chore: upgrade vite-plus to 0.3.0

[juliusmarminge]

Problem

vp run dev --share defaults T3CODE_BUNDLED_DEV=1 (scripts/dev-runner.ts, apps/web/vite.config.tsexperimental.bundledDev) so a remote browser on the tailnet pays a few chunk requests instead of a per-module round trip. On vite-plus 0.2.2, which bundles vite 8.1.2 / rolldown 1.1.4, that bundled dev mode is flaky for the shared origin: the page throws @vitejs/plugin-react can't detect preamble and Cannot read properties of undefined (reading 'Root') in <Collapsible> (a base-ui namespace import resolving to undefined in the bundled chunk), so the app never gets past the splash. Unbundled dev renders fine.

Fix

Bump the pinned toolchain to vite-plus 0.3.0, which bundles vite 8.2.2 / rolldown 1.2.5. The 8.1.2 to 8.2.2 range carries the batch of experimental.bundledDev fixes (vitejs/vite CHANGELOG):

• bundled-dev: reload once after rebuild instead of via the fallback page (#23106)
• bundled-dev: support worker file update accepted by HMR (#23068)
• bundled-dev: print build errors when an HMR update fails (#23024)
• bundled-dev: avoid stack overflow on import.meta.hot.invalidate() (#22797)
• bundled-dev: serve assets emitted during HMR / lazy compile (#22745)
• bundled-dev: skip plugin transform hooks for rolldown-lazy stub modules (#22778)
• bundled-dev: avoid duplicated buildEnd (#22931)

Plus from vite-plus itself:

• v0.2.9: vp dev no longer crashes at startup with ENOENT when experimental.bundledDev is enabled ([Feature]: Auto review for tool call safety #2384)
• v0.3.0: vite 8.2.2, rolldown 1.2.5, vitest 4.1.11, oxlint 1.79.0, oxfmt 0.64.0

What changed

Only the two catalog entries in pnpm-workspace.yaml (vite: npm:@voidzero-dev/vite-plus-core@0.3.0, vite-plus: 0.3.0). Every package consumes them via catalog:, so nothing else needed touching. No other 0.2.2 pin exists in package.json files, workflows, Dockerfiles, or tool-version files; CI uses voidzero-dev/setup-vp@v1 with no version pin, and the local vp reads the project version from the lockfile.

The lockfile delta is limited to the vite-plus toolchain: vite-plus, @voidzero-dev/vite-plus-core and platform binaries, vitest 4.1.9 to 4.1.11, oxlint 1.72.0 to 1.79.0, oxfmt 0.57.0 to 0.64.0, oxlint-tsgolint 0.24.0 to 7.0.2001, @oxc-project/* 0.138.0 to 0.146.0, and new transitive deps of vite-plus-core (lightningcss 1.33.0, yuku-parser / yuku-codegen 0.5.48, @oxlint/plugins 1.79.0). No unrelated major bumps.

experimental.bundledDev is still typed boolean in 8.2.2, so apps/web/vite.config.ts and scripts/dev-runner.ts need no change.

0.3.0 caveats

• XDG install layout for fresh installs only. Existing ~/.vite-plus installs keep their layout and vp upgrade leaves them in place. CI via setup-vp stays on the single-root layout, so the $HOME/.vite-plus/package_manager/pnpm/... paths in the mobile workflows keep working. A maintainer who does a fresh install will get ~/.local/share/vite-plus/bin instead; nothing in this repo hard-codes that path.
• @vitejs/devtools optional peer range widened to ^0.4.0 || ^0.5.0. We do not depend on it.
• Upgrade check moved out of process; commands no longer wait on the registry.
• Newer oxlint/oxfmt may flag new issues; the pre-commit vp fmt on the two changed files was clean.

Verification

• vp i refreshed the lockfile; vp --version inside apps/web reports vite-plus 0.3.0, vite 8.2.2, rolldown 1.2.5, vitest 4.1.11
• vp run build in apps/web: built in 12.7s, working tree clean afterward
• vp run typecheck in apps/web: exit 0
• vp test run --project unit src/lib/assistantCitationNavigation.test.ts src/lib/attachmentUploadState.test.ts in apps/web: 2 files, 8 tests passed on vitest 4.1.11
• T3CODE_BUNDLED_DEV=1 vp run dev in the worktree (no --share): after the first rolldown build the real index was served with <script type="module" src="/bundledDevClient.mjs"> and <script type="module" crossorigin src="/assets/index.js">; both returned 200 (text/javascript, 57 kB and 19.5 MB) with no error / ENOENT / preamble lines in the server log. Stopped by the process group of the PID captured at spawn.

Not verified here: the remote-browser --share path itself, which needs a second machine on the tailnet.

Model: Claude Fable 5 via Claude Code.

🤖 Generated with Claude Code

---

Note

Low Risk
Primarily a dev/build toolchain and auto-format refresh; application logic is unchanged aside from formatting, though a major vite/rolldown bump can still surface new build or lint issues in CI.

Overview
Bumps the workspace vite-plus / vite catalog pins from 0.2.2 to 0.3.0 and refreshes pnpm-lock.yaml, pulling in vite 8.2.2, rolldown 1.2.5, vitest 4.1.11, oxlint 1.79.0, oxfmt 0.64.0, and related native bindings (including new yuku-parser / yuku-codegen transitives).

Alongside the lockfile, many TypeScript files were reformatted only: multi-line .map / Effect.map / Effect.flatMap callbacks and similar patterns were collapsed to match oxfmt 0.64 (desktop, mobile, server, client-runtime, web, shared, generated schema.gen.ts). No intentional runtime or API behavior changes in those edits.

The upgrade targets more reliable experimental.bundledDev (used when shared dev defaults T3CODE_BUNDLED_DEV=1); app config in apps/web/vite.config.ts and scripts/dev-runner.ts is unchanged.

^{Reviewed by Cursor Bugbot for commit 21611ad. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Upgrade vite-plus to 0.3.0

Bumps the Vite alias and vite-plus catalog versions from 0.2.2 to 0.3.0 in pnpm-workspace.yaml and updates the lockfile. The rest of the PR is formatting-only reformatting across many provider, orchestration, and client modules triggered by the new formatter rules.

^{Macroscope summarized 21611ad.}

Follow-up commit: oxfmt 0.57 → 0.64 reformat

The toolchain bump also moves oxfmt, which now hugs a sole trailing arrow-function argument (Effect.map((x) => ({...})) on one line instead of breaking before the arrow) and breaks long string-literal unions onto leading | lines. vp check failed on 29 pre-existing files for that reason, so a second commit runs vp fmt over the repo. Verified per file that the change is whitespace, trailing commas, and | placement only (stripping those, old and new are byte-identical). Includes packages/effect-codex-app-server/src/_generated/schema.gen.ts, whose generator already runs vp fmt over its output, so it stays consistent from here.

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.2 KiB	13.4 KiB	+234 B (+1.7%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	−4 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.3 KiB	6.6 KiB	+238 B (+3.7%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.5 KiB	57.0 KiB	+1.5 KiB (+2.7%)	66.4 KiB	✅
Codex	Live turn messages	8	10	+2 (+25.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−48 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	+3 B (+0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−51 B (−0.8%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.3 KiB	−88 B (−0.2%)	66.4 KiB	✅
Claude	Live turn messages	10	8	−2 (−20.0%)	21	✅

Baseline: 3b2de9d · PR result: 21611ad · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This updates the shared Vite-plus build, bundling, testing, linting, and formatting toolchain across the monorepo, including the implementation of the existing bundled-development path. Although no application source changes are included, the upgrade can affect generated web/desktop/server artifacts and the unverified remote HMR path.

No code changes detected at 03de157. Prior analysis still applies.

