agent-comms

Coordinate two AI sessions to review and converge on a shared artifact.

Updated May 15, 2026

Search Tags:#code review #claude code #codex #convergence #agent collaboration #shared channel

Ready to run

npx skills add https://github.com/klaidliadon/claude-plugins --skill agent-comms-klaidliadon

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: agent-comms
Source: https://github.com/klaidliadon/claude-plugins/tree/main/plugins/agent-comms/skills/agent-comms
Command: npx skills add https://github.com/klaidliadon/claude-plugins --skill agent-comms-klaidliadon

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill lets two AI sessions review the same artifact, exchange findings through a shared channel, and converge on a result without the human manually relaying messages.

Core Features & Use Cases

Coordinates an author and reviewer role for autonomous iteration on documents, plans, or code changes.
Supports durable message passing, hashed artifact snapshots, and explicit terminal states so the loop can stop cleanly.
Ideal for cross-model reviews such as Claude and Codex debating a proposal, reconciling findings, or confirming a final draft.

Quick Start

Ask the agent-comms skill to set up a shared review channel and drive the two sessions until they agree on the artifact.

Frequently Asked Questions about agent-comms

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

FAQPage Schema

How do I set up an autonomous code review loop between Claude and Codex?▼

Channel-based messaging coordinates autonomous review loops by passing findings between agents through a shared channel, preventing human message relaying and ensuring durable communication across separate sessions.

How do I stop an autonomous agent collaboration loop without manual intervention?▼

Terminal state tags stop autonomous review loops cleanly by marking explicit convergence states, preventing endless iteration between the author and reviewer roles on a shared artifact.

Does agent collaboration work with separate AI sessions for document review?▼

Yes, agent collaboration supports separate AI sessions by coordinating an author and reviewer role to iterate on documents or code changes, using hashed artifact snapshots to track feedback without human message relaying.

What's the best way to resolve feedback between two AI sessions without relaying messages?▼

Guarded wrappers for non-interactive reviewer launches resolve feedback between AI sessions by enabling autonomous message exchange and reconciliation through a shared channel without manual relaying.

When do I need snapshot hashing for agent collaboration workflows?▼

Snapshot hashing is needed for agent collaboration workflows when tracking artifact states across autonomous review loops, ensuring both author and reviewer reference verified versions during convergence.