The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/ingest_literature_library.py uses subprocess.run() with a list of arguments to orchestrate the internal workflow by executing other Python scripts within the skill directory, which is a safe practice that prevents shell injection.
[EXTERNAL_DOWNLOADS]: The skill fetches metadata and PDF documents from established academic sources including arxiv.org, api.semanticscholar.org, api.openalex.org, and huggingface.co.
[EXTERNAL_DOWNLOADS]: The scripts/paddleocr_layout_to_markdown.py script sends PDF data to an external API at aistudio-app.com for layout analysis and OCR processing. This is a legitimate data flow associated with the skill's primary purpose.
[PROMPT_INJECTION]: The skill ingests untrusted external content from scientific papers. This introduces a surface for indirect prompt injection, where malicious instructions embedded in a paper's text could potentially influence the agent when it reads the converted Markdown output.

literature-pdf-ocr-library