experiments
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill follows a policy of least privilege by restricting its Bash tool access to a specific local helper script (
gb-call) and thesleepcommand. It also encourages secure secret management by suggesting the use of environment variables or a local configuration file for API keys, rather than hardcoding them. - [INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it ingests and processes free-text data from the GrowthBook API, such as experiment hypotheses, descriptions, and learning conclusions.
- Ingestion points: Metadata and text content retrieved from the GrowthBook API across several workflows, including
experiment-analyze.md,experiment-brainstorm.md, andlearnings.md. - Boundary markers: No specific delimiters or "ignore instructions" tags are used when presenting external data to the agent for interpretation.
- Capability inventory: The skill has the ability to stop experiments, modify feature flag targeting rules, and delete Learning records via the
gb-callscript. - Sanitization: The skill mitigates potential injection risks by explicitly instructing the agent to obtain human confirmation before executing any state-changing operations (POST, PUT, DELETE).
Audit Metadata