[getBoolean]

Merges upstream pingdotgg/t3code main into the fork: 5da53f87d9..8a072ca51d (about 1,160 commits, three merges on this branch: through d412871f27 on 2026-11-12, 9e19b7f7e1 and 8a072ca51d on 2026-11-14). Every merge is a merge commit, so upstream history is kept.

Conflicts and resolutions

First merge (17 files, through d412871f27)

• ci.yml / release.yml: took upstream's two-shard test_web matrix and kept the fork's T3CODE_LINUX_RUNNER / T3CODE_TEST_WEB_TIMEOUT_MINUTES variables.
• README.md: took upstream's new header and hero, with the fork's "What this fork adds" / "Installing this fork" sections right after the links.
• ProjectionStore.ts: took upstream's rewritten tail query and kept the fork's per-window turnItemLimit for tailRows.
• ProviderSessionManager.ts (+ test, ProviderReplayHarness.ts): the fork's accountOverlay goes inside upstream's installationGate.execute(...).
• PullRequestService.ts: kept the fork's account(operation, …) wrapper with upstream's renamed alwaysInteractive.
• GitVcsDriverCore.ts: upstream now ships the Git 2.56 "gitfile does not point to a valid repository" check. Took upstream's isNonRepositoryGitStderr and deleted the fork's gitNonRepository.ts, its test, its imports and its manifest entries.
• PullRequestReviewersScreen.tsx / ReviewSheet.tsx: took upstream's LegendList versions and dropped those manifest entries.
• Everything else: kept both sides' tests and imports, and re-measured agentContextBudget.json.

Second merge (32 files, through 9e19b7f7e1)

• Workflows: took upstream's macos-27 labels as the defaults behind T3CODE_MACOS_RUNNER. Combined PAUSE_NIGHTLY with T3CODE_RELEASES, and publish_cli's preview skip with T3CODE_NPM_PUBLISH. Updated those manifest anchors.
• ProviderTurnStartService.ts: the fork's MOVED_TO_OWN_SESSION retry wraps upstream's new withTurnStartReservation plus the workspace lease.
• server.ts: upstream's TerminalManager.layer(deriveProviderInstanceConfigMap), still provided with layerGitAccounts.
• GitVcsDriverCore.ts push: goes through executeGitWithStableDiagnostics with the fork's pushOptions (manifest anchor moved to the env spread).
• GitManager.ts: upstream's new push-remote argument to runPrStep, still wrapped by the fork's withGitAccount.
• ACP adapter: appliesAccountEnvironment sits next to upstream's renamed buildAttachmentPrompt. Pi adapter: upstream's extracted forkSessionIntoCwd now uses the account sessionEnvironment.
• AzureDevOpsCli.ts: upstream's invocation.env merged with the fork's AZURE_DEVOPS_EXT_PAT.
• McpHttpServer.ts: kept upstream's new theme toolkit and the fork's GitAccountSwitch for project tools.
• Mobile:
  • thread-file-navigator-pane uses upstream's SheetPane, with only the AppBar in the fork's InspectorCloseContext.
  • WorkspaceTabsContext keeps upstream's initialTab and the fork's selectedTab.
  • useOpenPullRequest follows upstream's removal of the Android selectDetail branch and keeps the fork's iPad action with isPad = deviceType === "pad" && Platform.OS === "ios".
  • The PR and usage screens keep renderNativeScrollView and take upstream's RefreshControl.
• vite.config.ts (server/web): union of imports. Kept the fork's update-repository settings and dropped a top-level version constant that upstream moved.
• Tests: kept both sides' cases everywhere.

Third merge (8a072ca51d, clean): brings in upstream pingdotgg#36638, which pins unenv to fix Release Smoke after unenv 2.0.0-rc.25 was published.

