prior-elicitation
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill uses standard scientific Python libraries (PyMC, ArviZ, PreliZ, SciPy, NumPy) for statistical modeling and prior elicitation workflow. All package references are to established libraries in the Bayesian ecosystem.
- [SAFE]: File operations are limited to saving simulation results (NetCDF) and plots (PNG) locally, which is expected behavior for the described data analysis workflow.
- [SAFE]: External references point to official documentation, source code repositories of well-known libraries on GitHub (arviz-devs, pymc-devs), and academic papers on ArXiv/DOI.
- [INDIRECT_PROMPT_INJECTION]: The skill describes a workflow involving expert elicitation (ingesting external judgments into model parameters). While this represents a data ingestion surface, the context is limited to statistical modeling and lacks high-risk capabilities like network exfiltration or system modification.
- [DYNAMIC_EXECUTION]: The documentation mentions experimental features (PreliZ PPE) that can generate code proposals. The skill explicitly advises users to review these proposals manually rather than executing them automatically, adhering to safe usage practices.
Audit Metadata