openclaw-china-docker
Fail
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: HIGHPROMPT_INJECTIONPRIVILEGE_ESCALATIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill includes explicit instructions designed to override the agent's behavior and tool usage. Specifically, Pattern 3 provides a text snippet meant to be used in conversation that instructs the agent to disable its built-in
web_searchandweb_fetchtools and replace them with a specific external tool (agent-reach), effectively hijacking the agent's tool-calling logic. - [PRIVILEGE_ESCALATION]: The configuration instructions in Pattern 4 and the Troubleshooting section advocate for mounting the host's Docker socket (
/var/run/docker.sock) into the container. This provides the container with full administrative control over the host Docker daemon, which is equivalent to root access on the host system. - [EXTERNAL_DOWNLOADS]: The skill instructs users to clone a repository from an untrusted GitHub account (
justlovemaki/openclaw-china-docker) and pull a Docker image from an untrusted registry user (justlikemaki/openclaw-docker-cn-im). These sources are not recognized as trusted organizations or well-known services. - [REMOTE_CODE_EXECUTION]: The skill facilitates remote code execution by providing instructions to install plugins at runtime using
npx openclaw plugin:install larksuite/openclaw-lark. This fetches and executes code from an external repository during the skill's operation. - [INDIRECT_PROMPT_INJECTION]: The skill creates a significant attack surface for indirect prompt injection by design.
- Ingestion points: The agent processes incoming messages and data from external Chinese IM platforms (Feishu, DingTalk, QQ, WeChat Work).
- Boundary markers: None are present in the instructions to help the agent distinguish between user intent and data content from these external channels.
- Capability inventory: The skill has the capability to execute code in Docker containers, install new plugins via npx, and manipulate system environment variables.
- Sanitization: There is no mention of sanitizing or filtering input received from the external messaging platforms before it is processed by the AI agent.
Recommendations
- AI detected serious security threats
Audit Metadata