^{You can add or adjust custom eligibility rules. Learn more.}

[juliusmarminge]

Problem

vp run dev --share defaults T3CODE_BUNDLED_DEV=1 (scripts/dev-runner.ts, apps/web/vite.config.tsexperimental.bundledDev) so a remote browser on the tailnet pays a few chunk requests instead of a per-module round trip. On vite-plus 0.2.2, which bundles vite 8.1.2 / rolldown 1.1.4, that bundled dev mode is flaky for the shared origin: the page throws @vitejs/plugin-react can't detect preamble and Cannot read properties of undefined (reading 'Root') in <Collapsible> (a base-ui namespace import resolving to undefined in the bundled chunk), so the app never gets past the splash. Unbundled dev renders fine.

Fix

Bump the pinned toolchain to vite-plus 0.3.0, which bundles vite 8.2.2 / rolldown 1.2.5. The 8.1.2 to 8.2.2 range carries the batch of experimental.bundledDev fixes (vitejs/vite CHANGELOG):

• bundled-dev: reload once after rebuild instead of via the fallback page (#23106)
• bundled-dev: support worker file update accepted by HMR (#23068)
• bundled-dev: print build errors when an HMR update fails (#23024)
• bundled-dev: avoid stack overflow on import.meta.hot.invalidate() (#22797)
• bundled-dev: serve assets emitted during HMR / lazy compile (#22745)
• bundled-dev: skip plugin transform hooks for rolldown-lazy stub modules (#22778)
• bundled-dev: avoid duplicated buildEnd (#22931)

Plus from vite-plus itself:

• v0.2.9: vp dev no longer crashes at startup with ENOENT when experimental.bundledDev is enabled ([Feature]: Auto review for tool call safety #2384)
• v0.3.0: vite 8.2.2, rolldown 1.2.5, vitest 4.1.11, oxlint 1.79.0, oxfmt 0.64.0

What changed

Only the two catalog entries in pnpm-workspace.yaml (vite: npm:@voidzero-dev/vite-plus-core@0.3.0, vite-plus: 0.3.0). Every package consumes them via catalog:, so nothing else needed touching. No other 0.2.2 pin exists in package.json files, workflows, Dockerfiles, or tool-version files; CI uses voidzero-dev/setup-vp@v1 with no version pin, and the local vp reads the project version from the lockfile.

The lockfile delta is limited to the vite-plus toolchain: vite-plus, @voidzero-dev/vite-plus-core and platform binaries, vitest 4.1.9 to 4.1.11, oxlint 1.72.0 to 1.79.0, oxfmt 0.57.0 to 0.64.0, oxlint-tsgolint 0.24.0 to 7.0.2001, @oxc-project/* 0.138.0 to 0.146.0, and new transitive deps of vite-plus-core (lightningcss 1.33.0, yuku-parser / yuku-codegen 0.5.48, @oxlint/plugins 1.79.0). No unrelated major bumps.

experimental.bundledDev is still typed boolean in 8.2.2, so apps/web/vite.config.ts and scripts/dev-runner.ts need no change.

0.3.0 caveats

• XDG install layout for fresh installs only. Existing ~/.vite-plus installs keep their layout and vp upgrade leaves them in place. CI via setup-vp stays on the single-root layout, so the $HOME/.vite-plus/package_manager/pnpm/... paths in the mobile workflows keep working. A maintainer who does a fresh install will get ~/.local/share/vite-plus/bin instead; nothing in this repo hard-codes that path.
• @vitejs/devtools optional peer range widened to ^0.4.0 || ^0.5.0. We do not depend on it.
• Upgrade check moved out of process; commands no longer wait on the registry.
• Newer oxlint/oxfmt may flag new issues; the pre-commit vp fmt on the two changed files was clean.

Verification

• vp i refreshed the lockfile; vp --version inside apps/web reports vite-plus 0.3.0, vite 8.2.2, rolldown 1.2.5, vitest 4.1.11
• vp run build in apps/web: built in 12.7s, working tree clean afterward
• vp run typecheck in apps/web: exit 0
• vp test run --project unit src/lib/assistantCitationNavigation.test.ts src/lib/attachmentUploadState.test.ts in apps/web: 2 files, 8 tests passed on vitest 4.1.11
• T3CODE_BUNDLED_DEV=1 vp run dev in the worktree (no --share): after the first rolldown build the real index was served with <script type="module" src="/bundledDevClient.mjs"> and <script type="module" crossorigin src="/assets/index.js">; both returned 200 (text/javascript, 57 kB and 19.5 MB) with no error / ENOENT / preamble lines in the server log. Stopped by the process group of the PID captured at spawn.

Not verified here: the remote-browser --share path itself, which needs a second machine on the tailnet.

Model: Claude Fable 5 via Claude Code.

🤖 Generated with Claude Code

---

Note

Low Risk
Primarily a dev/build toolchain and auto-format refresh; application logic is unchanged aside from formatting, though a major vite/rolldown bump can still surface new build or lint issues in CI.

Overview
Bumps the workspace vite-plus / vite catalog pins from 0.2.2 to 0.3.0 and refreshes pnpm-lock.yaml, pulling in vite 8.2.2, rolldown 1.2.5, vitest 4.1.11, oxlint 1.79.0, oxfmt 0.64.0, and related native bindings (including new yuku-parser / yuku-codegen transitives).

Alongside the lockfile, many TypeScript files were reformatted only: multi-line .map / Effect.map / Effect.flatMap callbacks and similar patterns were collapsed to match oxfmt 0.64 (desktop, mobile, server, client-runtime, web, shared, generated schema.gen.ts). No intentional runtime or API behavior changes in those edits.

The upgrade targets more reliable experimental.bundledDev (used when shared dev defaults T3CODE_BUNDLED_DEV=1); app config in apps/web/vite.config.ts and scripts/dev-runner.ts is unchanged.

^{Reviewed by Cursor Bugbot for commit 21611ad. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Upgrade vite-plus to 0.3.0

Bumps the Vite alias and vite-plus catalog versions from 0.2.2 to 0.3.0 in pnpm-workspace.yaml and updates the lockfile. The rest of the PR is formatting-only reformatting across many provider, orchestration, and client modules triggered by the new formatter rules.

^{Macroscope summarized 21611ad.}

Follow-up commit: oxfmt 0.57 → 0.64 reformat

The toolchain bump also moves oxfmt, which now hugs a sole trailing arrow-function argument (Effect.map((x) => ({...})) on one line instead of breaking before the arrow) and breaks long string-literal unions onto leading | lines. vp check failed on 29 pre-existing files for that reason, so a second commit runs vp fmt over the repo. Verified per file that the change is whitespace, trailing commas, and | placement only (stripping those, old and new are byte-identical). Includes packages/effect-codex-app-server/src/_generated/schema.gen.ts, whose generator already runs vp fmt over its output, so it stays consistent from here.

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.2 KiB	13.4 KiB	+234 B (+1.7%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	−4 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.3 KiB	6.6 KiB	+238 B (+3.7%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.5 KiB	57.0 KiB	+1.5 KiB (+2.7%)	66.4 KiB	✅
Codex	Live turn messages	8	10	+2 (+25.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−48 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	+3 B (+0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−51 B (−0.8%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.3 KiB	−88 B (−0.2%)	66.4 KiB	✅
Claude	Live turn messages	10	8	−2 (−20.0%)	21	✅

Baseline: 3b2de9d · PR result: 21611ad · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This updates the shared Vite-plus build, bundling, testing, linting, and formatting toolchain across the monorepo, including the implementation of the existing bundled-development path. Although no application source changes are included, the upgrade can affect generated web/desktop/server artifacts and the unverified remote HMR path.

No code changes detected at 03de157. Prior analysis still applies.

