The Agent Skills Directory

[SAFE]: The skill defines a structured methodology for knowledge persistence. It uses allowed tools like Bash and Write to manage documentation within the project's .claude/skills/ directory. The instructions include a 'Quality Gates' phase that specifically mandates the exclusion of sensitive information such as credentials or personal data.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect injection as it reads existing skills using grep to avoid duplication.
Ingestion points: Reads files in .claude/skills/ via grep and ls (SKILL.md).
Boundary markers: The workflow uses a specific YAML frontmatter and Markdown structure for new skills, providing some schema separation.
Capability inventory: The skill has Read, Write, and Bash access to the local project environment.
Sanitization: The 'Quality Gates' phase explicitly requires the agent to filter out sensitive information (credentials, personal data) and verify the solution.

learn