chore(upstream): sync marcode through 018d7f27

[maarco]

What Changed

Merges 59 upstream commits, f035a0f4..018d7f27 (pingdotgg/t3code@018d7f27), into Marcode as one two-parent merge commit (a940ce5c), plus one follow-up commit (0a265857) fixing a Marcode-owned test the merge invalidated. No force push, reset, restore, stash, clean, or wholesale ours/theirs resolution was used; MERGE_HEAD stayed at 018d7f27 for the whole resolution.

Supersedes #53, #55, and #56 — those draft syncs targeted older upstream SHAs (a3a8cbd6, f6f2be32) that are contained in this one. Merging this closes the delta they cover; they can be closed unmerged.

Conflicts resolved (24 paths)

Path	Decision
apps/mobile/app.config.ts	Took upstream's adaptive-launcher-icon fix (pingdotgg#4332) — release now uses the shared transparent foreground instead of the rounded-square universal export Android was masking twice. See the branding note below.
apps/mobile/src/lib/authClientMetadata.ts	Kept "Marcode Mobile"; took upstream's tablet/phone deviceType detection.
apps/mobile/src/components/CompactBrandTitle.tsx	Kept MarcodeMark and the conditional stage badge; took upstream's Uniwind className styling over useThemeColor + inline styles (the new lint rule forbids them).
apps/mobile/src/components/T3Wordmark.tsx	Kept Marcode's deletion; upstream only restyled it.
apps/mobile/src/features/home/HomeHeader.tsx	Kept MarcodeMark.
apps/mobile/src/lib/mobileBranding.test.ts	Upstream deleted it (pingdotgg#8397). Kept — it pins Marcode returning "" (no badge) where upstream returns "Alpha". Now says so in a comment.
oxlint-plugin-marcode/rules/no-mobile-uniwind-theme-escape-hatches.*	Landed upstream's new rule under Marcode's plugin dir; renamed the test's rule prefix t3code/ → marcode/ to match the plugin meta name.
vite.config.ts	Enabled the new rule at upstream's "error"; kept Marcode's prefixes and the no-native-title-tooltip"warn" seam.
apps/server/src/cloud/bootService.ts + test	Took upstream's PATH export (pingdotgg#8173, keeps provider CLIs on the service path); kept MARCODE_HOME, not T3CODE_HOME. Both guard tests now stand side by side.
apps/server/src/environment/ServerEnvironment.ts, packages/contracts/src/environment.ts	Kept both capability flags: workspaceLayoutMutations and the new threadPullRequestLinking.
apps/server/src/persistence/Migrations.ts	Marcode ids sit one higher (033 is ProjectWorkspaceLayout). Renamed upstream's 042/043 → 043/044 and fixed the new migration test's hardcoded toMigrationInclusive ids.
apps/web/src/components/sidebar/SidebarChrome.tsx	Kept Marcode's bare Electron drag strip. Upstream's only change was a project-settings case for the sidebar Back button (pingdotgg#8168) — Marcode's FloatingPillNav already shows Back to Workspace on every non-thread route, so that intent is covered.
apps/web/src/components/threadSidebarWidth.test.ts	Accepted upstream's prune of the trivial width assertions, but moved the Marcode brand-removal pin into its own components/sidebar/sidebarBrandRemoval.test.ts so a future upstream prune can't take it out silently.
apps/web/src/components/ThreadTerminalDrawer.tsx	Kept Marcode's xterm link provider. Upstream's isTerminalUrl fix (pingdotgg#7488) lands in the shared terminal-links.ts, which Marcode's match.kind classification already goes through.
apps/web/src/components/chat/OpenInPicker.tsx	Kept Marcode's FolderOpenFilled icon; took upstream's move of labels to the shared editorLabelForPlatform.
apps/web/src/components/ChatMarkdown.tsx	Took upstream's editorPreferences/editorLabels/remoteOpen imports; kept openFileInFloatingEditor. The fork seam routing agent file links to the floating editor survives — upstream's right-panel file viewer (pingdotgg#8098) is not reintroduced.
apps/web/src/components/ChatView.tsx	Took upstream's hoisted useNewThreadHandler/useThreadActions (needed by the new settle/pin keyboard shortcuts) and removed the duplicate later declaration. Kept Marcode's slim ChatHeader; dropped upstream's openProjectPullRequest, unreachable here because openThreadPullRequest already falls back to the project repository.
pnpm-lock.yaml	Kept Marcode's xterm entries, then regenerated with pnpm install --lockfile-only from the merged manifests.

Fork-boundary breaks that merged cleanly (no conflict) and were fixed here

These are the dangerous ones — they produce no conflict marker:

1. apps/web/src/connection/clientMetadata.ts arrived new from upstream with hardcoded "T3 Code Desktop" / "T3 Code Web" labels and replaced Marcode's inline metadata builder in platform.ts. Every client would have reported itself as T3 Code. Routed both labels through APP_BASE_NAME and pinned "Marcode Desktop" in clientMetadata.test.ts.
2. Three upstream oxlint-disable-next-line t3code/no-global-process-runtime comments (apps/desktop/src/preload.ts, apps/desktop/scripts/verify-preload-bundle.mjs, apps/server/src/process/externalLauncher.test.ts). Marcode's plugin is named marcode, so those disables match nothing and the rule — set to "error" — would have failed the Check job. Repointed to marcode/.

Swept and confirmed correct for Marcode: T3CODE_HOME (only guard tests/comments), t3code.service / com.t3tools.t3code launchd label, WORKTREE_BRANCH_PREFIX ("marcode"; the t3code/... branch strings in server/checkpoint tests are pre-existing arbitrary fixtures, not prefix-derived), t3code: storage keys, T3_BOOT_SERVICE_UNIT, T3CODE_DESKTOP_REUSE_RESOURCE_MONITOR, @t3tools/*, and the t3 package/CLI name — all left upstream-shaped on purpose.

Follow-up commit 0a265857

CI caught one thing the focused local run did not: apps/server/src/persistence/Migrations.test.ts pins the migration registry with slice(-10), so appending the two renumbered upstream migrations pushed id 33 — the id the test is named for — out of the asserted window.

Fixed by anchoring the slice at id 33 via findIndex rather than a fixed tail length, so the assertion keeps covering Marcode's owned migration however many upstream ones land after it. Without that, the pin would have silently stopped testing its own subject on the next sync.

Coupled vendored subtrees

pnpm-workspace.yaml moved, but only Clerk versions; effect stayed at 4.0.0-beta.103 and infra/relay/package.json did not change. No sync:repos run was needed.

Why

Daily upstream sync. Upstream had 59 unmerged commits including a Codex approval-callback recovery fix, a projection-bootstrap replay fix (pingdotgg#7538), the macOS service PATH fix, Grok reliability work, and the Clerk 6.30/0.0.37 bumps.

UI Changes

Two decisions need your eye — I could not drive a browser or simulator for them in this container:

• Android release launcher icon. Upstream's fix pairs a transparent light mark with a dark plate. Marcode's release previously used the black-on-white universal export as the foreground, which is exactly the double-masking bug upstream fixed. Taking the fix means the release Android launcher is now a light mark on #000000, diverging from Marcode's black-on-white rule for that one surface. Restoring black-on-white needs a Marcode-owned foreground SVG/PNG pair; no rasterizer is available here to generate one. Flagging for your call — the alternative was keeping the double-mask bug or shipping a white mark on a white plate.
• vp run icons:check could not run: it needs macOS Icon Composer.

Checklist

• This PR is small and focused — no, it is a 59-commit upstream sync; that is the unit of work
• I explained what changed and why
• I included before/after screenshots for any UI changes — no browser/simulator available in this environment
• I included a video for animation/interaction changes — n/a

CI on 0a265857

All four required checks green: Check ✅ · Test ✅ · Mobile Native Static Analysis ✅ · Release Smoke ✅. Also green: Test Server 1/2/3, Rust, Mobile Native Changes, claude-review, and the thread-transfer budget report (every metric within its ceiling). Native fingerprint diff sits queued, as it did on the previous head; it is not a required check.

Local verification before pushing (focused, per AGENTS.md — not the full suite)

• vp test run green on: clientMetadata, sidebarBrandRemoval, mobileBranding, no-mobile-uniwind-theme-escape-hatches, bootService, ServerEnvironment, Migrations, 043_ProjectionThreadLinkedPullRequest, 042_AuthSessionClientConnection, build-desktop-artifact, contracts/environment, contracts/settings, terminal-links, editorLabels, Sidebar.logic, versionSkew, upstream-sync-workflow, serviceLauncher, electron-launcher, generate-uniwind-themes, connection/platform.
• vp run --filter … typecheck clean for @t3tools/contracts, @t3tools/web, @t3tools/shared, @t3tools/client-runtime, t3 (server), @t3tools/mobile, @t3tools/desktop (only pre-existing Effect style suggestions, no errors).
• vp fmt --check clean on all 356 changed TS files.
• vp lint on the 82 changed directories: no errors; only warnings that are pre-existing on main.
• generate-uniwind-themes.mts --check passes, so the generated theme CSS matches Marcode's tokens.

One thing to review deliberately

apps/server/src/provider/ModelManifest.ts is new from upstream and fetches legacy model classification from https://raw.githubusercontent.com/pingdotgg/t3code/main/apps/server/src/provider/model-manifest.json. I left the URL pointing at upstream — they maintain that data and Marcode benefits from their updates — but it does mean Marcode servers read a file from upstream's repo at runtime. Repointing it at maarco/marcode is a one-line change if you'd rather own it. The in-repo model-manifest.json is the fallback either way.

Left deliberately as a draft: this sync is yours to review and mark ready.

[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	13.4 KiB	13.3 KiB	−122 B (−0.9%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+5 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.5 KiB	6.4 KiB	−127 B (−1.9%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.0 KiB	55.6 KiB	+558 B (+1.0%)	66.4 KiB	✅
Codex	Live turn messages	16	10	−6 (−37.5%)	21	✅
Claude	Total thread wire	13.4 KiB	13.3 KiB	−156 B (−1.1%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	+8 B (+0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.6 KiB	6.4 KiB	−164 B (−2.4%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	55.8 KiB	56.4 KiB	+532 B (+0.9%)	66.4 KiB	✅
Claude	Live turn messages	16	9	−7 (−43.8%)	21	✅

Baseline: 133f3f7 · PR result: 0a26585 · 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.

[maarco]

Two status changes since the description was written.

Native fingerprint diff is now cancelled, and it is not this PR's failure. It sat queued for exactly 24h (2026-08-28 15:42:11 → 2026-08-29 15:42:11) and GitHub killed it on the job timeout. Every run of mobile-fingerprint-check.yml on this fork does the same: I checked the last 30 runs and not one has ever completed — claude/gracious-bohr-1dtucn, -rlxi8l, -ie7aa6, -23bj7a, -x80z9c, -23l2b6, back to at least 2026-08-23, all queued-then-cancelled. It is a pre-existing fork-environment condition (the job never starts; the blacksmith-8vcpu-ubuntu-2404 label itself is fine, since Check, Test, and Rust all run on it here), and it is not in requiredPullRequestChecks. I am deliberately not spending the one re-run on it: 30 runs across 6 branches is stronger evidence than a 31st would be, and a re-run would just occupy a runner slot for another 24h. It needs a repo-settings fix, not a change to this PR.

The four required checks remain green on 0a265857: Check ✅ · Test ✅ · Mobile Native Static Analysis ✅ · Release Smoke ✅.

This PR's upstream coverage is now superseded by #61 (053affbe), opened by today's scheduled run. I verified 018d7f27 is an ancestor of #61's head, so #61 covers everything here.

But #61 branched from main independently rather than building on this branch, so the two are not interchangeable:

• chore(upstream): sync marcode through 053affbe #61is currently red — Test Server 2 and Test both failing — because it reintroduces the Migrations.test.ts problem this PR fixed in 0a265857. Its registry ends at id 44 while its pin still asserts slice(-10) equals ids 33–42.
• chore(upstream): sync marcode through 053affbe #61did independently make the other two fork-boundary fixes (the APP_BASE_NAME client labels and the marcode/ oxlint prefixes), so those are not at risk either way.

So the migration-pin fix in 0a265857 is the one piece of this PR that #61 still needs. Cherry-picking that single commit onto #61 should get it green, at which point #61 is the better merge candidate and this PR can be closed alongside #53, #55, and #56.

I have not pushed to #61 — it is not a PR I opened, and this session was not asked to drive it.

