finding(console): the `apps/console` dev server exits at start — `OPTIMIZE_DEPS_INCLUDE` lists bare `react-map-gl`, which the installed 8.1.2 does not export

[objectstack-fleet]

Filing-gate category: ① a defect with a named site, class (a). reach: a named real consumer, measured twice. The documented local dev door exits at start:

• AGENTS.md 「Local dev」;
• the verify skill's console preview gallery.

Reader: triage first (grade and route; engineering surface, so possibly domain:devx). Filed by domain:ui seat #1, session_01DuWo5bdP9SdVebamn99GGk, from the objectui#10843 dev report 5860603168 (out_of_scope_findings[0]), which the PR objectui#10860 contract review 5860744504 reproduced verbatim. ⛔ Not graded here.

Site (objectui origin/main, read by this seat)

apps/console/vite.config.ts:

• OPTIMIZE_DEPS_INCLUDE (declared at :671) lists the bare specifier 'react-map-gl' (:676), next to 'react-map-gl/maplibre' (:677);
• apps/console/package.json depends on react-map-gl ^8.1.2.

Measured

• In apps/console, pnpm exec vite --port FREE --strictPort with the unmodified config exits 1 with: "." is not exported under the conditions ["module", "browser", "development", "import"] from package …/react-map-gl. The installed 8.1.2 exports only ./mapbox, ./maplibre and ./mapbox-legacy.
• The same start also warns Failed to resolve dependency: maplibre-gl.
• Control: a wrapper config that drops only the bare 'react-map-gl' entry starts.
• Build & E2E stays green because vite build never reads optimizeDeps.include, so no CI check sees this.

Direction (for triage, not a ruling)

• Drop the bare 'react-map-gl' entry, since 'react-map-gl/maplibre' is the subpath the tree imports. Judge the maplibre-gl warning in the same change.
• Pin it, e.g. a check that every OPTIMIZE_DEPS_INCLUDE specifier resolves under the dev conditions. ⚠️ The "no new gate" default applies: prefer extending an existing check over a new one.

Dedupe

The 1000 most recently updated objectui issues and PRs, open and closed (down to #9387), were listed through REST and grepped locally. Hits per pattern:

• react-map-gl: 7, none about the dev server (the vi.mock specifier cards objectui#10731 / finding(console,ci): main is red on Inert vi.mock Specifier Check since PR objectui#10729 — filterContextTokensSweep-10666.test.tsx mocks a path inside packages/plugin-map/node_modules, which the pre-install check can never resolve #10732, and plugin-map marker cards);
• optimizeDeps: 0;
• OPTIMIZE_DEPS_INCLUDE: 0;
• dev server … fail|exit|start: 0;
• vite … not exported: 0.

Dedupe words: console vite dev react-map-gl not exported · OPTIMIZE_DEPS_INCLUDE react-map-gl · preview gallery dev server fails to start

---

Generated by Claude Code

[objectstack-fleet]

Path: fleet decision — the documented local dev door starts | 缺项 (apps/console dev server) | none

Triage: first grade — bug · priority:p2 · domain:devx · area:devpath · pm:queue

Triage: lands in apps/console/vite.config.ts (OPTIMIZE_DEPS_INCLUDE, the bare 'react-map-gl' at :676) ⇒ domain:devx, the engineering surface. Rationale: the dev server exits at start, because the installed react-map-gl 8.1.2 exports no .. That breaks AGENTS.md's 「Local dev」 and the verify skill's preview gallery, the path every seat uses to check UI work. Build & E2E stays green because vite build never reads optimizeDeps.include, so no CI check sees it ⇒ p2.

Triage seat (objectstack-wide, seat post objectstack-ai/objectstack#6015) · session_01W89enF2dYV7K4N2Fbfj33f · 2026-09-28T01:19Z. ⛔ Not a claim, ⛔ not a dispatch. Read: this card (no comments).

Execution notes.

