In the no-estate sweep, park skips a root that has no Speckit git overlay instead of counting it as a refusal #6

Lane: openreposhape-2 (openRepoShape-2)

Ruling: opensoft/openRepoShape#92 (comment) — Brett Heap, 2026-09-10, verbatim: "rule on the #92 open item: skip roots without the overlay."

What is wrong today

openRepoShape #91/#93 (landed there as d090a7a, carved here by its #94 and living in this repository since PR #1) made bare park — no <Name>, no --repo, no estate around the cwd — park EVERY estate under the projects directory, in name order, continuing past a refusal.

On Eagle that sweep exits 1 and always will: four roots (MedxEHR, MedxGlass, openDox, openXdox) carry no Speckit git overlay, so each one's own make park refuses with exit 2 —

make park needs the Speckit worktree overlay; install it with: setup-openspeckit

— which the sweep counts as a refusal. A root that was never given the overlay is not a park that FAILED; it is a park there was never anything to run. Counting it as a failure makes the one summary line that matters unreadable: a person who has one genuinely refused estate cannot see it among four roots that were never park-able.

The semantics to implement

1. The sweep only. In park_every_estate / run_one_estate, before running make park in a standalone root, test the overlay THE SAME WAY THE MAKEFILE TESTS IT — -x <root>/.specify/extensions/git/scripts/bash/park.sh, which is templates/assembly-root/Makefile's own test -x $(PARK) || … exit 2 with OVERLAY_MISSING. Absent: print ONE line in that estate's report block —

     SKIPPED (no Speckit overlay; install it with: setup-openspeckit)
   

   — count it under a new skipped tally, run nothing there, and do not let it touch the exit code. A family holder is run as it is today: its own templates/family-root/Makefile has no such test, it fans out to the members with siblings.py --make, and a member's own refusal is the holder's to report — so members are NOT pre-checked here, and the code says so in a comment.

2. Explicit is unchanged. park <Name>, park --repo <owner/name> and the walk-up inside an estate keep today's behaviour: the Makefile's exit 2 refusal is relayed, verbatim, naming setup-openspeckit. A person who NAMES a root gets the refusal; only the sweep, which named nothing, skips.

3. The summary gains the skipped count, and only when there is one, so the Crash-consistent lane worktree recovery: canonical inventory and RUNNING → SWAPPING → SWAPPED lifecycle #91 line is untouched on a workstation where every estate has the overlay:

   park: 3 estate(s) parked, 0 refused or with findings, 4 skipped (no overlay)
   

   The exit code is computed over NON-SKIPPED estates only: 0 when every non-skipped estate parked, 1 when any refused or was left with findings — which is Crash-consistent lane worktree recovery: canonical inventory and RUNNING → SWAPPING → SWAPPED lifecycle #91's rule with the skipped ones simply not counted. If EVERY estate was skipped: exit 0, and a line that says so and names the installer.

4. --dry-run prints would skip (no Speckit overlay; install it with: setup-openspeckit) for such a root, and creates nothing.

5. The documents. README.md's and AGENTS.md's bare-park sentences (from Crash-consistent lane worktree recovery: canonical inventory and RUNNING → SWAPPING → SWAPPED lifecycle #91) gain the skip clause, and so does the header comment in park. The byte-identical shared estate resolver between park and resume is NOT touched — test_the_two_verbs_carry_the_same_estate_resolver_byte_for_byte holds that still, and this change belongs above it, in park-everything's own section.

resume does not change: the ruling is park-only, and resume's bare form still refuses.

Tests

In tests/test_park_resume_commands.py, in the #91 section's style: a sweep over two roots WITH the overlay stub and two WITHOUT (both parked, both SKIPPED, exit 0, the summary line); the all-skipped case (exit 0, and it says so); a skipped root beside a REAL refusal (exit 1, both lines); --dry-run says would skip; explicit park <Name> on an overlay-less root still exits non-zero with the Makefile's own refusal; a family holder in the sweep is still RUN and never pre-checked. The existing #91 sweep tests get the overlay stub their standalone roots now need, which is the fixture telling the truth about what a park-able root is.

Where this lands

Here, in opensoft/openRepoTools — the command's home since openRepoShape #94 — never in openRepoShape. workBenches re-vendors park afterwards.