handoverlisted

# handover: the repo must speak for itself The test is always the same: **a competent stranger, any IDE, any model, gets only the repo. Can they continue within ten minutes?** If anything requires this conversation, your memory, or files outside the repo, the handover is broken. ## The audit Walk it as the stranger would: 1. `README.md` → do the copy/start instructions still hold for this project? 2. `AGENTS.md` routing table → does every row point at a file that exists and is current? 3. `docs/state/STATE.md` handoff block → status, phase, branch, gates, Now ▶, blockers. Is it literally true *right now*? Stale handoff is the most damaging lie in the system. 4. `docs/product/BRIEF.md` → does it still describe the product that exists? 5. Can the stranger run it? Setup steps, env vars (`.env.example` complete?), seed data, `docs/operations/` runbooks. Follow them literally; fix what's wrong, don't annotate around it. A fresh clone must also re-run `node checks/check.mjs --install-hooks`. Is that step where the stranger will see it (README/setup)? 6. Secrets and access: nothing secret in the repo, and a named list of what access a successor must be granted (where credentials live, who grants them) in `docs/operations/access.md`. 7. Open threads: in-flight specs at their real status; INTAKE triaged or explicitly queued; DEBT.md reconciled with `defer:` markers. Fix everything fixable now. What you can't fix, write down as an explicit gap in STATE.md. A name