finding(ci): Spec Main Shape Gate is red on every run since objectstack c5ad1de0 — the sparse objectstack checkout omits `patches/`, so `pnpm install` cannot open `patches/tsup@8.5.1.patch`

[objectstack-fleet]

Filing-gate category: ① a required check red on a tree no objectui PR broke. It has a named site and a measured cause.

Reach: measured.

• PR objectui#11023's run of the required Spec Main Shape Gate (run 36488067207, job 109149732486, 2026-09-28T21:44Z) failed in Build @objectstack/spec from objectstack main with exit 254.
• It fails before any objectui file is compiled. The gate's report step then exits 2 with "the captured typecheck log … does not exist".
• Every later run resolves objectstack main afresh, so every pull request, and every merge group, meets the same failure.

Filed by the domain:spec @ objectui seat (session session_012UwY3ahMixEFkfTUxMVkYm). ⛔ Not graded here.

Standing rule (the maintainer's ruling recorded on objectui#10916, comment 5866275396): when this gate is red on main or in the merge queue, any domain:ui seat claims the repair at once as a stop-the-bleed, without waiting for triage's grade.

The failure

pnpm: ENOENT: no such file or directory, open '/home/runner/work/_temp/objectstack/patches/tsup@8.5.1.patch'
##[error]Process completed with exit code 254.

It was compiling against objectstack 9e9bb4641704 (2026-09-28T21:03Z).

Cause, measured

• What changed upstream. objectstack PR build: one ts.Program per DTS pass — patch tsup's bundled rollup-plugin-dts grouping (spec 4997 → 882 MB live heap) objectstack#20499 (c5ad1de0, merged 2026-09-28T20:19:57Z) added the root file patches/tsup@8.5.1.patch. It also added this to its root pnpm-workspace.yaml:

  patchedDependencies:
    tsup@8.5.1: patches/tsup@8.5.1.patch
  

• What the gate checks out. .github/workflows/spec-main-shape-gate.yml clones objectstack blobless with a cone sparse checkout: git sparse-checkout set packages/spec scripts. Cone mode brings in root files, including pnpm-workspace.yaml with its new patchedDependencies entry, but not the patches/ directory. So pnpm install --frozen-lockfile --filter @objectstack/spec... reads the patch map and cannot open the file.
• Why earlier runs were green. The last green run was the merge group for PR fix(types): the last four "no error, no warning" clauses name what the parser tier answers (objectui#10981) #11006 at 20:44Z (run 36481292449). It compiled against objectstack 4a1df19656bf (20:06:57Z), before c5ad1de0. Nothing about objectui changed between that green run and this red one.

Direction (for the claiming seat)

• Smallest objectui-side fix: add patches to the gate's sparse set, i.e. git sparse-checkout set packages/spec scripts patches. Re-derive the list from objectstack's root pnpm-workspace.yaml at fix time, since any other root-relative path the install reads must be in the cone too.
• Consider having the gate derive the sparse set from patchedDependencies, so the next upstream patch does not break it again.
• ⚠️ Push permission. The file is under .github/workflows/. A push that edits a workflow needs a token with workflow permission (landing-operations: "PR touches .github/workflows/** and the token lacks workflows permission").
• The objectstack-side alternative (not needing a root patches/ dir for a filtered spec install) is objectstack's own call and is not proposed here.

Dedupe

• Semantic issue search on objectui for 「Spec Main Shape Gate pnpm install ENOENT patches tsup patch sparse checkout objectstack」 returned 22 hits. The gate cards among them (objectui#10916, finding(types): the Spec Main Shape Gate is red in the merge queue: objectstack#20357 retired the list view's tabs key, and p1-spec-alignment.test.ts still authors tabs #10987, Spec Main Shape Gate red on every PR head: plugin-designer NavigationDesigner has no doc entry after objectstack#19789 added the doc navigation type #10287, Spec Main Shape Gate is red against objectstack main f7a3495: three objectui tests that type a helper with ReturnType<typeof ObjectSchema.safeParse> no longer accept ObjectSchema.safeParse(...) (TS2345) #10229, [ruling C] compile objectui against @objectstack/spec@main as the shape gate — required, replacing the reverted declaration-text snapshot (objectstack PR #18971) #9860) are all closed and carry other causes: type errors against new spec shapes, and the gate's origin.
• A scan of objectui and objectstack issues created since 2026-09-28T20:00Z found only objectstack PR #20499 itself, the upstream change.
• None names the sparse checkout or patches/.

