The Agent Skills Directory

[SAFE]: The skill serves as a comprehensive framework for managing AI agent skills, emphasizing portable standards and rigorous auditing. The instructions and scripts demonstrate a strong focus on security and isolation.
[COMMAND_EXECUTION]: The skill includes several helper scripts (validate_skill.py, inspect_skill.py, fix_skill.py, scaffold_skill.py, run_uv.py) designed to automate skill management tasks. These scripts perform file system operations and execute the uv package manager.
Evidence: SKILL.md provides detailed instructions for running these scripts in isolated environments using python or uv.
Analysis: These are functional components of the skill. run_uv.py implements a defensive measure by scrubbing the environment of UV_* variables before execution to ensure host configurations do not interfere with the intended isolated run.
[DYNAMIC_EXECUTION]: An AST scan flagged a potentially unsafe YAML loading operation.
Evidence: assets/scripts/validate_skill.py lines 48-52.
Analysis: This is a false positive. The implementation uses a custom UniqueSafeLoader which inherits from yaml.SafeLoader, ensuring that YAML deserialization is performed safely and preventing duplicate key exploits.
[INDIRECT_PROMPT_INJECTION]: As an auditing tool, the skill processes untrusted markdown and YAML data from external skill folders.
Ingestion points: Reads SKILL.md, references/*.md, and other files in target skill directories during auditing and validation modes.
Boundary markers: No specific boundary markers are used for the content of processed files.
Capability inventory: fix_skill.py (file write), run_uv.py (subprocess execution).
Sanitization: scaffold_skill.py includes yaml_double_quoted to escape quotes; validate_skill.py uses SafeLoader. The risk is considered low given these defensive measures and the tool's intended purpose.

skill-architect