project-explorer-skill

Explores unfamiliar GitHub projects, installs and runs them, analyzes architecture, and generates comprehensive documentation guides

This skill is coherent in purpose but has meaningful gaps and gives the agent broad discretion to run untrusted code. Before installing or using it: 1) Require the agent to ask for explicit permission before cloning, installing, or executing anything. 2) Run any dynamic execution in an isolated sandbox/container or a disposable VM, not on your primary machine. 3) Never provide sensitive credentials; remove or sanitize .env files and avoid sharing secrets. 4) Prefer that the skill perform static analysis first and only run code after you review the exact commands. 5) Ask the skill author to declare required binaries/runtimes and to add explicit safety checks and confirmation prompts. If you cannot guarantee sandboxing or safe handling of secrets, use the skill for read-only analysis (file inspection, architecture mapping) rather than actually installing/running third-party projects.

Type: OpenClaw Skill Name: project-explorer Version: 1.0.0 The skill bundle, specifically in skill.md, instructs the AI agent to clone arbitrary GitHub repositories, install their dependencies, and execute the code on the user's machine to 'explore' them. This design creates a high risk of Remote Code Execution (RCE) and makes the agent vulnerable to indirect prompt injection or malware contained within the target repositories. While the stated intent is educational, the instruction to run untrusted code without sandboxing or safety constraints is a significant security risk.