Multi-agent war room review before key decisions

★ 7.5 · vibe-coding

ato-warroom is a Claude Code skill that convenes a war-room session before any material decision — sending a final code draft, opening a PR, pushing to a remote-tracking branch, committing more than 50 lines of behavior change, or delivering a strategic recommendation. It dispatches multiple agents in parallel via `ato dispatch`, where each seat fires standalone with no visibility into other seats' replies, so first-pass priors genuinely diverge. A failure-mode filter runs on every dispatch, covering wrong assumptions, overcomplexity, orthogonal edits, and imperative-over-declarative reasoning — Karpathy's four by default, swappable for your own. Every war-room round is recorded with a shared `war_room_id` UUID as an audit trail. Within the v2.16 stack, the division of responsibility is explicit: war-room decides, `ato-mission` executes, `ato-review` verifies the merged result.