knowfirst

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill calls membrane connection ensure with "https://www.knowfirst.ai/" at runtime, and the Membrane response can include clientAction.agentInstructions which would directly control the AI agent's instructions—so this runtime URL dependency can deliver prompts that control the agent.