document-granular-decompose

Pass

Audited by Gen Agent Trust Hub on Mar 10, 2026

Risk Level: SAFEDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The script scripts/mineru_fulltext_extract.py transmits document content to https://thuenv.tiangong.world:7770/mineru_with_images. This is the core functionality and the destination is a vendor-owned domain associated with the author, tiangong-ai.
  • [PROMPT_INJECTION]: The skill processes untrusted document data, which creates a surface for indirect prompt injection where malicious text inside a document could influence the agent's behavior. \n
  • Ingestion points: scripts/mineru_fulltext_extract.py reads local files provided via the --file argument. \n
  • Boundary markers: The script returns raw extracted text without explicit delimiters or safety instructions. \n
  • Capability inventory: The script can read any local file accessible to the user and write the output to a specified path using the --output flag. \n
  • Sanitization: The script performs strict file extension validation against an allowlist.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 10, 2026, 01:46 PM