chore(ops): version wip-sync.sh — the multi-remote v2 (leave/arrive/status)

[aarontrowbridge]

Important

Problem

wip-sync.sh — the code-repo layer of the fleet model (leave/arrive/status; only commits cross machines) — lives only on the mini at ~/.amico/ops/wip-sync.sh, unversioned. It is the last ops script outside the repo: #427/#453 versioned fleet-status, fleet-alert, papers-digest, and hunt.sh, but wip-sync shipped before that sweep. Its hand-applied v2 (multi-remote resilience, born of the 2026-08-20 qldpc-challenge 403) exists only on mini disk.

Approach

Follow the #453 pattern exactly: ops/wip-sync.sh becomes the versioned source of truth (the tested v2 currently deployed on the mini), ops/install.sh deploys it, ops/README.md documents the ritual (manual cadence — no launchd) and the v2 remote semantics.

Approaches Considered

• Version it in ops/ like its siblings (chosen) — same review trail, same deploy path, zero new mechanisms.
• Fold it into amico CLI as a verb — rejected: it's a shell ritual over many repos, not a package; a rewrite adds risk for no gain right now.
• Leave it mini-only — rejected: this is precisely the "disk failure loses it, no review trail" class chore: version the fleet ops scripts (fleet-status, fleet-alert, papers-digest) — unversioned on the mini today #427 closed.

Scope

In: versioning the script as-is (v2), install.sh row, README section, deploy reconciliation (mini copy already matches v2). Out: behavior changes beyond v2 (v2's semantics were incident-driven and sandbox-proven 2026-08-20); launchd wiring (it runs on demand, by hand); CLI-ization.

---

Acceptance Criteria

• ops/wip-sync.sh in the repo is byte-identical to the tested v2 on the mini (bash -n clean; the 5-scenario sandbox matrix ran against this exact content)
• ops/install.sh copies it to ~/.amico/ops/ idempotently, touching no state (.bak stays a mini-local recovery artifact)
• ops/README.md gains the wip-sync row (on-demand cadence) + a section covering the ritual and v2 semantics: multi-remote push (origin first, then forks), re-anchor over wip-only non-FF by stacking a fresh snapshot (never force, never amend), all-remote fetch, cross-remote wip-branch discovery
• After merge + deploy, repo and mini copies are in sync (they already are — the deploy is a no-op)

Key Decisions

• v2's re-anchor stacks a new snapshot commit on the incoming tip (chain grows, never forks) — an --amend variant was tested and rejected: it replaces the remote snapshot and stays non-FF.
• A remote carrying real (non-wip) incoming commits is genuine divergence: warn, never force — unchanged from v1.
• wip-sync.sh.bak on the mini is runtime recovery state, not repo content — install never touches it.

Prior Art / Patterns

• chore: version the fleet ops scripts (fleet-status, fleet-alert, papers-digest) — unversioned on the mini today #427 (closed by chore(ops): version the fleet ops scripts — fleet-status, fleet-alert, papers-digest #453) — the ops versioning sweep this completes; ops/README.md structure and ops/install.sh are the templates.
• ops/hunt.sh — later addition following the same pattern (Fleet dispatch hardening: retire raw-nohup hunts (timeout, heartbeat, durable logs, sweep-adoptable records) #426/Fleet dispatch hardening: retire raw-nohup hunts (timeout, heartbeat, durable logs, sweep-adoptable records) #454); wip-sync slots in identically.
• Incident record: 2026-08-20 qldpc-challenge leave stranded a snapshot (origin=unitaryfoundation read-only, pushable fork remote ignored) → the 403 that motivated v2.

Source

• Incident + v2 work: Amico session on the mini, 2026-08-20 (rebase sweep + wip-sync robustification); sandbox matrix: read-only-origin leave, competing-snapshot non-FF, cross-remote arrive, divergence guard, plain-repo regression.
• Related: chore: version the fleet ops scripts (fleet-status, fleet-alert, papers-digest) — unversioned on the mini today #427, chore(ops): version the fleet ops scripts — fleet-status, fleet-alert, papers-digest #453.