[lastobelus]

Outcome

PR landing in markover uses the global babysit skill, which replaces the diverged per-project copies in markover, lastCode and certificate-translations. The project copy is removed. Its markover-only rules now live in a ## Landing section of AGENTS.md, where the global skill reads them.

Linked work

• Installs the global skills: lasto-env #29. Merge this right after it, back to back. Until this lands, Codex sees two babysit skills here.
• Companion removals: lastCode #613, certificate-translations #299.

Scope

• Included:
  • Removes .agents/skills/babysit/ (SKILL.md, references/merge.md, agents/openai.yaml).
  • Adds the AGENTS.md Landing section:
    • the exact review-head and handled markers;
    • the Wait for PR Action rules: clean checkout of the target PR, list and ask if two match, and verify host status, exit code, PR, head and base;
    • the slice boundary and parallel claims;
    • adjunct triage before merge;
    • squash merge plus pullRequestStatus done, claim and issue bookkeeping.
  • Adds a one-line CLAUDE.md (@AGENTS.md) for Claude Code outside T3 Code.
  • Tests updated: test/agent-guidance.test.ts and test/start-issue-evals.test.ts read the removed files. Their babysit and merge-reference assertions now check the same behaviour in AGENTS.md. The list_project_actions / run_project_action_and_resume / "end the turn immediately" checks on the old skill are gone: the global skill holds those rules, and the same strings are still asserted in AGENTS.md's CI section. The references/merge.md link check is removed.
• Changed on purpose (proposal from the landing-stack program; the owner may revert):
  • The global churn budget replaces the old "three finding-bearing rounds freeze the boundary" rule. It is stricter: the judge is called at the second Codex round that still has an accepted blocker or should.
  • review-loop's Answer step replaces the five triage verbs.
  • The sleep 100 fallback is dropped.
• Deliberately excluded: application code, the Complexity brake section, and start-issue (it keeps writing the claim babysit reads).

Validation

• npm test: 990 of 990 passed, including the two rewritten tests.
• npm run check: not run separately. npm test builds with tsc, so the edited test files compile.
• Focused automated tests for changed behavior
• Relevant manual macOS validation, when applicable (n/a)
• Before-and-after screenshots for meaningful visual changes, when applicable (n/a)
• Documentation updated, when applicable
• Third-party notices regenerated for production dependency changes, when applicable (n/a)

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary: ✅ Completed

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-07T06:12:19.516549Z	1ca93c8	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.