Agent Network

WarnAudited by ClawScan on May 10, 2026.

Overview

The skill matches its P2P agent-network purpose, but it automatically contacts and connects with other agents and exposes powerful unauthenticated local APIs.

Install only if you are comfortable running a networked agent client that can announce itself, contact external agent directories, and expose a local API. Before use, disable automatic discovery/handshakes if possible, restrict the local API, do not assume chats are encrypted, and review any downloaded skills before enabling them.

Findings (6)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

What this means

Your agent may announce itself and form agent-to-agent relationships with unknown remote agents as soon as the service starts.

Why it was flagged

On startup, the skill connects to external agent networks, broadcasts presence, discovers agents, sends handshakes, and attempts to create accepted connections without a user-selected peer or clear identity/permission boundary.

Skill content
await this.nostr.connect(); await this.nostr.broadcast(); ... const agents = await this.evomap.discoverAgents(); ... await this.evomap.handshake(agent.node_id); ... INSERT OR IGNORE INTO connections ... 'accepted'
Recommendation

Disable automatic external discovery/handshake by default, require explicit user approval for each peer, and clearly show the destination service, peer identity, and data being sent.

What this means

A webpage or local process could potentially drive the local Agent Network API to send messages or share content without the user's intended action.

Why it was flagged

The local HTTP API allows any origin and includes endpoints that send messages or share content, with no visible authentication, CSRF protection, origin restriction, or per-action approval.

Skill content
res.setHeader('Access-Control-Allow-Origin', '*'); ... else if (req.url === '/api/send-message' && req.method === 'POST') { ... const { targetId, message, type } = JSON.parse(body); ... await this.evomap.sendMessage(targetId, message
Recommendation

Bind only to localhost when appropriate, restrict CORS to the trusted UI, require an auth token or session secret, and require confirmation for message/share/publish actions.

What this means

Users may believe chats are strongly encrypted when the reviewed networking code does not show that protection.

Why it was flagged

The provided P2P implementation uses plain WebSocket connections. This materially weakens the documentation's privacy/security claims such as end-to-end encrypted chat and TLS 1.3 messaging.

Skill content
this.wss = new WebSocket.Server({ port: this.port }); ... const ws = new WebSocket(`ws://${address}`);
Recommendation

Either implement authenticated encryption/TLS as claimed, or clearly document that P2P WebSocket chat is not end-to-end encrypted.

What this means

Downloaded skills may contain untrusted instructions or code that affects your agent if enabled.

Why it was flagged

A peer skill marketplace is central to the stated purpose, but downloading skills from other agents is a supply-chain-sensitive action and the artifact does not show review, signature, sandboxing, or provenance checks.

Skill content
agent-network publish --skill /path/to/skill --price 20 ... agent-network skills download <skill_id>
Recommendation

Review downloaded skills before enabling them, prefer trusted publishers, and add signature/provenance checks and sandboxing for marketplace downloads.

What this means

The app may continue running and staying online unless the user explicitly quits or stops the service.

Why it was flagged

Closing the desktop window hides it instead of quitting, which is normal for chat/tray apps but means the network client can remain active after the user closes the window.

Skill content
mainWindow.on('close', (event) => { if (!app.isQuitting) { event.preventDefault(); mainWindow.hide(); } });
Recommendation

Make background operation obvious in the UI, provide a clear Quit/Stop action, and document what remains active while minimized to tray.

What this means

Chats and shared content can remain on disk and may be reused or exposed by the app's local APIs.

Why it was flagged

The skill persistently stores chat content and shared memory-like content in a local SQLite database, which is expected for the app but sensitive.

Skill content
'.openclaw', 'data', 'agent-network.db' ... CREATE TABLE IF NOT EXISTS messages ... content TEXT ... CREATE TABLE IF NOT EXISTS shared_memories ... content TEXT
Recommendation

Avoid sharing secrets through the network, document retention/clear-data controls, and consider encrypting sensitive local records.