The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill implements an auto-import pattern that dynamically loads and executes Python modules from the filesystem.
The import_modules helper function in references/auto_import.md uses pkgutil.iter_modules to scan directories and importlib.import_module to load them.
This mechanism is triggered during package initialization in __init__.py files (e.g., src/data_module/dataset/__init__.py), meaning code execution happens automatically when the module is accessed.
[INDIRECT_PROMPT_INJECTION]: The auto-import architecture creates a vulnerability surface where untrusted data (new files) can lead to code execution.
Ingestion points: The local filesystem directories scanned by the factory pattern (e.g., src/data_module/dataset/, src/model_module/model/).
Boundary markers: The logic only filters out files starting with an underscore (_). There are no other delimiters or integrity checks for the content being imported.
Capability inventory: The skill possesses full shell execution capabilities via pipeline_example.sh and dynamic Python execution via importlib in auto_import.md.
Sanitization: No sanitization, signature verification, or content validation is performed on the modules before they are imported and executed.

architecture-design