managing-experiment-lifecycle
Pass
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from experiment metadata which can influence automated code changes through the GitHub cleanup feature.
- Ingestion points: Experiment names, descriptions, and variant keys retrieved through experiment resolution and migration tools.
- Boundary markers: Not explicitly defined in the instructions for metadata processing; however, user confirmation is required before shipping variants or opening PRs.
- Capability inventory: Modifies production feature flags via
experiment-ship-variant, deletes snapshot cohorts, and triggers the creation of draft Pull Requests in GitHub via theopen_cleanup_proption. - Sanitization: The instructions do not mention specific sanitization, validation, or escaping of metadata before it is used in tool calls or pull request generation.
- [EXTERNAL_DOWNLOADS]: The skill facilitates interaction with a team's connected GitHub repository to automate feature flag cleanup. GitHub is recognized as a well-known and trusted service, and this functionality is an intended feature of the author's platform.
Audit Metadata