claude-rescue

Forwards tasks from Codex CLI sessions to a Claude Code subprocess for independent diagnosis or implementation.

3|Updated Jun 15, 2026
One-click install
npx skills add https://github.com/patrickserrano/lacquer --skill claude-rescue-patrickserrano
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: claude-rescue
Source: https://github.com/patrickserrano/lacquer/tree/main/core/skills/claude-rescue
Command: npx skills add https://github.com/patrickserrano/lacquer --skill claude-rescue-patrickserrano

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve? When working in a Codex CLI session and stuck on a recurring error or a non-converging approach, this Skill hands the task to Claude Code — a genuinely different reasoning system — for a second, independent opinion instead of another Codex run. ## Core Features & Use Cases - Task Forwarding: Passes the task text verbatim to the Claude Code CLI via a companion shell script, with no independent analysis that would bias the second opinion. - Foreground or Background Execution: Runs Claude in blocking --wait mode for small bounded asks, or detached --background mode returning an agent id for long-running, multi-file work. - Session and Permission Control: Supports --resume to continue a prior rescue session, --read-only for diagnosis-only reviews, and --model selection on explicit request. - Use Case: A Codex session hits a build error that keeps recurring after several fix attempts; forward the exact error context to Claude Code in read-only mode and return its diagnosis verbatim. ## Quick Start Ask the assistant to forward the current stuck task to Claude Code using the claude-rescue script and return its output exactly as printed.

Frequently Asked Questions about claude-rescue

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I get a second opinion from Claude Code while using Codex CLI?▼

Run the claude-companion.sh script with the task text, which forwards it to the Claude Code CLI and prints the result. Use --wait for small bounded asks or --background for long-running work that returns an agent id immediately.

How to run Claude Code as a background agent from a script?▼

Pass --background to claude-companion.sh, which starts a detached Claude Code agent and returns an id. Manage it with claude agents, claude attach <id>, claude logs <id>, or claude stop <id>.

Can I continue a previous Claude Code session in the same repo?▼

Yes, pass --resume to the script, which invokes claude -c to continue the most recent Claude Code session in that repository. The default --fresh flag starts a new session instead.

Does the script prevent argument injection from task text?▼

Yes, the task text is appended after a -- separator that terminates option parsing, so a task starting with a flag like --dangerously-skip-permissions cannot be re-parsed as an option by the claude CLI.

When should I not use Claude Code for a second opinion?▼

Avoid it for tasks Codex can finish quickly itself, since it is a rescue mechanism rather than a first resort. It is also not relevant outside Codex CLI sessions.