^{You can add or adjust custom eligibility rules. Learn more.}

[juliusmarminge]

Problem

vp run dev --share defaults T3CODE_BUNDLED_DEV=1 (scripts/dev-runner.ts, apps/web/vite.config.tsexperimental.bundledDev) so a remote browser on the tailnet pays a few chunk requests instead of a per-module round trip. On vite-plus 0.2.2, which bundles vite 8.1.2 / rolldown 1.1.4, that bundled dev mode is flaky for the shared origin: the page throws @vitejs/plugin-react can't detect preamble and Cannot read properties of undefined (reading 'Root') in <Collapsible> (a base-ui namespace import resolving to undefined in the bundled chunk), so the app never gets past the splash. Unbundled dev renders fine.

Fix

Bump the pinned toolchain to vite-plus 0.3.0, which bundles vite 8.2.2 / rolldown 1.2.5. The 8.1.2 to 8.2.2 range carries the batch of experimental.bundledDev fixes (vitejs/vite CHANGELOG):

• bundled-dev: reload once after rebuild instead of via the fallback page (#23106)
• bundled-dev: support worker file update accepted by HMR (#23068)
• bundled-dev: print build errors when an HMR update fails (#23024)
• bundled-dev: avoid stack overflow on import.meta.hot.invalidate() (#22797)
• bundled-dev: serve assets emitted during HMR / lazy compile (#22745)
• bundled-dev: skip plugin transform hooks for rolldown-lazy stub modules (#22778)
• bundled-dev: avoid duplicated buildEnd (#22931)

Plus from vite-plus itself:

• v0.2.9: vp dev no longer crashes at startup with ENOENT when experimental.bundledDev is enabled ([Feature]: Auto review for tool call safety #2384)
• v0.3.0: vite 8.2.2, rolldown 1.2.5, vitest 4.1.11, oxlint 1.79.0, oxfmt 0.64.0

What changed

Only the two catalog entries in pnpm-workspace.yaml (vite: npm:@voidzero-dev/vite-plus-core@0.3.0, vite-plus: 0.3.0). Every package consumes them via catalog:, so nothing else needed touching. No other 0.2.2 pin exists in package.json files, workflows, Dockerfiles, or tool-version files; CI uses voidzero-dev/setup-vp@v1 with no version pin, and the local vp reads the project version from the lockfile.

The lockfile delta is limited to the vite-plus toolchain: vite-plus, @voidzero-dev/vite-plus-core and platform binaries, vitest 4.1.9 to 4.1.11, oxlint 1.72.0 to 1.79.0, oxfmt 0.57.0 to 0.64.0, oxlint-tsgolint 0.24.0 to 7.0.2001, @oxc-project/* 0.138.0 to 0.146.0, and new transitive deps of vite-plus-core (lightningcss 1.33.0, yuku-parser / yuku-codegen 0.5.48, @oxlint/plugins 1.79.0). No unrelated major bumps.

experimental.bundledDev is still typed boolean in 8.2.2, so apps/web/vite.config.ts and scripts/dev-runner.ts need no change.

0.3.0 caveats

• XDG install layout for fresh installs only. Existing ~/.vite-plus installs keep their layout and vp upgrade leaves them in place. CI via setup-vp stays on the single-root layout, so the $HOME/.vite-plus/package_manager/pnpm/... paths in the mobile workflows keep working. A maintainer who does a fresh install will get ~/.local/share/vite-plus/bin instead; nothing in this repo hard-codes that path.
• @vitejs/devtools optional peer range widened to ^0.4.0 || ^0.5.0. We do not depend on it.
• Upgrade check moved out of process; commands no longer wait on the registry.
• Newer oxlint/oxfmt may flag new issues; the pre-commit vp fmt on the two changed files was clean.

Verification

• vp i refreshed the lockfile; vp --version inside apps/web reports vite-plus 0.3.0, vite 8.2.2, rolldown 1.2.5, vitest 4.1.11
• vp run build in apps/web: built in 12.7s, working tree clean afterward
• vp run typecheck in apps/web: exit 0
• vp test run --project unit src/lib/assistantCitationNavigation.test.ts src/lib/attachmentUploadState.test.ts in apps/web: 2 files, 8 tests passed on vitest 4.1.11
• T3CODE_BUNDLED_DEV=1 vp run dev in the worktree (no --share): after the first rolldown build the real index was served with <script type="module" src="/bundledDevClient.mjs"> and <script type="module" crossorigin src="/assets/index.js">; both returned 200 (text/javascript, 57 kB and 19.5 MB) with no error / ENOENT / preamble lines in the server log. Stopped by the process group of the PID captured at spawn.

Not verified here: the remote-browser --share path itself, which needs a second machine on the tailnet.

Model: Claude Fable 5 via Claude Code.

🤖 Generated with Claude Code

---

Note

Low Risk
Primarily a dev/build toolchain and auto-format refresh; application logic is unchanged aside from formatting, though a major vite/rolldown bump can still surface new build or lint issues in CI.

Overview
Bumps the workspace vite-plus / vite catalog pins from 0.2.2 to 0.3.0 and refreshes pnpm-lock.yaml, pulling in vite 8.2.2, rolldown 1.2.5, vitest 4.1.11, oxlint 1.79.0, oxfmt 0.64.0, and related native bindings (including new yuku-parser / yuku-codegen transitives).

Alongside the lockfile, many TypeScript files were reformatted only: multi-line .map / Effect.map / Effect.flatMap callbacks and similar patterns were collapsed to match oxfmt 0.64 (desktop, mobile, server, client-runtime, web, shared, generated schema.gen.ts). No intentional runtime or API behavior changes in those edits.

The upgrade targets more reliable experimental.bundledDev (used when shared dev defaults T3CODE_BUNDLED_DEV=1); app config in apps/web/vite.config.ts and scripts/dev-runner.ts is unchanged.

^{Reviewed by Cursor Bugbot for commit 21611ad. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Upgrade vite-plus to 0.3.0

Bumps the Vite alias and vite-plus catalog versions from 0.2.2 to 0.3.0 in pnpm-workspace.yaml and updates the lockfile. The rest of the PR is formatting-only reformatting across many provider, orchestration, and client modules triggered by the new formatter rules.

^{Macroscope summarized 21611ad.}

Follow-up commit: oxfmt 0.57 → 0.64 reformat

The toolchain bump also moves oxfmt, which now hugs a sole trailing arrow-function argument (Effect.map((x) => ({...})) on one line instead of breaking before the arrow) and breaks long string-literal unions onto leading | lines. vp check failed on 29 pre-existing files for that reason, so a second commit runs vp fmt over the repo. Verified per file that the change is whitespace, trailing commas, and | placement only (stripping those, old and new are byte-identical). Includes packages/effect-codex-app-server/src/_generated/schema.gen.ts, whose generator already runs vp fmt over its output, so it stays consistent from here.

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.2 KiB	13.4 KiB	+234 B (+1.7%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	−4 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.3 KiB	6.6 KiB	+238 B (+3.7%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.5 KiB	57.0 KiB	+1.5 KiB (+2.7%)	66.4 KiB	✅
Codex	Live turn messages	8	10	+2 (+25.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−48 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	+3 B (+0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−51 B (−0.8%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.3 KiB	−88 B (−0.2%)	66.4 KiB	✅
Claude	Live turn messages	10	8	−2 (−20.0%)	21	✅

Baseline: 3b2de9d · PR result: 21611ad · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This updates the shared Vite-plus build, bundling, testing, linting, and formatting toolchain across the monorepo, including the implementation of the existing bundled-development path. Although no application source changes are included, the upgrade can affect generated web/desktop/server artifacts and the unverified remote HMR path.

No code changes detected at 03de157. Prior analysis still applies.

