feat(repo-fleet-hygiene): sync verb — move every canonical repo to its default branch, ff-pull, park divergent work in a worktree

[kyle-sexton]

Problem

Nothing in the marketplace moves a fleet of canonical checkouts to their default branch and fast-forwards them without risking work. repo-fleet-hygiene:audit is read-only by contract, apply only deletes merged branches and worktrees, and repo-hygiene:clean tree-batch is a destructive reset.

Proposal

A new repo-fleet-hygiene:sync skill:

• switches each canonical checkout to the remote's current default branch (git ls-remote --symref origin HEAD) and pull --ff-onlys it;
• parks divergent work (non-default branch with uncommitted changes, or dirty tree on the default branch) into a linked worktree via source-control's worktree-create.sh, which gains an existing-branch mode;
• infers scope from a six-rung ladder on bare invocation (explicit args, fleet conf, conversation-named paths, ghq roots if ghq is present, cwd context, else a transparent exit 3), via a shared resolver that audit also adopts as its no-scope fallback;
• bare invocation is a dry-run plan; mutation only behind --apply after one confirmation gate; non-fast-forward, dubious ownership, and partial stash apply are skip-and-report.

Brief and acceptance criteria: docs/topics/fleet-sync/PLAN.md on the task branch.

Out of scope (follow-ups)

• Zero-argument scope inference for repo-hygiene:clean tree-batch (separate plugin, destructive verb).
• Adding sync-related findings to audit-fleet.sh.

[kyle-sexton]

This was generated by AI during triage.

Human-gated. Open decision: approve implementation of the locked PLAN (C4: new skill + a source-control worktree-create mode + audit's shared resolver, spans two plugins)

[kyle-sexton]

From the 2026-09-29 audit of the unattended Cursor run.

What the audit found. The 2026-09-23 triage comment parked this issue as human-gated on "approve implementation of the locked PLAN", and no approval is recorded on the issue. PR #5129 (0.23.43) merged the sync verb anyway, under the owner's account, from the unattended run. The PLAN the audit quotes (feat/3992-fleet-sync:docs/topics/fleet-sync/PLAN.md) is not on origin, so this comment cites the audit findings, not a link. The verb that shipped also had defects and left part of the PLAN's scope out.

Owner decision needed (implement neither option here).

• Question: ratify the shipped sync verb, or park it as closed PR docs(repo-fleet-hygiene): park structural sync verb until funded (#3992) #5007 proposed (revert 0.23.43)?
• Option A, ratify: keep the verb; the fix PR closes its defects.
• Option B, park: revert 0.23.43 and keep the PLAN as the record.
• Recommendation: A. Basis: plugins/repo-fleet-hygiene/skills/sync/SKILL.md sets disable-model-invocation: true, and sync-fleet.sh:213-225 defaults to dry-run and requires --apply plus one confirmation; the fix branch's T1-T3 close the audited defects.
• Unblocks: whether the follow-up issue (PLAN scope left out of feat(repo-fleet-hygiene): sync canonical checkouts to default branch (#3992) #5129) is worth working.

Delivered on the fix branch (draft PR #5289, Refs #3992). Stash by marker SHA; nonzero exit with the remedy; handling of worktree-helper exit codes 3/4/5; sibling-plugin probe removed; helper-resolved worktree root; exit-3 remedies; pairing gate for sync; docs.

Still open (listed in the follow-up issue). --skip, --skip-from, --repos-from, the ancestor probe, ghq root --all, per-target rung output, and shared discovery so sync honors fleet.skip and fleet.skipAppend.

Reopened and kept needs-human; nothing else changes until the owner answers.