codex-rescuelisted

# Codex Rescue ... Second-Opinion Delegation ## Context The Codex plugin for Claude Code (`openai/codex-plugin-cc`) lets you hand tasks to OpenAI's Codex agent without leaving Claude Code. This skill covers the **rescue** use case only ... delegating work to a different model when Claude Code is stuck, looping, or burning iterations on a problem it can't solve. This applies to **manual Claude Code sessions only** (not automated pipeline execution). ## Prerequisites - Codex CLI installed globally: `npm install -g @openai/codex` - Codex authenticated: `codex login` (ChatGPT subscription or OpenAI API key) - Plugin installed in Claude Code: ``` /plugin marketplace add openai/codex-plugin-cc /plugin install codex@openai-codex ``` - Verify setup: `/codex:setup` ## When to Use Rescue Use `/codex:rescue` when Claude Code is **stuck** ... not as a default or first-pass tool. Triggers: 1. **Three-strikes rule fires.** Same problem recurred after 3 fix attempts. Before declaring "triggering deep dive" and stopping, consider whether Codex rescue with a fresh model perspective could break the loop. 2. **Max iterations burned.** Claude Code has exhausted its iteration budget without resolution. Rather than re-prompting with the same model, delegate the specific failing piece to Codex. 3. **Model blindspot suspected.** The failure pattern suggests a systematic model weakness (e.g., a library Claude consistently misuses, a pattern it keeps reverting to despite correction). A