reflect

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

• Third-party content exposure detected (high risk: 0.75). The REQUIRED runtime path reads the active transcript JSONL from agent-transcripts/ and inlines its quoted content into reviewer/synthesizer LLM context; that transcript can include outsider-authored free text (e.g., messages/comments from other participants) and is treated as untrusted data, creating indirect prompt-injection exposure.