Fork adaptations to upstream's newer tests

• GitActionsControl.tsx: upstream's commit action now returns whether it committed, so the fork's account guard returns false when it stops one.
• fakeEnvironment.ts: answers the fork's gitAccounts.threadStatus check, so upstream's new send tests don't hang.
• MarkdownThreadLink.tsx: the fork's navigable chip for scoped links applies only to connected machines, so upstream's "Thread on another machine" label shows otherwise.
• ProjectionStore.test: uses upstream's revision-bumping historyPage helper, because upstream now caches timeline indexes.
• GitManager.test: creates .git/hooks before writing the pre-push hook, because upstream's template-cloned fixtures have none.
• thread-file-navigator-pane.test: mocks the fork's inspector-context modules.
• ProjectSetupScriptRunner.test: asserts only a nonzero exit for a syntax error, because the fork also runs it on macOS, where sh exits 1.
• transferBudgetV2.integration.test: see the owner decision below.
• scripts/mobile-behavior/native-walk.ts: on phones, the source-control step scrolls to "Add account".

Verification

• CI: all 37 checks pass on 72e2df6c1f, including Git Account Tests on macOS, Linux and Windows, and thread transfer is within every ceiling. agent-context-increase is applied, as on chore: sync upstream (2026-11-11) #24, because the growth (+333 B Claude append, +1,675 B MCP tools) is upstream's own text.
• Locally: type checks are clean for server, web, mobile, client-runtime, contracts, ACP, Pi, Azure DevOps and Forgejo. Fork server suites plus every conflicted test file: 1,057/1,059. The two failures were the fork hooks test, since fixed, and upstream's macOS-only /private/var test. Also passing: web 15/15, mobile 104/104, client-runtime 2,039, Forgejo 79, Azure DevOps 155, ACP 106, Pi 222. bash scripts/check-fork-manifest.sh: 542 entries.

iOS 17

Passed. deploymentTarget stays 17.0.

• Build: Release simulator build with Xcode 26.6, React Native from source: succeeded. Deployment targets are 16.4 and 17.0.
• Symbols: libc++ check: 277 symbols, all exported by iOS 17.0. check_runtime_symbols.py --runtime 17.0: 10,970 imports, nothing missing. The new native dependency gt-react-native follows the app's minimum and uses only NSLocale / NSUserDefaults.
• Launch: iPhone 15 and iPad Air 5 on iOS 17.0 (21A328), both still running after 90 s. Install through simctl: agent-device's tar install drops the sparse hermesvm binary.
• Native walk (native-surfaces-open, its own throwaway server): 42/42 steps on both devices, with no crash, uncaught error, rejection or console error.

Release launch, iPhone 15 (iOS 17.0)	Release launch, iPad Air 5 (iOS 17.0)

Walk	iPhone 15	iPad Air 5
Compose
Source control (git accounts)
Thread
Review diff
Terminal

Owner decision (not blocking)

Upstream now collapses a long finished turn's work on the server when a thread opens (pingdotgg#34490, pingdotgg#35130, pingdotgg#35225), and reports that long threads open about 10x faster. The fork's 25-row opening page (THREAD_INITIAL_HISTORY_POLICY) ends inside such a turn, so the turn's expand handle sits above the page and its work arrives uncollapsed when you scroll up. This sync keeps the fork's behaviour and adapts upstream's wire-budget test to measure the older page instead. The manifest says to remove this fork limit once upstream has equivalent first-page limits. Consider retiring it in its own change.

---

Model: Claude Opus 5.5 (claude-opus-5-5), harness: Claude Code in T3 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	0 B (0.0%)	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	0 B (0.0%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	21.5 KiB	21.5 KiB	0 B (0.0%)	24.4 KiB	✅
Codex	Live turn messages	1	1	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	21.9 KiB	21.9 KiB	+1 B (+0.0%)	24.4 KiB	✅
Claude	Live turn messages	1	1	0 (0.0%)	8	✅

Baseline: 0c53205 · PR result: 72e2df6 · 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.