^{You can add or adjust custom eligibility rules. Learn more.}

[juliusmarminge]

Problem

vp run dev --share defaults T3CODE_BUNDLED_DEV=1 (scripts/dev-runner.ts, apps/web/vite.config.tsexperimental.bundledDev) so a remote browser on the tailnet pays a few chunk requests instead of a per-module round trip. On vite-plus 0.2.2, which bundles vite 8.1.2 / rolldown 1.1.4, that bundled dev mode is flaky for the shared origin: the page throws @vitejs/plugin-react can't detect preamble and Cannot read properties of undefined (reading 'Root') in <Collapsible> (a base-ui namespace import resolving to undefined in the bundled chunk), so the app never gets past the splash. Unbundled dev renders fine.

Fix

Bump the pinned toolchain to vite-plus 0.3.0, which bundles vite 8.2.2 / rolldown 1.2.5. The 8.1.2 to 8.2.2 range carries the batch of experimental.bundledDev fixes (vitejs/vite CHANGELOG):

• bundled-dev: reload once after rebuild instead of via the fallback page (#23106)
• bundled-dev: support worker file update accepted by HMR (#23068)
• bundled-dev: print build errors when an HMR update fails (#23024)
• bundled-dev: avoid stack overflow on import.meta.hot.invalidate() (#22797)
• bundled-dev: serve assets emitted during HMR / lazy compile (#22745)
• bundled-dev: skip plugin transform hooks for rolldown-lazy stub modules (#22778)
• bundled-dev: avoid duplicated buildEnd (#22931)

Plus from vite-plus itself:

• v0.2.9: vp dev no longer crashes at startup with ENOENT when experimental.bundledDev is enabled ([Feature]: Auto review for tool call safety #2384)
• v0.3.0: vite 8.2.2, rolldown 1.2.5, vitest 4.1.11, oxlint 1.79.0, oxfmt 0.64.0

What changed

Only the two catalog entries in pnpm-workspace.yaml (vite: npm:@voidzero-dev/vite-plus-core@0.3.0, vite-plus: 0.3.0). Every package consumes them via catalog:, so nothing else needed touching. No other 0.2.2 pin exists in package.json files, workflows, Dockerfiles, or tool-version files; CI uses voidzero-dev/setup-vp@v1 with no version pin, and the local vp reads the project version from the lockfile.

The lockfile delta is limited to the vite-plus toolchain: vite-plus, @voidzero-dev/vite-plus-core and platform binaries, vitest 4.1.9 to 4.1.11, oxlint 1.72.0 to 1.79.0, oxfmt 0.57.0 to 0.64.0, oxlint-tsgolint 0.24.0 to 7.0.2001, @oxc-project/* 0.138.0 to 0.146.0, and new transitive deps of vite-plus-core (lightningcss 1.33.0, yuku-parser / yuku-codegen 0.5.48, @oxlint/plugins 1.79.0). No unrelated major bumps.

experimental.bundledDev is still typed boolean in 8.2.2, so apps/web/vite.config.ts and scripts/dev-runner.ts need no change.

0.3.0 caveats

• XDG install layout for fresh installs only. Existing ~/.vite-plus installs keep their layout and vp upgrade leaves them in place. CI via setup-vp stays on the single-root layout, so the $HOME/.vite-plus/package_manager/pnpm/... paths in the mobile workflows keep working. A maintainer who does a fresh install will get ~/.local/share/vite-plus/bin instead; nothing in this repo hard-codes that path.
• @vitejs/devtools optional peer range widened to ^0.4.0 || ^0.5.0. We do not depend on it.
• Upgrade check moved out of process; commands no longer wait on the registry.
• Newer oxlint/oxfmt may flag new issues; the pre-commit vp fmt on the two changed files was clean.

Verification

• vp i refreshed the lockfile; vp --version inside apps/web reports vite-plus 0.3.0, vite 8.2.2, rolldown 1.2.5, vitest 4.1.11
• vp run build in apps/web: built in 12.7s, working tree clean afterward
• vp run typecheck in apps/web: exit 0
• vp test run --project unit src/lib/assistantCitationNavigation.test.ts src/lib/attachmentUploadState.test.ts in apps/web: 2 files, 8 tests passed on vitest 4.1.11
• T3CODE_BUNDLED_DEV=1 vp run dev in the worktree (no --share): after the first rolldown build the real index was served with <script type="module" src="/bundledDevClient.mjs"> and <script type="module" crossorigin src="/assets/index.js">; both returned 200 (text/javascript, 57 kB and 19.5 MB) with no error / ENOENT / preamble lines in the server log. Stopped by the process group of the PID captured at spawn.

Not verified here: the remote-browser --share path itself, which needs a second machine on the tailnet.

Model: Claude Fable 5 via Claude Code.

🤖 Generated with Claude Code

---

Note

Low Risk
Primarily a dev/build toolchain and auto-format refresh; application logic is unchanged aside from formatting, though a major vite/rolldown bump can still surface new build or lint issues in CI.

Overview
Bumps the workspace vite-plus / vite catalog pins from 0.2.2 to 0.3.0 and refreshes pnpm-lock.yaml, pulling in vite 8.2.2, rolldown 1.2.5, vitest 4.1.11, oxlint 1.79.0, oxfmt 0.64.0, and related native bindings (including new yuku-parser / yuku-codegen transitives).

Alongside the lockfile, many TypeScript files were reformatted only: multi-line .map / Effect.map / Effect.flatMap callbacks and similar patterns were collapsed to match oxfmt 0.64 (desktop, mobile, server, client-runtime, web, shared, generated schema.gen.ts). No intentional runtime or API behavior changes in those edits.

The upgrade targets more reliable experimental.bundledDev (used when shared dev defaults T3CODE_BUNDLED_DEV=1); app config in apps/web/vite.config.ts and scripts/dev-runner.ts is unchanged.

^{Reviewed by Cursor Bugbot for commit 21611ad. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Upgrade vite-plus to 0.3.0

Bumps the Vite alias and vite-plus catalog versions from 0.2.2 to 0.3.0 in pnpm-workspace.yaml and updates the lockfile. The rest of the PR is formatting-only reformatting across many provider, orchestration, and client modules triggered by the new formatter rules.

^{Macroscope summarized 21611ad.}

Follow-up commit: oxfmt 0.57 → 0.64 reformat

The toolchain bump also moves oxfmt, which now hugs a sole trailing arrow-function argument (Effect.map((x) => ({...})) on one line instead of breaking before the arrow) and breaks long string-literal unions onto leading | lines. vp check failed on 29 pre-existing files for that reason, so a second commit runs vp fmt over the repo. Verified per file that the change is whitespace, trailing commas, and | placement only (stripping those, old and new are byte-identical). Includes packages/effect-codex-app-server/src/_generated/schema.gen.ts, whose generator already runs vp fmt over its output, so it stays consistent from here.

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.2 KiB	13.4 KiB	+234 B (+1.7%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	−4 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.3 KiB	6.6 KiB	+238 B (+3.7%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.5 KiB	57.0 KiB	+1.5 KiB (+2.7%)	66.4 KiB	✅
Codex	Live turn messages	8	10	+2 (+25.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−48 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	+3 B (+0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−51 B (−0.8%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.3 KiB	−88 B (−0.2%)	66.4 KiB	✅
Claude	Live turn messages	10	8	−2 (−20.0%)	21	✅

Baseline: 3b2de9d · PR result: 21611ad · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This updates the shared Vite-plus build, bundling, testing, linting, and formatting toolchain across the monorepo, including the implementation of the existing bundled-development path. Although no application source changes are included, the upgrade can affect generated web/desktop/server artifacts and the unverified remote HMR path.

No code changes detected at 03de157. Prior analysis still applies.

