From codex-advisor
Runs official Codex code review on code changes, diffs, branches, or commits, then double-checks each finding by reading source code and classifying agree/disagree/nuance.
How this skill is triggered — by the user, by Claude, or both
Slash command
/codex-advisor:codex-reviewThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Invoke the Official Codex companion's review, then apply Claude's independent evaluation to every finding.
Invoke the Official Codex companion's review, then apply Claude's independent evaluation to every finding.
CODEX_COMPANION=$("${CLAUDE_PLUGIN_ROOT}/scripts/resolve-companion.sh")
If resolve fails: direct to /codex-setup immediately. Do NOT fall back to a Claude-only solo review.
Use the $CODEX_COMPANION resolved in Step 1:
node "$CODEX_COMPANION" review --wait $ARGUMENTS
If $ARGUMENTS is empty, pass no args — companion auto-detects scope (uncommitted or branch).
Timeout: 300000ms (5 minutes).
| Error | Action |
|---|---|
| "not authenticated" in stderr | Auth required → suggest codex login |
| Other error | Show raw error, don't retry silently |
Read ${CLAUDE_PLUGIN_ROOT}/references/evaluation.md.
For each finding in the Official review output:
Present to user:
mkdir -p "${CLAUDE_PLUGIN_DATA}/reviews"
Save to ${CLAUDE_PLUGIN_DATA}/reviews/review-<YYYYMMDD-HHMMSS>.md using format from evaluation.md.
--wait. Without it, companion prompts for foreground/background selection via AskUserQuestion, which disrupts our flow.npx claudepluginhub leejuoh/claude-code-zero --plugin codex-advisorRuns code review using OpenAI Codex CLI in parallel instances with fallback to Claude Code. Activates on 'review' commands or after implementation for quality checks.
Performs deep code review via Codex CLI with full disk access. Use for uncommitted changes and full codebase analysis. Outputs severity-grouped findings and merge gate.
Executes Codex adversarial review to skeptically challenge code changes for security and correctness issues, followed by Claude's verification of findings against actual code. Use on 'adversarial review' requests.