LocalUDPMessenger

Use when agents need to communicate over the local network — "send message to agent", "discover agents", "check for messages", "coordinate with other agents", "approve agent", "agent status", "add peer", "message log"

This plugin appears to do what it says, but check these before installing: (1) Review the npm/GitHub source (owner turfptax) to ensure you trust the publisher. (2) Do not set a relayServer to an untrusted host — enabling relay forwards copies of every message off your LAN. (3) Treat the hook token as sensitive: if set, it lets the plugin trigger agent turns via /hooks/agent; keep it secret and only enable hooks.enabled if you understand the implications. (4) Trusted peers are persisted to ~/.openclaw/extensions/openclaw-udp-messenger/trusted-peers.json — inspect that file and the plugin logs if you need an audit trail. (5) Prefer 'always-confirm' trust mode on untrusted networks and never instruct the agent to share secrets with peers. Finally, note a minor inconsistency: the registry marked the skill as 'instruction-only' while package files are present; confirm the exact install path you will use (npm vs. manual) and review the repository before enabling in production.

Type: OpenClaw Skill Name: udp-messenger Version: 1.6.1 The plugin is classified as suspicious primarily due to the `relayServer` feature in `index.ts` and `openclaw.plugin.json`, which transparently forwards all agent-to-agent message payloads to an external, configurable monitoring server. While documented as an optional human monitoring tool, this constitutes a direct data exfiltration vector that could be misused if configured to a malicious endpoint. Additionally, the `wakeAgent` mechanism and `udp_set_config` tool in `index.ts` provide powerful capabilities (triggering agent turns with message content and runtime configuration changes) that, despite explicit instructions in `SKILL.md` and `CLAUDE.md` to prevent prompt injection and data leakage, still present a significant attack surface for an AI agent if those instructions are bypassed.