---

Generated by Claude Code

[maarco]

What Changed

Merges 59 upstream commits, f035a0f4..018d7f27 (pingdotgg/t3code@018d7f27), into Marcode as one two-parent merge commit (a940ce5c), plus one follow-up commit (0a265857) fixing a Marcode-owned test the merge invalidated. No force push, reset, restore, stash, clean, or wholesale ours/theirs resolution was used; MERGE_HEAD stayed at 018d7f27 for the whole resolution.

Supersedes #53, #55, and #56 — those draft syncs targeted older upstream SHAs (a3a8cbd6, f6f2be32) that are contained in this one. Merging this closes the delta they cover; they can be closed unmerged.

Conflicts resolved (24 paths)

Path	Decision
apps/mobile/app.config.ts	Took upstream's adaptive-launcher-icon fix (pingdotgg#4332) — release now uses the shared transparent foreground instead of the rounded-square universal export Android was masking twice. See the branding note below.
apps/mobile/src/lib/authClientMetadata.ts	Kept "Marcode Mobile"; took upstream's tablet/phone deviceType detection.
apps/mobile/src/components/CompactBrandTitle.tsx	Kept MarcodeMark and the conditional stage badge; took upstream's Uniwind className styling over useThemeColor + inline styles (the new lint rule forbids them).
apps/mobile/src/components/T3Wordmark.tsx	Kept Marcode's deletion; upstream only restyled it.
apps/mobile/src/features/home/HomeHeader.tsx	Kept MarcodeMark.
apps/mobile/src/lib/mobileBranding.test.ts	Upstream deleted it (pingdotgg#8397). Kept — it pins Marcode returning "" (no badge) where upstream returns "Alpha". Now says so in a comment.
oxlint-plugin-marcode/rules/no-mobile-uniwind-theme-escape-hatches.*	Landed upstream's new rule under Marcode's plugin dir; renamed the test's rule prefix t3code/ → marcode/ to match the plugin meta name.
vite.config.ts	Enabled the new rule at upstream's "error"; kept Marcode's prefixes and the no-native-title-tooltip"warn" seam.
apps/server/src/cloud/bootService.ts + test	Took upstream's PATH export (pingdotgg#8173, keeps provider CLIs on the service path); kept MARCODE_HOME, not T3CODE_HOME. Both guard tests now stand side by side.
apps/server/src/environment/ServerEnvironment.ts, packages/contracts/src/environment.ts	Kept both capability flags: workspaceLayoutMutations and the new threadPullRequestLinking.
apps/server/src/persistence/Migrations.ts	Marcode ids sit one higher (033 is ProjectWorkspaceLayout). Renamed upstream's 042/043 → 043/044 and fixed the new migration test's hardcoded toMigrationInclusive ids.
apps/web/src/components/sidebar/SidebarChrome.tsx	Kept Marcode's bare Electron drag strip. Upstream's only change was a project-settings case for the sidebar Back button (pingdotgg#8168) — Marcode's FloatingPillNav already shows Back to Workspace on every non-thread route, so that intent is covered.
apps/web/src/components/threadSidebarWidth.test.ts	Accepted upstream's prune of the trivial width assertions, but moved the Marcode brand-removal pin into its own components/sidebar/sidebarBrandRemoval.test.ts so a future upstream prune can't take it out silently.
apps/web/src/components/ThreadTerminalDrawer.tsx	Kept Marcode's xterm link provider. Upstream's isTerminalUrl fix (pingdotgg#7488) lands in the shared terminal-links.ts, which Marcode's match.kind classification already goes through.
apps/web/src/components/chat/OpenInPicker.tsx	Kept Marcode's FolderOpenFilled icon; took upstream's move of labels to the shared editorLabelForPlatform.
apps/web/src/components/ChatMarkdown.tsx	Took upstream's editorPreferences/editorLabels/remoteOpen imports; kept openFileInFloatingEditor. The fork seam routing agent file links to the floating editor survives — upstream's right-panel file viewer (pingdotgg#8098) is not reintroduced.
apps/web/src/components/ChatView.tsx	Took upstream's hoisted useNewThreadHandler/useThreadActions (needed by the new settle/pin keyboard shortcuts) and removed the duplicate later declaration. Kept Marcode's slim ChatHeader; dropped upstream's openProjectPullRequest, unreachable here because openThreadPullRequest already falls back to the project repository.
pnpm-lock.yaml	Kept Marcode's xterm entries, then regenerated with pnpm install --lockfile-only from the merged manifests.

Fork-boundary breaks that merged cleanly (no conflict) and were fixed here

These are the dangerous ones — they produce no conflict marker:

1. apps/web/src/connection/clientMetadata.ts arrived new from upstream with hardcoded "T3 Code Desktop" / "T3 Code Web" labels and replaced Marcode's inline metadata builder in platform.ts. Every client would have reported itself as T3 Code. Routed both labels through APP_BASE_NAME and pinned "Marcode Desktop" in clientMetadata.test.ts.
2. Three upstream oxlint-disable-next-line t3code/no-global-process-runtime comments (apps/desktop/src/preload.ts, apps/desktop/scripts/verify-preload-bundle.mjs, apps/server/src/process/externalLauncher.test.ts). Marcode's plugin is named marcode, so those disables match nothing and the rule — set to "error" — would have failed the Check job. Repointed to marcode/.

Swept and confirmed correct for Marcode: T3CODE_HOME (only guard tests/comments), t3code.service / com.t3tools.t3code launchd label, WORKTREE_BRANCH_PREFIX ("marcode"; the t3code/... branch strings in server/checkpoint tests are pre-existing arbitrary fixtures, not prefix-derived), t3code: storage keys, T3_BOOT_SERVICE_UNIT, T3CODE_DESKTOP_REUSE_RESOURCE_MONITOR, @t3tools/*, and the t3 package/CLI name — all left upstream-shaped on purpose.

Follow-up commit 0a265857

CI caught one thing the focused local run did not: apps/server/src/persistence/Migrations.test.ts pins the migration registry with slice(-10), so appending the two renumbered upstream migrations pushed id 33 — the id the test is named for — out of the asserted window.

Fixed by anchoring the slice at id 33 via findIndex rather than a fixed tail length, so the assertion keeps covering Marcode's owned migration however many upstream ones land after it. Without that, the pin would have silently stopped testing its own subject on the next sync.

Coupled vendored subtrees

pnpm-workspace.yaml moved, but only Clerk versions; effect stayed at 4.0.0-beta.103 and infra/relay/package.json did not change. No sync:repos run was needed.

Why

Daily upstream sync. Upstream had 59 unmerged commits including a Codex approval-callback recovery fix, a projection-bootstrap replay fix (pingdotgg#7538), the macOS service PATH fix, Grok reliability work, and the Clerk 6.30/0.0.37 bumps.

UI Changes

Two decisions need your eye — I could not drive a browser or simulator for them in this container:

• Android release launcher icon. Upstream's fix pairs a transparent light mark with a dark plate. Marcode's release previously used the black-on-white universal export as the foreground, which is exactly the double-masking bug upstream fixed. Taking the fix means the release Android launcher is now a light mark on #000000, diverging from Marcode's black-on-white rule for that one surface. Restoring black-on-white needs a Marcode-owned foreground SVG/PNG pair; no rasterizer is available here to generate one. Flagging for your call — the alternative was keeping the double-mask bug or shipping a white mark on a white plate.
• vp run icons:check could not run: it needs macOS Icon Composer.

Checklist

• This PR is small and focused — no, it is a 59-commit upstream sync; that is the unit of work
• I explained what changed and why
• I included before/after screenshots for any UI changes — no browser/simulator available in this environment
• I included a video for animation/interaction changes — n/a

CI on 0a265857

All four required checks green: Check ✅ · Test ✅ · Mobile Native Static Analysis ✅ · Release Smoke ✅. Also green: Test Server 1/2/3, Rust, Mobile Native Changes, claude-review, and the thread-transfer budget report (every metric within its ceiling). Native fingerprint diff sits queued, as it did on the previous head; it is not a required check.

Local verification before pushing (focused, per AGENTS.md — not the full suite)

• vp test run green on: clientMetadata, sidebarBrandRemoval, mobileBranding, no-mobile-uniwind-theme-escape-hatches, bootService, ServerEnvironment, Migrations, 043_ProjectionThreadLinkedPullRequest, 042_AuthSessionClientConnection, build-desktop-artifact, contracts/environment, contracts/settings, terminal-links, editorLabels, Sidebar.logic, versionSkew, upstream-sync-workflow, serviceLauncher, electron-launcher, generate-uniwind-themes, connection/platform.
• vp run --filter … typecheck clean for @t3tools/contracts, @t3tools/web, @t3tools/shared, @t3tools/client-runtime, t3 (server), @t3tools/mobile, @t3tools/desktop (only pre-existing Effect style suggestions, no errors).
• vp fmt --check clean on all 356 changed TS files.
• vp lint on the 82 changed directories: no errors; only warnings that are pre-existing on main.
• generate-uniwind-themes.mts --check passes, so the generated theme CSS matches Marcode's tokens.

One thing to review deliberately

apps/server/src/provider/ModelManifest.ts is new from upstream and fetches legacy model classification from https://raw.githubusercontent.com/pingdotgg/t3code/main/apps/server/src/provider/model-manifest.json. I left the URL pointing at upstream — they maintain that data and Marcode benefits from their updates — but it does mean Marcode servers read a file from upstream's repo at runtime. Repointing it at maarco/marcode is a one-line change if you'd rather own it. The in-repo model-manifest.json is the fallback either way.

Left deliberately as a draft: this sync is yours to review and mark ready.

[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	13.4 KiB	13.3 KiB	−122 B (−0.9%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+5 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.5 KiB	6.4 KiB	−127 B (−1.9%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.0 KiB	55.6 KiB	+558 B (+1.0%)	66.4 KiB	✅
Codex	Live turn messages	16	10	−6 (−37.5%)	21	✅
Claude	Total thread wire	13.4 KiB	13.3 KiB	−156 B (−1.1%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	+8 B (+0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.6 KiB	6.4 KiB	−164 B (−2.4%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	55.8 KiB	56.4 KiB	+532 B (+0.9%)	66.4 KiB	✅
Claude	Live turn messages	16	9	−7 (−43.8%)	21	✅

Baseline: 133f3f7 · PR result: 0a26585 · 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.

[maarco]

Two status changes since the description was written.

Native fingerprint diff is now cancelled, and it is not this PR's failure. It sat queued for exactly 24h (2026-08-28 15:42:11 → 2026-08-29 15:42:11) and GitHub killed it on the job timeout. Every run of mobile-fingerprint-check.yml on this fork does the same: I checked the last 30 runs and not one has ever completed — claude/gracious-bohr-1dtucn, -rlxi8l, -ie7aa6, -23bj7a, -x80z9c, -23l2b6, back to at least 2026-08-23, all queued-then-cancelled. It is a pre-existing fork-environment condition (the job never starts; the blacksmith-8vcpu-ubuntu-2404 label itself is fine, since Check, Test, and Rust all run on it here), and it is not in requiredPullRequestChecks. I am deliberately not spending the one re-run on it: 30 runs across 6 branches is stronger evidence than a 31st would be, and a re-run would just occupy a runner slot for another 24h. It needs a repo-settings fix, not a change to this PR.

The four required checks remain green on 0a265857: Check ✅ · Test ✅ · Mobile Native Static Analysis ✅ · Release Smoke ✅.

This PR's upstream coverage is now superseded by #61 (053affbe), opened by today's scheduled run. I verified 018d7f27 is an ancestor of #61's head, so #61 covers everything here.

But #61 branched from main independently rather than building on this branch, so the two are not interchangeable:

• chore(upstream): sync marcode through 053affbe #61is currently red — Test Server 2 and Test both failing — because it reintroduces the Migrations.test.ts problem this PR fixed in 0a265857. Its registry ends at id 44 while its pin still asserts slice(-10) equals ids 33–42.
• chore(upstream): sync marcode through 053affbe #61did independently make the other two fork-boundary fixes (the APP_BASE_NAME client labels and the marcode/ oxlint prefixes), so those are not at risk either way.

So the migration-pin fix in 0a265857 is the one piece of this PR that #61 still needs. Cherry-picking that single commit onto #61 should get it green, at which point #61 is the better merge candidate and this PR can be closed alongside #53, #55, and #56.

I have not pushed to #61 — it is not a PR I opened, and this session was not asked to drive it.

