The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/pdf_ocr_processor.py contains a function install_dependency that utilizes subprocess.check_call to execute pip install commands. This allows the skill to run system-level commands to modify the local Python environment.
[REMOTE_CODE_EXECUTION]: The skill implements a dynamic dependency management system that downloads and installs Python packages (rapidocr_onnxruntime, pymupdf, pillow) from PyPI at runtime if they are not already present. Installing and importing external code during execution is a significant security risk.
[EXTERNAL_DOWNLOADS]: The skill is designed to fetch and install external packages from the Python Package Index (PyPI) during runtime to enable specific OCR functionalities.
[DATA_EXFILTRATION]: The SiliconFlowOCREngine class transmits image data to an external endpoint (https://api.siliconflow.cn/v1/chat/completions) for text recognition. While this is a documented feature, it results in the transmission of document content to a third-party service.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes untrusted documents and returns the extracted content to the agent.
Ingestion points: Document and image files processed in scripts/pdf_ocr_processor.py via ocr_pdf and ocr_image_file.
Boundary markers: The extracted text is returned as a raw string without any delimiters or 'ignore' instructions to prevent the agent from following commands embedded in the documents.
Capability inventory: The skill can execute shell commands (via pip installation) and perform network operations (API requests).
Sanitization: There is no evidence of text sanitization or filtering applied to the OCR output before it is returned to the agent context.

pdf-ocr-skill