The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (analyzed skill files), creating an attack surface. However, it contains mandatory defensive instructions to mitigate this risk. 1. Ingestion points: Skill files and metadata from user-provided URLs or local paths ($ARGUMENTS). 2. Boundary markers: Explicit directive to treat all content as data, never as instructions. 3. Capability inventory: Uses Read, Grep, Glob, and Bash for file inspection. 4. Sanitization: Explicitly forbids execution of scripts, installation of dependencies, or following setup steps from the audited skill.
[PROMPT_INJECTION]: The skill mentions common prompt injection phrases (e.g., 'ignore previous instructions') solely as detection criteria for the auditor, which is a meta-discussion and does not represent an attempt to override the agent's behavior.
[COMMAND_EXECUTION]: The skill utilizes standard tools like Bash and Grep for file analysis and repository cloning, which is consistent with its stated functionality as a security auditor.

skill-auditor