chore(upstream): sync marcode through 8b2838e0

[maarco]

What Changed

Merged pingdotgg/t3code@8b2838e0 — 33 upstream commits since 52b2bf77a941 — into a two-parent merge commit on this branch. Nine files conflicted; four more broke the fork boundary while merging perfectly cleanly.

Conflict decisions

Path	Decision
.devcontainer/devcontainer.json	Took upstream's rewrite that makes container setup work (pingdotgg#7875): ubuntu-24.04 base, node:2 + rust + gh features, on-create/update-content scripts, forwarded ports, cache volumes. Re-applied Marcode's identity — container name, the bun feature AGENTS.md needs for bun:sqlite test-data seeding, and MARCODE_HOME → .marcode in place of upstream's T3CODE_HOME → .t3.
apps/marketing/src/pages/95.astro	modify/delete. Marcode retired the upstream-branded /95 page and pins that in retiredRoutes.test.ts. Upstream's pingdotgg#10460 fix is confined to that file, so nothing was left to port. Kept deleted.
apps/server/src/project/AgentSessionScanner.test.ts	Upstream inserted two new exclusion tests where Marcode had renamed a fixture .t3 → .marcode. Kept all three. The scanner's /.marcode/worktrees/ seam was untouched by the merge.
apps/web/src/components/onboarding/WelcomeWizard.tsx	Upstream rewrote the wizard into a shared multi-computer dialog (pingdotgg#10465). Took the rewrite, re-applied Marcode's three intents: product identity in the copy, the brand mark (upstream's wordmark component doesn't exist in this fork — MarcodeMark replaced it), and the TerminalViewport prop contract, which here has no providerInstanceId/visible and requires an onAddTerminalContext sink.
apps/web/src/hooks/useTheme.ts, useTheme.test.ts	Upstream deleted the onboarding theme surface wholesale. Marcode's wizard is deliberately always dark (hotspot added in 082418a), so mountOnboardingTheme, its branches, and its tests stay.
apps/web/src/index.css	Both sides appended at the same two points. Kept upstream's AnimatedHeight collapsible rules and sidebar drag-label animation alongside Marcode's stage-blueprint tokens and chrome.
docs/user/welcome-wizard.md	Took upstream's new copy for multi-computer selection and repository-grouped import, in Marcode's product voice.
packages/shared/src/git.test.ts	Kept Marcode's T3Tools/Marcode fixture and added upstream's new ssh:// remote case against it.

Clean merges that would have silently broken the fork

None of these produce a conflict marker:

1. FirstRunGate.tsx lost its mountOnboardingTheme mount when upstream removed the feature. Restored — without it the wizard follows the saved light preference.
2. index.css lost html[data-onboarding-surface]:root and the matching contrast probe. Restored, so portaled menus and tooltips keep the wizard's palette.
3. docs/internals/devcontainer.md arrived new from upstream naming T3CODE_HOME, .t3, and T3CODE_DEV_ALLOWED_ORIGINS — none of which this fork reads. Corrected.
4. .devcontainer/on-create.sh writes the container's first-run notice in upstream's product voice. Corrected.

New apps/web/src/components/onboarding/onboardingTheme.seam.test.ts pins the dark-onboarding seam — the exported theme owner, both mounts, the CSS block, the seam comments, and that the wizard renders Marcode's brand glyph — so the next sync fails loudly instead of reverting it quietly. The matching removal (that upstream's wordmark component is absent from every source root) stays owned by scripts/brand-mark-ownership.test.ts, which already does it better.

Why

Daily upstream sync. Upstream's delta carries real correctness and operational work worth taking (a devcontainer that actually builds, a Cursor transport-failure fix, Codex Spark limit handling, a large mobile thread-lifecycle pass, project-import grouping), and none of it should be dropped because it landed in a customized file. Every conflict was resolved by combining both intents rather than taking a side, and the four no-conflict breaks are exactly the class the fork-boundary rules in AGENTS.md exist to catch.

Compatibility identifiers left upstream-shaped on purpose: @t3tools/*, the t3 CLI name and its npx t3 … commands, T3_BOOT_SERVICE_UNIT, T3_SERVICE_LAUNCHER_CONTEXT, T3CODE_DESKTOP_*, T3CODE_TRANSFER_*, and t3code-* cache volume names.

Neither coupled vendored subtree source (pnpm-workspace.yaml, infra/relay/package.json) moved, so no sync:repos run was needed.

UI Changes

The onboarding wizard changes shape (upstream's redesign) and the chat composer swaps a generic file icon for Pierre icons, but this environment is headless — no browser or simulator — so no before/after captures were taken. Live verification of the wizard at 390px and 820px is still owed before this leaves draft, per the sync runbook's runtime-verification step for apps/web/** hotspots. That is the one thing blocking this from being ready for review.

Checklist

• This PR is small and focused — one concern: the upstream merge
• I explained what changed and why
• I included before/after screenshots for any UI changes — not capturable here, see above
• I included a video for animation/interaction changes — same

Verification

CI on 0c81071f: all four required checks green — Check, Test, Mobile Native Static Analysis, Release Smoke — plus Test Server 1/2/3, Rust, Native fingerprint diff, and Mobile Native Changes. Thread-transfer impact is within every ceiling. No merge conflict against main.

claude-review is red and is not this PR's failure: the review app's own credentials. It has failed on every run since 2026-09-07 02:30 UTC, including seven on PR #75's unrelated branch, and its re-run here failed identically. Details in the comments below. It is not a required check.

Local, before pushing:

• Full apps/web suite: 356 files, 4462 tests, all passing.
• vp run -r typecheck: clean across all 15 packages, 0 errors.
• Focused tests for every resolved path: AgentSessionScanner (81), useTheme, git, forkIdentity, branding, retiredRoutes, brand-mark-ownership, the new seam test, plus the unified-workspace, right-panel, chat-ambient, sidebar-drag/pointer, ChatView-logic and project-import suites.
• vp fmt --check and vp lint clean on the changed scope.

