deep-review

name	deep-review
description	Deep review — random exploration + cross-agent review, alternating. Finds integration bugs that self-review misses. Run every 30–60 min during active implementation.

Use the verbatim prompts at .flywheel/prompts/deep-review.md, alternating Round A (random exploration) and Round B (cross-agent review).

Run this on 1–2 agents who just finished a bead, not the whole swarm. Continue until two consecutive rounds from different agents come back clean.

If agents keep finding bugs after 4+ rounds, the real issue is a pattern — go back to bead space and create fix-beads for the bug class.