---

Generated by Claude Code

[maarco]

What Changed

Merges 59 upstream commits, f035a0f4..018d7f27 (pingdotgg/t3code@018d7f27), into Marcode as one two-parent merge commit (a940ce5c), plus one follow-up commit (0a265857) fixing a Marcode-owned test the merge invalidated. No force push, reset, restore, stash, clean, or wholesale ours/theirs resolution was used; MERGE_HEAD stayed at 018d7f27 for the whole resolution.

Supersedes #53, #55, and #56 — those draft syncs targeted older upstream SHAs (a3a8cbd6, f6f2be32) that are contained in this one. Merging this closes the delta they cover; they can be closed unmerged.

Conflicts resolved (24 paths)

Path	Decision
apps/mobile/app.config.ts	Took upstream's adaptive-launcher-icon fix (pingdotgg#4332) — release now uses the shared transparent foreground instead of the rounded-square universal export Android was masking twice. See the branding note below.
apps/mobile/src/lib/authClientMetadata.ts	Kept "Marcode Mobile"; took upstream's tablet/phone deviceType detection.
apps/mobile/src/components/CompactBrandTitle.tsx	Kept MarcodeMark and the conditional stage badge; took upstream's Uniwind className styling over useThemeColor + inline styles (the new lint rule forbids them).
apps/mobile/src/components/T3Wordmark.tsx	Kept Marcode's deletion; upstream only restyled it.
apps/mobile/src/features/home/HomeHeader.tsx	Kept MarcodeMark.
apps/mobile/src/lib/mobileBranding.test.ts	Upstream deleted it (pingdotgg#8397). Kept — it pins Marcode returning "" (no badge) where upstream returns "Alpha". Now says so in a comment.
oxlint-plugin-marcode/rules/no-mobile-uniwind-theme-escape-hatches.*	Landed upstream's new rule under Marcode's plugin dir; renamed the test's rule prefix t3code/ → marcode/ to match the plugin meta name.
vite.config.ts	Enabled the new rule at upstream's "error"; kept Marcode's prefixes and the no-native-title-tooltip"warn" seam.
apps/server/src/cloud/bootService.ts + test	Took upstream's PATH export (pingdotgg#8173, keeps provider CLIs on the service path); kept MARCODE_HOME, not T3CODE_HOME. Both guard tests now stand side by side.
apps/server/src/environment/ServerEnvironment.ts, packages/contracts/src/environment.ts	Kept both capability flags: workspaceLayoutMutations and the new threadPullRequestLinking.
apps/server/src/persistence/Migrations.ts	Marcode ids sit one higher (033 is ProjectWorkspaceLayout). Renamed upstream's 042/043 → 043/044 and fixed the new migration test's hardcoded toMigrationInclusive ids.
apps/web/src/components/sidebar/SidebarChrome.tsx	Kept Marcode's bare Electron drag strip. Upstream's only change was a project-settings case for the sidebar Back button (pingdotgg#8168) — Marcode's FloatingPillNav already shows Back to Workspace on every non-thread route, so that intent is covered.
apps/web/src/components/threadSidebarWidth.test.ts	Accepted upstream's prune of the trivial width assertions, but moved the Marcode brand-removal pin into its own components/sidebar/sidebarBrandRemoval.test.ts so a future upstream prune can't take it out silently.
apps/web/src/components/ThreadTerminalDrawer.tsx	Kept Marcode's xterm link provider. Upstream's isTerminalUrl fix (pingdotgg#7488) lands in the shared terminal-links.ts, which Marcode's match.kind classification already goes through.
apps/web/src/components/chat/OpenInPicker.tsx	Kept Marcode's FolderOpenFilled icon; took upstream's move of labels to the shared editorLabelForPlatform.
apps/web/src/components/ChatMarkdown.tsx	Took upstream's editorPreferences/editorLabels/remoteOpen imports; kept openFileInFloatingEditor. The fork seam routing agent file links to the floating editor survives — upstream's right-panel file viewer (pingdotgg#8098) is not reintroduced.
apps/web/src/components/ChatView.tsx	Took upstream's hoisted useNewThreadHandler/useThreadActions (needed by the new settle/pin keyboard shortcuts) and removed the duplicate later declaration. Kept Marcode's slim ChatHeader; dropped upstream's openProjectPullRequest, unreachable here because openThreadPullRequest already falls back to the project repository.
pnpm-lock.yaml	Kept Marcode's xterm entries, then regenerated with pnpm install --lockfile-only from the merged manifests.

Fork-boundary breaks that merged cleanly (no conflict) and were fixed here

These are the dangerous ones — they produce no conflict marker:

1. apps/web/src/connection/clientMetadata.ts arrived new from upstream with hardcoded "T3 Code Desktop" / "T3 Code Web" labels and replaced Marcode's inline metadata builder in platform.ts. Every client would have reported itself as T3 Code. Routed both labels through APP_BASE_NAME and pinned "Marcode Desktop" in clientMetadata.test.ts.
2. Three upstream oxlint-disable-next-line t3code/no-global-process-runtime comments (apps/desktop/src/preload.ts, apps/desktop/scripts/verify-preload-bundle.mjs, apps/server/src/process/externalLauncher.test.ts). Marcode's plugin is named marcode, so those disables match nothing and the rule — set to "error" — would have failed the Check job. Repointed to marcode/.

Swept and confirmed correct for Marcode: T3CODE_HOME (only guard tests/comments), t3code.service / com.t3tools.t3code launchd label, WORKTREE_BRANCH_PREFIX ("marcode"; the t3code/... branch strings in server/checkpoint tests are pre-existing arbitrary fixtures, not prefix-derived), t3code: storage keys, T3_BOOT_SERVICE_UNIT, T3CODE_DESKTOP_REUSE_RESOURCE_MONITOR, @t3tools/*, and the t3 package/CLI name — all left upstream-shaped on purpose.

Follow-up commit 0a265857

CI caught one thing the focused local run did not: apps/server/src/persistence/Migrations.test.ts pins the migration registry with slice(-10), so appending the two renumbered upstream migrations pushed id 33 — the id the test is named for — out of the asserted window.

Fixed by anchoring the slice at id 33 via findIndex rather than a fixed tail length, so the assertion keeps covering Marcode's owned migration however many upstream ones land after it. Without that, the pin would have silently stopped testing its own subject on the next sync.

Coupled vendored subtrees

pnpm-workspace.yaml moved, but only Clerk versions; effect stayed at 4.0.0-beta.103 and infra/relay/package.json did not change. No sync:repos run was needed.

Why

Daily upstream sync. Upstream had 59 unmerged commits including a Codex approval-callback recovery fix, a projection-bootstrap replay fix (pingdotgg#7538), the macOS service PATH fix, Grok reliability work, and the Clerk 6.30/0.0.37 bumps.

UI Changes

Two decisions need your eye — I could not drive a browser or simulator for them in this container:

• Android release launcher icon. Upstream's fix pairs a transparent light mark with a dark plate. Marcode's release previously used the black-on-white universal export as the foreground, which is exactly the double-masking bug upstream fixed. Taking the fix means the release Android launcher is now a light mark on #000000, diverging from Marcode's black-on-white rule for that one surface. Restoring black-on-white needs a Marcode-owned foreground SVG/PNG pair; no rasterizer is available here to generate one. Flagging for your call — the alternative was keeping the double-mask bug or shipping a white mark on a white plate.
• vp run icons:check could not run: it needs macOS Icon Composer.

Checklist

• This PR is small and focused — no, it is a 59-commit upstream sync; that is the unit of work
• I explained what changed and why
• I included before/after screenshots for any UI changes — no browser/simulator available in this environment
• I included a video for animation/interaction changes — n/a

CI on 0a265857

All four required checks green: Check ✅ · Test ✅ · Mobile Native Static Analysis ✅ · Release Smoke ✅. Also green: Test Server 1/2/3, Rust, Mobile Native Changes, claude-review, and the thread-transfer budget report (every metric within its ceiling). Native fingerprint diff sits queued, as it did on the previous head; it is not a required check.

Local verification before pushing (focused, per AGENTS.md — not the full suite)

• vp test run green on: clientMetadata, sidebarBrandRemoval, mobileBranding, no-mobile-uniwind-theme-escape-hatches, bootService, ServerEnvironment, Migrations, 043_ProjectionThreadLinkedPullRequest, 042_AuthSessionClientConnection, build-desktop-artifact, contracts/environment, contracts/settings, terminal-links, editorLabels, Sidebar.logic, versionSkew, upstream-sync-workflow, serviceLauncher, electron-launcher, generate-uniwind-themes, connection/platform.
• vp run --filter … typecheck clean for @t3tools/contracts, @t3tools/web, @t3tools/shared, @t3tools/client-runtime, t3 (server), @t3tools/mobile, @t3tools/desktop (only pre-existing Effect style suggestions, no errors).
• vp fmt --check clean on all 356 changed TS files.
• vp lint on the 82 changed directories: no errors; only warnings that are pre-existing on main.
• generate-uniwind-themes.mts --check passes, so the generated theme CSS matches Marcode's tokens.

One thing to review deliberately

apps/server/src/provider/ModelManifest.ts is new from upstream and fetches legacy model classification from https://raw.githubusercontent.com/pingdotgg/t3code/main/apps/server/src/provider/model-manifest.json. I left the URL pointing at upstream — they maintain that data and Marcode benefits from their updates — but it does mean Marcode servers read a file from upstream's repo at runtime. Repointing it at maarco/marcode is a one-line change if you'd rather own it. The in-repo model-manifest.json is the fallback either way.

Left deliberately as a draft: this sync is yours to review and mark ready.

[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	13.4 KiB	13.3 KiB	−122 B (−0.9%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+5 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.5 KiB	6.4 KiB	−127 B (−1.9%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.0 KiB	55.6 KiB	+558 B (+1.0%)	66.4 KiB	✅
Codex	Live turn messages	16	10	−6 (−37.5%)	21	✅
Claude	Total thread wire	13.4 KiB	13.3 KiB	−156 B (−1.1%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	+8 B (+0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.6 KiB	6.4 KiB	−164 B (−2.4%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	55.8 KiB	56.4 KiB	+532 B (+0.9%)	66.4 KiB	✅
Claude	Live turn messages	16	9	−7 (−43.8%)	21	✅

Baseline: 133f3f7 · PR result: 0a26585 · 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.

[maarco]

Two status changes since the description was written.

Native fingerprint diff is now cancelled, and it is not this PR's failure. It sat queued for exactly 24h (2026-08-28 15:42:11 → 2026-08-29 15:42:11) and GitHub killed it on the job timeout. Every run of mobile-fingerprint-check.yml on this fork does the same: I checked the last 30 runs and not one has ever completed — claude/gracious-bohr-1dtucn, -rlxi8l, -ie7aa6, -23bj7a, -x80z9c, -23l2b6, back to at least 2026-08-23, all queued-then-cancelled. It is a pre-existing fork-environment condition (the job never starts; the blacksmith-8vcpu-ubuntu-2404 label itself is fine, since Check, Test, and Rust all run on it here), and it is not in requiredPullRequestChecks. I am deliberately not spending the one re-run on it: 30 runs across 6 branches is stronger evidence than a 31st would be, and a re-run would just occupy a runner slot for another 24h. It needs a repo-settings fix, not a change to this PR.

The four required checks remain green on 0a265857: Check ✅ · Test ✅ · Mobile Native Static Analysis ✅ · Release Smoke ✅.

This PR's upstream coverage is now superseded by #61 (053affbe), opened by today's scheduled run. I verified 018d7f27 is an ancestor of #61's head, so #61 covers everything here.

But #61 branched from main independently rather than building on this branch, so the two are not interchangeable:

• chore(upstream): sync marcode through 053affbe #61is currently red — Test Server 2 and Test both failing — because it reintroduces the Migrations.test.ts problem this PR fixed in 0a265857. Its registry ends at id 44 while its pin still asserts slice(-10) equals ids 33–42.
• chore(upstream): sync marcode through 053affbe #61did independently make the other two fork-boundary fixes (the APP_BASE_NAME client labels and the marcode/ oxlint prefixes), so those are not at risk either way.

So the migration-pin fix in 0a265857 is the one piece of this PR that #61 still needs. Cherry-picking that single commit onto #61 should get it green, at which point #61 is the better merge candidate and this PR can be closed alongside #53, #55, and #56.

I have not pushed to #61 — it is not a PR I opened, and this session was not asked to drive it.

