codex-review
Codex Review
Independent gpt-5.6 review of local changes through the Codex CLI. Read-only - never lets Codex modify the repo. Works identically from the main loop, subagents, and workflow stages (no plugin runtime required).
Reviews run on gpt-5.6-luna at "max" (MODELS.md: Luna max is the Codex
default for almost everything, review included). Escalate to gpt-5.6-terra at
"max" when the review is the load-bearing check - a one-way door, a
security or money path, or an adversarial pass whose verdict you will act on
directly. Drop to gpt-5.6-luna at "high" only for a fast look at a small
diff where latency matters more than depth.
Pin BOTH the model and the effort on every command below rather than inheriting
CLI defaults. For codex review, -m is a GLOBAL flag and must come before the
subcommand. Paths written as <skill-dir> mean this skill's base directory
(provided when the skill is invoked).