ds-evaluation-design

Pass

Audited by Gen Agent Trust Hub on Aug 19, 2026

Risk Level: SAFE
Full Analysis
  • Synthetic Data Generation: The skill explicitly instructs the agent to avoid using real customer data, credentials, or personal identifiers in generated datasets, mandating the use of representative synthetic content to protect privacy.
  • Instructional Integrity: It includes a directive to treat all supplied transcripts and documents as data rather than instructions, which serves as a defensive measure against prompt injection within processed inputs.
  • Human-in-the-loop Protocol: The workflow requires explicit user confirmation of summaries and sample reviews before finalizing any durable artifacts, ensuring human oversight of the generated content.
  • Restricted Write Operations: The instructions require all file writes to pass through an external 'scan gate' (data-workstream-foundation), providing a layer of governance for outputting data to customer locations.
  • Authoritative Source Referencing: The skill references official Microsoft Azure and Copilot Studio documentation for evaluator definitions while maintaining a 'cite-only' posture to ensure data currency and licensing compliance.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 19, 2026, 11:39 PM
Security Audit — agent-trust-hub — ds-evaluation-design