^{You can add or adjust custom eligibility rules. Learn more.}

[juliusmarminge]

Problem

vp run dev --share defaults T3CODE_BUNDLED_DEV=1 (scripts/dev-runner.ts, apps/web/vite.config.tsexperimental.bundledDev) so a remote browser on the tailnet pays a few chunk requests instead of a per-module round trip. On vite-plus 0.2.2, which bundles vite 8.1.2 / rolldown 1.1.4, that bundled dev mode is flaky for the shared origin: the page throws @vitejs/plugin-react can't detect preamble and Cannot read properties of undefined (reading 'Root') in <Collapsible> (a base-ui namespace import resolving to undefined in the bundled chunk), so the app never gets past the splash. Unbundled dev renders fine.

Fix

Bump the pinned toolchain to vite-plus 0.3.0, which bundles vite 8.2.2 / rolldown 1.2.5. The 8.1.2 to 8.2.2 range carries the batch of experimental.bundledDev fixes (vitejs/vite CHANGELOG):

• bundled-dev: reload once after rebuild instead of via the fallback page (#23106)
• bundled-dev: support worker file update accepted by HMR (#23068)
• bundled-dev: print build errors when an HMR update fails (#23024)
• bundled-dev: avoid stack overflow on import.meta.hot.invalidate() (#22797)
• bundled-dev: serve assets emitted during HMR / lazy compile (#22745)
• bundled-dev: skip plugin transform hooks for rolldown-lazy stub modules (#22778)
• bundled-dev: avoid duplicated buildEnd (#22931)

Plus from vite-plus itself:

• v0.2.9: vp dev no longer crashes at startup with ENOENT when experimental.bundledDev is enabled ([Feature]: Auto review for tool call safety #2384)
• v0.3.0: vite 8.2.2, rolldown 1.2.5, vitest 4.1.11, oxlint 1.79.0, oxfmt 0.64.0

What changed

Only the two catalog entries in pnpm-workspace.yaml (vite: npm:@voidzero-dev/vite-plus-core@0.3.0, vite-plus: 0.3.0). Every package consumes them via catalog:, so nothing else needed touching. No other 0.2.2 pin exists in package.json files, workflows, Dockerfiles, or tool-version files; CI uses voidzero-dev/setup-vp@v1 with no version pin, and the local vp reads the project version from the lockfile.

The lockfile delta is limited to the vite-plus toolchain: vite-plus, @voidzero-dev/vite-plus-core and platform binaries, vitest 4.1.9 to 4.1.11, oxlint 1.72.0 to 1.79.0, oxfmt 0.57.0 to 0.64.0, oxlint-tsgolint 0.24.0 to 7.0.2001, @oxc-project/* 0.138.0 to 0.146.0, and new transitive deps of vite-plus-core (lightningcss 1.33.0, yuku-parser / yuku-codegen 0.5.48, @oxlint/plugins 1.79.0). No unrelated major bumps.

experimental.bundledDev is still typed boolean in 8.2.2, so apps/web/vite.config.ts and scripts/dev-runner.ts need no change.

0.3.0 caveats

• XDG install layout for fresh installs only. Existing ~/.vite-plus installs keep their layout and vp upgrade leaves them in place. CI via setup-vp stays on the single-root layout, so the $HOME/.vite-plus/package_manager/pnpm/... paths in the mobile workflows keep working. A maintainer who does a fresh install will get ~/.local/share/vite-plus/bin instead; nothing in this repo hard-codes that path.
• @vitejs/devtools optional peer range widened to ^0.4.0 || ^0.5.0. We do not depend on it.
• Upgrade check moved out of process; commands no longer wait on the registry.
• Newer oxlint/oxfmt may flag new issues; the pre-commit vp fmt on the two changed files was clean.

Verification

• vp i refreshed the lockfile; vp --version inside apps/web reports vite-plus 0.3.0, vite 8.2.2, rolldown 1.2.5, vitest 4.1.11
• vp run build in apps/web: built in 12.7s, working tree clean afterward
• vp run typecheck in apps/web: exit 0
• vp test run --project unit src/lib/assistantCitationNavigation.test.ts src/lib/attachmentUploadState.test.ts in apps/web: 2 files, 8 tests passed on vitest 4.1.11
• T3CODE_BUNDLED_DEV=1 vp run dev in the worktree (no --share): after the first rolldown build the real index was served with <script type="module" src="/bundledDevClient.mjs"> and <script type="module" crossorigin src="/assets/index.js">; both returned 200 (text/javascript, 57 kB and 19.5 MB) with no error / ENOENT / preamble lines in the server log. Stopped by the process group of the PID captured at spawn.

Not verified here: the remote-browser --share path itself, which needs a second machine on the tailnet.

Model: Claude Fable 5 via Claude Code.

🤖 Generated with Claude Code

---

Note

Low Risk
Primarily a dev/build toolchain and auto-format refresh; application logic is unchanged aside from formatting, though a major vite/rolldown bump can still surface new build or lint issues in CI.

Overview
Bumps the workspace vite-plus / vite catalog pins from 0.2.2 to 0.3.0 and refreshes pnpm-lock.yaml, pulling in vite 8.2.2, rolldown 1.2.5, vitest 4.1.11, oxlint 1.79.0, oxfmt 0.64.0, and related native bindings (including new yuku-parser / yuku-codegen transitives).

Alongside the lockfile, many TypeScript files were reformatted only: multi-line .map / Effect.map / Effect.flatMap callbacks and similar patterns were collapsed to match oxfmt 0.64 (desktop, mobile, server, client-runtime, web, shared, generated schema.gen.ts). No intentional runtime or API behavior changes in those edits.

The upgrade targets more reliable experimental.bundledDev (used when shared dev defaults T3CODE_BUNDLED_DEV=1); app config in apps/web/vite.config.ts and scripts/dev-runner.ts is unchanged.

^{Reviewed by Cursor Bugbot for commit 21611ad. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Upgrade vite-plus to 0.3.0

Bumps the Vite alias and vite-plus catalog versions from 0.2.2 to 0.3.0 in pnpm-workspace.yaml and updates the lockfile. The rest of the PR is formatting-only reformatting across many provider, orchestration, and client modules triggered by the new formatter rules.

^{Macroscope summarized 21611ad.}

Follow-up commit: oxfmt 0.57 → 0.64 reformat

The toolchain bump also moves oxfmt, which now hugs a sole trailing arrow-function argument (Effect.map((x) => ({...})) on one line instead of breaking before the arrow) and breaks long string-literal unions onto leading | lines. vp check failed on 29 pre-existing files for that reason, so a second commit runs vp fmt over the repo. Verified per file that the change is whitespace, trailing commas, and | placement only (stripping those, old and new are byte-identical). Includes packages/effect-codex-app-server/src/_generated/schema.gen.ts, whose generator already runs vp fmt over its output, so it stays consistent from here.

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.2 KiB	13.4 KiB	+234 B (+1.7%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	−4 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.3 KiB	6.6 KiB	+238 B (+3.7%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.5 KiB	57.0 KiB	+1.5 KiB (+2.7%)	66.4 KiB	✅
Codex	Live turn messages	8	10	+2 (+25.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−48 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	+3 B (+0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−51 B (−0.8%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.3 KiB	−88 B (−0.2%)	66.4 KiB	✅
Claude	Live turn messages	10	8	−2 (−20.0%)	21	✅

Baseline: 3b2de9d · PR result: 21611ad · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This updates the shared Vite-plus build, bundling, testing, linting, and formatting toolchain across the monorepo, including the implementation of the existing bundled-development path. Although no application source changes are included, the upgrade can affect generated web/desktop/server artifacts and the unverified remote HMR path.

No code changes detected at 03de157. Prior analysis still applies.

