multi-agent-communication

Coordinates parallel agent sessions over a project message bus with broadcast events and ack discipline.

2|Updated Jul 18, 2026
One-click install
npx skills add https://github.com/Arasz/ai-badger --skill multi-agent-communication-arasz
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: multi-agent-communication
Source: https://github.com/Arasz/ai-badger/tree/main/features/common/skills/multi-agent-communication
Command: npx skills add https://github.com/Arasz/ai-badger --skill multi-agent-communication-arasz

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? When multiple agent sessions share one project, they overwrite each other's files, duplicate work, and miss review requests because there is no shared channel. This Skill defines a coordination protocol over the project message bus so sessions announce work, request reviews, and react to merges without colliding. ## Core Features & Use Cases - Broadcast protocol: Defines six event types (starting, pr-opened, review-request, review-feedback, merged, blocked) with a one-line message shape sent via the send-message transport. - Ack-without-reply rule: Every non-ack message gets exactly one ack, and acks are terminal, preventing infinite reply loops that burn LLM turns per session. - Delivery guidance: Covers per-harness delivery seams (pi message-bus extension, Claude turn-end injection, Hermes pre-call delivery) and warns against blocking on acks. - Use Case: Two agents work in the same repo on separate branches; one announces merged: #412 to main, the other acks and rebases its lane before merging, avoiding a conflict. ## Quick Start Announce my current task, opened PRs, and merge status to the other agent sessions in this project over the message bus, and ask them for review.

Frequently Asked Questions about multi-agent-communication

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

FAQPage Schema
How do I coordinate multiple AI agents working in the same repository?

Broadcast lifecycle events over the project message bus: announce starting work with files touched, pr-opened, review-request, and merged. Other sessions ack each message once and adjust, for example rebasing after a peer's merge.

How do I send a message to other agent sessions in a project?

Run the send_message.py script with --project-id read from .ai-badger/project-id and content formatted as [<taskId>] <event>: <detail>. Under pi with the message-bus extension, use the message-bus send tool with the project id instead.

Why do agent message loops burn tokens and how do I stop them?

Loops happen when sessions reply to acknowledgments, restarting the conversation forever with a full LLM turn per cycle. The rule is: ack every non-ack message exactly once, and never reply to a message starting with ack:.

Does the message bus deliver instantly to idle agent sessions?

No. An idle Claude session fires no hook until its next turn, and pi machine-broadcasts do not wake sessions unless AI_BADGER_PI_BUS_WAKE=all. Never block waiting for an ack; announce and continue working.

When should agents not use the message bus?

Skip the bus when no other session is active in the project and just do the work. Also avoid narrating every commit or push; broadcast once per milestone at package boundaries, since the PR UI already narrates commits.