is proof sufficiency for that runtime behavior.
• Options:
  • Require live cron proof (recommended): Ask for redacted scheduled-cron terminal output, copied live output, or runtime logs showing setup progress surviving past the old 60-second cutoff before merge.
  • Override with focused regression proof: A maintainer can explicitly accept the fake-timer regression proof as sufficient for this narrow watchdog contract after the type/lint blocker is fixed.
  • Pick a sibling contract instead: Pause or close this PR if maintainers prefer a narrower before_agent_reply phase-emission fix from a sibling PR.

Security
Cleared: Cleared: the diff changes cron watchdog logic and regression tests only, with no dependency, CI, secret, package, or external code-execution surface changed.

Review findings

• [P2] Use the stage lookup in the new guard — src/cron/service/agent-watchdog.ts:118-120

Addressed the current review blocker around runner_entered.

What changed in 60413c7ed7:

• runner_entered no longer clears the pre-execution watchdog by itself.
• Only phases after runner entry, or firstModelCallStarted, move the watchdog to normal execution timeout.
• Added regression coverage that runner_entered alone still fails as stalled before execution start (last phase: runner-entered).
• Existing setup-progress and fallback-progress coverage still passes, so workspace / runtime_plugins / model_resolution / auth / context_engine continue to count as meaningful progress.

Validation after the narrower fix:

node scripts/run-vitest.mjs src/cron/service/timer.regression.test.ts
Test Files  1 passed (1)
Tests       65 passed (65)

Additional checks:

corepack pnpm exec oxfmt --check --threads=1 src/cron/service/agent-watchdog.ts src/cron/service/timer.regression.test.ts
All matched files use the correct format.

git diff --check
# no output

@clawsweeper re-review

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

This pull request has been automatically marked as stale due to inactivity.
Please add updates or it will be closed.

ClawSweeper status: review started.

I am starting a fresh review of this pull request: fix(cron): treat isolated setup phases as progress This is item 1/1 in the current shard. Shard 3/4.

This placeholder means the worker is alive and reading the current context. I will edit this same comment with the actual review when the claws are done clicking.

Crustacean status: shell secured, claws on keyboard, evidence pebbles being sorted.