braintrust-monitor-production-evals
Pass
Audited by Gen Agent Trust Hub on Aug 20, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill documentation includes an 'Interaction Contract' that specifically addresses prompt injection risks. It instructs the agent to treat traces and logs as data rather than instructions, explicitly warning against 'Directive-following' attacks where data might contain commands like 'ignore previous instructions'.
- [SAFE]: The skill guidelines prioritize data privacy and security by requiring PII scrubbing, documented retention windows, and access controls to be established at the point of data collection.
- [SAFE]: References to tools such as
braintrust-deploy-evaluatorand specific API endpoints represent standard usage of the vendor's platform and do not involve unauthorized command execution or remote code vulnerabilities. - [SAFE]: The skill provides structured templates for sampling plans and alert specifications, emphasizing human-in-the-loop review and documented provenance for all thresholds and metrics.
Audit Metadata