diagnosing-experiment-results
Pass
Audited by Gen Agent Trust Hub on Jul 24, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill provides structured SQL templates for the agent to execute via the
execute-sqltool. These queries are designed to pull diagnostic data such as exposure counts, identity fragmentation patterns, and custom event properties from the PostHog events database to facilitate troubleshooting. - [DATA_EXFILTRATION]: Access to data and project configurations is appropriately restricted to integrated PostHog tools. The skill analyzes telemetry and experiment metadata for its stated diagnostic purpose and does not attempt to exfiltrate data to external or untrusted destinations.
- [PROMPT_INJECTION]: No patterns attempting to bypass safety filters or override agent behavior were detected. The instructions are focused on guiding the agent through specific diagnostic steps based on user-reported symptoms.
- [SAFE]: The skill is authored by the vendor (PostHog) and its behavior is entirely consistent with its purpose of assisting users in the technical interpretation of experiment results. All external links point to official documentation.
Audit Metadata