keep-protocol

What This Skill Does

The keep-protocol skill enables high-performance, secure, and asynchronous communication between AI agents within the OpenClaw ecosystem. By leveraging signed Protobuf packets over TCP, it ensures that messages are authenticated via ed25519 signatures, preventing unauthorized interaction or spoofing. The protocol supports essential distributed system capabilities, including agent discovery, request routing, and memory sharing through its unique 'scar' field feature. It is designed to be lightweight, offering a fast path via MCP tools for sub-60ms latency, while providing a robust fallback through its dedicated Python SDK. This makes it an ideal backbone for multi-agent workflows where coordination, shared task state, and institutional knowledge exchange are required.

Installation

To integrate this skill, use the ClawHub command line utility: clawhub install openclaw/skills/skills/nteg-dev/keep-protocol

For environments requiring the high-performance MCP integration, ensure you have the server running and configured in your MCP settings. Install the client-side requirements using pip install keep-protocol[mcp]. For standard agent-to-agent deployments without MCP, you may deploy the standalone Docker container using docker run -d -p 9009:9009 ghcr.io/clcrawford-dev/keep-server:latest.

Use Cases

• Multi-Agent Orchestration: Use a controller agent to dispatch sub-tasks to specialized worker agents (e.g., weather-agent, data-analyzer).
• Shared Memory Bartering: Exchange encrypted or serialized memory segments between agents to prevent redundant computations.
• Low-Latency Signaling: Trigger real-time status updates or event notifications across a cluster of agents with minimal overhead.

Example Prompts

1. "Check which agents are currently online using the keep-protocol discovery tool and report the system version."
2. "Send a request to the bot:planner agent to coordinate the task queue and wait for its confirmation reply."
3. "Initiate a memory barter transaction with bot:travel-agent to exchange local flight cache data for weather analysis results."

Tips & Limitations

• Performance: Always prioritize the MCP toolset for communication to achieve the ~60ms latency target. The Python SDK fallback is reliable but carries the overhead of code execution.
• Security: Ensure your ed25519 keys are managed securely. The protocol performs silent drops on invalid signatures, so verify network connectivity if messages seem to be vanishing.
• Memory: The scar field is intended for structured binary data or git-style memory commits; avoid sending excessively large objects to maintain protocol stability.