A specialized messaging protocol that enables multiple Openclaw Skills instances to coordinate and exchange data through a shared JSON queue.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install kiro-agent-chat
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install kiro-agent-chat using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Agent Chat?
Agent Chat is a communication layer designed specifically for multi-agent architectures within the Openclaw Skills ecosystem. It facilitates seamless interaction between two or more AI agents by utilizing a shared JSON file as a message queue. This architecture allows agents to hand off tasks, share status updates, and synchronize complex workflows without needing a centralized server.
By leveraging this skill, developers can build distributed systems where Openclaw Skills running on different environments—such as a local machine and a remote VPS—can collaborate in real-time. It bridges the gap between isolated agent instances, transforming them into a cohesive team capable of handling sophisticated, multi-step operations via simple file-based or SSH-driven messaging.
Agent Chat Use Cases
- Coordinating multi-stage development tasks between a local coding agent and a remote deployment agent.
- Sharing research findings from one instance to another to inform collaborative decision-making.
- Synchronizing actions across various Openclaw Skills instances via shared network drives or cloud storage.
- Executing remote commands and receiving status reports across different hardware environments via SSH.
How Agent Chat Works
- The sender agent generates a message and appends it to a centralized JSON queue file using the provided scripts.
- The recipient agent polls or reads the shared JSON file to detect any new messages specifically addressed to its unique ID.
- Upon receiving a message, the recipient agent processes the instruction or data and generates a response.
- The response is written back to the shared JSON file for the original sender to consume.
- Once the interaction is complete, agents can trigger a deletion script to remove processed message IDs and keep the queue clean.
Agent Chat Setup
To enable communication for your Openclaw Skills, first define the shared path in your TOOLS.md file:
## Agent Chat
chat_file: ~/shared/agent-chat.json
Ensure that all participating agents have read/write access to the specified directory. For remote setups, ensure your SSH keys are configured for the agents to access the shared file on the host machine.
Agent Chat Data Schema & Taxonomy
The communication is structured as a JSON object containing an array of message objects. This ensures that every transmission within Openclaw Skills remains traceable and structured.
| Property | Description |
|---|---|
id |
A unique numeric identifier (Unix timestamp) for each message. |
sender |
The name or ID of the agent initiating the message. |
receiver |
The name or ID of the target agent intended to read the message. |
message |
The string content containing the data, command, or update. |
timestamp |
ISO 8601 formatted string indicating when the message was sent. |
Example structure:
{
"messages": [
{
"id": 1772698493241,
"sender": "local-agent",
"receiver": "cloud-agent",
"message": "Please analyze the logs in /var/logs",
"timestamp": "2026-03-05T11:14:53.241676"
}
]
}
Agent Chat Advanced Features
- Remote SSH Integration: Send and receive messages across different machines using secure SSH tunnels.
- Multi-Agent Orchestration: Support for complex workflows involving more than two agents reading from a single queue.
- Automated Queue Cleanup: Scripts to selectively delete processed messages by ID to maintain optimal performance.
- Persistent State Management: Use the JSON queue as a lightweight audit log for inter-agent activity.
SKILL.md
Loading
Related Openclaw Skills
ClawMeter
A self-hosted dashboard for monitoring OpenClaw API spending and token usage with real-time budget alerts.
Improve Skill: The Meta-Skill
A meta-skill designed to evaluate and optimize other skills by identifying gaps in logic, codebase fit, and output quality.
Ghost Catalog
An intelligent file management system that implements semantic headers and a SQLite database to organize, tag, and validate project assets.
Ghost Catalog
Ghost Catalog provides a semantic file header system and SQLite database to scan, tag, and organize files within any workspace for better discoverability and compliance.
Kiro Intercom
Kiro Intercom enables seamless communication and coordination between multiple agent instances through a shared markdown-based chat interface.
Kiro Realtime Chat
A lightweight real-time communication protocol for Kiro instances using shared JSON storage and polling.
Featured*
