ai-pair

What This Skill Does

The ai-pair skill brings multi-model collaboration directly into your OpenClaw environment. By leveraging the native Agent Teams capability within Claude Code, it orchestrates a specialized workforce to tackle complex tasks. The architecture follows a 'one-creator, two-reviewer' model: a primary agent (Developer or Author) executes the core task, while two independent reviewers (Codex and Gemini) analyze the output through distinct cognitive lenses. This dual-reviewer approach ensures that your code or content is scrutinized for both technical integrity and stylistic excellence simultaneously, reducing the 'blind spots' inherent in relying on a single model.

Installation

To integrate this skill into your workflow, run the following command in your terminal:

clawhub install openclaw/skills/skills/axtonliu/ai-pair

Ensure that you have the Claude Code agent runtime, as well as the CLI tools for Codex and Gemini, pre-installed and authenticated on your local machine. The skill relies on these external binaries to pipe feedback back to your central Team Lead session.

Use Cases

• Software Development: Use this to build robust features, where Codex checks for edge cases, security vulnerabilities, and performance bottlenecks, while Gemini critiques your choice of design patterns and long-term codebase maintainability.
• Content Creation: Ideal for drafting newsletters, video scripts, or technical articles. While the author generates the draft, Codex ensures logical accuracy and factual consistency, and Gemini refines the tone, style, and audience engagement.
• Complex Planning: Use a pair team to review architectural proposals or project roadmaps by forcing competing models to find flaws in your initial logic.

Example Prompts

1. "/ai-pair dev-team PortfolioWebsite - Implement a React-based navigation bar with authentication state management."
2. "/ai-pair content-team Q4-Marketing-Plan - Draft an email newsletter announcement about our new AI integration features."
3. "/ai-pair team-stop"

Tips & Limitations

• Stay in Control: This skill is designed for human-in-the-loop workflows. You must manually approve results between the creation and review phases to ensure the agent doesn't enter an unwanted loop.
• Authentication: Always verify your Codex and Gemini CLI tokens before starting a session, as missing credentials will prevent the reviewers from initializing.
• Context Windows: Complex, massive projects may require you to break tasks into smaller sub-tasks, as the aggregated output of three agents can quickly consume your terminal session's history buffer.