academic-plotting
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface since it is designed to ingest external data and text fields to generate visualizations.
- Ingestion points: Processes untrusted inputs such as full paper drafts, section drafts, paragraph descriptions, and raw CSV/JSON experiment logs as specified in
SKILL.md. - Boundary markers: Lacks explicit boundary markers or instructions to isolate or treat user text as untrusted data within the generated prompt templates.
- Capability inventory: Involves local python execution capabilities to run plotting scripts and perform network API communication via
google-genai. - Sanitization: No sanitization or escaping mechanisms are specified for incoming research context or data streams.
- [DYNAMIC_EXECUTION]: The workflow requires the agent to generate and execute Python scripts locally (
figures/gen_fig_<name>.py) to produce figures via matplotlib/seaborn or communicate with the Gemini API. This template-driven script generation is expected behavior given the skill's primary visualization purpose.
Audit Metadata