pair-coding

Coordinate Claude Code and Codex pair-programming with proposer/reviewer sign-offs.

8|1|Updated Feb 22, 2026
One-click install
npx skills add https://github.com/ccomkhj/skills --skill pair-coding
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: pair-coding
Source: https://github.com/ccomkhj/skills/tree/main/pair-coding
Command: npx skills add https://github.com/ccomkhj/skills --skill pair-coding

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

It solves the problem of getting end-to-end, two-way accountability in code generation by forcing both Claude Code and Codex to propose and then review each other’s work at every stage.

Core Features & Use Cases

  • Mutual review at each stage: Claude/Codex alternate as proposer and reviewer for spec, plan, and each code step, preventing silent acceptance of flawed decisions.
  • Shared workflow state in .pair/: The session’s artifacts and disagreements live in .pair/ (PROGRESS, CONTEXT, DISCUSSION, SPEC/PLAN/STEP-N), so the loop is resumable and transparent.
  • Structured handoffs and human steering: Detached pair_handoff runs keep your terminal free, while pair_watch, pair_inject, and pair_takeover let you monitor or intervene safely.

Quick Start

Use pair-coding when you want two AI agents to mutually review a coding task from spec through implemented steps in a single coordinated session.

Frequently Asked Questions about pair-coding

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

FAQPage Schema
How do I set up mutual code review between Claude Code and Codex for a coding task?

Mutual code review between Claude Code and Codex is set up by enforcing alternating proposer and reviewer sign-offs for the spec, plan, and each code step, storing all shared workflow state and artifacts in a `.pair/` directory.

Can I resume an interrupted pair-programming session between two LLM agents?

Yes, you can resume an interrupted pair-programming session because all session artifacts, progress, context, and disagreements are stored as resumable shared state in a `.pair/` working directory.

How does cross-agent sign-off work for spec, plan, and incremental code steps?

Cross-agent sign-off works by forcing both LLM agents to alternate as proposer and reviewer at every stage, gating progression to the next step only after the reviewing agent provides explicit sign-off on the current spec, plan, or code.

What is the best way to handle disagreements between Claude and Codex during pair programming?

Disagreements between Claude and Codex are handled via a structured DISCUSSION log, allowing open disagreements to be tracked and resolved as decisions before being recorded as resolved context for future steps.

How do I monitor or intervene in a detached LLM agent coding workflow without blocking my terminal?

You can monitor or intervene in a detached coding workflow safely using file-based prompts, STATUS-driven handoffs, and specific commands like `pair_watch` to monitor, `pair_inject` to intervene, or `pair_takeover` to steer.

Do I need any external dependencies to coordinate Claude and Codex for two-way accountability in code generation?

No external dependencies are required to coordinate Claude and Codex for two-way accountability, as the skill relies entirely on file-based artifacts and shared state within a local `.pair/` directory to manage the workflow.