shap
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill allows for the discovery and execution of external data tasks via SandBase tools (e.g.,
sandbase_discover,sandbase_run). Security is managed through mandatory disclosure of API endpoints and pricing, coupled with an explicit requirement for user confirmation before any paid operation. - [INDIRECT_PROMPT_INJECTION]: The skill processes local data and user-provided inputs for SHAP analysis (Ingestion). Risk is mitigated by requiring the agent to restate outcomes and constraints (Boundaries) and by implementing validation gates for data shapes, types, and units (Sanitization). The primary capabilities involve local modeling and controlled external data calls.
- [DATA_EXFILTRATION]: The instructions mandate sending only the 'minimum necessary data' to external services, serving as a safeguard against the over-exposure of local datasets during remote inference steps.
Audit Metadata