nature-figure
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is well-structured, providing clear boundaries for execution and focusing on the scientific integrity of generated figures.
- [EXTERNAL_DOWNLOADS]: The skill includes links to official journal guidelines from Nature, ScienceDirect, and Rockefeller University Press, along with a link to a public academic repository on GitHub for reference examples. These references are used solely for technical and stylistic guidance.
- [INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface as it processes user datasets to create plot labels. It manages this risk by requiring a 'figure contract' that defines the scientific claim and evidence logic before generating any visualization, reducing the likelihood of processing adversarial instructions embedded in data.
- Ingestion points: Processes user datasets and metadata in CSV, TSV, or DataFrame formats.
- Boundary markers: Implements a mandatory 'figure contract' to define claim and logic (contract.md).
- Capability inventory: Generates plotting scripts utilizing standard scientific libraries (matplotlib, ggplot2).
- Sanitization: Relies on the agent's internal reasoning and the established figure contract to ensure data is treated as content rather than instructions.
Audit Metadata