lane-end close-out gate: refuse to end a lane whose inventory still has worktrees, branches, scratch or caches — the dry-run sweep first, lane-end --sweep --yes to retire them in the same run; landing retires its own tree

[brettheap]

On Brett Heap's word of 2026-10-05 ("fold the three into 162 and file the close-out gate"), from the finding that workspace creation outpaces workspace closeout (the xFactory cleanup analysis: completed worktrees left after landing; branches outliving their purpose; scratch and caches beside repositories).

The gate

lane-end <lane> (and --retire) runs #162's sweep as a DRY RUN first and refuses to end the lane while its inventory still has anything to retire — worktrees (#97's sidecar inventory), their branches, local branches with no worktree that the lane created, lane scratch, caches — exactly as it refuses today while the lane holds open objects in its log. The refusal prints the disposition table and the one command that clears it:

lane-end: openRepoTools-3 still owns 3 worktrees (2 clean+merged; 1 dirty) and 5 branches — not ended.
  retire them:  lane-end openRepoTools-3 --sweep --yes      (rescue branches first; see the table)
  or review:    lane-worktrees sweep openRepoTools-3 --dry-run

lane-end --sweep --yes performs #162's dispositions (rescue before any removal; never a live writer; FOREIGN trees left alone), writes one register NOTED line per tree, and then ends the lane in the same run. --force-end ends the lane WITHOUT retiring (a word, logged with its reason) for the cases a person decides to keep a tree alive past the lane — the trees then show as orphaned in lanes --index --worktrees (#161) with the ended lane as owner.

Landing retires too

The shipped landing path — whatever command squashes a lane's PR with proof (today each lane's private squash-land.sh; a lane-land is the natural home, filed when that script is adopted) — removes the PR's worktree and both branches when the tree is clean at the landed head, as the private script does now, and records it. A landing that leaves a dirty tree says so and leaves it to the sweep.

Why a gate and not a reminder

The cleanup records show landing and retirement happening as separate acts for months, and the separate act not happening. lane-end already has the shape — refuse while open objects remain — so the gate adds one more thing it counts, from the inventory #97 already keeps.

Tests

Shell-suite cases: a lane with a clean merged tree refuses to end, ends with --sweep --yes, the tree and branches are gone, the register line names them; a dirty tree gets its rescue branch on the fake origin before removal; a live writer's tree is never touched and the lane refuses to end; --force-end ends and the orphan shows in the inventory; the refusal's printed command is the one that clears it.

Refs #162, #161, #97.

Lane: openRepoTools-3

🤖 Generated with Claude Code

Folded in on Brett Heap's word of 2026-10-05 ("file the not prevented fixes") — the inventory-only door and the guard checks

From the re-analysis of the ten cleanup causes: standalone clones (cause 2) and temporary work inside the estate (cause 4) were DETECTED by the sweep, not prevented. Two additions make the inventory the only door:

1. lane-worktree add <slice> [--branch <name>] [--from <ref>] — the one verb a lane uses to make a worktree: it runs git worktree add AND records the tree in the lane's Crash-consistent lane worktree recovery: a lane now says WHERE its session stopped — RUNNING → SWAPPING → SWAPPED with a generation/operation fence, a machine-readable worktree inventory taken at every handoff, and a resume reconciliation that reports and resets nothing #97 inventory in one act (path, branch, base, lane, UTC). A worktree made by hand is still legal git, but it is FOREIGN to every sweep until lane-worktree adopt <path> records it.
2. Two guard checks (in the UserPromptSubmit guard, through the register service of Amendment 22 act 2: the SQL register under lanes-edit.sh — mode=primary writers in one transaction each, the estate switch and lanes/STORE, the write journal and the renderer (one mirror commit per write), exit 69, fail-closed readers, lanes-index import/rollback, SERIALIZABLE claim order, and the agreement tests #165 once it exists, directly until then; both refuse with exit 2 and print the one remedy):
   • the session's cwd is a git repository under the estate that NO lane's inventory knows (a clone, or a worktree nobody recorded) → "this tree is not a lane's: lane-worktree adopt <path> or work in lane-worktree add …";
   • untracked, non-ignored paths exist directly under .lane-worktrees/<lane>/ or the estate root that are not inventory worktrees → "scratch belongs in ${XDG_STATE_HOME:-$HOME/.local/state}/openRepoTools/lanes/<lane>/scratch/ or the harness scratchpad; move it or lane-worktrees sweep --include-scratch".
     Both checks read only the inventory and git status --porcelain --ignored=no of the cwd's parent; neither reads the index (Amendment 14 clause (b)).

Tests: a hand-made clone under the estate refuses the prompt and the remedy adopts it; a stray scratch dir refuses until moved; an inventory worktree passes; the checks add under 50 ms to the guard.

Lane: openRepoTools-3

Folded in on Brett Heap's word of 2026-10-05 ("fold them") — cause 18, ignored scratch escaping git status

The guard's scratch check (and the sweep's) reads git status --porcelain --ignored — ignored material included — not the default porcelain that hides it. Ignored directories directly under .lane-worktrees/<lane>/, the estate root or a worktree root above a size threshold (default 50 MB; sweep.conf) refuse the prompt the same way untracked scratch does, naming the directory and its size and the remedy (lane-worktrees sweep --include-caches for caches/venvs; the scratch directory for the rest). The nightly report lists ignored directories over the threshold estate-wide.

Lane: openRepoTools-3

[brettheap]

CLAIMED — lane openRepoTools-3, session 83619980-06bb-4469-ae73-2a8c5e61981a@Eagle, 2026-10-05T13:04:31Z, for #163

Logged in lanes/log/openRepoTools-3.md at 0fa63ec1fa05076564701299ee3dcf4857b05051.

The three reads (lane-collision-protocol Rule 1):

1. existing `CLAIMED —` comments on opensoft/openRepoTools#163:
   none
2. `gh pr list --repo opensoft/openRepoTools --state all --search "#163"`, kept where the row or its body names `#163`:
   none
3. `git ls-remote --heads git@github.com:opensoft/openRepoTools` branches naming `163` as a whole token:
   none