The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses importlib.import_module and importlib.util.spec_from_file_location to dynamically load and execute Python code from other skill directories at runtime based on the skill name provided.
Evidence: Found in scripts/skill_imports.py within the skill_import function and SkillImportFinder class.
[DYNAMIC_EXECUTION]: The skill modifies the global Python process environment by inserting directories into sys.path and registering a custom MetaPathFinder in sys.meta_path to handle custom import resolution.
Evidence: Found in setup_skill_path and _ensure_finder_installed functions in scripts/skill_imports.py.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates an attack surface where malicious Python code residing in a skill directory could be executed if the agent is induced to import it.
Ingestion points: Local file system directories containing a SKILL.md file (analyzed in scripts/discover.py).
Boundary markers: None; the skill assumes any directory with the appropriate marker file is a valid source for code imports.
Capability inventory: Dynamic code execution via importlib and file writing via Path.write_text in generate_registry.
Sanitization: The skill performs basic name conversion (dash to underscore) but lacks robust validation to ensure the integrity of the imported code.

inspecting-skills