What problem does it solve?
Claude Code sessions sometimes loop on the same issue, exhaust their iteration quota, or reveal blind spots that stall progress in manual workflows. This guide explains how to identify those trapped states, why rescue is a second-opinion intervention, and how to safely hand off the specific failing piece to Codex while keeping the manual session in control.
Core Features & Use Cases
- Command-driven delegation: Use
/codex:rescue with --background, --wait, --resume, --fresh, --model, and --effort to describe the precise failure and request the right level of investigation or fix.
- Lifecycle tracking: Monitor and manage jobs with
/codex:status, /codex:result, and /codex:cancel, remembering that the rescue runs on the same checkout and that you should review Codex output before resuming Claude Code.
- Use Case guidance: Follow the three-strike rule, iteration budget guardrail, targeted investigations, or model-switch patterns, and optionally set project-level defaults in
.codex/config.toml (or user defaults) to preselect models, effort levels, and behavior.
Quick Start
Describe the stuck Claude Code work to Codex and request a rescue with the appropriate flags so it can investigate or continue while you monitor the output before resuming the session.