What problem does it solve?
Before opening a pull request, Elixir/Phoenix developers often want an external, cross-model code review to catch issues their own analysis missed. Manually invoking the Codex CLI, parsing its findings, fixing approved items, verifying with compile and tests, and re-reviewing is repetitive and error-prone, especially with quota costs per review round.
Core Features & Use Cases
- Bounded critic-refiner loop: Runs Codex CLI review, triages findings by priority (P0-P3), applies approved fixes, verifies with
mix compile --warnings-as-errors and tests, then re-reviews until clean or max rounds (default 3) are reached.
- Flexible diff modes: Review against a base branch, uncommitted changes (staged, unstaged, untracked), or auto-detected default branch state.
- Iron Law enforcement: Declines Codex suggestions that violate project Iron Laws, never commits or pushes, and reports every finding as fixed, declined, or deferred.
- Use Case: After implementing a Phoenix feature, run the loop with
--auto to auto-approve P0/P1/P2 fixes, get a clean external review verdict, and then open your PR with confidence.
Quick Start
Ask the assistant to run the codex-loop skill on your current branch diff so Codex reviews the changes and approved findings get fixed until the review comes back clean.