Codex review inspects the GitHub pull-request diff and evidence, then emits tabellio-agent-review/v0.1 for the durable review cycle. It does not replace deterministic checks.
Run review against the exact GitHub pull-request head:
CODEX_HOME=/tmp/codex-review-home codex review --base mainUse an isolated CODEX_HOME when local MCP configuration could add noise or private integrations.
Serialize actionable findings into the agent-review contract, validate them with scripts/check-tabellio-agent-review.mjs, then import them with tabellio-review. The review ledger keeps provider feedback, triage, fixes, head remapping, and validation readiness under refs/tabellio/reviews.
Give Codex these facts from the change request and local ledgers:
- task source
- intended behavior
- evidence artifact path
- commands run
- known skipped checks
- protected side effects requested or blocked
Codex review should flag correctness risks, missing tests, stale evidence, unsafe side effects, and unclear PR scope.
Codex review is not:
- branch protection
- dependency update automation
- deployment approval
- substitute for human ownership
- proof that agent output is secure
When the repository commits a product-validation manifest, review begins only after the exact pull-request head has a passed v0.3 decision. A blocked decision means the evidence is incomplete or untrustworthy; it is not a product pass and should not be converted into reviewer guesswork.