The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill's primary purpose is to identify and access sensitive credentials. The documentation explicitly mentions secrets such as BROWSER_PASSWORD and SOME_API_KEY.
[DATA_EXFILTRATION]: Although the provided script llm-secrets.js only lists key names, the SKILL.md file provides direct instructions for the agent to retrieve the underlying secret values using echo $KEY_NAME. This pattern facilitates the exposure of sensitive system environment variables to the LLM context, which can then be exfiltrated via subsequent tool calls or network requests.

llm-secrets