daily-news-digest

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

• Third-party content exposure detected (high risk: 0.85). Outsider free text can enter the LLM context because STEP 4/5 requires reading full note bodies from references/ (and other vault folders) for clustering/synthesis, and those notes may contain outsider-authored content; the collector only provides snippets but the subagents “open every note in the cluster” and feed the resulting text into the model.