ra-local-talking-head-cut

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

• Third-party content exposure detected (high risk: 0.70). Outsider free text can flow into the LLM context via the runtime-created user-facing review artifacts (review.md, corrected-script.md, and uncertain-terms.md) derived from the source video’s ASR/transcript; these scripts embed untrusted ASR-derived text directly into Markdown files that are then presented to/handled by the agent, enabling indirect prompt-injection.