The second commit here (0c81071f) fixes two defects CI caught in the seam test I added — a banned node:fs import in apps/web, and two references to upstream's wordmark name that brand-mark-ownership.test.ts correctly rejects. Neither touched a conflict resolution. Both are explained in the commit message and in the comments.

🤖 Generated with Claude Code

https://claude.ai/code/session_01LE6a6L6CrRksuzTjFLR9YM

[maarco]

claude-review is failing, and it is not this PR's failure. Flagging it so it doesn't read as something the merge introduced.

What's failing. Job 101794066018 on 4f87747. The action gets through OIDC, the app token, the Claude Code install, the marketplace and plugin install, then the SDK run itself dies:

{"type":"result","subtype":"success","is_error":true,
 "duration_ms":2039,"num_turns":1,"total_cost_usd":0,"modelUsage":{}}
##[error]Claude execution failed: result is_error:true

One turn, two seconds, zero cost, empty modelUsage — it never reached the model. ANTHROPIC_API_KEY is empty in the job env. Full output is hidden for security (show_full_output: true or debug mode would print the underlying error).

Why it isn't this PR's. Claude Code Review has failed on every run since run pingdotgg#84 (2026-09-07 02:30 UTC) and last succeeded at run pingdotgg#83 (2026-09-06 22:42 UTC). Runs pingdotgg#84–pingdotgg#90 are all on chore/upstream-52b2bf77a941 (PR #75) across six different head SHAs, and pingdotgg#89 was already a re-run attempt that failed the same way. This branch didn't exist for any of them. It also reproduces on a diff that shares nothing with this one.

Two further reasons the diff can't be the cause: the workflow restores .claude, .mcp.json, .gitmodules, and CLAUDE.md from origin/main before running ("PR head is untrusted"), so it isn't even reading this PR's copies — and this PR touches none of those files anyway.

No fix ported, because none exists in the repository. This is a credentials/account condition on the review app, not a code path. Nothing I could add to this branch would turn it green, and widening the PR to guess at workflow config would be worse than leaving it red. Someone with access to the repo's Claude secrets should check whether the app's credential expired or was revoked around 2026-09-07 02:30 UTC.

Not a required check. Branch protection requires Check, Test, Mobile Native Static Analysis, and Release Smoke. Release Smoke, Rust, Mobile Native Changes, and Native fingerprint diff are already green on this head; the rest are still running. claude-review is advisory and its absence costs a review pass, not correctness.

I'm re-running the job once to confirm it fails identically on this head SHA, and will keep watching the PR until the required checks are terminal.

---

Generated by 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	13.6 KiB	13.6 KiB	+12 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.0 KiB	7.0 KiB	+2 B (+0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.6 KiB	6.6 KiB	+10 B (+0.1%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	57.1 KiB	57.1 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	10	10	0 (0.0%)	21	✅
Claude	Total thread wire	13.6 KiB	13.6 KiB	−16 B (−0.1%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.1 KiB	7.1 KiB	−2 B (−0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.5 KiB	6.5 KiB	−14 B (−0.2%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	57.8 KiB	57.8 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	9	8	−1 (−11.1%)	21	✅

Baseline: b25b24b · PR result: 0c81071 · 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: 113.9 KiB
• Claude decoded thread snapshot: 114.6 KiB

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

[maarco]

Update on the three red checks.

claude-review — confirmed not this PR's. The re-run (job 101794656294) failed identically on the same head SHA, matching the seven prior failures on PR #75's branch. That's my one re-run spent; the conclusion in the comment above stands, and it still needs someone with access to the repo's Claude credentials.

Check and Test — both mine, both fixed. Two real defects, and both were in the seam test I added rather than in the merge resolution:

1. Check → onboardingTheme.seam.test.ts(19,25): error TS377057 effect(nodeBuiltinImport). apps/web forbids node:fs. I used it to read index.css, because a ?raw import of a stylesheet resolves to an empty string under the test CSS pipeline. Fixed with the // @effect-diagnostics nodeBuiltinImport:off pragma, which is the convention scripts/brand-mark-ownership.test.ts already uses for exactly this — a fork-boundary assertion that has to read a checked-in file.

2. Test → brand-mark-ownership.test.ts flagged two references to upstream's wordmark component, in WelcomeWizard.tsx and in my seam test. That test is right and I was wrong: it scans every file under apps/{web,mobile,desktop}/src for the literal name and assembles the string at runtime so it cannot match itself. My seam comment spelled the name out, and my seam test asserted on its absence — duplicating a check that file already owns, better. The comment now describes the component instead of naming it, and the seam test asserts the positive (MarcodeMark is imported and rendered) and leaves the removal to brand-mark-ownership.test.ts.

Why local verification missed both. I installed dependencies with --ignore-scripts, so the repo's prepare step never ran effect-tsgo patch and the Effect language-service diagnostics were inactive in my typecheck. I've since applied the patch, reproduced TS377057 locally, and confirmed it clears. The brand-mark-ownership miss was simpler: I ran focused tests for the paths I resolved and didn't run the one in scripts/ that guards the identifier I'd just typed.

Everything else on this head is green, including all the other required checks — Release Smoke, Mobile Native Static Analysis, Test Server 1/2/3, Rust, Native fingerprint diff. Note that Test aborted after the scripts failure, so apps/web, apps/marketing, packages/shared and the oxlint plugin never ran in it; I'm running the full apps/web suite and a repo-wide typecheck locally before pushing, rather than sending up another speculative commit.

---

Generated by Claude Code