diagnosing-experiment-health
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is a legitimate analytical tool designed to help users diagnose and understand their PostHog experiment results. It contains no malicious instructions or bypass attempts.
- [COMMAND_EXECUTION]: The skill provides structured SQL templates for the agent to use with project-specific tools like
execute-sql. These queries are used strictly for data retrieval and analysis of experiment events and properties, such as variant assignments and metric conversions. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external data, including experiment configurations and survey responses retrieved from the PostHog API. It treats this data as information to be analyzed rather than instructions to be executed. The documentation specifically advises the agent to treat survey response text as untrusted data to avoid potential instruction injection through qualitative feedback.
- [DYNAMIC_EXECUTION]: The skill generates SQL queries at runtime by interpolating experiment IDs and feature flag keys into predefined templates. This is a standard and safe implementation for data analysis skills within the PostHog environment.
Audit Metadata