academic-plotting
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided research context, including paper sections, descriptions, and data tables, which are interpolated into LLM prompts and visualization scripts.
- Ingestion points: User-supplied research text, results paragraphs, and CSV/JSON data files used to drive diagram and chart generation (SKILL.md, Step 0).
- Boundary markers: The provided prompt and script templates do not employ explicit boundary markers or "ignore embedded instructions" warnings for the user-supplied data.
- Capability inventory: The skill performs file system writes (generating .py, .pdf, and .png files) and network operations (Gemini API calls via the google-genai library).
- Sanitization: There is no mentioned sanitization or validation of the input data before it is interpolated into prompts or generated scripts.
- [DYNAMIC_EXECUTION]: The core workflow involves the agent generating Python scripts based on templates and then executing them to produce figures. While standard for this application, the practice of generating and running code that incorporates external, potentially untrusted input represents a persistent attack surface.
Audit Metadata