strands-evals

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

• Third-party content exposure detected (high risk: 0.70). Chaos testing uses ChaosPlugin to inject chaos into tool-result text blocks (JSON/text) via Strands hook events (after_tool_call → _apply_to_blocks), and those corrupted tool outputs are then included in the evaluation trajectory/LLM-judge context; the injected content originates from the outsider-controlled tool response payloads (i.e., tool outputs are not authored by the operating user).