The Agent Skills Directory

[SAFE]: The skill provides comprehensive safety guardrails for refactoring, including mandatory characterization tests and a strict 'one boundary per step' rule to prevent unintended side effects.
[SAFE]: It contains explicit warnings against using refactoring as a cover for malicious behavior, specifically mentioning that context extraction must not be used to obscure exfiltration, secret logging, or new network endpoints.
[SAFE]: The verification protocol requires the agent to provide 'Observed output' from actual command runs (mix test), which serves as a defense against hallucinated or fabricated results.
[SAFE]: External tool references (Credo, Sobelow, Swoosh) are industry-standard for Elixir development and are used in a defensive context to maintain code quality and security.

refactor-code