Add RESTORE.md — runbook to rebuild the whole team from this repo

[yihanzhu]

Goal / problem

Goal #2 in CLAUDE.md is full backup — everything needed to reconstruct the team
if the live setup is lost. Today that promise is implicit: the prompts, roles, and CI
all live here, but there is no single document that says how to turn this repo back
into a running team. If the live Claude Routines, the Codex connection, the labels, or
CI disappeared, someone would have to reverse-engineer the wiring from scattered files.

Add a top-level RESTORE.md — a disaster-recovery runbook that walks a human through
rebuilding the whole team from this repo, in order, from nothing.

Acceptance criteria

RESTORE.md at the repo root, ordered so a human can follow it top to bottom:

1. Prerequisites — accounts/access needed: a Claude plan with Routines, Codex
   (OpenAI) for PR review, GitHub access + gh CLI authenticated. State the
   personal config that must be supplied (target repo name(s)) — keep it parameterized,
   no hardcoded personal values per CLAUDE.md reusability rule.
2. Recreate Faber (manager) — paste manager/CLAUDE.md into a Claude Code project;
   note this is the only human-facing surface.
3. Recreate the routines — one step per routines/*.md, each with its trigger:
   • coder.md → issues.labeled (ready)
   • coder-revision.md → pull_request_review.submitted
   • brief.md → daily cron
     Each step says "paste this file's contents + set this trigger."
4. Recreate the Codex reviewer — connect Codex to each target repo, comments-only /
   read-only, using reviewer/codex-review.md as the prompt.
5. Recreate labels + branch protection + CI per target repo — reuse
   templates/repo-setup.md (labels, branch protection, the ready/round-*/needs-human
   set) and note CI comes from .github/workflows/ci.yml. Link to it; do not duplicate
   its steps.
6. Smoke test / verification — run one trivial issue through the full loop
   (issue → label ready → PR → Codex review → merge) to confirm the rebuilt team is alive.
7. Restate the safety rails that must survive any rebuild: reviewer stays
   read-only/comments-only, no auto-merge in Phase 1, rounds cap + needs-human
   escalation intact (per CLAUDE.md self-modification safety).
8. README "Layout" section updated to list RESTORE.md.

Likely files

• RESTORE.md (new)
• README.md (add to Layout)
• references only (do not duplicate): manager/CLAUDE.md, routines/*.md,
  reviewer/codex-review.md, templates/repo-setup.md, .github/workflows/ci.yml

Test expectations

• CI stays green: structure check + shellcheck. Any shell snippets in RESTORE.md
  must be shellcheck-clean (or fenced as non-executed examples consistent with how
  repo-setup.md does it).
• Every file/path the runbook references must actually exist in the repo (no dangling
  links).

Out of scope

• Automating the restore (scripts/tooling) — this issue is the human runbook only;
  a future issue can script it.
• Changing any routine/prompt behavior — this is documentation, no prompt edits.
• Per-target-repo setup details already covered by templates/repo-setup.md — link, don't copy.