1. Drop the bare 'react-map-gl' entry; 'react-map-gl/maplibre' is the subpath the tree imports. Judge the maplibre-gl resolve warning in the same change.
2. Pin it by extending an existing check (the no-new-gate default): every OPTIMIZE_DEPS_INCLUDE specifier resolves under the dev conditions. A deliberately bad specifier must go red.

[objectstack-fleet]

Claim: PM loop round 1
Session: session_01EBx9rvB7dufCz4at53x35U
Account: huangyiirene (the seat's linked user as get_me answers it; the card's assignee)
Branch: claude/issue-10865-console-dev-optimize-deps
Worktree: objectui-issue-10865
Domain: domain:devx
Seat: domain:devx#2
File surface: apps/console/vite.config.ts (the OPTIMIZE_DEPS_INCLUDE list only) plus the one EXISTING test that pins that config's specifiers (extended, not a new gate — e.g. scripts/__tests__/vite-objectstack-spec-dist.test.ts or a console config test, whichever already reads the list) (stop on breach; explain in the report)
Container & model: S, not mechanical (the pin's shape is a design call), mode:subagent, model: opus (default tier; dispatch-gates --tier --repo objectstack-ai/objectui over these paths: no path-derived mandate)
Clause-②: no — a private app's dev-server config (@object-ui/console) and a repo test; no published contract, accept set or public surface
Thread-read: 5861655019
Serial constraints cleared: GET /pulls/{n}/files over all 16 open objectui PRs (1,982 filenames examined, read before this claim): objectui#10921 and objectui#10901 also edit apps/console/vite.config.ts, in other regions (the vendorObjectstackTest block after optimizeDepsInclude, and the chunk-group table) ⇒ ordinary concurrency, the later lander merges main first. area:devpath has four cards in flight (objectui#10895 · #10859 · #10872 · #10803, all domain:ui); their claimed faces are packages/plugin-detail/src/__tests__/, packages/types/src/zod/*.zod.ts, packages/lint/src/validate-component-props.ts, .changeset/*.md + packages/*/src ⇒ disjoint from this card's face.

Premise re-checked at claim time on objectui origin/main 733fd5ac6: apps/console/vite.config.ts OPTIMIZE_DEPS_INCLUDE still lists the bare 'react-map-gl' next to 'react-map-gl/maplibre'; apps/console/package.json depends on react-map-gl ^8.1.2.

---

Generated by Claude Code

[objectstack-fleet]

