overnight-review-client-deliverylisted

# Overnight Review Panel + Client Delivery ## Problem Client deliverables (slide decks, reports, methodology notes, Google Doc updates) are usually written by an author who is too close to the work to spot factual errors, terminology slips, and miscommunication risks. Synchronous review is slow and blocks the author. Most overnight automation runs a single agent on a single task, missing the "many eyes catch what one misses" advantage. And when authors do polish deliverables overnight, they tend to add archive banners + partial header refreshes to stale content rather than regenerating it cleanly, which leaves readers confused about what's current vs historical. This skill structures an overnight autonomous session as **three phases running mostly in parallel**: 1. **Phase A (content work, foreground)** — Author the actual deliverables 2. **Phase B (review panel, parallel background)** — Launch 8 specialised review agents simultaneously to audit code, methodology, numbers, ops, tests, architecture, and compliance 3. **Phase C (morning hand-off, user-gated)** — Synthesize findings, fix P0 errors before the user wakes up, prepare a clean morning summary with actionable next steps The key insight: Phase A and Phase B run **concurrently**, not sequentially. The review agents are launched in a single message with multiple `Agent` tool calls (parallel batch), they run in the background while Phase A content work continues, and their findings land in `docs/reviews/` over the cou