eval-generator
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFE
Full Analysis
- Functional Data Processing: The skill is designed to process user-provided evaluation planning workbooks (.xlsx) and narrative descriptions to generate structured test cases. This is a standard functional behavior for evaluation tooling and does not involve suspicious data handling patterns.
- File Generation Capability: The skill generates CSV files for import into Copilot Studio and .docx reports for manifest documentation. These operations are within the stated purpose of the skill and utilize standard formatting (2-column CSV, UTF-8 encoding) without including executable content or hidden scripts.
- Instructional Integrity: The provided instructions focus on maintaining rigorous testing methodologies, such as separating capability sets from trust and safety sets, and ensuring human review of AI-generated content through specific markers (
[VERIFY]). These are best practices for AI safety and evaluation quality. - No Dangerous Capabilities: The skill does not request or utilize tools for network exfiltration, arbitrary command execution, or sensitive file system access. It operates on data provided within the session context or via explicit user attachments.
Audit Metadata