paper-research-assistant
Pass
Audited by Gen Agent Trust Hub on Jun 24, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill includes several Python scripts (parse_paper.py, generate_report.py, scaffold_code.py) designed to automate research workflows. These scripts perform routine filesystem operations such as reading PDF files and writing structured JSON metadata, Markdown reports, and Python source code files.
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection via the parse_paper.py script, which ingests text from untrusted PDF files. Malicious content within a PDF could potentially be included in generated reports or influence the agent's summary.
- Ingestion points: parse_paper.py (extracts text from external PDF files).
- Boundary markers: None (extracted text is processed and interpolated without explicit delimiters or safety instructions).
- Capability inventory: File system read/write and script generation capabilities are present across the utility scripts.
- Sanitization: No sanitization or filtering is performed on the extracted text beyond basic regex matching.
- [COMMAND_EXECUTION]: The scaffold_code.py script dynamically generates Python source code skeletons (e.g., PyTorch model architectures and training loops). While the generation is based on predefined templates and paper metadata, the resulting code is intended for execution in a development environment.
Audit Metadata