feat(web): express Dimidium as a runtime theme entry

[deanjstone]

Closes #21.

Summary

• Moves the Dimidium palette out of apps/web/src/index.css (where it was a hand-edited :root override) and into packages/shared/src/themePalettes.ts as a built-in theme, selectable in Settings → Appearance.
• index.css no longer touches upstream's palette declarations at all — 57 lines removed, leaving only --info/--success.
• Restores warm sidebar surfaces, which main currently gets wrong.

Why

index.css was 6 of the 21 conflicts in the 2026-09-05 sync (#20) — the worst single file in a 1295-commit merge, and for the least functionally load-bearing thing the fork carries. Upstream develops that file constantly and has since built a runtime theming system; a theme entry uses it as intended instead of fighting it. Future syncs have nothing to conflict on there.

Three constraints the codebase enforced

Each was caught by an existing guard rather than assumed:

• Adding a built-in widens the mobile Uniwind theme union — the RN app generates a theme per built-in, so MobileUniwindThemeName stopped matching. Regenerated generated-uniwind-* and uniwind-types.d.ts via the real generators rather than hand-editing files marked "should not be edited manually."
• Built-in palettes must be canonical OKLCH (themePalette.test.ts). The initial hex/rgb values failed that invariant, so they were converted with the repo's own toCanonicalThemeColor — no hand-computed color maths.
• Values must be concrete — these roles feed React Native tokens as well as web CSS, and RN cannot parse color-mix()/var(). Upstream's derivations (toolbar follows chrome, message surfaces follow accent, code background mixes card into canvas) are resolved to literals.

Behaviour notes

• Dimidium is not the boot default, by choice. Select it once per client in Settings → Appearance. This keeps divergence to this file plus generated mobile artifacts, with nothing in index.css or index.html — making it the boot default would have required adding ~60 lines to index.html's inlined palette table to avoid a first-paint flash, i.e. trading one conflict surface for another.
• Sidebar fix: the nine sidebar roles are now explicit. main renders them in upstream's cool zinc because the sync dropped the dead [data-sidebar-version] blocks and --sidebar: var(--card) went with them. This branch restores the intended warm surfaces.
• --info/--success remain in index.css: upstream deliberately keeps success, info, provider and channel identity colors out of the themeable role set, so a theme entry cannot carry them.

Test plan

• typecheck — exit 0, 0 errors across all packages
• vp check — 0 errors (593 pre-existing warnings)
• Tests — 506/507 across web, mobile and shared; theme suites specifically 73/73
• @t3tools/web build — exit 0
• Build output inspected: Dimidium's canvas value ships in the JS bundle and dimidium appears 0 times in the built CSS, confirming the values are runtime-injected rather than baked into the stylesheet
• Not verified: on-device appearance. Worth eyeballing light/dark once after selecting the theme, particularly the restored sidebar surfaces

Known unrelated failure: packages/shared/src/Net.test.ts > findAvailablePort fails locally (expected port 51553, got 50753). Confirmed pre-existing by stashing this branch's changes and re-running on a clean tree, where it fails identically. Environmental port contention, not caused by this change.

🤖 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	13.6 KiB	13.5 KiB	−54 B (−0.4%)	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	6.5 KiB	−56 B (−0.8%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	57.1 KiB	57.0 KiB	−88 B (−0.2%)	66.4 KiB	✅
Codex	Live turn messages	10	8	−2 (−20.0%)	21	✅
Claude	Total thread wire	13.6 KiB	13.5 KiB	−83 B (−0.6%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.0 KiB	7.0 KiB	−3 B (−0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.6 KiB	6.5 KiB	−80 B (−1.2%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	57.9 KiB	57.8 KiB	−88 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	10	8	−2 (−20.0%)	21	✅

Baseline: 11ce39d · PR result: 0b2d1f2 · 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: 113.8 KiB
• Claude decoded thread snapshot: 114.5 KiB

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

[deanjstone]

CI caught two real defects in the first push. Both fixed in 0b2d1f24f.

1. Translucent roles broke contrast checking (NaN). Dimidium's surfaces were carried over from the CSS override as alpha values (rgb(0 0 0 / 4%) and similar). That works in CSS, where the browser composites them — but the mobile variable generator emits rgba() for those, and relativeLuminance parses hex, so any contrast assertion touching one produced NaN.

Worth noting for anyone adding a theme later: no other built-in uses alpha at all. All 38 alpha values in themePalettes.ts were Dimidium's. Opaque values are the unstated convention. Each of the 20 translucent roles is now flattened over the surface it actually sits on — app-level surfaces over canvas, sidebar roles over the sidebar, terminal roles over the terminal background.

2. A genuine WCAG AA failure, previously invisible. Once parsing worked, placeholder-on-input measured 4.07:1 light and 3.23:1 dark against the suite's 4.5:1 floor. This predates the port: the CSS override was never subject to a contrast gate, so the palette shipped this unnoticed. Placeholder now measures 4.60:1 in both, adjusted in lightness only — hue and chroma unchanged.

So this PR now also fixes an accessibility bug that was live in the fork.

On the first push's test claim. The PR body cited 506/507 passing. That was obtained with an over-narrow --filter that never actually exercised the mobile app — CI runs 1230 mobile tests where my local run covered a fraction. The claim was wrong, not just unlucky. Re-verified properly this time:

• typecheck — exit 0, 0 errors
• vp check — exit 0, 0 errors (593 pre-existing warnings)
• Full mobile suite — 1230/1230, matching CI's own count

The one visible consequence: placeholder text is slightly darker in light mode and lighter in dark. Correct for contrast, but it is a deliberate change to the palette's appearance and worth a look when reviewing.