umap-learn
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill enforces a local-first approach to data analysis, ensuring that core computation remains within the authorized environment and only uses external resources when explicitly directed by the user.
- [SAFE]: The integration with SandBase tools (sandbase_run, sandbase_inspect) includes robust safety protocols, requiring the agent to disclose parameters, unit prices, and total cost estimates for user approval before any network operations.
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion and processing of external datasets, which is an inherent vector for indirect prompt injection. However, the instructions mandate thorough data inspection (types, shapes, leakage) and validation against baselines, and the agent lacks high-privilege execution capabilities, keeping the risk level low and the overall posture safe.
Audit Metadata