---

Generated by Claude Code

[maarco]

What Changed

Merges 59 upstream commits, f035a0f4..018d7f27 (pingdotgg/t3code@018d7f27), into Marcode as one two-parent merge commit (a940ce5c), plus one follow-up commit (0a265857) fixing a Marcode-owned test the merge invalidated. No force push, reset, restore, stash, clean, or wholesale ours/theirs resolution was used; MERGE_HEAD stayed at 018d7f27 for the whole resolution.

Supersedes #53, #55, and #56 — those draft syncs targeted older upstream SHAs (a3a8cbd6, f6f2be32) that are contained in this one. Merging this closes the delta they cover; they can be closed unmerged.

Conflicts resolved (24 paths)

Path	Decision
apps/mobile/app.config.ts	Took upstream's adaptive-launcher-icon fix (pingdotgg#4332) — release now uses the shared transparent foreground instead of the rounded-square universal export Android was masking twice. See the branding note below.
apps/mobile/src/lib/authClientMetadata.ts	Kept "Marcode Mobile"; took upstream's tablet/phone deviceType detection.
apps/mobile/src/components/CompactBrandTitle.tsx	Kept MarcodeMark and the conditional stage badge; took upstream's Uniwind className styling over useThemeColor + inline styles (the new lint rule forbids them).
apps/mobile/src/components/T3Wordmark.tsx	Kept Marcode's deletion; upstream only restyled it.
apps/mobile/src/features/home/HomeHeader.tsx	Kept MarcodeMark.
apps/mobile/src/lib/mobileBranding.test.ts	Upstream deleted it (pingdotgg#8397). Kept — it pins Marcode returning "" (no badge) where upstream returns "Alpha". Now says so in a comment.
oxlint-plugin-marcode/rules/no-mobile-uniwind-theme-escape-hatches.*	Landed upstream's new rule under Marcode's plugin dir; renamed the test's rule prefix t3code/ → marcode/ to match the plugin meta name.
vite.config.ts	Enabled the new rule at upstream's "error"; kept Marcode's prefixes and the no-native-title-tooltip"warn" seam.
apps/server/src/cloud/bootService.ts + test	Took upstream's PATH export (pingdotgg#8173, keeps provider CLIs on the service path); kept MARCODE_HOME, not T3CODE_HOME. Both guard tests now stand side by side.
apps/server/src/environment/ServerEnvironment.ts, packages/contracts/src/environment.ts	Kept both capability flags: workspaceLayoutMutations and the new threadPullRequestLinking.
apps/server/src/persistence/Migrations.ts	Marcode ids sit one higher (033 is ProjectWorkspaceLayout). Renamed upstream's 042/043 → 043/044 and fixed the new migration test's hardcoded toMigrationInclusive ids.
apps/web/src/components/sidebar/SidebarChrome.tsx	Kept Marcode's bare Electron drag strip. Upstream's only change was a project-settings case for the sidebar Back button (pingdotgg#8168) — Marcode's FloatingPillNav already shows Back to Workspace on every non-thread route, so that intent is covered.
apps/web/src/components/threadSidebarWidth.test.ts	Accepted upstream's prune of the trivial width assertions, but moved the Marcode brand-removal pin into its own components/sidebar/sidebarBrandRemoval.test.ts so a future upstream prune can't take it out silently.
apps/web/src/components/ThreadTerminalDrawer.tsx	Kept Marcode's xterm link provider. Upstream's isTerminalUrl fix (pingdotgg#7488) lands in the shared terminal-links.ts, which Marcode's match.kind classification already goes through.
apps/web/src/components/chat/OpenInPicker.tsx	Kept Marcode's FolderOpenFilled icon; took upstream's move of labels to the shared editorLabelForPlatform.
apps/web/src/components/ChatMarkdown.tsx	Took upstream's editorPreferences/editorLabels/remoteOpen imports; kept openFileInFloatingEditor. The fork seam routing agent file links to the floating editor survives — upstream's right-panel file viewer (pingdotgg#8098) is not reintroduced.
apps/web/src/components/ChatView.tsx	Took upstream's hoisted useNewThreadHandler/useThreadActions (needed by the new settle/pin keyboard shortcuts) and removed the duplicate later declaration. Kept Marcode's slim ChatHeader; dropped upstream's openProjectPullRequest, unreachable here because openThreadPullRequest already falls back to the project repository.
pnpm-lock.yaml	Kept Marcode's xterm entries, then regenerated with pnpm install --lockfile-only from the merged manifests.

Fork-boundary breaks that merged cleanly (no conflict) and were fixed here

These are the dangerous ones — they produce no conflict marker:

1. apps/web/src/connection/clientMetadata.ts arrived new from upstream with hardcoded "T3 Code Desktop" / "T3 Code Web" labels and replaced Marcode's inline metadata builder in platform.ts. Every client would have reported itself as T3 Code. Routed both labels through APP_BASE_NAME and pinned "Marcode Desktop" in clientMetadata.test.ts.
2. Three upstream oxlint-disable-next-line t3code/no-global-process-runtime comments (apps/desktop/src/preload.ts, apps/desktop/scripts/verify-preload-bundle.mjs, apps/server/src/process/externalLauncher.test.ts). Marcode's plugin is named marcode, so those disables match nothing and the rule — set to "error" — would have failed the Check job. Repointed to marcode/.

Swept and confirmed correct for Marcode: T3CODE_HOME (only guard tests/comments), t3code.service / com.t3tools.t3code launchd label, WORKTREE_BRANCH_PREFIX ("marcode"; the t3code/... branch strings in server/checkpoint tests are pre-existing arbitrary fixtures, not prefix-derived), t3code: storage keys, T3_BOOT_SERVICE_UNIT, T3CODE_DESKTOP_REUSE_RESOURCE_MONITOR, @t3tools/*, and the t3 package/CLI name — all left upstream-shaped on purpose.

Follow-up commit 0a265857

CI caught one thing the focused local run did not: apps/server/src/persistence/Migrations.test.ts pins the migration registry with slice(-10), so appending the two renumbered upstream migrations pushed id 33 — the id the test is named for — out of the asserted window.

Fixed by anchoring the slice at id 33 via findIndex rather than a fixed tail length, so the assertion keeps covering Marcode's owned migration however many upstream ones land after it. Without that, the pin would have silently stopped testing its own subject on the next sync.

Coupled vendored subtrees

pnpm-workspace.yaml moved, but only Clerk versions; effect stayed at 4.0.0-beta.103 and infra/relay/package.json did not change. No sync:repos run was needed.

Why

Daily upstream sync. Upstream had 59 unmerged commits including a Codex approval-callback recovery fix, a projection-bootstrap replay fix (pingdotgg#7538), the macOS service PATH fix, Grok reliability work, and the Clerk 6.30/0.0.37 bumps.

UI Changes

Two decisions need your eye — I could not drive a browser or simulator for them in this container:

• Android release launcher icon. Upstream's fix pairs a transparent light mark with a dark plate. Marcode's release previously used the black-on-white universal export as the foreground, which is exactly the double-masking bug upstream fixed. Taking the fix means the release Android launcher is now a light mark on #000000, diverging from Marcode's black-on-white rule for that one surface. Restoring black-on-white needs a Marcode-owned foreground SVG/PNG pair; no rasterizer is available here to generate one. Flagging for your call — the alternative was keeping the double-mask bug or shipping a white mark on a white plate.
• vp run icons:check could not run: it needs macOS Icon Composer.

Checklist

• This PR is small and focused — no, it is a 59-commit upstream sync; that is the unit of work
• I explained what changed and why
• I included before/after screenshots for any UI changes — no browser/simulator available in this environment
• I included a video for animation/interaction changes — n/a

CI on 0a265857

All four required checks green: Check ✅ · Test ✅ · Mobile Native Static Analysis ✅ · Release Smoke ✅. Also green: Test Server 1/2/3, Rust, Mobile Native Changes, claude-review, and the thread-transfer budget report (every metric within its ceiling). Native fingerprint diff sits queued, as it did on the previous head; it is not a required check.

Local verification before pushing (focused, per AGENTS.md — not the full suite)

• vp test run green on: clientMetadata, sidebarBrandRemoval, mobileBranding, no-mobile-uniwind-theme-escape-hatches, bootService, ServerEnvironment, Migrations, 043_ProjectionThreadLinkedPullRequest, 042_AuthSessionClientConnection, build-desktop-artifact, contracts/environment, contracts/settings, terminal-links, editorLabels, Sidebar.logic, versionSkew, upstream-sync-workflow, serviceLauncher, electron-launcher, generate-uniwind-themes, connection/platform.
• vp run --filter … typecheck clean for @t3tools/contracts, @t3tools/web, @t3tools/shared, @t3tools/client-runtime, t3 (server), @t3tools/mobile, @t3tools/desktop (only pre-existing Effect style suggestions, no errors).
• vp fmt --check clean on all 356 changed TS files.
• vp lint on the 82 changed directories: no errors; only warnings that are pre-existing on main.
• generate-uniwind-themes.mts --check passes, so the generated theme CSS matches Marcode's tokens.

One thing to review deliberately

apps/server/src/provider/ModelManifest.ts is new from upstream and fetches legacy model classification from https://raw.githubusercontent.com/pingdotgg/t3code/main/apps/server/src/provider/model-manifest.json. I left the URL pointing at upstream — they maintain that data and Marcode benefits from their updates — but it does mean Marcode servers read a file from upstream's repo at runtime. Repointing it at maarco/marcode is a one-line change if you'd rather own it. The in-repo model-manifest.json is the fallback either way.

Left deliberately as a draft: this sync is yours to review and mark ready.

[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	13.4 KiB	13.3 KiB	−122 B (−0.9%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+5 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.5 KiB	6.4 KiB	−127 B (−1.9%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.0 KiB	55.6 KiB	+558 B (+1.0%)	66.4 KiB	✅
Codex	Live turn messages	16	10	−6 (−37.5%)	21	✅
Claude	Total thread wire	13.4 KiB	13.3 KiB	−156 B (−1.1%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	+8 B (+0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.6 KiB	6.4 KiB	−164 B (−2.4%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	55.8 KiB	56.4 KiB	+532 B (+0.9%)	66.4 KiB	✅
Claude	Live turn messages	16	9	−7 (−43.8%)	21	✅

Baseline: 133f3f7 · PR result: 0a26585 · 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.

[maarco]

Two status changes since the description was written.

Native fingerprint diff is now cancelled, and it is not this PR's failure. It sat queued for exactly 24h (2026-08-28 15:42:11 → 2026-08-29 15:42:11) and GitHub killed it on the job timeout. Every run of mobile-fingerprint-check.yml on this fork does the same: I checked the last 30 runs and not one has ever completed — claude/gracious-bohr-1dtucn, -rlxi8l, -ie7aa6, -23bj7a, -x80z9c, -23l2b6, back to at least 2026-08-23, all queued-then-cancelled. It is a pre-existing fork-environment condition (the job never starts; the blacksmith-8vcpu-ubuntu-2404 label itself is fine, since Check, Test, and Rust all run on it here), and it is not in requiredPullRequestChecks. I am deliberately not spending the one re-run on it: 30 runs across 6 branches is stronger evidence than a 31st would be, and a re-run would just occupy a runner slot for another 24h. It needs a repo-settings fix, not a change to this PR.

The four required checks remain green on 0a265857: Check ✅ · Test ✅ · Mobile Native Static Analysis ✅ · Release Smoke ✅.

This PR's upstream coverage is now superseded by #61 (053affbe), opened by today's scheduled run. I verified 018d7f27 is an ancestor of #61's head, so #61 covers everything here.

But #61 branched from main independently rather than building on this branch, so the two are not interchangeable:

• chore(upstream): sync marcode through 053affbe #61is currently red — Test Server 2 and Test both failing — because it reintroduces the Migrations.test.ts problem this PR fixed in 0a265857. Its registry ends at id 44 while its pin still asserts slice(-10) equals ids 33–42.
• chore(upstream): sync marcode through 053affbe #61did independently make the other two fork-boundary fixes (the APP_BASE_NAME client labels and the marcode/ oxlint prefixes), so those are not at risk either way.

So the migration-pin fix in 0a265857 is the one piece of this PR that #61 still needs. Cherry-picking that single commit onto #61 should get it green, at which point #61 is the better merge candidate and this PR can be closed alongside #53, #55, and #56.

I have not pushed to #61 — it is not a PR I opened, and this session was not asked to drive it.

