The Agent Skills Directory

[SAFE]: The skill instructions define a read-only auditing process. No commands for exfiltration, persistence, or unauthorized data access were identified.
[INDIRECT_PROMPT_INJECTION]: The skill processes content from entire codebases, which represents an ingestion surface for untrusted data. However, the risk is mitigated by the skill's limited capabilities. * Ingestion points: Reads the full repository tree as specified in the description (SKILL.md). * Boundary markers: None explicitly defined to separate untrusted code from instructions. * Capability inventory: The skill is limited to generating a one-shot text report; it contains no file-write, network, or subprocess capabilities. * Sanitization: No specific sanitization or filtering of the ingested code is mentioned.

ponytail-audit