The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses uvx to download and execute the disambiguate package from the Python Package Index (PyPI) to lint glossary files.
[COMMAND_EXECUTION]: Executes shell commands such as uvx disambiguate --lint to validate the structure and cross-references of the generated documentation.
[PROMPT_INJECTION]: The skill processes untrusted user input to generate domain model documentation, creating a surface for indirect prompt injection.
Ingestion points: User-provided domain terms, definitions, and scenarios recorded in docs/glossary/ and docs/adr/.
Boundary markers: The skill specifies a rigid Markdown format (H2 for canonical names, specific link styles) but lacks explicit delimiters to separate untrusted content from agent instructions.
Capability inventory: Includes file system write access to create documentation and architectural decision records, and command execution for linting.
Sanitization: The skill uses a linting tool (uvx disambiguate --lint) to verify the file format, although it does not filter the semantic content of the definitions provided by the user.

domain-modeling