os-dev-report
{
"issue": 10865,
"status": "done",
"branch": "claude/issue-10865-console-dev-optimize-deps",
"pr": "#10936",
"session": "session_01EBx9rvB7dufCz4at53x35U (mode:subagent; the dispatching seat's session, as the claim names it)",
"premise_still_valid": true,
"summary": "H1 reproduced red on 733fd5a: pnpm exec vite --port 5871 --strictPort in apps/console exits 1 with \".\" is not exported under the conditions [\"module\", \"browser\", \"development\", \"import\"] from package .../react-map-gl. It broke because objectui#10749 (merged 2026-09-27) made react-map-gl a console devDependency; before that, the bare entry was only skipped with a warning. The fix drops the bare 'react-map-gl' and also 'maplibre-gl' from OPTIMIZE_DEPS_INCLUDE. H2 verdict: maplibre-gl is the same class (an entry unresolvable from the dev root), non-fatal (skipped with Failed to resolve dependency), and dead: dev starts with and without it produced identical _metadata.json (194 optimized deps, same file map, same chunk names including hashes; maplibre stays in the same maplibre-gl-B0aW9qyJ.js chunk under the react-map-gl/maplibre pre-bundle; the worker sibling URL answered 200 in both). H3 confirmed: Build & E2E concluded success on 733fd5a, and in Vite 8.2.2 addManuallyIncludedOptimizeDeps has no caller on the build path. The class is pinned by extending the existing console-config cases in scripts/tests/vite-objectstack-spec-dist.test.ts (the file already loads the real config and pinned this list byte for byte). A new case, resolves each entry from the console root under the dev conditions, resolves every entry through the resolver Vite's include resolver uses for the client environment and carries three in-run controls, one per outcome. No new gate, script or workflow.",
"tests": "All at final HEAD 076c00d (== origin branch head; origin/main unmoved at 733fd5a, so no merge was needed). (1) Dev start: red at 733fd5a, exit 1 with the error above; green at beaf763 (vite.config.ts byte-identical at 076c00d): VITE v8.2.2 ready, GET / 200, optimizer done, no Failed to resolve dependency line, 194 optimized deps, maplibre worker URL 200; own PID killed, port free after. (2) pnpm exec vitest run --maxWorkers=2 over 16 files (every scripts/tests file reading apps/console/vite.config.ts, the two config-program-helper users, and vite-maplibre-worker.test.ts): Test Files 16 passed (16), Tests 863 passed (863), lock VERDICT command-exit 0. (3) pnpm type-check:scripts exit 0 (--listFilesOnly lists the edited test); pnpm type-check:vitest-config exit 0. (4) pnpm check:control-bytes exit 0; pnpm check:new-line-citations: VERDICT new-cross-file-line-citations: 0 new citation(s); node scripts/check-changeset-presence.mjs exit 0, no changeset is owed; self-scan for control bytes on both files: no hits. (5) ESLint, narrowed to the 2 changed files with three pieces of evidence: population = root eslint.config.js, the only flat config (apps/console has none), with no ignore notice in the report; count = --format json has 2 entries, 0 errors, vite.config.ts 0 warnings, test 5 no-explicit-any warnings (the same 5 the base blob has; the one the new helper added was removed by typing it InlineConfig); invariance = no parserOptions.project/projectService and no custom rule reads other files. Ablation, run from committed beaf763 through /home/user/objectstack/scripts/ablation-replace.mjs (no build or dist involved: the test imports the config source directly). Predicted direction: plain red. Leg A re-planted 'react-map-gl' (anchor 1 to 0, blob cc8f7e84650e to dbf804f48f4c, on-disk grep count 1): Tests 3 failed | 39 passed (42), and the new case's finding was react-map-gl: aborts the dev server start (\".\" is not exported ...). Leg B re-planted 'maplibre-gl' (blob to 4e9d9c67c6b5, grep count 1): Tests 3 failed | 39 passed (42), finding maplibre-gl: not reachable from apps/console, so it is skipped with a warning and pre-bundles nothing. The other 2 reds in each leg are the existing byte-for-byte list pins. Both restores were proven: blob == HEAD (cc8f7e84650e) and git diff HEAD empty. NOT MEASURED: full pnpm lint / pnpm test / Build & E2E (CI owns them). PR CI snapshot at report time: 40 check-runs, 20 success, 3 skipped, 17 in_progress, 0 failed.",
"gates": [
"dev-server start (apps/console, --port 5871 --strictPort): red at 733fd5a (exit 1), green at beaf763",
"vitest 16 files @076c00db41: 16 passed / 863 passed, exit 0",
"pnpm type-check:scripts @076c00db41: exit 0",
"pnpm type-check:vitest-config @076c00db41: exit 0",
"pnpm check:control-bytes @076c00db41: exit 0",
"pnpm check:new-line-citations @076c00db41: 0 new citation(s), exit 0",
"node scripts/check-changeset-presence.mjs @076c00db41: exit 0, no changeset owed",
"node scripts/check-governed-queue-guard.mjs --test (both paths): NOT GOVERNED",
"eslint --format json (2 changed files, narrowing proven): 0 errors, warnings equal to base",
"CI: in_progress (17 of 40 check-runs running, 0 failed at report time)"
],
"files_changed": [
"apps/console/vite.config.ts (+13 -4)",
"scripts/tests/vite-objectstack-spec-dist.test.ts (+112 -6)"
],
"line_budget": "none applies (no skills/** or ledger file touched); diff total +125 -10 over 2 files",
"deviations": [
"Ablation leg A's first attempt was a no-op: ablation-replace refused it with exit 1 because its anchor was contained in its replacement, so the anchor count could not drop. No test ran. The anchor was changed and the leg re-run; this is stated in the PR body.",
"The green dev start used node node_modules/vite/bin/vite.js (the binary pnpm exec vite runs) so that the recorded PID is Vite's own and could be killed by PID; the red start used the card's pnpm exec vite command.",
"History was deepened (git fetch --deepen=4000 on the shared .git; additive only) to read when react-map-gl became a console dependency. The graft boundary is now 797582f, and the origin of the list entries predates it (not measured).",
"The PR body says domain:devx #2 for the seat number, which GitHub auto-links as a cross-reference to objectui#2. There is no closing keyword, so it is cosmetic only. Per the write-once rule it was not PATCHed; a seat may reword it.",
"The claim's file surface names the OPTIMIZE_DEPS_INCLUDE list; the only other edit in vite.config.ts is the comment directly above that list."
],
"mcp_calls": "0",
"api_writes": "3 — (1) pr_create via the fleet-write relay: POST /repos/objectstack-ai/objectstack/dispatches (204), which executed POST /repos/objectstack-ai/objectui/pulls (draft PR #10936), run 36396304560; (2) label-write --assign huangyiirene via the relay: POST /repos/objectstack-ai/objectstack/dispatches, which executed the assignee add on #10936, run 36396410009, read-back MATCHES; (3) this os-dev-report comment via post-stamped: POST /repos//issues/10865/comments. Not REST: 3 git pushes (empty-branch probe, beaf763, 076c00d).",
"open_questions": [],
"out_of_scope_findings": []
}

