Dual AI cross-validation: Codex (GPT) reviews Claude Code output as a Senior Reviewer.

/plugin marketplace add anthhub/codex-reviewer
/plugin install codex-reviewer

• Codex CLI (npm i -g @openai/codex)
• jq (brew install jq)

/cx-check

1. /cx or /cxx launches a subagent (isolated context, no overflow)
2. Subagent collects evidence:
   • git diff and recent commit history
   • Upstream callers — who imports/depends on the changed code
   • Downstream dependencies — what the changed code depends on
   • Related test files — matched by filename and symbol references
   • Test results, build status — runs project tests and build automatically
3. Passes all context via temp files to codex-reviewer.sh
4. Codex CLI reviews with configurable reasoning effort and returns structured result with score + verdict
5. Review focuses on core path stability — upstream/downstream breakage is weighted highest

With the codex-review skill, Claude automatically invokes Codex review for:

• Security-related changes
• 3+ files or 100+ lines modified
• Architecture refactoring
• Bug fixes or new core features

Optional .claude/reviewer.json:

{
  "model": "gpt-5.2-codex",
  "timeout": 300,
  "advancedTimeout": 600,
  "scoreThreshold": 80,
  "lang": "en",
  "saveHistory": true,
  "bypassApproval": true,
  "reasoningEffort": {
    "quick": "low",
    "standard": "medium",
    "advanced": "high"
  }
}

MIT

anthhub