^{You can add or adjust custom eligibility rules. Learn more.}

[juliusmarminge]

Problem

vp run dev --share defaults T3CODE_BUNDLED_DEV=1 (scripts/dev-runner.ts, apps/web/vite.config.tsexperimental.bundledDev) so a remote browser on the tailnet pays a few chunk requests instead of a per-module round trip. On vite-plus 0.2.2, which bundles vite 8.1.2 / rolldown 1.1.4, that bundled dev mode is flaky for the shared origin: the page throws @vitejs/plugin-react can't detect preamble and Cannot read properties of undefined (reading 'Root') in <Collapsible> (a base-ui namespace import resolving to undefined in the bundled chunk), so the app never gets past the splash. Unbundled dev renders fine.

Fix

Bump the pinned toolchain to vite-plus 0.3.0, which bundles vite 8.2.2 / rolldown 1.2.5. The 8.1.2 to 8.2.2 range carries the batch of experimental.bundledDev fixes (vitejs/vite CHANGELOG):

• bundled-dev: reload once after rebuild instead of via the fallback page (#23106)
• bundled-dev: support worker file update accepted by HMR (#23068)
• bundled-dev: print build errors when an HMR update fails (#23024)
• bundled-dev: avoid stack overflow on import.meta.hot.invalidate() (#22797)
• bundled-dev: serve assets emitted during HMR / lazy compile (#22745)
• bundled-dev: skip plugin transform hooks for rolldown-lazy stub modules (#22778)
• bundled-dev: avoid duplicated buildEnd (#22931)

Plus from vite-plus itself:

• v0.2.9: vp dev no longer crashes at startup with ENOENT when experimental.bundledDev is enabled ([Feature]: Auto review for tool call safety #2384)
• v0.3.0: vite 8.2.2, rolldown 1.2.5, vitest 4.1.11, oxlint 1.79.0, oxfmt 0.64.0

What changed

Only the two catalog entries in pnpm-workspace.yaml (vite: npm:@voidzero-dev/vite-plus-core@0.3.0, vite-plus: 0.3.0). Every package consumes them via catalog:, so nothing else needed touching. No other 0.2.2 pin exists in package.json files, workflows, Dockerfiles, or tool-version files; CI uses voidzero-dev/setup-vp@v1 with no version pin, and the local vp reads the project version from the lockfile.

The lockfile delta is limited to the vite-plus toolchain: vite-plus, @voidzero-dev/vite-plus-core and platform binaries, vitest 4.1.9 to 4.1.11, oxlint 1.72.0 to 1.79.0, oxfmt 0.57.0 to 0.64.0, oxlint-tsgolint 0.24.0 to 7.0.2001, @oxc-project/* 0.138.0 to 0.146.0, and new transitive deps of vite-plus-core (lightningcss 1.33.0, yuku-parser / yuku-codegen 0.5.48, @oxlint/plugins 1.79.0). No unrelated major bumps.

experimental.bundledDev is still typed boolean in 8.2.2, so apps/web/vite.config.ts and scripts/dev-runner.ts need no change.

0.3.0 caveats

• XDG install layout for fresh installs only. Existing ~/.vite-plus installs keep their layout and vp upgrade leaves them in place. CI via setup-vp stays on the single-root layout, so the $HOME/.vite-plus/package_manager/pnpm/... paths in the mobile workflows keep working. A maintainer who does a fresh install will get ~/.local/share/vite-plus/bin instead; nothing in this repo hard-codes that path.
• @vitejs/devtools optional peer range widened to ^0.4.0 || ^0.5.0. We do not depend on it.
• Upgrade check moved out of process; commands no longer wait on the registry.
• Newer oxlint/oxfmt may flag new issues; the pre-commit vp fmt on the two changed files was clean.

Verification

• vp i refreshed the lockfile; vp --version inside apps/web reports vite-plus 0.3.0, vite 8.2.2, rolldown 1.2.5, vitest 4.1.11
• vp run build in apps/web: built in 12.7s, working tree clean afterward
• vp run typecheck in apps/web: exit 0
• vp test run --project unit src/lib/assistantCitationNavigation.test.ts src/lib/attachmentUploadState.test.ts in apps/web: 2 files, 8 tests passed on vitest 4.1.11
• T3CODE_BUNDLED_DEV=1 vp run dev in the worktree (no --share): after the first rolldown build the real index was served with <script type="module" src="/bundledDevClient.mjs"> and <script type="module" crossorigin src="/assets/index.js">; both returned 200 (text/javascript, 57 kB and 19.5 MB) with no error / ENOENT / preamble lines in the server log. Stopped by the process group of the PID captured at spawn.

Not verified here: the remote-browser --share path itself, which needs a second machine on the tailnet.

Model: Claude Fable 5 via Claude Code.

🤖 Generated with Claude Code

---

Note

Low Risk
Primarily a dev/build toolchain and auto-format refresh; application logic is unchanged aside from formatting, though a major vite/rolldown bump can still surface new build or lint issues in CI.

Overview
Bumps the workspace vite-plus / vite catalog pins from 0.2.2 to 0.3.0 and refreshes pnpm-lock.yaml, pulling in vite 8.2.2, rolldown 1.2.5, vitest 4.1.11, oxlint 1.79.0, oxfmt 0.64.0, and related native bindings (including new yuku-parser / yuku-codegen transitives).

Alongside the lockfile, many TypeScript files were reformatted only: multi-line .map / Effect.map / Effect.flatMap callbacks and similar patterns were collapsed to match oxfmt 0.64 (desktop, mobile, server, client-runtime, web, shared, generated schema.gen.ts). No intentional runtime or API behavior changes in those edits.

The upgrade targets more reliable experimental.bundledDev (used when shared dev defaults T3CODE_BUNDLED_DEV=1); app config in apps/web/vite.config.ts and scripts/dev-runner.ts is unchanged.

^{Reviewed by Cursor Bugbot for commit 21611ad. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Upgrade vite-plus to 0.3.0

Bumps the Vite alias and vite-plus catalog versions from 0.2.2 to 0.3.0 in pnpm-workspace.yaml and updates the lockfile. The rest of the PR is formatting-only reformatting across many provider, orchestration, and client modules triggered by the new formatter rules.

^{Macroscope summarized 21611ad.}

Follow-up commit: oxfmt 0.57 → 0.64 reformat

The toolchain bump also moves oxfmt, which now hugs a sole trailing arrow-function argument (Effect.map((x) => ({...})) on one line instead of breaking before the arrow) and breaks long string-literal unions onto leading | lines. vp check failed on 29 pre-existing files for that reason, so a second commit runs vp fmt over the repo. Verified per file that the change is whitespace, trailing commas, and | placement only (stripping those, old and new are byte-identical). Includes packages/effect-codex-app-server/src/_generated/schema.gen.ts, whose generator already runs vp fmt over its output, so it stays consistent from here.

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.2 KiB	13.4 KiB	+234 B (+1.7%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	−4 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.3 KiB	6.6 KiB	+238 B (+3.7%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.5 KiB	57.0 KiB	+1.5 KiB (+2.7%)	66.4 KiB	✅
Codex	Live turn messages	8	10	+2 (+25.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−48 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	+3 B (+0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−51 B (−0.8%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.3 KiB	−88 B (−0.2%)	66.4 KiB	✅
Claude	Live turn messages	10	8	−2 (−20.0%)	21	✅

Baseline: 3b2de9d · PR result: 21611ad · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This updates the shared Vite-plus build, bundling, testing, linting, and formatting toolchain across the monorepo, including the implementation of the existing bundled-development path. Although no application source changes are included, the upgrade can affect generated web/desktop/server artifacts and the unverified remote HMR path.

No code changes detected at 03de157. Prior analysis still applies.

