AIpandadreams

Multi Agent Protocol — AI skill for Claude Code

AI community

A four-party collaboration protocol for running AI agent teams on real work — auditable authorization, durable cross-session state, and independent review gating every change.

How to install Multi Agent Protocol

This entry records only its repository, not the path inside it, so there is no exact command to give. Open AIpandadreams/multi-agent-protocol and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Multi Agent Protocol does

A four-party collaboration protocol for running AI agent teams on real work — auditable authorization, durable cross-session state, and independent review gating every change.

Alternatives in AI

Debate — AI Debate Hub - Structured four-way debates between Claude, Sonnet, Gemini, and Codex 2.8k ★
Meme Rush — Track real-time meme token lists from launchpads (Pump.fun, Four.meme) and AI-powered hot market topics ranked 483 ★
Empirica — Make AI agents and AI workflows measurably reliable 244 ★

README

multi-agent-protocol

**A multi-agent collaboration protocol for running AI agent teams on real work — without losing control of authorization, history, or quality.** Five default participants; four of them hold authority (principal, owner, builder, reviewer) — the orchestrator is an interface, never an authority.

[![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](LICENSE) [![CI](https://github.com/AIpandadreams/multi-agent-protocol/actions/workflows/mirror-check.yml/badge.svg)](https://github.com/AIpandadreams/multi-agent-protocol/actions/workflows/mirror-check.yml) [![Release](https://img.shields.io/github/v/release/AIpandadreams/multi-agent-protocol?sort=semver)](https://github.com/AIpandadreams/multi-agent-protocol/releases) [![Protocol](https://img.shields.io/badge/protocol-v3.1-informational)](docs/PROTOCOL.md) [![Python](https://img.shields.io/badge/python-3.10%2B-blue)](https://www.python.org/downloads/) [![Claude Code plugin](https://img.shields.io/badge/Claude%20Code-plugin-8A2BE2)](docs/QUICKSTART.md#1-install-the-plugin) [![No telemetry](https://img.shields.io/badge/telemetry-none-brightgreen)](#trust-properties)

(The release number and the `PROTOCOL vX.Y` stamp in the badges above are **different counters** that move independently — see [PROTOCOL.md § Version spaces](docs/PROTOCOL.md#version-spaces).)

Most multi-agent setups fail the same three ways: an agent quietly does something it was never authorized to do, a session dies and takes its state with it, or agents review their own work and call it done. This protocol is the distillation of a system that ran for months on real production work — every rule in it exists because its absence caused a real defect.

The parties

party	who	job
Principal	you, the human	holds ALL authorization gates. Authorization is only ever your word
Orchestrator	a Claude Code session	your single point of contact: runs the queue, dispatches the workers, bri