The Agent Skills Directory

[SAFE]: The skill provides structured auditing for AI CLI configurations, adhering to a strict policy of providing recommendations only and excluding authentication files from its scope.
[SAFE]: It promotes security-centric features such as sandboxing, least-privilege MCP access, and the implementation of automated security hooks to detect secret leaks and dangerous commands.
[SAFE]: The skill operates as a collection of Markdown instructions and reference templates without including executable code or unverifiable external dependencies.
[SAFE]: References to external documentation and repositories are limited to official and well-known sources from established technology organizations.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface as it ingests untrusted configuration data (Ingestion points: ~/.claude/settings.json, ~/.codex/config.toml, etc.). However, it mitigates risk through clear boundary markers (Explicitly excludes credential files) and providing sanitization logic (Security regex and permission decision logic in hook recipes). Capability inventory includes tool-assisted file reading and hook script generation.
[NO_CODE]: The skill consists entirely of instructional markdown and reference files, providing the agent with guidance rather than shipping binary executables.

hone