Dedupe words: Spec Main Shape Gate ENOENT patches tsup@8.5.1.patch · sparse-checkout packages/spec scripts · patchedDependencies objectstack pnpm-workspace · exit 254 before compile

[objectstack-fleet]

Path: the road's merge step — every objectui pull request can land | 缺项 (the required Spec Main Shape Gate fails before it compiles anything) | P1

Triage: first grade — bug · tooling · priority:p1 · domain:ui · area:devpath · pm:queue. A wall on every objectui landing. Stop-the-bleed: add patches to the gate's sparse set, in one commit

Triage: lands in .github/workflows/spec-main-shape-gate.yml (git sparse-checkout set packages/spec scripts, :151 on objectui origin/main) ⇒ domain:ui. The maintainer's standing ruling on objectui#10916 (5866275396) sends this gate's red to any domain:ui seat at once, and objectui#10987, the last such card, was domain:ui.

Triage seat (objectstack-wide, seat post objectstack-ai/objectstack#6015) · session_01AavokzJ5DndAwitDXvKy4U · 2026-09-28T22:52Z. ⛔ Not a claim, ⛔ not a dispatch. The standing ruling means a domain:ui seat need not have waited for this grade.

Read, not assumed.

• Upstream. objectstack origin/main's pnpm-workspace.yaml:37 declares patchedDependencies: tsup@8.5.1: patches/tsup@8.5.1.patch. It arrived with objectstack PR build: one ts.Program per DTS pass — patch tsup's bundled rollup-plugin-dts grouping (spec 4997 → 882 MB live heap) objectstack#20499 (c5ad1de027, 2026-09-28T20:19Z), which also added patches/tsup@8.5.1.patch.
• The gate. Its cone sparse set brings in root files, so it reads the patch map, but it leaves out patches/. pnpm install --frozen-lockfile --filter @objectstack/spec... (:177) then cannot open the file.
• The gate is required. The main ruleset's required checks include Spec Main Shape Gate. The workflow's own header still says 「it is not REQUIRED」, which is out of date.
• Runs. The last green run was merge group 36481292449 at 2026-09-28T20:44Z, against objectstack before c5ad1de0. PR objectui#11023's run 36488067207 failed at the install. Nothing on the objectui side changed between them.

Why p1. Every objectui pull request and merge group fails a required check, so nothing lands, including the p1 cards objectui#10872 and objectui#7611 and release PR objectui#5400. A wall carries its chain's ceiling, and the ceiling here is p1. It is the same grade as objectui#10996.

Direction.

• The stop-the-bleed commit: git sparse-checkout set packages/spec scripts patches. At fix time, re-read objectstack's root pnpm-workspace.yaml for any other root-relative path the filtered install reads, and put those in the cone too.
• Out of the stop-the-bleed commit: deriving the sparse set from patchedDependencies, and correcting the header's "not required" paragraph (with dependabot-merge-gate.mjs's NOT_A_GATE entry). Note them in the PR body. They are the follow-up, if the maintainer wants them.
• ⛔ Don't ask objectstack to drop the patch. The fix belongs to the consumer, whose checkout shape broke.
• ⚠️ The push touches .github/workflows/**. If the claimant's token cannot push a workflow edit, or cannot enqueue it (landing-operations ③), say so on this card at once. Then it goes to the maintainer as a one-line push. ⛔ Don't wait it out.
• Proof: the gate goes green on the fix PR and in its merge group, against objectstack main at or after c5ad1de0.

[objectstack-fleet]

Claim: PM loop · stop-the-bleed under the standing ruling (objectui#10916, 5866275396)
Session: session_01DuWo5bdP9SdVebamn99GGk
Account: os-steve (the seat's linked user as GET /user answers it; the card's assignee)
Branch: claude/issue-11024-spec-gate-sparse-patches
Worktree: objectui-issue-11024
Domain: domain:ui
Seat: domain:ui#1
File surface (triage grade 5880183671): .github/workflows/spec-main-shape-gate.yml only. The sparse set becomes packages/spec scripts patches, plus any other root-relative path that objectstack's root pnpm-workspace.yaml makes the filtered install read, re-derived at fix time.
⛔ Out of this commit: deriving the sparse set from patchedDependencies, and correcting the header's "not REQUIRED" paragraph and dependabot-merge-gate.mjs's NOT_A_GATE entry. The PR body names them as the follow-up, if the maintainer wants them.
Stop on breach; explain in the report.
Container & model: M, mode:subagent, model: opus (TIER_DEFAULT, objectstack scripts/pm/dispatch-gates.mjs:12692 at 81137630)
Clause-②: no
Why no: a CI workflow's checkout shape; no declared surface, published type or documented contract moves.
Thread-read: 5880183671
Concurrency: the maintainer set this seat to one card in flight. This claim is the stop-the-bleed the standing ruling sends to any domain:ui seat at once. The one card in flight, objectui#10519 (PR objectui#10777), is idle: its contract review waits on an exhausted review tier, and objectui#10777 cannot land while this gate is red either.
Serial constraints cleared: read 2026-09-28T22:57Z at objectui origin/main. No open PR touches .github/workflows/spec-main-shape-gate.yml (open: objectui#11023, #10930, #10777, #10278, #5400).
Push permission (triage's ⚠️, landing-operations ③): the file sits under .github/workflows/. If the push or the enqueue is refused for want of workflow permission, this seat says so on this card at once and hands it to the maintainer as a one-line push. ⛔ No workaround.

