The Agent Skills Directory

[SAFE]: The skill establishes a structured and defensive methodology for software auditing. It includes explicit operating principles that require validating every claim against primary sources and obtaining user consent for choices affecting security posture.
[PROMPT_INJECTION]: No evidence of malicious prompt injection or attempts to bypass safety filters was found. The skill is designed to improve the quality of technical responses through structured routing.
[DATA_EXFILTRATION]: No sensitive file access or network exfiltration patterns were detected. The use of configuration paths like ~/.claude is documented for legitimate stack profiling purposes.
[EXTERNAL_DOWNLOADS]: While the methodology mentions referencing official documentation and advisories, it does not contain URLs or commands for automated, unverified downloads.
[COMMAND_EXECUTION]: The skill lists numerous security tools (e.g., gitleaks, trufflehog, semgrep) to be used during audits, but it does not perform automated execution of these tools or any other system commands.
[SAFE]: In the context of indirect prompt injection, the skill identifies ingestion points from user requests and repository code but mitigates risks by mandating that findings cite specific evidence (file:line) and be validated against primary sources.

sota