ae-experiment-design
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided business objectives, success criteria, and technical schemas, which constitutes a potential surface for indirect prompt injection. Ingestion points are primarily located in the Workflow section of SKILL.md and the Creation contract gate in references/experiment-creation.md. The risk is mitigated by explicit instructions to verify all data against platform schemas retrieved via the ae-cli tool and by enforcing strict validation gates defined in metric-readiness.md. The capability inventory includes the ae-cli for platform interaction and the local calculation script scripts/calculate_experiment_plan.py.
- [SAFE]: The skill includes a local Python script for experiment planning that performs purely mathematical calculations using standard libraries. It does not perform network operations, access sensitive files, or use dynamic code execution functions like eval or exec.
- [SAFE]: All external URLs and SDK dependency references point to official ThinkingData resources and established package registries (Maven, CocoaPods). These are confirmed as legitimate vendor-owned resources for providing integration guidance as detailed in sdk-index.md.
Audit Metadata