openclaw-china-docker

Warn

Audited by Gen Agent Trust Hub on May 17, 2026

Risk Level: MEDIUMPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill documentation includes an explicit instruction in Pattern 3 (provided in Chinese) that directs the agent to ignore its default web search capabilities in favor of a specific third-party plugin.
  • Ingestion points: User conversation (Pattern 3 instructions meant for chat input).
  • Boundary markers: Absent from the recommended prompt string.
  • Capability inventory: Host system access via Docker socket, persistent storage via volume mounts, and network access.
  • Sanitization: No sanitization or safety delimiters are suggested for the injected instructions.
  • [COMMAND_EXECUTION]: The documentation and example configurations (Pattern 4 and Troubleshooting) recommend mounting the host's Docker socket (/var/run/docker.sock) into the container. This configuration provides the container—and any code executed by the agent within it—administrative privileges over the host's Docker daemon, representing a significant privilege escalation risk and enabling container escape.
  • [EXTERNAL_DOWNLOADS]: The skill relies on external code and assets from various sources, with notable naming inconsistencies between the GitHub repository (justlovemaki) and the DockerHub image (justlikemaki).
  • Evidence: Repository URL github.com/justlovemaki/openclaw-china-docker.git vs. Docker image justlikemaki/openclaw-docker-cn-im.
  • Plugin Installation: Uses a custom command npx openclaw plugin:install to fetch and execute third-party code (larksuite/openclaw-lark) at runtime.
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 17, 2026, 09:12 AM
Security Audit — agent-trust-hub — openclaw-china-docker