^{You can add or adjust custom eligibility rules. Learn more.}

[juliusmarminge]

Problem

vp run dev --share defaults T3CODE_BUNDLED_DEV=1 (scripts/dev-runner.ts, apps/web/vite.config.tsexperimental.bundledDev) so a remote browser on the tailnet pays a few chunk requests instead of a per-module round trip. On vite-plus 0.2.2, which bundles vite 8.1.2 / rolldown 1.1.4, that bundled dev mode is flaky for the shared origin: the page throws @vitejs/plugin-react can't detect preamble and Cannot read properties of undefined (reading 'Root') in <Collapsible> (a base-ui namespace import resolving to undefined in the bundled chunk), so the app never gets past the splash. Unbundled dev renders fine.

Fix

Bump the pinned toolchain to vite-plus 0.3.0, which bundles vite 8.2.2 / rolldown 1.2.5. The 8.1.2 to 8.2.2 range carries the batch of experimental.bundledDev fixes (vitejs/vite CHANGELOG):

• bundled-dev: reload once after rebuild instead of via the fallback page (#23106)
• bundled-dev: support worker file update accepted by HMR (#23068)
• bundled-dev: print build errors when an HMR update fails (#23024)
• bundled-dev: avoid stack overflow on import.meta.hot.invalidate() (#22797)
• bundled-dev: serve assets emitted during HMR / lazy compile (#22745)
• bundled-dev: skip plugin transform hooks for rolldown-lazy stub modules (#22778)
• bundled-dev: avoid duplicated buildEnd (#22931)

Plus from vite-plus itself:

• v0.2.9: vp dev no longer crashes at startup with ENOENT when experimental.bundledDev is enabled ([Feature]: Auto review for tool call safety #2384)
• v0.3.0: vite 8.2.2, rolldown 1.2.5, vitest 4.1.11, oxlint 1.79.0, oxfmt 0.64.0

What changed

Only the two catalog entries in pnpm-workspace.yaml (vite: npm:@voidzero-dev/vite-plus-core@0.3.0, vite-plus: 0.3.0). Every package consumes them via catalog:, so nothing else needed touching. No other 0.2.2 pin exists in package.json files, workflows, Dockerfiles, or tool-version files; CI uses voidzero-dev/setup-vp@v1 with no version pin, and the local vp reads the project version from the lockfile.

The lockfile delta is limited to the vite-plus toolchain: vite-plus, @voidzero-dev/vite-plus-core and platform binaries, vitest 4.1.9 to 4.1.11, oxlint 1.72.0 to 1.79.0, oxfmt 0.57.0 to 0.64.0, oxlint-tsgolint 0.24.0 to 7.0.2001, @oxc-project/* 0.138.0 to 0.146.0, and new transitive deps of vite-plus-core (lightningcss 1.33.0, yuku-parser / yuku-codegen 0.5.48, @oxlint/plugins 1.79.0). No unrelated major bumps.

experimental.bundledDev is still typed boolean in 8.2.2, so apps/web/vite.config.ts and scripts/dev-runner.ts need no change.

0.3.0 caveats

• XDG install layout for fresh installs only. Existing ~/.vite-plus installs keep their layout and vp upgrade leaves them in place. CI via setup-vp stays on the single-root layout, so the $HOME/.vite-plus/package_manager/pnpm/... paths in the mobile workflows keep working. A maintainer who does a fresh install will get ~/.local/share/vite-plus/bin instead; nothing in this repo hard-codes that path.
• @vitejs/devtools optional peer range widened to ^0.4.0 || ^0.5.0. We do not depend on it.
• Upgrade check moved out of process; commands no longer wait on the registry.
• Newer oxlint/oxfmt may flag new issues; the pre-commit vp fmt on the two changed files was clean.

Verification

• vp i refreshed the lockfile; vp --version inside apps/web reports vite-plus 0.3.0, vite 8.2.2, rolldown 1.2.5, vitest 4.1.11
• vp run build in apps/web: built in 12.7s, working tree clean afterward
• vp run typecheck in apps/web: exit 0
• vp test run --project unit src/lib/assistantCitationNavigation.test.ts src/lib/attachmentUploadState.test.ts in apps/web: 2 files, 8 tests passed on vitest 4.1.11
• T3CODE_BUNDLED_DEV=1 vp run dev in the worktree (no --share): after the first rolldown build the real index was served with <script type="module" src="/bundledDevClient.mjs"> and <script type="module" crossorigin src="/assets/index.js">; both returned 200 (text/javascript, 57 kB and 19.5 MB) with no error / ENOENT / preamble lines in the server log. Stopped by the process group of the PID captured at spawn.

Not verified here: the remote-browser --share path itself, which needs a second machine on the tailnet.

Model: Claude Fable 5 via Claude Code.

🤖 Generated with Claude Code

---

Note

Low Risk
Primarily a dev/build toolchain and auto-format refresh; application logic is unchanged aside from formatting, though a major vite/rolldown bump can still surface new build or lint issues in CI.

Overview
Bumps the workspace vite-plus / vite catalog pins from 0.2.2 to 0.3.0 and refreshes pnpm-lock.yaml, pulling in vite 8.2.2, rolldown 1.2.5, vitest 4.1.11, oxlint 1.79.0, oxfmt 0.64.0, and related native bindings (including new yuku-parser / yuku-codegen transitives).

Alongside the lockfile, many TypeScript files were reformatted only: multi-line .map / Effect.map / Effect.flatMap callbacks and similar patterns were collapsed to match oxfmt 0.64 (desktop, mobile, server, client-runtime, web, shared, generated schema.gen.ts). No intentional runtime or API behavior changes in those edits.

The upgrade targets more reliable experimental.bundledDev (used when shared dev defaults T3CODE_BUNDLED_DEV=1); app config in apps/web/vite.config.ts and scripts/dev-runner.ts is unchanged.

^{Reviewed by Cursor Bugbot for commit 21611ad. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Upgrade vite-plus to 0.3.0

Bumps the Vite alias and vite-plus catalog versions from 0.2.2 to 0.3.0 in pnpm-workspace.yaml and updates the lockfile. The rest of the PR is formatting-only reformatting across many provider, orchestration, and client modules triggered by the new formatter rules.

^{Macroscope summarized 21611ad.}

Follow-up commit: oxfmt 0.57 → 0.64 reformat

The toolchain bump also moves oxfmt, which now hugs a sole trailing arrow-function argument (Effect.map((x) => ({...})) on one line instead of breaking before the arrow) and breaks long string-literal unions onto leading | lines. vp check failed on 29 pre-existing files for that reason, so a second commit runs vp fmt over the repo. Verified per file that the change is whitespace, trailing commas, and | placement only (stripping those, old and new are byte-identical). Includes packages/effect-codex-app-server/src/_generated/schema.gen.ts, whose generator already runs vp fmt over its output, so it stays consistent from here.

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.2 KiB	13.4 KiB	+234 B (+1.7%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	−4 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.3 KiB	6.6 KiB	+238 B (+3.7%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.5 KiB	57.0 KiB	+1.5 KiB (+2.7%)	66.4 KiB	✅
Codex	Live turn messages	8	10	+2 (+25.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−48 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	+3 B (+0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−51 B (−0.8%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.3 KiB	−88 B (−0.2%)	66.4 KiB	✅
Claude	Live turn messages	10	8	−2 (−20.0%)	21	✅

Baseline: 3b2de9d · PR result: 21611ad · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This updates the shared Vite-plus build, bundling, testing, linting, and formatting toolchain across the monorepo, including the implementation of the existing bundled-development path. Although no application source changes are included, the upgrade can affect generated web/desktop/server artifacts and the unverified remote HMR path.

No code changes detected at 03de157. Prior analysis still applies.

