manage-experiment
Pass
Audited by Gen Agent Trust Hub on Aug 11, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill follows established safety patterns for AI agents, such as requiring explicit user confirmation ('CONFIRM') for destructive or irreversible actions like launching or concluding experiments.
- [SAFE]: The skill implements a 'read-merge-write' pattern for settings updates, which prevents accidental data loss or the silent removal of security configuration and integrity checks (e.g., ensuring Sample Ratio Mismatch (SRM) and QA exclusion settings are preserved).
- [SAFE]: All external interactions are restricted to the Mixpanel platform via dedicated MCP tools, and the skill includes logic to correctly resolve projects and experiments within the user's authorized environment.
- [SAFE]: The instructions include deterministic pre-launch 'pitfall' checks that prioritize blockers (e.g., insufficient statistical power) and warnings, ensuring that users are alerted to configuration risks before committing to experimental changes.
Audit Metadata