turn-eta

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

• Third-party content exposure detected (medium risk: 0.30). In hooks/eta-gate.js, the gate reads the runtime transcript JSONL from input.transcript_path (deriving assistant/user free-text and tool-result text) via planShown(), so outsider-authored message content can affect whether the agent is blocked/retried.