glmv-doc-based-writing
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes document files (PDF/DOCX) that could contain malicious instructions designed to hijack the model's output.
- Ingestion points: External documents provided via the --files argument in scripts/doc_based_writing.py are ingested and sent to the LLM.
- Boundary markers: The skill lacks explicit boundary markers or instructions telling the model to ignore commands embedded within the input documents.
- Capability inventory: The skill has the capability to perform network requests (requests.post) and write output to the file system (open().write()).
- Sanitization: No sanitization or filtering is performed on the content extracted from the documents before processing.
- [EXTERNAL_DOWNLOADS]: The skill communicates with the Zhipu AI platform API (open.bigmodel.cn) to perform content generation. It also references the PyMuPDF package for local file processing.
Audit Metadata