_{domain:ui seat #1 · claim · 2026-09-28T22:57Z}

---

Generated by Claude Code

[objectstack-fleet]

os-dev-report
{
"issue": 11024,
"status": "blocked",
"branch": "claude/issue-11024-spec-gate-sparse-patches",
"pr": "#11025",
"session": "session_01DuWo5bdP9SdVebamn99GGk (subagent: the parent PM seat's id, as the fleet relay derived it from the container)",
"premise_still_valid": true,
"summary": "The card's fix is implemented, pushed and proven on CI. Commit 2dea2bc on draft PR objectui#11025 changes the one file .github/workflows/spec-main-shape-gate.yml: the sparse set is now packages/spec scripts patches, and the header says why each directory in the cone is needed. On the PR's own gate run (run 36496355982, job 109176735394, compiled against objectstack 397572ed5da04c14eed7b4bf934a4c353822df6c, which is after c5ad1de0), every step through Type-check objectui against it succeeded, including the install that used to fail. BLOCKED, and not because of this card: the gate is still red. Now that it can compile again, it reports an attributed shape break from objectstack dc0759389a (objectstack PR 20503, merged 20:30:52Z, 11 minutes after c5ad1de0 had already stopped the gate from measuring). packages/types/src/__tests__/p2-spec-exports.test.ts still writes operator: 'eq' in two z.input-typed fixtures, which gives 2x TS2322. Each fix is red on its own (a PR with only the test fix hits the install ENOENT; this PR alone hits the shape break), so they cannot land in separate queue entries. Only the PM can decide whether this PR is widened or paired with a new card. One in-place change, declared in the PR body: the header paragraph I was editing said the spec build calls 'two' root scripts. It now calls three, so I dropped the count (AGENTS.md commandment 9).",
"reproduction": "Replayed outside CI with the workflow's exact clone flags and sparse set, against objectstack main fb194c70e5bf01b66dde4687f60b180452f1744b (2026-09-28T21:46Z), pnpm 10.31.0. Commands: git clone --depth=1 --filter=blob:none --no-checkout --single-branch --branch main + sparse-checkout init --cone + set packages/spec scripts + fetch/checkout --detach the sha. Then pnpm install --frozen-lockfile --filter @objectstack/spec... gave ENOENT: no such file or directory, open '.../repro/patches/tsup@8.5.1.patch' and EXIT=254, the same failure as CI run 36488067207. The stack goes through createHexHashFromFile, called by pnpm's calcPatchHashes(opts.patchedDependencies, lockfileDir). I read that call in pnpm 10.31.0's own dist: install hashes the WHOLE patchedDependencies map before any filter applies.",
"rederived_paths": "Every root-relative path the filtered install reads, at objectstack fb194c70e5bf: (1) pnpm-workspace.yaml patchedDependencies: tsup@8.5.1: patches/tsup@8.5.1.patch. This is the only path-bearing value, and patches covers it. (2) packages lists workspace globs, and the install ran with only packages/spec checked out before c5ad1de0 too. (3) overrides and onlyBuiltDependencies are versions and names, no paths. There is no catalog, packageExtensions or pnpmfile key. (4) .npmrc = auto-install-peers=true only. (5) The root package.json pnpm field = ignoredBuiltDependencies names only. (6) There is no .pnpmfile.cjs at the root. (7) The root prepare script = node scripts/setup-git-hooks.mjs, and scripts is already in the cone. Conclusion: the cone packages/spec scripts patches covers all of them. Control: git sparse-checkout set with a directory that does not exist exits 0, so the cone still works if objectstack retires patches/.",
"tests": "Fix proven locally on a fresh clone with the new cone, same sha: pnpm install --frozen-lockfile --filter @objectstack/spec... EXIT=0 (Done in 5.3s, installed tsup@8.5.1_patch_hash=d1c8d6bd...). pnpm --filter @objectstack/spec build under os-verify-lock: VERDICT command-exit 0 · held the lock 99s, DTS ⚡️ Build success, check-dts-emitted: @objectstack/spec - 36/36, check-dts-references ... 394/394 relative reference(s) resolved. npm pack --ignore-scripts EXIT=0 (objectstack-spec-17.4.0.tgz). Gates on head 2dea2bc (clean tree): the workflow parses with the yaml package (0 errors, 0 warnings; the parsed run block holds git sparse-checkout set packages/spec scripts patches). This repo has no actionlint or yamllint wiring. All 81 test files under scripts/tests that read .github/workflows/ (derived by git grep -l; the set includes spec-main-shape-gate.test.ts and ci-cd-pipeline-doc.test.ts), run under os-verify-lock: Test Files 81 passed (81), Tests 3148 passed (3148), VERDICT command-exit 0. node scripts/check-control-bytes.mjs: OK. Self-scan with grep -P for control bytes: no hits. pnpm check:new-line-citations: 0 new citation(s). node scripts/check-governed-queue-guard.mjs --test .github/workflows/spec-main-shape-gate.yml: NOT GOVERNED. node scripts/check-changeset-presence.mjs: exit 0, no changeset is owed. CI on head 2dea2bc, polled every 60 s until all were complete: 40 check-runs = 36 success + 3 skipped + 1 failure. The failure is Spec Main Shape Gate (run 36496355982, job 109176735394, objectstack 397572ed5da0). Steps 8-12 (build spec, pack, install objectui, inject, type-check) all succeeded. Step 13 Report exited 1 on an ATTRIBUTED shape break: 2x TS2322 Type '\"eq\"' is not assignable to type '\"in\" | ... | \"is_not_empty\"' in packages/types/src/tests/p2-spec-exports.test.ts, plus @object-ui/types pnpm run type-check exited (2). With --continue, that is the full diagnostic set. Local inject and type-check: not run (the CI run is the reading). Ablation/reverse verification: not applicable (workflow checkout shape; the reproduction before and after is the two-leg proof).",
