braintrust-design-eval-instrumentation

Pass

Audited by Gen Agent Trust Hub on Aug 20, 2026

Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [PROMPT_INJECTION]: The skill includes defensive instructions in references/interaction-contract.md (Section 9) that explicitly warn the agent to treat trace data as evidence rather than instructions. This is designed to prevent the agent from following malicious commands like 'ignore previous instructions' when processing external logs.
  • [DATA_EXFILTRATION]: The skill promotes secure data handling by instructing users to scrub sensitive values at the point of emission and explicitly limiting data collection to fields with a defined retention purpose. It includes metadata templates specifically for tracking PII and secret handling.
  • [SAFE]: No malicious code, obfuscation, or suspicious dependencies were identified. The skill is purely instructional and follows established security best practices for observability and data lifecycle management.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 20, 2026, 10:16 PM