nsfc-roadmap
Pass
Audited by Gen Agent Trust Hub on Apr 25, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill transmits extracted research content from user-provided proposal files (.tex, .md, .txt) to the Gemini API (generativelanguage.googleapis.com) for image generation in 'Nano Banana' mode. This is a primary feature of the skill and uses a well-known service domain.
- [DATA_EXFILTRATION]: The skill includes a feature to report bugs to the author's repository (huangwb8/bensz-bugs) using the GitHub CLI tool. This action is documented and only triggered by explicit user request.
- [COMMAND_EXECUTION]: The script 'scripts/render_roadmap.py' executes the draw.io desktop CLI to export diagrams to PNG, SVG, and PDF formats. Commands are constructed using lists to prevent shell injection vulnerabilities.
- [PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it ingests content from external scientific proposal files and incorporates them into prompts sent to the Gemini API. The skill attempts to mitigate this by applying 'guardrail' constraints regarding typography and layout in the final prompt generation logic.
Audit Metadata