Peer AI — AI skill for Claude Code

peer-ai

**AI-to-AI peer consultation.** Let Claude Code, Codex CLI, and Antigravity CLI (agy) ask each other for a second opinion, code review, or deep analysis — with one command.

[](https://www.npmjs.com/package/@pilosite/peer-ai) [](LICENSE) [](https://nodejs.org)

---

Why

You're using Claude Code. You reach a conclusion that feels load-bearing — a security-sensitive fix, an architecture call, a regex nobody understands. Before you ship it, you want a **second opinion from a different model family**: Codex for typed-language rigor, Antigravity (Google) for a fresh perspective, another Claude session for independent validation.

Today you'd open another terminal, copy-paste the diff, write a brief, paste the response back. Multiply by 10 reviews a week.

**peer-ai** installs a symmetric skill/extension in each AI CLI you use, so any of them can call the others with one line:

/peer-ai codex review the last 2 commits
/peer-ai claude second opinion on this auth flow
/peer-ai antigravity what's wrong with this SQL

The source AI assembles a smart brief (diff, relevant files, your question), calls the target CLI in read-only mode, and relays the response verbatim. Bounded to **5 rounds per target per exchange chain** so you don't get stuck in an AI-to-AI loop. The cap is per-dialogue (auto-resets after 30 min of inactivity for that target), not per CLI session — see [Managing the cap](#managing-the-cap) for details.

What it is

peer-ai is a **Node zero-dependency installer** that scaffolds one skill file per AI CLI you pick:

Source AI	What gets installed	Invocation
Claude Code	`~/.claude/comm