diagnosing-experiment-results

Pass

Audited by Gen Agent Trust Hub on Sep 4, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user complaints and experiment metadata (such as flag keys, variant names, and properties) which are then interpolated into SQL queries and used to determine diagnostic outcomes. The agent has the capability to execute modifications via the experiment-ship-variant tool.
  • Ingestion points: Retrieves experiment configuration via experiment-get and event data via execute-sql and experiment-results-get.
  • Boundary markers: The instructions lack explicit directives for the agent to treat retrieved metadata as untrusted or to use secure boundary markers when processing this data.
  • Capability inventory: The skill utilizes execute-sql for database queries and experiment-ship-variant for modifying feature flag distributions.
  • Sanitization: There are no specified sanitization routines for the data pulled from the PostHog API before it is used in further tool calls or analysis.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 4, 2026, 09:39 AM
Security Audit — agent-trust-hub — diagnosing-experiment-results