When a story breaks against your company, the war room watches Twitter and the news sites — text media, minute by minute. Video runs on a lag and a different physics: the first explainer drops within hours, the big commentary channels weigh in by evening, and each upload carries its own comment section — thousands of unprompted reactions telling you which version of the story is actually landing.

Video is where the crisis gets its durable narrative. The tweet cycle burns out in a day; the "what happened with <your company>" explainer gets recommended for months. Watching that narrative form — who's framing it, with what claims, to what audience reaction — is the difference between responding to the story and responding to last night's version of it.

deep_search is the sweep: your company or crisis keywords, upload window pinned to today or this_week, sorted newest first. Each pass shows what's new since the last one — which channels picked it up, what the transcripts actually claim (evidence snippets mark where you're mentioned), and whether the coverage is reporting, opinion or pile-on, per the content-type labels.

run_youtube_tool reads the reaction layer: comment-sentiment and comment-summarizer on the biggest videos show how audiences are receiving each framing — angry, split, bored, moving on. That's the input your response actually needs: not just what's being said about you, but which parts the public believes. Re-run the pair every few hours and the war room has a video desk that never sleeps.