The Agent Skills Directory

[SAFE]: The skill body provides comprehensive guidance for authoring and reviewing Agent Skills, emphasizing best practices, portability, and security.
[SAFE]: An explicit safety rule (Rule 20) in SKILL.md prohibits the creation of skills for unauthorized access, exfiltration, or deception.
[SAFE]: In references/client-extensions-and-paths.md, the skill identifies and warns against dangerous host-specific extensions such as dynamic shell injection and forced file attachments, classifying them as anti-patterns and recommending safer alternatives like relative paths.
[SAFE]: The skill uses and recommends standard ecosystem tools (npx skills, npx skills-ref) for validating and managing skills, which is consistent with the described domain.
[SAFE]: Evaluation fixtures in evals/files/ contain examples of poorly authored or insecure skills to test the agent's ability to identify and remediate these issues during a review.
[SAFE]: All external references in NOTICE.md and SOURCES.yaml are to well-known service documentation and trusted open-source repositories for the purpose of attribution and technical reference.

skill-authoring