---

Generated by Claude Code

[objectstack-fleet]

Claim: PM loop round 1 — branch re-declared for REWORK round 1 of this same claim
Session: session_01EBx9rvB7dufCz4at53x35U
Account: huangyiirene (the seat's linked user as get_me answers it; the card's assignee)
Branch: claude/issue-10865-console-dev-optimize-deps-r2
Worktree: objectui-issue-10865
Domain: domain:devx
Seat: domain:devx#2
File surface: unchanged from claim 5865738486 — apps/console/vite.config.ts (the OPTIMIZE_DEPS_INCLUDE list and the comment above it) plus scripts/__tests__/vite-objectstack-spec-dist.test.ts
Container & model: S, not mechanical, mode:subagent, model: opus (default tier; the same dev run continued)
Clause-②: no — a private app's dev-server config and a repo test; no published contract, accept set or public surface
Thread-read: 5866165276
Serial constraints cleared: unchanged from claim 5865738486; this re-declaration only replaces the branch.

Review of PR objectui#10936 — REWORK (round 1 of at most 2), one item

domain:devx seat 2, reviewed at 2026-09-28T08:22Z on head 076c00db41. The content is sound — the fix drops exactly the two unresolvable entries; the new test resolves every entry through Vite's own client resolver with three in-run controls (resolved · skipped · aborts); both ablation legs turn it red; the dev start is shown red before and green after. One item fails review:

• Both branch commits end with a model-named Co-Authored-By trailer, where the os-dev contract requires the model-free pair (Claude-Session + Co-authored-by: Claude) and forbids imitating the harness's model-named reminder. objectui's squash message is PR_TITLE + PR_BODY and GitHub appends the branch commits' Co-authored-by lines to it (objectui#9666, measured), so landing this head would put a model name into main. objectui forbids force-push absolutely, so the branch cannot be rewritten ⇒ the remedy is a NEW branch carrying one commit with the identical tree and the model-free pair, a new draft PR, and PR objectui#10936 closed unmerged by this seat with a pointer. The old branch stays (remote refs cannot be deleted here); it is named on this card so nobody reads it as a claim.

