liveperson

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.80). The SKILL.md explicitly tells the agent to fetch and act on Liveperson data via Membrane (e.g., "membrane request CONNECTION_ID /path/to/endpoint" and membrane action list/run), which returns user-generated/untrusted conversation content that the agent is expected to read and which could influence subsequent actions.