[andrewcai8]

Syncs upstream T3 Code through 680152e, upstream's newest commit, which has green CI. That is 162 commits (8ecb607..680152e), and 51 files conflicted.

Notable upstream changes:

• About 11k low-signal tests deleted (test: delete 11k low-signal tests so CI runs faster and flakes less pingdotgg/t3code#22557 and follow-ups). New lint rules ban wall-clock reads, real-clock waits, polling, per-test timeouts and frame waits in tests, and browser tests must use the disposable render.
• Web refactors split ChatView, Sidebar, CommandPalette, MessagesTimeline, ChatComposer, ConnectionsSettings and the pull request page into smaller files.
• t3 thread CLI (feat(cli): start, message, and read threads from the terminal with t3 thread pingdotgg/t3code#22732). Webhook event filters plus replay dedupe and coalescing (feat(scheduled-tasks): event filters for webhook deliveries pingdotgg/t3code#22194, feat(scheduled-tasks): replay dedupe and coalescing for webhook deliveries pingdotgg/t3code#22234, migration 67). Mermaid diagrams render with pele (feat(web): render Mermaid diagrams with pele pingdotgg/t3code#21472).
• orchestrator_capabilities returns a short model list and takes search (fix(server): agents get a short, current model list instead of a 55k-character catalog dump pingdotgg/t3code#22889).
• The composer stays usable during approvals (feat(web): composer stays usable during approvals, and sending declines and redirects pingdotgg/t3code#21503). Limits bars show the even-pace mark as a tick in a gap (fix(web,mobile): Limits bars show the even-pace mark as a tick in a gap pingdotgg/t3code#22408).
• pnpm 12, Vite+ 1.3 and Electron 44.9.3.

Conflicts keep upstream's change and re-apply the fork's behavior. Where upstream moved code to a new file, the fork's edit moved with it.

Web

• ChatView: the cloud box disconnect guard and "cloud machine unavailable" banner now live in useSystemComposerBanners. The Run-on pick (onEnvironmentPicked) and needsLoadBalancedPick live in useDraftEnvironmentRouting. Held cloud sends keep their message id and time through upstream's planComposerTurn. ChatView reads useProjects() again for useCloudChat. Opening a related thread from Home reads thread refs on click, so ChatView does not subscribe to every thread again.
• Sidebar: the cloud draft rows went to sidebar/SidebarDraftBlock.tsx. The machine status ranking, the expired-row touch exit and the Home row went to sidebar/SidebarThreadRow.tsx.
• MessagesTimeline: "Watch report from T3 Code" and "Sent by Home" went to TimelineUserMessageRow.tsx.
• ChatComposer: the cloud and auto-pick picker entries go to upstream's ComposerModelPicker.
• ConnectionsSettings: pairingUrlHost went to ConnectionsSettings.logic.ts and the "Waking up..." status to SavedEnvironmentRows.tsx.
• CommandPalette: Home's folder stays out of upstream's buildProjectSearchItems and buildNewThreadInProjectItems.
• Pooled Limits: upstream's pace gap and tick stay off each account's segment, as the fork chose in f224b7c. Upstream's version stays on the single-account rows.

Server

• OrchestratorMcpService / orchestrator tools: upstream's short model list and search now feed a shared providerCapabilities. Home's remote capabilities use it too, and the Fleet capabilities operation takes search. Tool text keeps the fork's Home environmentId and cloud-only scheduling sentences.
• ScheduledTaskService: runTask keeps the fork's options object (resumedAt, webhook) and adds upstream's dispatched flag. A cloud fire still skips releaseStuckRun, which now takes the pending delivery id.
• project/handlers: Home's watch-before-launch runs alongside upstream's project-default runtime mode.

Mobile

• modelOptions: auto-pick stand-ins sit on top of upstream's continuation-group filter and provider badges. A stand-in shows no account badge.
• thread-list-v2-items and useThreadListActions: both sides kept. That is the machine status labels plus upstream's duration and undo.

Tests and config

• Upstream deleted or pruned 27 test files the fork had edited. Fork-only tests were kept: Home, cloud targets, account routing, continuation prompt, the rejected-limit window, background-task notifications and Claude wake/roster. Edits that only adapted upstream's own tests were dropped with them.
• The kept Claude adapter tests use upstream's change signal instead of the removed awaitUntil poller.
• knip.jsonc merges both entry lists. pnpm-lock.yaml is upstream's lock re-resolved with pnpm 12 for the fork's dependencies.

Other fixes the merge needed:

• Upstream stopped exporting providersWithLimits, isLoopbackHostname and getAutoBootstrapThreadModelSelection. Fork code uses them, so they are exported again.
• getSlowRpcAckRequests is back for the fork's cloud resume latency test.
• Knip flagged three cloud re-exports and usePaceMark as unused, so those exports are gone.
• Upstream's new test lint rules failed 57 times in fork-only tests. Server cloud tests now use TestClock, fixed epochs, frozen Date and Deferred handshakes. Web browser tests use the disposable render, and boxDemand flushes its registry.
• One reasoned disable remains, in NamespaceProvisionRuntime.test.ts. The fake Mac boot races the command's real AbortSignal.timeout, which no fake clock drives.
• release.yml and mobile-eas-production.yml each had two if: keys on fork main, which is invalid YAML and hid resolve-nightly-release.ts from knip. The fork's broader condition is kept.
• FleetService.test expects upstream's rule that a label repeating the id is omitted. ForkMigrations.test lists migration 67.

Verification:

• tsc-rs --noEmit is clean for contracts, shared, client-runtime, provider-core, provider-cursor, server, web, mobile, desktop and scripts.
• knip:check passes. vp lint has no errors and vp fmt --check is clean on every file that differs from upstream (624 TS files).
• Server: the conflicted test files, scheduledTasks, home, environmentControl, mcp, account routing, runtimeLayer, ProviderAccountSwitch, forkExec, RestartContinuation and ForkMigrations all pass, about 1900 tests.
• One server test fails only on this Mac: AgentSessionScanner "excludes sandboxes reached through a symlink" (the macOS /var symlink case noted in chore: merge upstream pingdotgg/t3code through 67c0fc0cda #236).
• Web unit tests for the touched files pass (203), and the three fixed browser tests pass (15). Mobile tests for the touched features pass (282). All client-runtime (1347) and contracts (145) tests pass.

Not verified: the web, desktop and mobile UI were not run.

🤖 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	5.0 KiB	5.0 KiB	−7 B (−0.1%)	6.8 KiB	✅
Codex	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Codex	Live turn WebSocket wire	1.2 KiB	1.2 KiB	−7 B (−0.6%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	20.9 KiB	20.9 KiB	0 B (0.0%)	29.3 KiB	✅
Codex	Live turn messages	2	2	0 (0.0%)	8	✅
Claude	Total thread wire	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Claude	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Claude	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	21.2 KiB	21.2 KiB	0 B (0.0%)	29.3 KiB	✅
Claude	Live turn messages	2	2	0 (0.0%)	8	✅

Baseline: 36750d4 · PR result: 31fd632 · 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: 108.5 KiB
• Claude decoded thread snapshot: 108.8 KiB

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