---

Generated by Claude Code

[maarco]

What Changed

Merges 59 upstream commits, f035a0f4..018d7f27 (pingdotgg/t3code@018d7f27), into Marcode as one two-parent merge commit (a940ce5c), plus one follow-up commit (0a265857) fixing a Marcode-owned test the merge invalidated. No force push, reset, restore, stash, clean, or wholesale ours/theirs resolution was used; MERGE_HEAD stayed at 018d7f27 for the whole resolution.

Supersedes #53, #55, and #56 — those draft syncs targeted older upstream SHAs (a3a8cbd6, f6f2be32) that are contained in this one. Merging this closes the delta they cover; they can be closed unmerged.

Conflicts resolved (24 paths)

Path	Decision
apps/mobile/app.config.ts	Took upstream's adaptive-launcher-icon fix (pingdotgg#4332) — release now uses the shared transparent foreground instead of the rounded-square universal export Android was masking twice. See the branding note below.
apps/mobile/src/lib/authClientMetadata.ts	Kept "Marcode Mobile"; took upstream's tablet/phone deviceType detection.
apps/mobile/src/components/CompactBrandTitle.tsx	Kept MarcodeMark and the conditional stage badge; took upstream's Uniwind className styling over useThemeColor + inline styles (the new lint rule forbids them).
apps/mobile/src/components/T3Wordmark.tsx	Kept Marcode's deletion; upstream only restyled it.
apps/mobile/src/features/home/HomeHeader.tsx	Kept MarcodeMark.
apps/mobile/src/lib/mobileBranding.test.ts	Upstream deleted it (pingdotgg#8397). Kept — it pins Marcode returning "" (no badge) where upstream returns "Alpha". Now says so in a comment.
oxlint-plugin-marcode/rules/no-mobile-uniwind-theme-escape-hatches.*	Landed upstream's new rule under Marcode's plugin dir; renamed the test's rule prefix t3code/ → marcode/ to match the plugin meta name.
vite.config.ts	Enabled the new rule at upstream's "error"; kept Marcode's prefixes and the no-native-title-tooltip"warn" seam.
apps/server/src/cloud/bootService.ts + test	Took upstream's PATH export (pingdotgg#8173, keeps provider CLIs on the service path); kept MARCODE_HOME, not T3CODE_HOME. Both guard tests now stand side by side.
apps/server/src/environment/ServerEnvironment.ts, packages/contracts/src/environment.ts	Kept both capability flags: workspaceLayoutMutations and the new threadPullRequestLinking.
apps/server/src/persistence/Migrations.ts	Marcode ids sit one higher (033 is ProjectWorkspaceLayout). Renamed upstream's 042/043 → 043/044 and fixed the new migration test's hardcoded toMigrationInclusive ids.
apps/web/src/components/sidebar/SidebarChrome.tsx	Kept Marcode's bare Electron drag strip. Upstream's only change was a project-settings case for the sidebar Back button (pingdotgg#8168) — Marcode's FloatingPillNav already shows Back to Workspace on every non-thread route, so that intent is covered.
apps/web/src/components/threadSidebarWidth.test.ts	Accepted upstream's prune of the trivial width assertions, but moved the Marcode brand-removal pin into its own components/sidebar/sidebarBrandRemoval.test.ts so a future upstream prune can't take it out silently.
apps/web/src/components/ThreadTerminalDrawer.tsx	Kept Marcode's xterm link provider. Upstream's isTerminalUrl fix (pingdotgg#7488) lands in the shared terminal-links.ts, which Marcode's match.kind classification already goes through.
apps/web/src/components/chat/OpenInPicker.tsx	Kept Marcode's FolderOpenFilled icon; took upstream's move of labels to the shared editorLabelForPlatform.
apps/web/src/components/ChatMarkdown.tsx	Took upstream's editorPreferences/editorLabels/remoteOpen imports; kept openFileInFloatingEditor. The fork seam routing agent file links to the floating editor survives — upstream's right-panel file viewer (pingdotgg#8098) is not reintroduced.
apps/web/src/components/ChatView.tsx	Took upstream's hoisted useNewThreadHandler/useThreadActions (needed by the new settle/pin keyboard shortcuts) and removed the duplicate later declaration. Kept Marcode's slim ChatHeader; dropped upstream's openProjectPullRequest, unreachable here because openThreadPullRequest already falls back to the project repository.
pnpm-lock.yaml	Kept Marcode's xterm entries, then regenerated with pnpm install --lockfile-only from the merged manifests.

Fork-boundary breaks that merged cleanly (no conflict) and were fixed here

These are the dangerous ones — they produce no conflict marker:

1. apps/web/src/connection/clientMetadata.ts arrived new from upstream with hardcoded "T3 Code Desktop" / "T3 Code Web" labels and replaced Marcode's inline metadata builder in platform.ts. Every client would have reported itself as T3 Code. Routed both labels through APP_BASE_NAME and pinned "Marcode Desktop" in clientMetadata.test.ts.
2. Three upstream oxlint-disable-next-line t3code/no-global-process-runtime comments (apps/desktop/src/preload.ts, apps/desktop/scripts/verify-preload-bundle.mjs, apps/server/src/process/externalLauncher.test.ts). Marcode's plugin is named marcode, so those disables match nothing and the rule — set to "error" — would have failed the Check job. Repointed to marcode/.

Swept and confirmed correct for Marcode: T3CODE_HOME (only guard tests/comments), t3code.service / com.t3tools.t3code launchd label, WORKTREE_BRANCH_PREFIX ("marcode"; the t3code/... branch strings in server/checkpoint tests are pre-existing arbitrary fixtures, not prefix-derived), t3code: storage keys, T3_BOOT_SERVICE_UNIT, T3CODE_DESKTOP_REUSE_RESOURCE_MONITOR, @t3tools/*, and the t3 package/CLI name — all left upstream-shaped on purpose.

Follow-up commit 0a265857

CI caught one thing the focused local run did not: apps/server/src/persistence/Migrations.test.ts pins the migration registry with slice(-10), so appending the two renumbered upstream migrations pushed id 33 — the id the test is named for — out of the asserted window.

Fixed by anchoring the slice at id 33 via findIndex rather than a fixed tail length, so the assertion keeps covering Marcode's owned migration however many upstream ones land after it. Without that, the pin would have silently stopped testing its own subject on the next sync.

Coupled vendored subtrees

pnpm-workspace.yaml moved, but only Clerk versions; effect stayed at 4.0.0-beta.103 and infra/relay/package.json did not change. No sync:repos run was needed.

Why

Daily upstream sync. Upstream had 59 unmerged commits including a Codex approval-callback recovery fix, a projection-bootstrap replay fix (pingdotgg#7538), the macOS service PATH fix, Grok reliability work, and the Clerk 6.30/0.0.37 bumps.

UI Changes

Two decisions need your eye — I could not drive a browser or simulator for them in this container:

• Android release launcher icon. Upstream's fix pairs a transparent light mark with a dark plate. Marcode's release previously used the black-on-white universal export as the foreground, which is exactly the double-masking bug upstream fixed. Taking the fix means the release Android launcher is now a light mark on #000000, diverging from Marcode's black-on-white rule for that one surface. Restoring black-on-white needs a Marcode-owned foreground SVG/PNG pair; no rasterizer is available here to generate one. Flagging for your call — the alternative was keeping the double-mask bug or shipping a white mark on a white plate.
• vp run icons:check could not run: it needs macOS Icon Composer.

Checklist

• This PR is small and focused — no, it is a 59-commit upstream sync; that is the unit of work
• I explained what changed and why
• I included before/after screenshots for any UI changes — no browser/simulator available in this environment
• I included a video for animation/interaction changes — n/a

CI on 0a265857

All four required checks green: Check ✅ · Test ✅ · Mobile Native Static Analysis ✅ · Release Smoke ✅. Also green: Test Server 1/2/3, Rust, Mobile Native Changes, claude-review, and the thread-transfer budget report (every metric within its ceiling). Native fingerprint diff sits queued, as it did on the previous head; it is not a required check.

Local verification before pushing (focused, per AGENTS.md — not the full suite)

• vp test run green on: clientMetadata, sidebarBrandRemoval, mobileBranding, no-mobile-uniwind-theme-escape-hatches, bootService, ServerEnvironment, Migrations, 043_ProjectionThreadLinkedPullRequest, 042_AuthSessionClientConnection, build-desktop-artifact, contracts/environment, contracts/settings, terminal-links, editorLabels, Sidebar.logic, versionSkew, upstream-sync-workflow, serviceLauncher, electron-launcher, generate-uniwind-themes, connection/platform.
• vp run --filter … typecheck clean for @t3tools/contracts, @t3tools/web, @t3tools/shared, @t3tools/client-runtime, t3 (server), @t3tools/mobile, @t3tools/desktop (only pre-existing Effect style suggestions, no errors).
• vp fmt --check clean on all 356 changed TS files.
• vp lint on the 82 changed directories: no errors; only warnings that are pre-existing on main.
• generate-uniwind-themes.mts --check passes, so the generated theme CSS matches Marcode's tokens.

One thing to review deliberately

apps/server/src/provider/ModelManifest.ts is new from upstream and fetches legacy model classification from https://raw.githubusercontent.com/pingdotgg/t3code/main/apps/server/src/provider/model-manifest.json. I left the URL pointing at upstream — they maintain that data and Marcode benefits from their updates — but it does mean Marcode servers read a file from upstream's repo at runtime. Repointing it at maarco/marcode is a one-line change if you'd rather own it. The in-repo model-manifest.json is the fallback either way.

Left deliberately as a draft: this sync is yours to review and mark ready.

[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	13.4 KiB	13.3 KiB	−122 B (−0.9%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+5 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.5 KiB	6.4 KiB	−127 B (−1.9%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.0 KiB	55.6 KiB	+558 B (+1.0%)	66.4 KiB	✅
Codex	Live turn messages	16	10	−6 (−37.5%)	21	✅
Claude	Total thread wire	13.4 KiB	13.3 KiB	−156 B (−1.1%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	+8 B (+0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.6 KiB	6.4 KiB	−164 B (−2.4%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	55.8 KiB	56.4 KiB	+532 B (+0.9%)	66.4 KiB	✅
Claude	Live turn messages	16	9	−7 (−43.8%)	21	✅

Baseline: 133f3f7 · PR result: 0a26585 · 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.

[maarco]

Two status changes since the description was written.

Native fingerprint diff is now cancelled, and it is not this PR's failure. It sat queued for exactly 24h (2026-08-28 15:42:11 → 2026-08-29 15:42:11) and GitHub killed it on the job timeout. Every run of mobile-fingerprint-check.yml on this fork does the same: I checked the last 30 runs and not one has ever completed — claude/gracious-bohr-1dtucn, -rlxi8l, -ie7aa6, -23bj7a, -x80z9c, -23l2b6, back to at least 2026-08-23, all queued-then-cancelled. It is a pre-existing fork-environment condition (the job never starts; the blacksmith-8vcpu-ubuntu-2404 label itself is fine, since Check, Test, and Rust all run on it here), and it is not in requiredPullRequestChecks. I am deliberately not spending the one re-run on it: 30 runs across 6 branches is stronger evidence than a 31st would be, and a re-run would just occupy a runner slot for another 24h. It needs a repo-settings fix, not a change to this PR.

The four required checks remain green on 0a265857: Check ✅ · Test ✅ · Mobile Native Static Analysis ✅ · Release Smoke ✅.

This PR's upstream coverage is now superseded by #61 (053affbe), opened by today's scheduled run. I verified 018d7f27 is an ancestor of #61's head, so #61 covers everything here.

But #61 branched from main independently rather than building on this branch, so the two are not interchangeable:

• chore(upstream): sync marcode through 053affbe #61is currently red — Test Server 2 and Test both failing — because it reintroduces the Migrations.test.ts problem this PR fixed in 0a265857. Its registry ends at id 44 while its pin still asserts slice(-10) equals ids 33–42.
• chore(upstream): sync marcode through 053affbe #61did independently make the other two fork-boundary fixes (the APP_BASE_NAME client labels and the marcode/ oxlint prefixes), so those are not at risk either way.

So the migration-pin fix in 0a265857 is the one piece of this PR that #61 still needs. Cherry-picking that single commit onto #61 should get it green, at which point #61 is the better merge candidate and this PR can be closed alongside #53, #55, and #56.

I have not pushed to #61 — it is not a PR I opened, and this session was not asked to drive it.