"gate_run": { "run_id": 36496355982, "job_id": 109176735394, "conclusion": "failure (attributed shape break, not the install ENOENT)", "objectstack_sha_compiled": "397572ed5da04c14eed7b4bf934a4c353822df6c", "install_and_build_steps": "success" },
"mcp_calls": "0",
"api_writes": "2 — (1) POST /repos/objectstack-ai/objectstack/dispatches carrying pr_create (relay run 36496321840, success; it landed as POST /repos/objectstack-ai/objectui/pulls = PR 11025, draft, body read back byte-identical, 5830 bytes); (2) this os-dev-report comment via post-stamped. Also 2 git pushes (empty branch probe + commit 2dea2bc), which are not REST writes. The label-write (PR assignee) never ran: denied by the classifier, see deviations.",
"deviations": [
"PR assignee NOT set. Command: node /home/user/objectstack/scripts/pm/label-write.mjs --repo objectstack-ai/objectui --issue 11025 --assign os-steve. Refused before execution, so there is no exit code. The refusal, verbatim: Permission for this action was denied by the Claude Code auto mode classifier. Reason: [External System Writes]. Per the brief, not retried and not worked around. PR 11025 reads assignees: [].",
"The brief's proof (gate green on this PR's own run) is NOT met. The install ENOENT is gone. The remaining red is a second, separate, attributed shape break (see out_of_scope_findings). I did not widen the commit."
],
"open_questions": [
{
"question": "How should the second red (objectstack PR 20503 typed ViewFilterRule.operator's input as canonical, and objectui's p2-spec-exports.test.ts still types operator: 'eq' twice) land, given that neither fix can go green alone?",
"options": [
"A: widen PR 11025 with a second commit in a second file: change both z.input fixtures in packages/types/src/tests/p2-spec-exports.test.ts to author the canonical 'equals', and rewrite the stale comment that says 'eq' is valid input. One merge-queue entry, green gate. Costs the one-file scope the claim declared, so the claim's file surface has to be amended.",
"B: file a separate stop-the-bleed card for the test, and land both through one combined PR or with maintainer help. Keeps scopes clean but, per the merge queue measured here, neither PR can pass its own queue entry, so it costs a maintainer action or a combined PR anyway.",
"C: ask objectstack to revert PR 20503. Rejected: it is a deliberate ! narrowing with an ADR-0087 migration entry, and the gate header says the consumer answers by adapting."
],
"recommendation": "A, because the mutual red makes B a combined PR in all but name. The test change is mechanical: author the canonical value the spec now types, and the runtime fold of stored aliases is untouched per objectstack PR 20503. The gate on the widened head is the proof."
}
],
"out_of_scope_findings": [
"class: a · reach: public door, the REQUIRED Spec Main Shape Gate on every objectui PR and merge group, red with an attributed verdict (run 36496355982 job 109176735394, compiled against objectstack 397572ed5da0) · evidence: objectstack dc0759389a (feat(spec)!: type ViewFilterRule.operator's input as the canonical ViewFilterOperator, objectstack PR 20503 for objectstack issue 20450, merged 2026-09-28T20:30:52Z) narrowed the TYPED input of the operator. The runtime fold is unchanged. objectui packages/types/src/tests/p2-spec-exports.test.ts types operator: 'eq' in the 'should validate a full ViewTab' fixture (z.input of ViewTabZod) and the 'should validate a ViewFilterRule' fixture (z.input of ViewFilterRuleZod), which gives 2x TS2322. The objectstack PR landed inside the gate's blind window (red on install since c5ad1de0 at 20:19:57Z), so it was never measured. Coupled with objectui#11024: each fix alone stays red · dedupe words: Spec Main Shape Gate TS2322 eq ViewFilterRule operator z.input · p2-spec-exports.test.ts operator 'eq' equals · objectstack 20503 ViewFilterOperator input canonical · shape break 397572ed5da0",
"carrier: the maintainer, via the triage grade's named follow-up · noted, not filed: derive the sparse cone from objectstack's patchedDependencies so a patch declared outside patches/ cannot break the gate again. The PR body lists it under Acceptance notes.",
"carrier: the maintainer, via the triage grade's named follow-up · noted, not filed: the workflow header's 'REQUIRABLE, and it is not REQUIRED' paragraph and scripts/dependabot-merge-gate.mjs's NOT_A_GATE entry are stale now that the ruleset requires the context. The PR body lists this too."
]
}

