deep-reviewlisted

# Deep Review (post-build review loop) Reviews a session's own output before it is considered done: findings verified against code, validated-expired legacy deleted, and all state written into a tracker another agent can resume from. Distinct from `functionality-audit` (does the PRODUCT work end to end) and `repo-audit-with-execution-batches` (surface scan of a repo at rest). ## When to run Before closing any session that produced multiple commits — features, fixes, refactors. Also on request ("review tonight's work", "check what we just built"). The unit of review is a commit RANGE, not a repo. ## Non-negotiables (owner-set) 1. **One review agent, sequential — never a fan-out.** A single general-purpose agent walks the whole range; the orchestrating agent runs the simplify pass itself. Two workers total, maximum. 2. **Findings must be verified against actual code before reporting.** The reviewer reads the files, not just the diff. Real bugs only: correctness, security, behavioral regressions, broken wiring. No style nits, no hypotheticals. 3. **Every finding is either fixed in-session or written into the backlog tracker.** Nothing lives only in the conversation. 4. **Cleanup is part of review, not a separate task** — see the remove-don't-archive protocol below. 5. **Track as you finish.** Every unit of work (review agent's findings applied, cleanup batch, spec built) appends one status line to the repo's live session tracker IN THE SAME COMMIT