langgraph-adk

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

• Third-party content exposure detected (high risk: 0.90). At runtime, the assistant action’s multi-turn path passes user-authored free text from the Ink <SessionApp> (history.current built from TextInput submissions) into ctx.llm.invoke([systemMessage(), ...messages]), so the LLM context includes outsider-authored input (the end-user’s messages are not authored by the operating user of the skill).