OpenClaw Agent Mesh

Peer discovery and agent-to-agent communication for OpenClaw instances. Use when the user wants nearby OpenClaw nodes to discover each other, request contact...

This skill implements the described mesh functionality, but it contains a serious implementation issue: it constructs filenames using sender-controlled fields (request_id, message_id) without sanitization. A remote peer that sends a specially crafted contact request could cause the service to write files outside the intended state directory (including absolute paths), enabling arbitrary file writes as the user running the server. Before installing or running the server, consider: 1) Do not run the server on public/untrusted networks; restrict binding to localhost or a trusted interface. 2) Do not run the server as a privileged user — run under a dedicated low-privilege account. 3) Require the maintainer to fix the code: validate/sanitize request_id and message_id (allow only safe characters like [A-Za-z0-9_.-], ban path separators and leading slashes), and avoid using unsanitized payload fields as path components. 4) Ensure 'openssl' is present and trusted (skill relies on it though it's not declared). 5) Consider running the scripts in a sandbox or container and review all incoming payload handling. If you cannot get a patched version that sanitizes filenames, treat the skill as unsafe to expose to untrusted peers.

Type: OpenClaw Skill Name: openclaw-agent-mesh Version: 1.0.0 The skill implements a functional peer-to-peer communication and discovery layer for OpenClaw agents. It includes scripts for identity generation (using Ed25519 via OpenSSL), LAN scanning, and signed message exchange. The implementation in `scripts/mesh.py` and `scripts/server.py` follows the documented protocol, requiring explicit manual approval of contact requests before establishing trust. No evidence of data exfiltration, unauthorized execution, or malicious prompt injection was found.