braintrust-size-eval-dataset

Pass

Audited by Gen Agent Trust Hub on Aug 20, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: No malicious injection patterns were found. The skill includes defensive instructions in references/interaction-contract.md to prevent the agent from following directives embedded in trace data (e.g., 'ignore previous instructions'). It specifically addresses the risk of indirect prompt injection by identifying ingestion points (traces, logs), setting clear boundaries for data processing, and restricting capabilities to reporting and analysis.
  • [DATA_EXFILTRATION]: There are no hardcoded credentials or sensitive file path accesses. References to Braintrust platform mechanics and API endpoints (e.g., DELETE /v1/experiment/{id}) are documented as part of the intended platform functionality for the skill's author.
  • [REMOTE_CODE_EXECUTION]: No remote script downloads or untrusted package installations were detected. Citations in the documentation refer to academic and technical sources for statistical methods.
  • [OBFUSCATION]: No base64 encoding, homoglyphs, or hidden text patterns were found.
  • [SAFE]: No security issues were detected. The skill follows best practices for data handling and statistical modeling.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 20, 2026, 10:17 PM