The Agent Skills Directory

[SAFE]: No malicious patterns or behaviors were detected. The skill is designed as a static security auditor and includes extensive documentation and references for identifying vulnerabilities.
[PROMPT_INJECTION]: The static analysis warning regarding instruction overrides is a false positive. The strings identified (e.g., "ignore previous instructions") are part of the references/vulnerability-patterns.md file, which contains regex patterns used by the agent to detect prompt injection in the target code it scans. These are not instructions for the agent itself to bypass its own safety guidelines.
[REMOTE_CODE_EXECUTION]: While the skill mentions various CLI tools like npx, syft, and trivy, these are provided as examples of commands for the user to generate SBOMs or as recommendations for security controls. The skill does not attempt to execute these commands automatically on the host system.
[EXTERNAL_DOWNLOADS]: The skill references several well-known security resources and organizations, including OWASP, CISA, Veracode, and GitGuardian. These references are used for academic and professional context and do not involve downloading or executing untrusted code.
[DATA_EXFILTRATION]: The skill includes instructions for detecting hardcoded secrets (using regex and entropy analysis) but does not contain any logic to transmit these secrets to external servers. It explicitly follows a "Never commit secrets" policy.

sentinel