The Agent Skills Directory

PROMPT_INJECTION (LOW): The skill processes frontmatter from external skill files to generate manifests and perform linting, creating an indirect injection surface.\n- Ingestion points: generate_manifest.py, skill_lint.py, and quick_validate.py read frontmatter from SKILL.md files.\n- Boundary markers: Absent. Content is interpolated directly into manifest files and linting logs.\n- Capability inventory: subprocess.run for running local linting tasks and file writes for generating manifests and zipping skills.\n- Sanitization: Uses yaml.safe_load and enforces character length limits in quick_validate.py.\n- COMMAND_EXECUTION (SAFE): skill_lint.py executes local Python scripts using subprocess.run in a controlled, non-dynamic manner.\n- DATA_EXFILTRATION (SAFE): No unauthorized file access or network communication patterns were identified.

skill-creator