The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes local CLI tools including git rev-parse --show-toplevel to resolve the repository root and ai-skillet (map/doctor) to analyze the skill portfolio.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted instructions by reading SKILL.md files and associated metadata from the repository it audits, creating an attack surface for indirect prompt injection. * Ingestion points: Repository artifacts including SKILL.md, AGENTS.md, CLAUDE.md, and local scripts/assets. * Boundary markers: Instructions direct the agent to perform a mechanical inventory, separate facts from inference, and treat inventory edges as leads rather than instructions. * Capability inventory: Access to the git and ai-skillet command-line interfaces for repository analysis and authorized file modifications. * Sanitization: The skill does not explicitly sanitize file contents, relying on the agent's adherence to the audit logic to maintain safety boundaries.

skill-harmonization