---

Generated by Claude Code

[maarco]

What Changed

Merges 59 upstream commits, f035a0f4..018d7f27 (pingdotgg/t3code@018d7f27), into Marcode as one two-parent merge commit (a940ce5c), plus one follow-up commit (0a265857) fixing a Marcode-owned test the merge invalidated. No force push, reset, restore, stash, clean, or wholesale ours/theirs resolution was used; MERGE_HEAD stayed at 018d7f27 for the whole resolution.

Supersedes #53, #55, and #56 — those draft syncs targeted older upstream SHAs (a3a8cbd6, f6f2be32) that are contained in this one. Merging this closes the delta they cover; they can be closed unmerged.

Conflicts resolved (24 paths)

Path	Decision
apps/mobile/app.config.ts	Took upstream's adaptive-launcher-icon fix (pingdotgg#4332) — release now uses the shared transparent foreground instead of the rounded-square universal export Android was masking twice. See the branding note below.
apps/mobile/src/lib/authClientMetadata.ts	Kept "Marcode Mobile"; took upstream's tablet/phone deviceType detection.
apps/mobile/src/components/CompactBrandTitle.tsx	Kept MarcodeMark and the conditional stage badge; took upstream's Uniwind className styling over useThemeColor + inline styles (the new lint rule forbids them).
apps/mobile/src/components/T3Wordmark.tsx	Kept Marcode's deletion; upstream only restyled it.
apps/mobile/src/features/home/HomeHeader.tsx	Kept MarcodeMark.
apps/mobile/src/lib/mobileBranding.test.ts	Upstream deleted it (pingdotgg#8397). Kept — it pins Marcode returning "" (no badge) where upstream returns "Alpha". Now says so in a comment.
oxlint-plugin-marcode/rules/no-mobile-uniwind-theme-escape-hatches.*	Landed upstream's new rule under Marcode's plugin dir; renamed the test's rule prefix t3code/ → marcode/ to match the plugin meta name.
vite.config.ts	Enabled the new rule at upstream's "error"; kept Marcode's prefixes and the no-native-title-tooltip"warn" seam.
apps/server/src/cloud/bootService.ts + test	Took upstream's PATH export (pingdotgg#8173, keeps provider CLIs on the service path); kept MARCODE_HOME, not T3CODE_HOME. Both guard tests now stand side by side.
apps/server/src/environment/ServerEnvironment.ts, packages/contracts/src/environment.ts	Kept both capability flags: workspaceLayoutMutations and the new threadPullRequestLinking.
apps/server/src/persistence/Migrations.ts	Marcode ids sit one higher (033 is ProjectWorkspaceLayout). Renamed upstream's 042/043 → 043/044 and fixed the new migration test's hardcoded toMigrationInclusive ids.
apps/web/src/components/sidebar/SidebarChrome.tsx	Kept Marcode's bare Electron drag strip. Upstream's only change was a project-settings case for the sidebar Back button (pingdotgg#8168) — Marcode's FloatingPillNav already shows Back to Workspace on every non-thread route, so that intent is covered.
apps/web/src/components/threadSidebarWidth.test.ts	Accepted upstream's prune of the trivial width assertions, but moved the Marcode brand-removal pin into its own components/sidebar/sidebarBrandRemoval.test.ts so a future upstream prune can't take it out silently.
apps/web/src/components/ThreadTerminalDrawer.tsx	Kept Marcode's xterm link provider. Upstream's isTerminalUrl fix (pingdotgg#7488) lands in the shared terminal-links.ts, which Marcode's match.kind classification already goes through.
apps/web/src/components/chat/OpenInPicker.tsx	Kept Marcode's FolderOpenFilled icon; took upstream's move of labels to the shared editorLabelForPlatform.
apps/web/src/components/ChatMarkdown.tsx	Took upstream's editorPreferences/editorLabels/remoteOpen imports; kept openFileInFloatingEditor. The fork seam routing agent file links to the floating editor survives — upstream's right-panel file viewer (pingdotgg#8098) is not reintroduced.
apps/web/src/components/ChatView.tsx	Took upstream's hoisted useNewThreadHandler/useThreadActions (needed by the new settle/pin keyboard shortcuts) and removed the duplicate later declaration. Kept Marcode's slim ChatHeader; dropped upstream's openProjectPullRequest, unreachable here because openThreadPullRequest already falls back to the project repository.
pnpm-lock.yaml	Kept Marcode's xterm entries, then regenerated with pnpm install --lockfile-only from the merged manifests.

Fork-boundary breaks that merged cleanly (no conflict) and were fixed here

These are the dangerous ones — they produce no conflict marker:

1. apps/web/src/connection/clientMetadata.ts arrived new from upstream with hardcoded "T3 Code Desktop" / "T3 Code Web" labels and replaced Marcode's inline metadata builder in platform.ts. Every client would have reported itself as T3 Code. Routed both labels through APP_BASE_NAME and pinned "Marcode Desktop" in clientMetadata.test.ts.
2. Three upstream oxlint-disable-next-line t3code/no-global-process-runtime comments (apps/desktop/src/preload.ts, apps/desktop/scripts/verify-preload-bundle.mjs, apps/server/src/process/externalLauncher.test.ts). Marcode's plugin is named marcode, so those disables match nothing and the rule — set to "error" — would have failed the Check job. Repointed to marcode/.

Swept and confirmed correct for Marcode: T3CODE_HOME (only guard tests/comments), t3code.service / com.t3tools.t3code launchd label, WORKTREE_BRANCH_PREFIX ("marcode"; the t3code/... branch strings in server/checkpoint tests are pre-existing arbitrary fixtures, not prefix-derived), t3code: storage keys, T3_BOOT_SERVICE_UNIT, T3CODE_DESKTOP_REUSE_RESOURCE_MONITOR, @t3tools/*, and the t3 package/CLI name — all left upstream-shaped on purpose.

Follow-up commit 0a265857

CI caught one thing the focused local run did not: apps/server/src/persistence/Migrations.test.ts pins the migration registry with slice(-10), so appending the two renumbered upstream migrations pushed id 33 — the id the test is named for — out of the asserted window.

Fixed by anchoring the slice at id 33 via findIndex rather than a fixed tail length, so the assertion keeps covering Marcode's owned migration however many upstream ones land after it. Without that, the pin would have silently stopped testing its own subject on the next sync.

Coupled vendored subtrees

pnpm-workspace.yaml moved, but only Clerk versions; effect stayed at 4.0.0-beta.103 and infra/relay/package.json did not change. No sync:repos run was needed.

Why

Daily upstream sync. Upstream had 59 unmerged commits including a Codex approval-callback recovery fix, a projection-bootstrap replay fix (pingdotgg#7538), the macOS service PATH fix, Grok reliability work, and the Clerk 6.30/0.0.37 bumps.

UI Changes

Two decisions need your eye — I could not drive a browser or simulator for them in this container:

• Android release launcher icon. Upstream's fix pairs a transparent light mark with a dark plate. Marcode's release previously used the black-on-white universal export as the foreground, which is exactly the double-masking bug upstream fixed. Taking the fix means the release Android launcher is now a light mark on #000000, diverging from Marcode's black-on-white rule for that one surface. Restoring black-on-white needs a Marcode-owned foreground SVG/PNG pair; no rasterizer is available here to generate one. Flagging for your call — the alternative was keeping the double-mask bug or shipping a white mark on a white plate.
• vp run icons:check could not run: it needs macOS Icon Composer.

Checklist

• This PR is small and focused — no, it is a 59-commit upstream sync; that is the unit of work
• I explained what changed and why
• I included before/after screenshots for any UI changes — no browser/simulator available in this environment
• I included a video for animation/interaction changes — n/a

CI on 0a265857

All four required checks green: Check ✅ · Test ✅ · Mobile Native Static Analysis ✅ · Release Smoke ✅. Also green: Test Server 1/2/3, Rust, Mobile Native Changes, claude-review, and the thread-transfer budget report (every metric within its ceiling). Native fingerprint diff sits queued, as it did on the previous head; it is not a required check.

Local verification before pushing (focused, per AGENTS.md — not the full suite)

• vp test run green on: clientMetadata, sidebarBrandRemoval, mobileBranding, no-mobile-uniwind-theme-escape-hatches, bootService, ServerEnvironment, Migrations, 043_ProjectionThreadLinkedPullRequest, 042_AuthSessionClientConnection, build-desktop-artifact, contracts/environment, contracts/settings, terminal-links, editorLabels, Sidebar.logic, versionSkew, upstream-sync-workflow, serviceLauncher, electron-launcher, generate-uniwind-themes, connection/platform.
• vp run --filter … typecheck clean for @t3tools/contracts, @t3tools/web, @t3tools/shared, @t3tools/client-runtime, t3 (server), @t3tools/mobile, @t3tools/desktop (only pre-existing Effect style suggestions, no errors).
• vp fmt --check clean on all 356 changed TS files.
• vp lint on the 82 changed directories: no errors; only warnings that are pre-existing on main.
• generate-uniwind-themes.mts --check passes, so the generated theme CSS matches Marcode's tokens.

One thing to review deliberately

apps/server/src/provider/ModelManifest.ts is new from upstream and fetches legacy model classification from https://raw.githubusercontent.com/pingdotgg/t3code/main/apps/server/src/provider/model-manifest.json. I left the URL pointing at upstream — they maintain that data and Marcode benefits from their updates — but it does mean Marcode servers read a file from upstream's repo at runtime. Repointing it at maarco/marcode is a one-line change if you'd rather own it. The in-repo model-manifest.json is the fallback either way.

Left deliberately as a draft: this sync is yours to review and mark ready.

[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	13.4 KiB	13.3 KiB	−122 B (−0.9%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+5 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.5 KiB	6.4 KiB	−127 B (−1.9%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.0 KiB	55.6 KiB	+558 B (+1.0%)	66.4 KiB	✅
Codex	Live turn messages	16	10	−6 (−37.5%)	21	✅
Claude	Total thread wire	13.4 KiB	13.3 KiB	−156 B (−1.1%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	+8 B (+0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.6 KiB	6.4 KiB	−164 B (−2.4%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	55.8 KiB	56.4 KiB	+532 B (+0.9%)	66.4 KiB	✅
Claude	Live turn messages	16	9	−7 (−43.8%)	21	✅

Baseline: 133f3f7 · PR result: 0a26585 · 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.

[maarco]

Two status changes since the description was written.

Native fingerprint diff is now cancelled, and it is not this PR's failure. It sat queued for exactly 24h (2026-08-28 15:42:11 → 2026-08-29 15:42:11) and GitHub killed it on the job timeout. Every run of mobile-fingerprint-check.yml on this fork does the same: I checked the last 30 runs and not one has ever completed — claude/gracious-bohr-1dtucn, -rlxi8l, -ie7aa6, -23bj7a, -x80z9c, -23l2b6, back to at least 2026-08-23, all queued-then-cancelled. It is a pre-existing fork-environment condition (the job never starts; the blacksmith-8vcpu-ubuntu-2404 label itself is fine, since Check, Test, and Rust all run on it here), and it is not in requiredPullRequestChecks. I am deliberately not spending the one re-run on it: 30 runs across 6 branches is stronger evidence than a 31st would be, and a re-run would just occupy a runner slot for another 24h. It needs a repo-settings fix, not a change to this PR.

The four required checks remain green on 0a265857: Check ✅ · Test ✅ · Mobile Native Static Analysis ✅ · Release Smoke ✅.

This PR's upstream coverage is now superseded by #61 (053affbe), opened by today's scheduled run. I verified 018d7f27 is an ancestor of #61's head, so #61 covers everything here.

But #61 branched from main independently rather than building on this branch, so the two are not interchangeable:

• chore(upstream): sync marcode through 053affbe #61is currently red — Test Server 2 and Test both failing — because it reintroduces the Migrations.test.ts problem this PR fixed in 0a265857. Its registry ends at id 44 while its pin still asserts slice(-10) equals ids 33–42.
• chore(upstream): sync marcode through 053affbe #61did independently make the other two fork-boundary fixes (the APP_BASE_NAME client labels and the marcode/ oxlint prefixes), so those are not at risk either way.

So the migration-pin fix in 0a265857 is the one piece of this PR that #61 still needs. Cherry-picking that single commit onto #61 should get it green, at which point #61 is the better merge candidate and this PR can be closed alongside #53, #55, and #56.

I have not pushed to #61 — it is not a PR I opened, and this session was not asked to drive it.

