The Agent Skills Directory

[SAFE]: The skill explicitly mandates that all untrusted input must be parsed and validated at the boundary before being converted into trusted, typed data structures like frozen dataclasses. This is a primary security defense against many classes of injection attacks.
[SAFE]: It enforces a 'standard-library-first' policy, requiring that any new third-party dependencies be pure Python and explicitly approved for the runtime requirements list, which minimizes the supply chain attack surface.
[COMMAND_EXECUTION]: The instructions direct the agent to execute local project commands, including just check, just bundle, and just typecheck, which are standard tasks for a repository-local development workflow.
[INDIRECT_PROMPT_INJECTION]: The skill involves reading local repository files (e.g., source code, configurations, and calls) to inform its actions, which creates a surface where the agent could potentially encounter and process instructions embedded in the codebase.
[DYNAMIC_EXECUTION]: The skill suggests using importlib.import_module for build-only modules to help the type checker resolve paths for components that are not statically imported, which is a managed use of dynamic loading for development purposes.
[EXTERNAL_DOWNLOADS]: The skill references uvx basedpyright as a fallback method for type-checking, which involves fetching the basedpyright package from the standard Python Package Index (PyPI).

python-authoring