The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/pdf_to_images.py utilizes subprocess.run to execute the pdftoppm command-line utility. This tool is used to convert PDF pages into images for visual processing. The command is constructed using a list of arguments, which prevents shell injection.
[EXTERNAL_DOWNLOADS]: The skill can download PDF files from external URLs provided by the user via the scripts/pdf_to_images.py script using the standard urllib.request.urlopen method. This behavior is central to the skill's purpose of analyzing remote documents.
[DYNAMIC_EXECUTION]: Several orchestration scripts, including scripts/run_document_parsing.py, use the importlib module to dynamically load and execute other Python scripts located within the skill's own scripts/ directory. This is a modular design pattern used to coordinate the various analysis workflows.
[INDIRECT_PROMPT_INJECTION]: As a tool designed to process external and potentially untrusted visual data (images and PDFs), the skill has an inherent surface for indirect prompt injection.
Ingestion points: External PDF URLs in scripts/pdf_to_images.py and image URLs in scripts/qianfan_ocr_cli.py.
Boundary markers: The skill uses predefined prompt templates in the references/ directory that include strict instructions for the AI to avoid inference and follow specific Markdown formatting.
Capability inventory: The skill has the ability to execute shell commands (pdftoppm), perform network requests to the Baidu API and user-provided URLs, and write files to the local filesystem (results, assets, and .env configuration).
Sanitization: The skill implements filename sanitization using regular expressions in scripts/pdf_to_images.py to ensure local file paths are safe.

qianfanocr-document-intelligence