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

paddleocr-text-recognition

Pass

Audited by Gen Agent Trust Hub on Apr 21, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: Indirect Prompt Injection Surface. \n
  • Ingestion points: scripts/ocr_caller.py accepts untrusted data via the --file-url and --file-path arguments. \n
  • Boundary markers: Absent; the skill does not use delimiters or include instructions to ignore embedded content in the extracted text. \n
  • Capability inventory: scripts/lib.py performs network POST requests and reads local files; scripts/ocr_caller.py writes output JSON to the filesystem. \n
  • Sanitization: Absent; the skill returns extracted text strings directly from the API response without validation or filtering. \n- [EXTERNAL_DOWNLOADS]: Functional network operations. \n
  • scripts/smoke_test.py fetches a sample image from the vendor's official cloud storage (bcebos.com). \n
  • scripts/lib.py communicates with the user-configured API endpoint (PADDLEOCR_OCR_API_URL) to perform text recognition.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 21, 2026, 04:48 AM