The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses sys.path.insert in SKILL.md to dynamically include a shared library directory (../../scripts) at runtime. This allows the skill to load core engine components such as extract.py, dataclean.py, and ingest.py which reside outside the skill's immediate folder.
[DYNAMIC_EXECUTION]: The skill features a reuse mechanism via the emit_runner function in dataclean.py. This functionality generates standalone Python scripts and Markdown documentation cards in the user's working directory to facilitate reproducible extraction workflows.
[EXTERNAL_DOWNLOADS]: The image_extract.py script performs outbound network requests to OpenAI-compatible vision API endpoints (defaulting to api.openai.com) to process chart and diagram data. This communication is documented as a necessary step for image-based extraction and uses standard environment variables for authentication.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the ingestion of untrusted document and image content.
Ingestion points: The image_extract.py script and ingest.py engine read external files including PDFs, Word documents, and various image formats.
Boundary markers: The skill employs specific prompt engineering strategies and a corrective retry mechanism in image_extract.py to attempt to constrain the vision model's output to valid Markdown tables.
Capability inventory: The skill has capabilities for network communication (API calls), local file reading (ingestion), and file writing (Excel and report generation).
Sanitization: External content from vision models is parsed for Markdown table structures, providing a layer of validation against unexpected prose or instructions.

data-extract