exploring-the-wizard

Pass

Audited by Gen Agent Trust Hub on Jul 12, 2026

Risk Level: SAFECOMMAND_EXECUTIONCREDENTIALS_UNSAFEDATA_EXFILTRATION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The skill instructs the agent to ask for a PostHog key file path. This file is required for the open_app and run_agent tools to authenticate with PostHog services. The skill explicitly warns against printing or committing the key to logs or repositories.
  • [COMMAND_EXECUTION]: The skill utilizes the run_agent tool to execute integration logic in the background. This is a core feature designed to simulate real-world wizard behavior during end-to-end testing.
  • [DATA_EXFILTRATION]: The run_agent tool creates real resources, such as dashboards and insights, within the target PostHog project. While this involves sending data to an external service, it is the documented and intended primary behavior of the PostHog wizard being tested.
  • [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it reads state and screen content from a target application. Malicious content within the target app could potentially influence the agent's decisions, although no specific malicious instructions are present in the skill itself.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 12, 2026, 10:15 PM
Security Audit — agent-trust-hub — exploring-the-wizard