Skills
skills/paddlepaddle/paddleocr/paddleocr-text-recognition/Gen Agent Trust Hub

paddleocr-text-recognition

Pass

Audited by Gen Agent Trust Hub on Mar 14, 2026

Risk Level: SAFEDATA_EXFILTRATIONEXTERNAL_DOWNLOADSPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill transmits file content or URLs to a remote API endpoint for processing.\n
  • Evidence: scripts/lib.py uses httpx.post to send base64-encoded image data.\n
  • Configuration: The endpoint is set via the PADDLEOCR_OCR_API_URL environment variable.\n- [EXTERNAL_DOWNLOADS]: The smoke test script references an external sample image.\n
  • Evidence: scripts/smoke_test.py uses a URL from the PaddlePaddle GitHub repository to verify connectivity.\n- [PROMPT_INJECTION]: Vulnerable to indirect prompt injection from content within scanned files.\n
  • Ingestion points: scripts/lib.py (via ocr function reading local files or URLs).\n
  • Boundary markers: None identified in instructions to the agent.\n
  • Capability inventory: Network access (httpx) and file system access (Path.read_bytes).\n
  • Sanitization: No filtering or sanitization of recognized text is performed.\n- [COMMAND_EXECUTION]: The skill is implemented as a set of Python scripts intended for local execution.\n
  • Evidence: SKILL.md specifies the use of scripts/ocr_caller.py, scripts/configure.py, and scripts/smoke_test.py.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 14, 2026, 01:36 AM