| name | review-loop |
| description | Use when the user wants the current code proactively reviewed and fixed (e.g. "review-loop", "/review-loop", "review what I have and fix it"). Self + cross-model review → triage → fix → update PR, loop until clean. |
| metadata | {"short-description":"Proactive self + Claude review → triage → fix → update PR"} |
Review Loop (Codex)
Proactively review the current code and fix it, Codex-side — the proactive
counterpart to reacting to CI/Bugbot on a PR. Same workflow as Claude's
/review-loop.
Canonical procedure: read ~/.codex/skills/skyhook-skills-commands/review-loop.md and follow it
(scope → review → triage → fix → update PR → loop; caps; safety). Codex
translations:
- Self review ("/review", "/simple") → YOU review your own diff hard:
correctness bugs, security, silent failures, races, logic errors, breaking API
— and over-engineering / dead weight (options no caller sets, one-impl
interfaces, tests that assert the implementation). Don't ping-pong with prior
reviewers.
- Cross-model review ("/cross-review") → the claude-review skill — from
Codex the secondary reviewer stays Claude (it reviews your work). Nontrivial
changes only — don't ceremony-ize a typo.
- Scenario-sensitive work (copy, diagnostics/remediation, detector precision,
error classification, security/permissions, UI states) → build a scenario
ledger before calling the review complete: scenario, expected final behavior or
copy, source-of-truth evidence, self-review verdict, Claude status, tests/live
proof, open decision. Claude status is either the verdict or
skipped: <reason> when cross-review was intentionally skipped. "Review loop ran" is not
a completion claim.