^{You can add or adjust custom eligibility rules. Learn more.}

[juliusmarminge]

Problem

vp run dev --share defaults T3CODE_BUNDLED_DEV=1 (scripts/dev-runner.ts, apps/web/vite.config.tsexperimental.bundledDev) so a remote browser on the tailnet pays a few chunk requests instead of a per-module round trip. On vite-plus 0.2.2, which bundles vite 8.1.2 / rolldown 1.1.4, that bundled dev mode is flaky for the shared origin: the page throws @vitejs/plugin-react can't detect preamble and Cannot read properties of undefined (reading 'Root') in <Collapsible> (a base-ui namespace import resolving to undefined in the bundled chunk), so the app never gets past the splash. Unbundled dev renders fine.

Fix

Bump the pinned toolchain to vite-plus 0.3.0, which bundles vite 8.2.2 / rolldown 1.2.5. The 8.1.2 to 8.2.2 range carries the batch of experimental.bundledDev fixes (vitejs/vite CHANGELOG):

• bundled-dev: reload once after rebuild instead of via the fallback page (#23106)
• bundled-dev: support worker file update accepted by HMR (#23068)
• bundled-dev: print build errors when an HMR update fails (#23024)
• bundled-dev: avoid stack overflow on import.meta.hot.invalidate() (#22797)
• bundled-dev: serve assets emitted during HMR / lazy compile (#22745)
• bundled-dev: skip plugin transform hooks for rolldown-lazy stub modules (#22778)
• bundled-dev: avoid duplicated buildEnd (#22931)

Plus from vite-plus itself:

• v0.2.9: vp dev no longer crashes at startup with ENOENT when experimental.bundledDev is enabled ([Feature]: Auto review for tool call safety #2384)
• v0.3.0: vite 8.2.2, rolldown 1.2.5, vitest 4.1.11, oxlint 1.79.0, oxfmt 0.64.0

What changed

Only the two catalog entries in pnpm-workspace.yaml (vite: npm:@voidzero-dev/vite-plus-core@0.3.0, vite-plus: 0.3.0). Every package consumes them via catalog:, so nothing else needed touching. No other 0.2.2 pin exists in package.json files, workflows, Dockerfiles, or tool-version files; CI uses voidzero-dev/setup-vp@v1 with no version pin, and the local vp reads the project version from the lockfile.

The lockfile delta is limited to the vite-plus toolchain: vite-plus, @voidzero-dev/vite-plus-core and platform binaries, vitest 4.1.9 to 4.1.11, oxlint 1.72.0 to 1.79.0, oxfmt 0.57.0 to 0.64.0, oxlint-tsgolint 0.24.0 to 7.0.2001, @oxc-project/* 0.138.0 to 0.146.0, and new transitive deps of vite-plus-core (lightningcss 1.33.0, yuku-parser / yuku-codegen 0.5.48, @oxlint/plugins 1.79.0). No unrelated major bumps.

experimental.bundledDev is still typed boolean in 8.2.2, so apps/web/vite.config.ts and scripts/dev-runner.ts need no change.

0.3.0 caveats

• XDG install layout for fresh installs only. Existing ~/.vite-plus installs keep their layout and vp upgrade leaves them in place. CI via setup-vp stays on the single-root layout, so the $HOME/.vite-plus/package_manager/pnpm/... paths in the mobile workflows keep working. A maintainer who does a fresh install will get ~/.local/share/vite-plus/bin instead; nothing in this repo hard-codes that path.
• @vitejs/devtools optional peer range widened to ^0.4.0 || ^0.5.0. We do not depend on it.
• Upgrade check moved out of process; commands no longer wait on the registry.
• Newer oxlint/oxfmt may flag new issues; the pre-commit vp fmt on the two changed files was clean.

Verification

• vp i refreshed the lockfile; vp --version inside apps/web reports vite-plus 0.3.0, vite 8.2.2, rolldown 1.2.5, vitest 4.1.11
• vp run build in apps/web: built in 12.7s, working tree clean afterward
• vp run typecheck in apps/web: exit 0
• vp test run --project unit src/lib/assistantCitationNavigation.test.ts src/lib/attachmentUploadState.test.ts in apps/web: 2 files, 8 tests passed on vitest 4.1.11
• T3CODE_BUNDLED_DEV=1 vp run dev in the worktree (no --share): after the first rolldown build the real index was served with <script type="module" src="/bundledDevClient.mjs"> and <script type="module" crossorigin src="/assets/index.js">; both returned 200 (text/javascript, 57 kB and 19.5 MB) with no error / ENOENT / preamble lines in the server log. Stopped by the process group of the PID captured at spawn.

Not verified here: the remote-browser --share path itself, which needs a second machine on the tailnet.

Model: Claude Fable 5 via Claude Code.

🤖 Generated with Claude Code

---

Note

Low Risk
Primarily a dev/build toolchain and auto-format refresh; application logic is unchanged aside from formatting, though a major vite/rolldown bump can still surface new build or lint issues in CI.

Overview
Bumps the workspace vite-plus / vite catalog pins from 0.2.2 to 0.3.0 and refreshes pnpm-lock.yaml, pulling in vite 8.2.2, rolldown 1.2.5, vitest 4.1.11, oxlint 1.79.0, oxfmt 0.64.0, and related native bindings (including new yuku-parser / yuku-codegen transitives).

Alongside the lockfile, many TypeScript files were reformatted only: multi-line .map / Effect.map / Effect.flatMap callbacks and similar patterns were collapsed to match oxfmt 0.64 (desktop, mobile, server, client-runtime, web, shared, generated schema.gen.ts). No intentional runtime or API behavior changes in those edits.

The upgrade targets more reliable experimental.bundledDev (used when shared dev defaults T3CODE_BUNDLED_DEV=1); app config in apps/web/vite.config.ts and scripts/dev-runner.ts is unchanged.

^{Reviewed by Cursor Bugbot for commit 21611ad. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Upgrade vite-plus to 0.3.0

Bumps the Vite alias and vite-plus catalog versions from 0.2.2 to 0.3.0 in pnpm-workspace.yaml and updates the lockfile. The rest of the PR is formatting-only reformatting across many provider, orchestration, and client modules triggered by the new formatter rules.

^{Macroscope summarized 21611ad.}

Follow-up commit: oxfmt 0.57 → 0.64 reformat

The toolchain bump also moves oxfmt, which now hugs a sole trailing arrow-function argument (Effect.map((x) => ({...})) on one line instead of breaking before the arrow) and breaks long string-literal unions onto leading | lines. vp check failed on 29 pre-existing files for that reason, so a second commit runs vp fmt over the repo. Verified per file that the change is whitespace, trailing commas, and | placement only (stripping those, old and new are byte-identical). Includes packages/effect-codex-app-server/src/_generated/schema.gen.ts, whose generator already runs vp fmt over its output, so it stays consistent from here.

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.2 KiB	13.4 KiB	+234 B (+1.7%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	−4 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.3 KiB	6.6 KiB	+238 B (+3.7%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.5 KiB	57.0 KiB	+1.5 KiB (+2.7%)	66.4 KiB	✅
Codex	Live turn messages	8	10	+2 (+25.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−48 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	+3 B (+0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−51 B (−0.8%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.3 KiB	−88 B (−0.2%)	66.4 KiB	✅
Claude	Live turn messages	10	8	−2 (−20.0%)	21	✅

Baseline: 3b2de9d · PR result: 21611ad · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This updates the shared Vite-plus build, bundling, testing, linting, and formatting toolchain across the monorepo, including the implementation of the existing bundled-development path. Although no application source changes are included, the upgrade can affect generated web/desktop/server artifacts and the unverified remote HMR path.

No code changes detected at 03de157. Prior analysis still applies.

^{You can add or adjust custom eligibility rules. Learn more.}