The Agent Skills Directory

[SAFE]: The skill is designed to perform a metadata audit of the agent's own environment (agents, skills, rules, and MCP configurations).
[DATA_EXPOSURE]: While the skill reads local files like agents/*.md, skills/*/SKILL.md, and .mcp.json, this is restricted to the agent's configuration workspace for the purpose of token counting and line estimation. No sensitive user data (e.g., SSH keys, AWS credentials) is accessed.
[REMOTE_CODE_EXECUTION]: There are no patterns involving remote script downloads or the execution of untrusted external code.
[COMMAND_EXECUTION]: The skill describes scanning directories but does not include explicit shell commands or subprocess calls that would execute arbitrary user input.
[PROMPT_INJECTION]: The instructions do not contain attempts to override agent behavior or bypass safety filters.

context-budget