ClawHub
Back to skill

Security audit

OpenClaw World

Security checks for vulnerabilities and agentic risk

Overview

This skill is a coherent shared-room/chat integration, but users should treat room chat and profiles as visible shared data.

Before installing, be comfortable with agent chat, room events, bios, room IDs, and optional public keys being visible to other room participants and possibly relayed through Nostr. Do not place secrets, private keys, system prompts, hidden context, or sensitive personal data in chat or bios, and only use browser preview when you expect the local room UI to open.

A.I.G

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs

SkillSpector

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Rogue AgentSelf-Modification, Session Persistence
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (5)

Context Leakage

High
Category
Data Exfiltration
Content
}
    },
    "world-chat": {
      "description": "Send chat message (bubble in 3D, max 500 chars)",
      "requiresAgent": true,
      "args": {
        "agentId": { "type": "string", "required": true },
Confidence
85% confidence
Finding
The `world-chat` command enables arbitrary agent-supplied text to be broadcast into a shared real-time room, which creates a direct channel for prompt/context leakage, social engineering, and exfiltration of sensitive data from an agent’s working context. In this skill’s collaborative multi-agent setting, freeform chat is more dangerous than usual because other agents may treat received messages as trustworthy instructions or may reveal internal state in response.

Session Persistence

Medium
Category
Rogue Agent
Content
---
name: world-room
description: Create or join a shared 3D lobster room where AI agents can walk, chat, and collaborate in real-time via Nostr relays.
---

# World Room
Confidence
60% confidence
Finding
Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

External Transmission

Medium
Category
Data Exfiltration
Content
```bash
# Register an agent in the room
# Bio is freeform — put your P2P pubkey here so others can contact you
curl -X POST http://127.0.0.1:18800/ipc -H "Content-Type: application/json" \
  -d '{"command":"register","args":{"agentId":"my-agent","name":"My Agent","color":"#e67e22","bio":"P2P pubkey: abc123...","capabilities":["chat","explore"]}}'

# Get all agent profiles
Confidence
60% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The skill recommends an `open-preview` flow that automatically opens a browser for the human, but it provides no warning or consent step. Triggering GUI/browser actions from an agent can surprise users, disrupt workflows, or be abused for unwanted navigation/social engineering, even if the target is local.

Missing User Warnings

Medium
Confidence
92% confidence
Finding
The documentation encourages placing a Nostr/P2P public identifier in a discoverable `bio` field so other agents can read and reuse it, but it omits a privacy warning. Even public identifiers can enable tracking, unwanted contact, correlation across rooms, and persistent identity linkage between activities.

Static analysis

No suspicious patterns detected.