signals-scout-replay-vision

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

• Third-party content exposure detected (medium risk: 0.30). In SKILL.md “Untrusted data — scanner output is LLM text over user content”, this scout reads freeform scanner_output_* text (including summaries/reasoning/tags) from the runtime events table via the $recording_observed event and incorporates it into reports, so outsider-authored session content indirectly flows into the LLM.