paddleocr-doc-parsing
Pass
Audited by Gen Agent Trust Hub on Apr 14, 2026
Risk Level: SAFE
Full Analysis
- [DATA_EXFILTRATION]: The skill is designed to send document data to a user-configured external API endpoint (PaddleOCR). This behavior is the primary intended function of the skill and is clearly documented. It uses an access token provided by the user through environment variables to authenticate these requests.
- [COMMAND_EXECUTION]: The skill provides several Python scripts (
vl_caller.py,optimize_file.py,split_pdf.py) which the agent is instructed to execute to perform document parsing, image optimization, and PDF page splitting. These scripts perform standard file and network operations required for OCR tasks. - [EXTERNAL_DOWNLOADS]: The skill requires standard, well-known Python packages including
httpxfor API communication,python-dotenvfor configuration, andPillowandpypdfium2for processing images and PDFs. No untrusted or suspicious remote code execution patterns were detected. - [PROMPT_INJECTION]: The instructions in
SKILL.mduse strong instructional language to ensure the agent uses the specialized PaddleOCR API instead of its own vision models for complex documents. These instructions are designed to improve output accuracy and do not attempt to bypass core AI safety guidelines.
Audit Metadata