pair-analytics
Pass
Audited by Gen Agent Trust Hub on Sep 24, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill defines a 'Hard Rule' that strictly prohibits the agent from connecting to any external warehouse or using credentials, even if available on the host machine. This architecture ensures the agent remains an offline query writer.
- [SAFE]: The generated SQL queries and notebooks are written to the local filesystem (
datasets/analytics/) for the user to review and execute, maintaining a human-in-the-loop safety model. - [INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it processes data from dataset indexes and query results provided by the user.
- Ingestion points: Table profiles, README files in the
datasets/directory, user-pasted text, and CSV files. - Boundary markers: None defined; the agent processes these inputs as direct context for further analysis.
- Capability inventory: The agent's actions are restricted to writing SQL and Python (notebook) files. It does not have access to network tools or arbitrary command execution.
- Sanitization: No explicit sanitization of ingested warehouse data is mentioned.
Audit Metadata