The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructions in SKILL.md direct the AI to proactively search for management keys in sensitive locations, specifically the MANAGEMENT_PASSWORD environment variable and a config.yaml file. This pattern encourages the agent to expose and handle high-privilege credentials.
[DATA_EXFILTRATION]: The script scripts/cliproxy_api.py includes a configurable --url parameter and methods to retrieve API keys (e.g., Gemini, Claude, OpenAI) from the service. An attacker could potentially manipulate the agent into transmitting these harvested keys to a malicious external server.
[EXTERNAL_DOWNLOADS]: The Python script scripts/cliproxy_api.py imports and relies on the requests library for all network operations.
[DATA_EXPOSURE]: The put command in the Python script contains logic that checks if the input data is a local file path (os.path.isfile(data)). If it is, the script reads the file content and sends it to the API, which could be exploited to leak local file contents if the AI is tricked into passing sensitive paths as arguments.

cliproxy-manager