smart-ocr

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (images and PDFs) which could contain hidden malicious instructions intended for the agent. Ingestion point: The ocr_pdf function in SKILL.md saves and processes temporary image files from PDF pages. Boundary markers: No specific delimiters or "ignore instructions" prompts are implemented in the code snippets. Capability inventory: The skill utilizes requests for network fetches and os.remove for file deletion. Sanitization: The provided examples lack sanitization of the extracted text before it is returned to the agent context.
  • [EXTERNAL_DOWNLOADS]: The skill documentation provides instructions to install standard Python libraries via pip and references official PaddleOCR repositories. This is consistent with the skill's primary function.
  • [DATA_EXFILTRATION]: Includes code snippets demonstrating the use of requests.get to fetch image content from URLs. While a standard feature for OCR, it represents a data ingestion vector.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 03:59 PM
Security Audit — agent-trust-hub — smart-ocr