chore: merge upstream pingdotgg/t3code

[andrewcai8]

Merges 19 upstream commits through pingdotgg/t3code@b33eda1399. Merge with Create a merge commit, never squash. A squash drops upstream's history, and then the next sync conflicts on every file again.

Upstream changes

• chore(mobile): upgrade to Expo SDK 58 and React Native 0.88 RC pingdotgg/t3code#12045. Mobile moves to Expo SDK 58 and React Native 0.88.0-rc.3, with React 19.3, Reanimated 4.7, and Screens 4.28. SDK 58 prebuild now generates the iOS scene delegate, so the custom scene plugin is gone. This changes the native fingerprint, so installed builds need a new native build (see below).
• feat(mobile): render the Android subscription widget with expo-widgets pingdotgg/t3code#12046, feat(mobile): degrade the agent Live Activity once its content goes stale pingdotgg/t3code#12047, feat(notifications): stack agent alerts by thread on both platforms pingdotgg/t3code#12048, chore(mobile): drive dev-menu suppression from the dev-client launch URL pingdotgg/t3code#12049, fix(mobile): read display scale and width from the view's scene, not UIScreen.main pingdotgg/t3code#12050, refactor(mobile): adopt Expo Modules 2.0 for function-only native members pingdotgg/t3code#12051, feat(mobile): suppress only the on-screen thread's alert on Android pingdotgg/t3code#12052. The Android subscription widget renders through expo-widgets. Locally started Live Activities go stale after ten minutes. Agent alerts stack by thread on both platforms. Android silences only the alert for the thread on screen. The stack also includes the dev-menu, scene geometry, and Expo Modules 2.0 follow-ups.
• feat(web): update providers on every machine with one click pingdotgg/t3code#14678. Settings → Providers has Update all, which updates every outdated provider on every connected machine at once.
• fix(server): background GitHub polling uses ~74% fewer calls with batched GraphQL pingdotgg/t3code#14673. Background GitHub polling batches branch lookups into GraphQL and reads the right rate-limit counter. Upstream measured about 74% fewer calls.
• feat(web): beta Working section hides busy threads until they need you pingdotgg/t3code#13926. A beta Working section (Settings → General) folds working and monitoring threads away until they need you. It is off by default.
• fix: cloned projects show their favicon instead of a monogram pingdotgg/t3code#14635. Cloned projects show their favicon once the clone lands, instead of a monogram.
• fix(web): show the Docker icon on dockerfile code blocks pingdotgg/t3code#14675, fix(ci): pin eas-cli so mobile PR previews deploy again pingdotgg/t3code#14659, docs: server features are services, and handlers stay thin pingdotgg/t3code#14613, test(web): remove duplicate sidebar ordering tests pingdotgg/t3code#14558, docs: update user count in AGENTS.md pingdotgg/t3code#11413, feat(marketing): replace the screenshot crop with a designed social card pingdotgg/t3code#13487. The Docker icon on dockerfile code blocks, a pinned eas-cli for mobile previews, docs, a test cleanup, and marketing updates.

Conflicts

apps/web/src/components/Sidebar.logic.test.ts. Both sides appended a describe block at the end of the file. I kept the fork's sidebarDraftRows block, then upstream's Working shelf block.

third-party-licenses.config.json. Both sides only added entries. I kept the fork's @namespacelabs/sdk, @bufbuild/protobuf, @connectrpc/*, and khroma entries, then appended upstream's nine noxcturnal entries. licenses:sync regenerates the notices with no diff.

pnpm-lock.yaml. I took upstream's lockfile and ran vp i. The result is upstream's lockfile plus the fork's @namespacelabs/sdk, e2b, and mermaid trees. iconv-lite is no longer optional, because mermaid depends on it.

Fork code upstream reached

The merge changes only files upstream changed. It changes nothing under packages/client-runtime/src/connection (#137, #139, #142) or apps/server/src/environmentControl. The 140 files only upstream touched match upstream exactly. Sixteen more files changed on both sides and git merged them cleanly. I read upstream's side of the ones that meet cloud code: mobile CloudEnvironmentRows.tsx (React Native 0.88 event type), web Sidebar.tsx and Sidebar.logic.ts (the Working section beside the fork's cloud draft rows), ws.ts and assets.ts (the clone-aware favicon beside the fork's gateway URL prefix), and settings.ts. The other mobile screens take React Native 0.88 type renames and notification routing, and CI's typecheck covers them.

Update all and cloud boxes. Update all reads useEnvironments(), which leaves out cloud boxes, so it updates the host and other user machines but never a box. It does not conflict with guestProviderInstall.ts. That file reinstalls each provider CLI at @latest whenever a box is prepared or resumed, so a box is already current when a chat starts on it.

Verification

Targeted vp test run:

• apps/mobile, the whole package, ran 194 files and 1782 tests, with 2 failures. Both are macOS-only native tests, and CI skips them on Linux. notification-center-manager.test.ts no longer compiles Expo 58's NotificationCenterManager.swift (cannot find 'Mutex' in scope). permissions-service.test.ts reports a ThreadSanitizer race. Both fail the same way on unmodified upstream main and pass on fork main, so they come from upstream's SDK 58 upgrade.
• packages/client-runtime (connection, cloud, state) ran 848 tests, all passing.
• apps/web (Sidebar.logic, ProviderUpdateLaunchNotification.logic, src/cloud, src/components/chat, src/hooks) ran 1263 tests, all passing.
• apps/server (server.test.ts, src/environmentControl, src/git, src/sourceControl) ran 1112 tests, all passing.
• packages/contracts (settings, server, rpc) ran 163 tests, and scripts third-party-licenses.test.ts ran 18, all passing.

vp lint on the 18 source files both sides changed reports 0 errors. Typecheck, mobile included, runs in CI, because this machine's policy blocks local typecheck.

🤖 Generated with Claude Code

[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	13.5 KiB	13.5 KiB	+1 B (+0.0%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.1 KiB	7.1 KiB	+2 B (+0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−1 B (−0.0%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	56.2 KiB	56.2 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	9	9	0 (0.0%)	21	✅
Claude	Total thread wire	13.5 KiB	13.5 KiB	−27 B (−0.2%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.1 KiB	7.1 KiB	−4 B (−0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−23 B (−0.3%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	57.0 KiB	57.0 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	9	8	−1 (−11.1%)	21	✅

Baseline: ec89414 · PR result: 176320d · 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: 114.0 KiB
• Claude decoded thread snapshot: 114.6 KiB

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

[andrewcai8]

Verdict: PASS. Reviewed by an agent that did not write this change. There are three additive conflicts (tests, license entries, lockfile), all resolved keeping both sides. Fork cloud code (client-runtime/src/connection, environmentControl) is untouched, and files only upstream changed match upstream exactly. pingdotgg#14678 "Update all" doesn't reach boxes, which reinstall CLIs on prepare and resume. CI is green (22 pass). The fork's deploy-relay workflow is skipped on this repo. Note: the mobile native fingerprint changed (Expo 58 / RN 0.88 RC), so native builds are needed before installed apps can take this JS. Merging as a merge commit, per #122/#131.