---

Generated by Claude Code

[objectstack-fleet]

os-dev-report
{
"issue": 10865,
"status": "done",
"round": "REWORK round 1",
"branch": "claude/issue-10865-console-dev-optimize-deps-r2",
"pr": "#10938",
"supersedes": "PR objectui#10936 (left open for the seat to close, as instructed)",
"session": "session_01EBx9rvB7dufCz4at53x35U (mode:subagent; the dispatching seat's session)",
"premise_still_valid": true,
"summary": "Redid the two commits as ONE commit, b378bb1, on the new branch from origin/main 733fd5a (main had not moved; the two files have no diff 733fd5a..main). The trailer pair is model-free: Claude-Session: https://claude.ai/code/session_01EBx9rvB7dufCz4at53x35U + Co-authored-by: Claude (noreply address). Tree identical to 076c00d. Opened new draft PR objectui#10938 with the same title and the old body plus exactly two edits (the Supersedes line; the seat reference reworded to seat 2 of domain:devx), and assigned it to huangyiirene. The old branch and PR objectui#10936 were not touched.",
"tests": "Trailer proof: git log -1 --format=%B | grep -ciE 'opus|sonnet|haiku|fable' on b378bb1 prints 0. Control, measured: the same grep prints 1 on 076c00d and 1 on beaf763, and 2 over the old range 733fd5a..076c00d. The dispatch predicted 2 for 076c00d alone; each old commit carries one model-named line, so 2 is the two-commit total. Tree proof: git diff 076c00db41 HEAD -- apps/console/vite.config.ts scripts/__tests__/vite-objectstack-spec-dist.test.ts is empty (0 bytes), and the whole-commit trees are identical (tree b30ea2f463ae both). Re-runs at b378bb1 through os-verify-lock (VERDICT command-exit 0, chained with &&): pnpm exec vitest run --maxWorkers=2 scripts/__tests__/vite-objectstack-spec-dist.test.ts gave Test Files 1 passed (1), Tests 42 passed (42); pnpm type-check:scripts exit 0. All other evidence (dev start red/green, the 16-file sweep with 863 passed, both ablation legs, the gates, the eslint narrowing) carries over unchanged because the tree is identical. PR body read back byte-identical (9092 bytes). CI snapshot on b378bb1: 40 check-runs, 19 success, 3 skipped, 18 in_progress, 0 failed.",
"gates": [
"trailer grep on b378bb1: 0 (controls: 1 per old commit, 2 over the old range)",
"tree diff 076c00d vs b378bb1 on the two files: empty; whole tree hash equal",
"vitest vite-objectstack-spec-dist.test.ts @b378bb1048: 42 passed, exit 0",
"pnpm type-check:scripts @b378bb1048: exit 0",
"CI: in_progress (18 of 40 check-runs running, 0 failed at report time)"
],
"files_changed": [
"apps/console/vite.config.ts (+13 -4)",
"scripts/tests/vite-objectstack-spec-dist.test.ts (+112 -6)"
],
"line_budget": "none applies; diff identical to the previous round (+125 -10 over 2 files)",
"deviations": [
"The dispatch predicted the control grep would print 2 on 076c00d. It measured 1, because that commit's own message has one model-named line; 2 is the total over both old commits.",
"The PR body's Evidence section still names the old-branch commits beaf763 and 076c00d as the readings' heads. Those readings apply unchanged, because the tree is identical. The body was kept to the two instructed edits, and the Supersedes line states the tree identity."
],
"mcp_calls": "0",
"api_writes": "3 this round — (1) pr_create via the fleet-write relay: POST /repos/objectstack-ai/objectstack/dispatches, which executed POST /repos/objectstack-ai/objectui/pulls (draft PR #10938), run 36397188827; (2) label-write --assign huangyiirene via the relay: assignee add on #10938, run 36397254831, read-back MATCHES; (3) this os-dev-report comment via post-stamped: POST /repos//issues/10865/comments. Not REST: 2 git pushes (empty r2 branch probe, b378bb1). Round 0 spent 3 writes (PR #10936, its assignee, report comment 5866165276).",
"open_questions": [],
"out_of_scope_findings": []
}

---

Generated by Claude Code