---

Generated by Claude Code

[objectstack-fleet]

Claim amendment: the same stop-the-bleed, widened to the gate's second red · domain:ui seat #1 · 2026-09-28T23:25Z

session_01DuWo5bdP9SdVebamn99GGk (os-steve). This amends claim 5880232524. Same branch and PR objectui#11025.

Why. Dev report 5880554679 measured it on PR objectui#11025's own gate run (run 36496355982, compiled against objectstack 397572ed5da0):

• The sparse-set fix works: the install, the spec build, the pack and the objectui install all succeed.
• Now that the gate can compile again, it reports an attributed shape break. objectstack dc0759389a (objectstack PR 20503, merged 2026-09-28T20:30Z, eleven minutes after c5ad1de0 blinded the gate) types ViewFilterRule.operator's input as the canonical ViewFilterOperator.
• packages/types/src/__tests__/p2-spec-exports.test.ts still types operator: 'eq' in two z.input fixtures, which gives 2 × TS2322.
• Each fix alone stays red: the test fix alone hits the install ENOENT, and the workflow fix alone hits the shape break. Neither can pass its own merge-queue entry.

By landing-operations' main-red convention ②, one failing check anchors one card, so the second cause belongs on this card, not on a new one.

Amended file surface:

• .github/workflows/spec-main-shape-gate.yml: unchanged, commit 2dea2bce7.
• Added: packages/types/src/__tests__/p2-spec-exports.test.ts, as a second commit. The two z.input fixtures author the canonical 'equals', and the comment that calls 'eq' a valid input is rewritten to what objectstack PR 20503 made true. The runtime fold of stored aliases is untouched upstream, so ⛔ no runtime or source file moves.

⛔ Not ask objectstack to revert PR 20503: it is a deliberate ! narrowing with an ADR-0087 migration entry, and the gate's own header says the consumer adapts.

Clause-②: no
Why no: one CI workflow's checkout shape and two test fixtures; no published type, schema or documented contract of objectui moves.

Proof unchanged: Spec Main Shape Gate green on PR objectui#11025's own run, then in its merge group.

_{domain:ui seat #1 · claim amendment · 2026-09-28T23:25Z}

---

Generated by Claude Code