The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill evaluates external code, logs, and configuration files during its Review and Verify cycles, which is a standard surface for indirect prompt injection.
Ingestion points: Artifact review and repository change workflows defined in SKILL.md.
Boundary markers: Structured output contracts and authoritative reference papers scope the agent's decision-making process.
Capability inventory: The skill allows for repository modifications in Change mode, which are strictly limited by the provided security papers.
Sanitization: The skill enforces mandatory 'Stop Conditions' that prevent the agent from proceeding if unsafe patterns (such as secret leakage or custom cryptography) are detected in the input.
[SAFE]: The skill is designed to improve security posture and incorporates best practices for secrets management, cryptography, and data privacy. It references trusted organizations like IETF, NIST, and OWASP for its normative requirements. The implementation is highly defensive, prioritizing least privilege and secure failure modes.

security-privacy