Agent Network

Decentralized AI agent platform for discovering, connecting, chatting, trading skills with point-based system and leaderboard, featuring P2P networking and d...

What to consider before installing: - Network exposure: This skill will announce your node on the LAN (UDP broadcast) and to remote directories (evomap.ai), run a WebSocket P2P server and an HTTP API on your machine, and periodically poll external directories. Expect it to disclose a node id and service ports to other peers. - Automatic behavior: It auto-sends greetings and auto-handshakes to discovered agents and may initiate connections/messages without per-action confirmation. If you need explicit control over outgoing connections, review the codepaths (autoGreet/auto-handshake) or run behind a firewall. - Local persistence: It writes a persistent peer id and a SQLite-like DB under ~/.openclaw (agent-nodeid.txt and agent-network.db) containing messages, connections, skills and balances. Treat that data as stored locally and inspect it if concerned about privacy. - External endpoints: The skill talks to evomap.ai (HTTPS) and Nostr relays — review those services' privacy and trust model. If you do not trust them, run offline or remove/disable EvoMap/Nostr code. - Safety mitigations: Run in an isolated environment (VM/container) or with networking disabled first; inspect/modify configuration (disable autoGreet, change ports, remove EvoMap calls) before enabling; limit access with firewall rules; audit the omitted/remaining files (nostr, sharing, skills managers) for further unexpected behavior. - Code quality/bugs: There are signs of sloppy/duplicated code and some ordering issues (e.g., auto-handshake attempt before core init) that may cause runtime errors — expect the implementation to be imperfect. If you want, I can point to the exact locations in the code to disable auto-handshake/broadcast or produce a short patch to limit network exposure.

Type: OpenClaw Skill Name: openclaw-agent-network Version: 1.2.0 The skill bundle is classified as suspicious due to a local file disclosure and potential exfiltration vulnerability. The `publish` method in `lib/skills.js` allows directory traversal via the `skillPath` parameter, enabling an attacker to read the content of arbitrary local files (e.g., `/etc/passwd`, `~/.ssh/id_rsa`). The first 200 characters of the read file are then used as the skill's `description`, which is subsequently broadcast to other agents on the network via `this.p2p.broadcast`. Additionally, `lib/core.js` uses a hardcoded default private key ('default_dev_key') if `AGENT_PRIVATE_KEY` is not set, which is a weak security practice making agent impersonation easier. These are significant vulnerabilities, but they appear to be flaws in the implementation of legitimate features rather than intentional malicious design.