hf-papers-reporter
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The shell scripts
video-project/scripts/generate_audio.shandvideo-project/scripts/render.shexecute system commands (python3, npx, ffmpeg) using hardcoded absolute paths that include the author's local username (/Users/xdrshjr/...). This practice exposes the internal environment structure and reduces the portability of the skill. - [EXTERNAL_DOWNLOADS]: The skill fetches PDF documents and images from
arxiv.organdhuggingface.co. These resources are recognized as well-known academic and technology services. - [INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it processes external scientific papers.
- Ingestion points: PDF text extraction (Abstract and Introduction) occurs in
scripts/process_papers.pyandscripts/process_papers_filtered.pyusing PyMuPDF (fitz). - Boundary markers: The extracted content is interpolated directly into document headings and paragraphs without boundary delimiters or safety instructions to the agent.
- Capability inventory: The skill has file system write access (creating and saving .docx files) and network read access.
- Sanitization: A
clean_xmlfunction is implemented to remove control characters for XML/Word compatibility, but it does not filter for potential malicious instructions embedded in the paper text.
Audit Metadata