The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local Python scripts (check-env.py and lint-template.py) and the docling-graph CLI for operational tasks. In scripts/check-env.py, subprocess.run is used to execute docling-graph --help to verify the command-line environment.
[DYNAMIC_EXECUTION]: The scripts/check-env.py utility uses importlib.import_module to verify that user-defined Pydantic templates can be successfully imported and validated. This is a necessary functional component for checking graph schema integrity.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to extract data from untrusted document sources which may contain adversarial text intended to influence agent behavior.
Ingestion points: PDF, HTML, DOCX, and images are listed as document sources in SKILL.md and references/pipeline-configuration.md.
Boundary markers: Not explicitly defined in the provided workflow instructions, though the skill relies on structured Pydantic schemas for extraction.
Capability inventory: File system access and subprocess execution are present in the helper scripts and the target docling-graph tool.
Sanitization: No explicit sanitization or filtering of ingested document content is performed by the provided skill scripts.

docling-graph