The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface by ingesting and processing external data from the filesystem while maintaining modification capabilities.\n
Ingestion points: The scripts scan_skills.py, repair_skill.py, and analyze_similarity.py read metadata and content from various SKILL.md files and the skills.json registry.\n
Boundary markers: There are no clear delimiters or safety instructions used when processing and reporting on the content of these external files.\n
Capability inventory: The skill has significant capabilities, including creating new directories, modifying existing SKILL.md files, updating the central skills.json registry, and packaging files into ZIP archives using init_skill.py, repair_skill.py, and package_skill.py.\n
Sanitization: The skill uses yaml.safe_load() for YAML parsing, but it does not perform robust sanitization or validation of the natural language descriptions and metadata extracted from the skill files before including them in generated reports or registries.

skill-creator-doctor