unlimited-ocr-document-parsing

Pass

Audited by Gen Agent Trust Hub on Aug 15, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted document data from local files or remote URLs (scripts/client.py). It mitigates potential injection attacks by providing explicit instructions in SKILL.md to treat all extracted text as untrusted data and to ignore instructions within. The skill's capabilities include reading local files, making network requests to OCR providers, and writing results to the filesystem. Error messages are sanitized before being returned.
  • [DATA_EXPOSURE_AND_EXFILTRATION]: API keys and secrets are handled securely through environment variables. The skill implements URL validation to prevent the inclusion of embedded credentials in endpoints.
  • [EXTERNAL_DOWNLOADS]: Fetches document processing results from Baidu's cloud infrastructure. It includes safety limits on download sizes (64 MiB) and enforces HTTPS for remote connections.
  • [UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: Uses standard, well-known Python packages managed via inline script metadata. No suspicious remote code execution patterns were identified.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 15, 2026, 03:04 AM
Security Audit — agent-trust-hub — unlimited-ocr-document-parsing