---

Generated by Claude Code

[maarco]

What Changed

Merges 59 upstream commits, f035a0f4..018d7f27 (pingdotgg/t3code@018d7f27), into Marcode as one two-parent merge commit (a940ce5c), plus one follow-up commit (0a265857) fixing a Marcode-owned test the merge invalidated. No force push, reset, restore, stash, clean, or wholesale ours/theirs resolution was used; MERGE_HEAD stayed at 018d7f27 for the whole resolution.

Supersedes #53, #55, and #56 — those draft syncs targeted older upstream SHAs (a3a8cbd6, f6f2be32) that are contained in this one. Merging this closes the delta they cover; they can be closed unmerged.

Conflicts resolved (24 paths)

Path	Decision
apps/mobile/app.config.ts	Took upstream's adaptive-launcher-icon fix (pingdotgg#4332) — release now uses the shared transparent foreground instead of the rounded-square universal export Android was masking twice. See the branding note below.
apps/mobile/src/lib/authClientMetadata.ts	Kept "Marcode Mobile"; took upstream's tablet/phone deviceType detection.
apps/mobile/src/components/CompactBrandTitle.tsx	Kept MarcodeMark and the conditional stage badge; took upstream's Uniwind className styling over useThemeColor + inline styles (the new lint rule forbids them).
apps/mobile/src/components/T3Wordmark.tsx	Kept Marcode's deletion; upstream only restyled it.
apps/mobile/src/features/home/HomeHeader.tsx	Kept MarcodeMark.
apps/mobile/src/lib/mobileBranding.test.ts	Upstream deleted it (pingdotgg#8397). Kept — it pins Marcode returning "" (no badge) where upstream returns "Alpha". Now says so in a comment.
oxlint-plugin-marcode/rules/no-mobile-uniwind-theme-escape-hatches.*	Landed upstream's new rule under Marcode's plugin dir; renamed the test's rule prefix t3code/ → marcode/ to match the plugin meta name.
vite.config.ts	Enabled the new rule at upstream's "error"; kept Marcode's prefixes and the no-native-title-tooltip"warn" seam.
apps/server/src/cloud/bootService.ts + test	Took upstream's PATH export (pingdotgg#8173, keeps provider CLIs on the service path); kept MARCODE_HOME, not T3CODE_HOME. Both guard tests now stand side by side.
apps/server/src/environment/ServerEnvironment.ts, packages/contracts/src/environment.ts	Kept both capability flags: workspaceLayoutMutations and the new threadPullRequestLinking.
apps/server/src/persistence/Migrations.ts	Marcode ids sit one higher (033 is ProjectWorkspaceLayout). Renamed upstream's 042/043 → 043/044 and fixed the new migration test's hardcoded toMigrationInclusive ids.
apps/web/src/components/sidebar/SidebarChrome.tsx	Kept Marcode's bare Electron drag strip. Upstream's only change was a project-settings case for the sidebar Back button (pingdotgg#8168) — Marcode's FloatingPillNav already shows Back to Workspace on every non-thread route, so that intent is covered.
apps/web/src/components/threadSidebarWidth.test.ts	Accepted upstream's prune of the trivial width assertions, but moved the Marcode brand-removal pin into its own components/sidebar/sidebarBrandRemoval.test.ts so a future upstream prune can't take it out silently.
apps/web/src/components/ThreadTerminalDrawer.tsx	Kept Marcode's xterm link provider. Upstream's isTerminalUrl fix (pingdotgg#7488) lands in the shared terminal-links.ts, which Marcode's match.kind classification already goes through.
apps/web/src/components/chat/OpenInPicker.tsx	Kept Marcode's FolderOpenFilled icon; took upstream's move of labels to the shared editorLabelForPlatform.
apps/web/src/components/ChatMarkdown.tsx	Took upstream's editorPreferences/editorLabels/remoteOpen imports; kept openFileInFloatingEditor. The fork seam routing agent file links to the floating editor survives — upstream's right-panel file viewer (pingdotgg#8098) is not reintroduced.
apps/web/src/components/ChatView.tsx	Took upstream's hoisted useNewThreadHandler/useThreadActions (needed by the new settle/pin keyboard shortcuts) and removed the duplicate later declaration. Kept Marcode's slim ChatHeader; dropped upstream's openProjectPullRequest, unreachable here because openThreadPullRequest already falls back to the project repository.
pnpm-lock.yaml	Kept Marcode's xterm entries, then regenerated with pnpm install --lockfile-only from the merged manifests.

Fork-boundary breaks that merged cleanly (no conflict) and were fixed here

These are the dangerous ones — they produce no conflict marker:

1. apps/web/src/connection/clientMetadata.ts arrived new from upstream with hardcoded "T3 Code Desktop" / "T3 Code Web" labels and replaced Marcode's inline metadata builder in platform.ts. Every client would have reported itself as T3 Code. Routed both labels through APP_BASE_NAME and pinned "Marcode Desktop" in clientMetadata.test.ts.
2. Three upstream oxlint-disable-next-line t3code/no-global-process-runtime comments (apps/desktop/src/preload.ts, apps/desktop/scripts/verify-preload-bundle.mjs, apps/server/src/process/externalLauncher.test.ts). Marcode's plugin is named marcode, so those disables match nothing and the rule — set to "error" — would have failed the Check job. Repointed to marcode/.

Swept and confirmed correct for Marcode: T3CODE_HOME (only guard tests/comments), t3code.service / com.t3tools.t3code launchd label, WORKTREE_BRANCH_PREFIX ("marcode"; the t3code/... branch strings in server/checkpoint tests are pre-existing arbitrary fixtures, not prefix-derived), t3code: storage keys, T3_BOOT_SERVICE_UNIT, T3CODE_DESKTOP_REUSE_RESOURCE_MONITOR, @t3tools/*, and the t3 package/CLI name — all left upstream-shaped on purpose.

Follow-up commit 0a265857

CI caught one thing the focused local run did not: apps/server/src/persistence/Migrations.test.ts pins the migration registry with slice(-10), so appending the two renumbered upstream migrations pushed id 33 — the id the test is named for — out of the asserted window.

Fixed by anchoring the slice at id 33 via findIndex rather than a fixed tail length, so the assertion keeps covering Marcode's owned migration however many upstream ones land after it. Without that, the pin would have silently stopped testing its own subject on the next sync.

Coupled vendored subtrees

pnpm-workspace.yaml moved, but only Clerk versions; effect stayed at 4.0.0-beta.103 and infra/relay/package.json did not change. No sync:repos run was needed.

Why

Daily upstream sync. Upstream had 59 unmerged commits including a Codex approval-callback recovery fix, a projection-bootstrap replay fix (pingdotgg#7538), the macOS service PATH fix, Grok reliability work, and the Clerk 6.30/0.0.37 bumps.

UI Changes

Two decisions need your eye — I could not drive a browser or simulator for them in this container:

• Android release launcher icon. Upstream's fix pairs a transparent light mark with a dark plate. Marcode's release previously used the black-on-white universal export as the foreground, which is exactly the double-masking bug upstream fixed. Taking the fix means the release Android launcher is now a light mark on #000000, diverging from Marcode's black-on-white rule for that one surface. Restoring black-on-white needs a Marcode-owned foreground SVG/PNG pair; no rasterizer is available here to generate one. Flagging for your call — the alternative was keeping the double-mask bug or shipping a white mark on a white plate.
• vp run icons:check could not run: it needs macOS Icon Composer.

Checklist

• This PR is small and focused — no, it is a 59-commit upstream sync; that is the unit of work
• I explained what changed and why
• I included before/after screenshots for any UI changes — no browser/simulator available in this environment
• I included a video for animation/interaction changes — n/a

CI on 0a265857

All four required checks green: Check ✅ · Test ✅ · Mobile Native Static Analysis ✅ · Release Smoke ✅. Also green: Test Server 1/2/3, Rust, Mobile Native Changes, claude-review, and the thread-transfer budget report (every metric within its ceiling). Native fingerprint diff sits queued, as it did on the previous head; it is not a required check.

Local verification before pushing (focused, per AGENTS.md — not the full suite)

• vp test run green on: clientMetadata, sidebarBrandRemoval, mobileBranding, no-mobile-uniwind-theme-escape-hatches, bootService, ServerEnvironment, Migrations, 043_ProjectionThreadLinkedPullRequest, 042_AuthSessionClientConnection, build-desktop-artifact, contracts/environment, contracts/settings, terminal-links, editorLabels, Sidebar.logic, versionSkew, upstream-sync-workflow, serviceLauncher, electron-launcher, generate-uniwind-themes, connection/platform.
• vp run --filter … typecheck clean for @t3tools/contracts, @t3tools/web, @t3tools/shared, @t3tools/client-runtime, t3 (server), @t3tools/mobile, @t3tools/desktop (only pre-existing Effect style suggestions, no errors).
• vp fmt --check clean on all 356 changed TS files.
• vp lint on the 82 changed directories: no errors; only warnings that are pre-existing on main.
• generate-uniwind-themes.mts --check passes, so the generated theme CSS matches Marcode's tokens.

One thing to review deliberately

apps/server/src/provider/ModelManifest.ts is new from upstream and fetches legacy model classification from https://raw.githubusercontent.com/pingdotgg/t3code/main/apps/server/src/provider/model-manifest.json. I left the URL pointing at upstream — they maintain that data and Marcode benefits from their updates — but it does mean Marcode servers read a file from upstream's repo at runtime. Repointing it at maarco/marcode is a one-line change if you'd rather own it. The in-repo model-manifest.json is the fallback either way.

Left deliberately as a draft: this sync is yours to review and mark ready.

[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	13.4 KiB	13.3 KiB	−122 B (−0.9%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+5 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.5 KiB	6.4 KiB	−127 B (−1.9%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.0 KiB	55.6 KiB	+558 B (+1.0%)	66.4 KiB	✅
Codex	Live turn messages	16	10	−6 (−37.5%)	21	✅
Claude	Total thread wire	13.4 KiB	13.3 KiB	−156 B (−1.1%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	+8 B (+0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.6 KiB	6.4 KiB	−164 B (−2.4%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	55.8 KiB	56.4 KiB	+532 B (+0.9%)	66.4 KiB	✅
Claude	Live turn messages	16	9	−7 (−43.8%)	21	✅

Baseline: 133f3f7 · PR result: 0a26585 · 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.

[maarco]

Two status changes since the description was written.

Native fingerprint diff is now cancelled, and it is not this PR's failure. It sat queued for exactly 24h (2026-08-28 15:42:11 → 2026-08-29 15:42:11) and GitHub killed it on the job timeout. Every run of mobile-fingerprint-check.yml on this fork does the same: I checked the last 30 runs and not one has ever completed — claude/gracious-bohr-1dtucn, -rlxi8l, -ie7aa6, -23bj7a, -x80z9c, -23l2b6, back to at least 2026-08-23, all queued-then-cancelled. It is a pre-existing fork-environment condition (the job never starts; the blacksmith-8vcpu-ubuntu-2404 label itself is fine, since Check, Test, and Rust all run on it here), and it is not in requiredPullRequestChecks. I am deliberately not spending the one re-run on it: 30 runs across 6 branches is stronger evidence than a 31st would be, and a re-run would just occupy a runner slot for another 24h. It needs a repo-settings fix, not a change to this PR.

The four required checks remain green on 0a265857: Check ✅ · Test ✅ · Mobile Native Static Analysis ✅ · Release Smoke ✅.

This PR's upstream coverage is now superseded by #61 (053affbe), opened by today's scheduled run. I verified 018d7f27 is an ancestor of #61's head, so #61 covers everything here.

But #61 branched from main independently rather than building on this branch, so the two are not interchangeable:

• chore(upstream): sync marcode through 053affbe #61is currently red — Test Server 2 and Test both failing — because it reintroduces the Migrations.test.ts problem this PR fixed in 0a265857. Its registry ends at id 44 while its pin still asserts slice(-10) equals ids 33–42.
• chore(upstream): sync marcode through 053affbe #61did independently make the other two fork-boundary fixes (the APP_BASE_NAME client labels and the marcode/ oxlint prefixes), so those are not at risk either way.

So the migration-pin fix in 0a265857 is the one piece of this PR that #61 still needs. Cherry-picking that single commit onto #61 should get it green, at which point #61 is the better merge candidate and this PR can be closed alongside #53, #55, and #56.

I have not pushed to #61 — it is not a PR I opened, and this session was not asked to drive it.

