The Agent Skills Directory

[SAFE]: The skill implements a robust security and quality posture by requiring behavior-preserving refactors to be validated through characterization tests and build-green gates prior to completion.
[INDIRECT_PROMPT_INJECTION]: The skill involves reading and analyzing repository content, which could include untrusted instructions in code comments or documentation. The risk is mitigated by the requirement for a structured 'Candidate Card' and the enforcement of objective validation commands (tests and builds) that prevent arbitrary or malicious instructions from being executed without verification. Evidence Chain: Ingestion points (Repository instructions, documentation, and source code files); Boundary markers (Structured 'Candidate Card' template); Capability inventory (File system modification and execution of local test/build tools in SKILL.md sections 6.5, 7, and 8); Sanitization (None).

safe-refactor