The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides Python utility scripts (assets/scripts/) intended to be executed by the agent to automate skill development tasks. These scripts perform local file system operations (read/write) restricted to a user-provided directory.
[DYNAMIC_EXECUTION]: assets/scripts/validate_skill.py utilizes yaml.load() for frontmatter parsing. The implementation uses a custom UniqueSafeLoader derived from yaml.SafeLoader, which follows security best practices by preventing arbitrary code execution or object instantiation during YAML parsing.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process other skill files (SKILL.md and references). Ingesting untrusted markdown content presents an inherent surface for indirect prompt injection.
Ingestion points: Targeted skill files in the current working directory or specified path are read using Path.read_text() in validate_skill.py and inspect_skill.py.
Boundary markers: Regex patterns are used to identify Markdown headers and YAML frontmatter blocks.
Capability inventory: The skill allows for local file reading, regex-based analysis, and limited file writing via the fix_skill.py and scaffold_skill.py utilities.
Sanitization: Employs yaml.SafeLoader for configuration data and provides a yaml_double_quoted helper in scaffold_skill.py to escape values interpolated into new skill templates.
[EXTERNAL_DOWNLOADS]: The reference files (references/comparative-study.md) neutrally cite and link to official public repositories and specifications from organizations including Anthropic, OpenAI, Vercel, and Microsoft for comparative study and benchmarking purposes.

skill-architect