What problem does it solve? Getting a genuine second opinion on code changes is hard: self-review misses blind spots, and switching tools loses context. This Skill wraps the OpenAI Codex CLI so an AI coding session can request an independent diff review, an adversarial try-to-break-it pass, or a free-form consultation without leaving the workflow. ## Core Features & Use Cases - Review mode: Runs codex review --base <branch> against the current diff with a fail-closed pass/fail gate keyed on P0/P1 severity markers, plus a custom-instructions path via codex exec when a focus area is given. - Challenge mode: Sends the branch diff to Codex with an adversarial prompt that hunts edge cases, race conditions, security holes, and failure modes, streaming reasoning traces via JSONL parsing. - Consult mode: Asks Codex free-form questions or plan reviews with session continuity (resume via saved thread ID), embedding plan content directly since Codex is sandboxed to the repo root. - Use Case: Before merging a feature branch, run the review mode to get a gated verdict on the diff, then run challenge mode to attack the riskiest changes, and compare findings against the host model's own review with the cross-model analysis output. ## Quick Start Ask the assistant to run a codex review of the current branch diff against the base branch and report the gate verdict.
