The Agent Skills Directory

[SAFE]: The skill is designed as a security auditing tool and follows a structured, multi-phase threat modeling process. It uses deterministic Python scripts to perform technical scans and restricts LLM involvement to final report synthesis, reducing the risk of model-driven errors during detection.
[SAFE]: The workflow includes a critical human-in-the-loop gate in Phase 2, ensuring that security policy changes (deny-lists) are reviewed by an operator before being finalized. This prevents automated privilege escalation or policy bypass.
[SAFE]: The supply chain audit in Phase 3 proactively scans for advanced attack vectors such as zero-width characters, homoglyphs, and instruction-injection patterns, treating these as blocking critical findings.
[SAFE]: The tool adheres to the principle of least privilege by defaulting to a dry-run mode for database operations (Phase 4), requiring explicit operator confirmation and specific flags to execute destructive actions.
[SAFE]: All file system operations are scoped to a local security/ output directory for reporting, and the skill respects platform-native configuration files for its discovery process without attempting to bypass standard security boundaries.

security-threat-model