---

Generated by Claude Code

[maarco]

What Changed

Merges 59 upstream commits, f035a0f4..018d7f27 (pingdotgg/t3code@018d7f27), into Marcode as one two-parent merge commit (a940ce5c), plus one follow-up commit (0a265857) fixing a Marcode-owned test the merge invalidated. No force push, reset, restore, stash, clean, or wholesale ours/theirs resolution was used; MERGE_HEAD stayed at 018d7f27 for the whole resolution.

Supersedes #53, #55, and #56 — those draft syncs targeted older upstream SHAs (a3a8cbd6, f6f2be32) that are contained in this one. Merging this closes the delta they cover; they can be closed unmerged.

Conflicts resolved (24 paths)

Path	Decision
apps/mobile/app.config.ts	Took upstream's adaptive-launcher-icon fix (pingdotgg#4332) — release now uses the shared transparent foreground instead of the rounded-square universal export Android was masking twice. See the branding note below.
apps/mobile/src/lib/authClientMetadata.ts	Kept "Marcode Mobile"; took upstream's tablet/phone deviceType detection.
apps/mobile/src/components/CompactBrandTitle.tsx	Kept MarcodeMark and the conditional stage badge; took upstream's Uniwind className styling over useThemeColor + inline styles (the new lint rule forbids them).
apps/mobile/src/components/T3Wordmark.tsx	Kept Marcode's deletion; upstream only restyled it.
apps/mobile/src/features/home/HomeHeader.tsx	Kept MarcodeMark.
apps/mobile/src/lib/mobileBranding.test.ts	Upstream deleted it (pingdotgg#8397). Kept — it pins Marcode returning "" (no badge) where upstream returns "Alpha". Now says so in a comment.
oxlint-plugin-marcode/rules/no-mobile-uniwind-theme-escape-hatches.*	Landed upstream's new rule under Marcode's plugin dir; renamed the test's rule prefix t3code/ → marcode/ to match the plugin meta name.
vite.config.ts	Enabled the new rule at upstream's "error"; kept Marcode's prefixes and the no-native-title-tooltip"warn" seam.
apps/server/src/cloud/bootService.ts + test	Took upstream's PATH export (pingdotgg#8173, keeps provider CLIs on the service path); kept MARCODE_HOME, not T3CODE_HOME. Both guard tests now stand side by side.
apps/server/src/environment/ServerEnvironment.ts, packages/contracts/src/environment.ts	Kept both capability flags: workspaceLayoutMutations and the new threadPullRequestLinking.
apps/server/src/persistence/Migrations.ts	Marcode ids sit one higher (033 is ProjectWorkspaceLayout). Renamed upstream's 042/043 → 043/044 and fixed the new migration test's hardcoded toMigrationInclusive ids.
apps/web/src/components/sidebar/SidebarChrome.tsx	Kept Marcode's bare Electron drag strip. Upstream's only change was a project-settings case for the sidebar Back button (pingdotgg#8168) — Marcode's FloatingPillNav already shows Back to Workspace on every non-thread route, so that intent is covered.
apps/web/src/components/threadSidebarWidth.test.ts	Accepted upstream's prune of the trivial width assertions, but moved the Marcode brand-removal pin into its own components/sidebar/sidebarBrandRemoval.test.ts so a future upstream prune can't take it out silently.
apps/web/src/components/ThreadTerminalDrawer.tsx	Kept Marcode's xterm link provider. Upstream's isTerminalUrl fix (pingdotgg#7488) lands in the shared terminal-links.ts, which Marcode's match.kind classification already goes through.
apps/web/src/components/chat/OpenInPicker.tsx	Kept Marcode's FolderOpenFilled icon; took upstream's move of labels to the shared editorLabelForPlatform.
apps/web/src/components/ChatMarkdown.tsx	Took upstream's editorPreferences/editorLabels/remoteOpen imports; kept openFileInFloatingEditor. The fork seam routing agent file links to the floating editor survives — upstream's right-panel file viewer (pingdotgg#8098) is not reintroduced.
apps/web/src/components/ChatView.tsx	Took upstream's hoisted useNewThreadHandler/useThreadActions (needed by the new settle/pin keyboard shortcuts) and removed the duplicate later declaration. Kept Marcode's slim ChatHeader; dropped upstream's openProjectPullRequest, unreachable here because openThreadPullRequest already falls back to the project repository.
pnpm-lock.yaml	Kept Marcode's xterm entries, then regenerated with pnpm install --lockfile-only from the merged manifests.

Fork-boundary breaks that merged cleanly (no conflict) and were fixed here

These are the dangerous ones — they produce no conflict marker:

1. apps/web/src/connection/clientMetadata.ts arrived new from upstream with hardcoded "T3 Code Desktop" / "T3 Code Web" labels and replaced Marcode's inline metadata builder in platform.ts. Every client would have reported itself as T3 Code. Routed both labels through APP_BASE_NAME and pinned "Marcode Desktop" in clientMetadata.test.ts.
2. Three upstream oxlint-disable-next-line t3code/no-global-process-runtime comments (apps/desktop/src/preload.ts, apps/desktop/scripts/verify-preload-bundle.mjs, apps/server/src/process/externalLauncher.test.ts). Marcode's plugin is named marcode, so those disables match nothing and the rule — set to "error" — would have failed the Check job. Repointed to marcode/.

Swept and confirmed correct for Marcode: T3CODE_HOME (only guard tests/comments), t3code.service / com.t3tools.t3code launchd label, WORKTREE_BRANCH_PREFIX ("marcode"; the t3code/... branch strings in server/checkpoint tests are pre-existing arbitrary fixtures, not prefix-derived), t3code: storage keys, T3_BOOT_SERVICE_UNIT, T3CODE_DESKTOP_REUSE_RESOURCE_MONITOR, @t3tools/*, and the t3 package/CLI name — all left upstream-shaped on purpose.

Follow-up commit 0a265857

CI caught one thing the focused local run did not: apps/server/src/persistence/Migrations.test.ts pins the migration registry with slice(-10), so appending the two renumbered upstream migrations pushed id 33 — the id the test is named for — out of the asserted window.

Fixed by anchoring the slice at id 33 via findIndex rather than a fixed tail length, so the assertion keeps covering Marcode's owned migration however many upstream ones land after it. Without that, the pin would have silently stopped testing its own subject on the next sync.

Coupled vendored subtrees

pnpm-workspace.yaml moved, but only Clerk versions; effect stayed at 4.0.0-beta.103 and infra/relay/package.json did not change. No sync:repos run was needed.

Why

Daily upstream sync. Upstream had 59 unmerged commits including a Codex approval-callback recovery fix, a projection-bootstrap replay fix (pingdotgg#7538), the macOS service PATH fix, Grok reliability work, and the Clerk 6.30/0.0.37 bumps.

UI Changes

Two decisions need your eye — I could not drive a browser or simulator for them in this container:

• Android release launcher icon. Upstream's fix pairs a transparent light mark with a dark plate. Marcode's release previously used the black-on-white universal export as the foreground, which is exactly the double-masking bug upstream fixed. Taking the fix means the release Android launcher is now a light mark on #000000, diverging from Marcode's black-on-white rule for that one surface. Restoring black-on-white needs a Marcode-owned foreground SVG/PNG pair; no rasterizer is available here to generate one. Flagging for your call — the alternative was keeping the double-mask bug or shipping a white mark on a white plate.
• vp run icons:check could not run: it needs macOS Icon Composer.

Checklist

• This PR is small and focused — no, it is a 59-commit upstream sync; that is the unit of work
• I explained what changed and why
• I included before/after screenshots for any UI changes — no browser/simulator available in this environment
• I included a video for animation/interaction changes — n/a

CI on 0a265857

All four required checks green: Check ✅ · Test ✅ · Mobile Native Static Analysis ✅ · Release Smoke ✅. Also green: Test Server 1/2/3, Rust, Mobile Native Changes, claude-review, and the thread-transfer budget report (every metric within its ceiling). Native fingerprint diff sits queued, as it did on the previous head; it is not a required check.

Local verification before pushing (focused, per AGENTS.md — not the full suite)

• vp test run green on: clientMetadata, sidebarBrandRemoval, mobileBranding, no-mobile-uniwind-theme-escape-hatches, bootService, ServerEnvironment, Migrations, 043_ProjectionThreadLinkedPullRequest, 042_AuthSessionClientConnection, build-desktop-artifact, contracts/environment, contracts/settings, terminal-links, editorLabels, Sidebar.logic, versionSkew, upstream-sync-workflow, serviceLauncher, electron-launcher, generate-uniwind-themes, connection/platform.
• vp run --filter … typecheck clean for @t3tools/contracts, @t3tools/web, @t3tools/shared, @t3tools/client-runtime, t3 (server), @t3tools/mobile, @t3tools/desktop (only pre-existing Effect style suggestions, no errors).
• vp fmt --check clean on all 356 changed TS files.
• vp lint on the 82 changed directories: no errors; only warnings that are pre-existing on main.
• generate-uniwind-themes.mts --check passes, so the generated theme CSS matches Marcode's tokens.

One thing to review deliberately

apps/server/src/provider/ModelManifest.ts is new from upstream and fetches legacy model classification from https://raw.githubusercontent.com/pingdotgg/t3code/main/apps/server/src/provider/model-manifest.json. I left the URL pointing at upstream — they maintain that data and Marcode benefits from their updates — but it does mean Marcode servers read a file from upstream's repo at runtime. Repointing it at maarco/marcode is a one-line change if you'd rather own it. The in-repo model-manifest.json is the fallback either way.

Left deliberately as a draft: this sync is yours to review and mark ready.

[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	13.4 KiB	13.3 KiB	−122 B (−0.9%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+5 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.5 KiB	6.4 KiB	−127 B (−1.9%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.0 KiB	55.6 KiB	+558 B (+1.0%)	66.4 KiB	✅
Codex	Live turn messages	16	10	−6 (−37.5%)	21	✅
Claude	Total thread wire	13.4 KiB	13.3 KiB	−156 B (−1.1%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	+8 B (+0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.6 KiB	6.4 KiB	−164 B (−2.4%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	55.8 KiB	56.4 KiB	+532 B (+0.9%)	66.4 KiB	✅
Claude	Live turn messages	16	9	−7 (−43.8%)	21	✅

Baseline: 133f3f7 · PR result: 0a26585 · 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.

[maarco]

Two status changes since the description was written.

Native fingerprint diff is now cancelled, and it is not this PR's failure. It sat queued for exactly 24h (2026-08-28 15:42:11 → 2026-08-29 15:42:11) and GitHub killed it on the job timeout. Every run of mobile-fingerprint-check.yml on this fork does the same: I checked the last 30 runs and not one has ever completed — claude/gracious-bohr-1dtucn, -rlxi8l, -ie7aa6, -23bj7a, -x80z9c, -23l2b6, back to at least 2026-08-23, all queued-then-cancelled. It is a pre-existing fork-environment condition (the job never starts; the blacksmith-8vcpu-ubuntu-2404 label itself is fine, since Check, Test, and Rust all run on it here), and it is not in requiredPullRequestChecks. I am deliberately not spending the one re-run on it: 30 runs across 6 branches is stronger evidence than a 31st would be, and a re-run would just occupy a runner slot for another 24h. It needs a repo-settings fix, not a change to this PR.

The four required checks remain green on 0a265857: Check ✅ · Test ✅ · Mobile Native Static Analysis ✅ · Release Smoke ✅.

This PR's upstream coverage is now superseded by #61 (053affbe), opened by today's scheduled run. I verified 018d7f27 is an ancestor of #61's head, so #61 covers everything here.

But #61 branched from main independently rather than building on this branch, so the two are not interchangeable:

• chore(upstream): sync marcode through 053affbe #61is currently red — Test Server 2 and Test both failing — because it reintroduces the Migrations.test.ts problem this PR fixed in 0a265857. Its registry ends at id 44 while its pin still asserts slice(-10) equals ids 33–42.
• chore(upstream): sync marcode through 053affbe #61did independently make the other two fork-boundary fixes (the APP_BASE_NAME client labels and the marcode/ oxlint prefixes), so those are not at risk either way.

So the migration-pin fix in 0a265857 is the one piece of this PR that #61 still needs. Cherry-picking that single commit onto #61 should get it green, at which point #61 is the better merge candidate and this PR can be closed alongside #53, #55, and #56.

I have not pushed to #61 — it is not a PR I opened, and this session was not asked to drive it.

---

Generated by Claude Code