TypeScript 6 UI engineering team for Claude Code. Six Opus specialists that design, review, optimize, and improve UI code so it looks like a senior human design team shipped it -- not AI.
Four operations on UI code, each backed by specialized Opus agents:
| Operation | Skill | What happens |
|---|---|---|
| Design | /typescript-ui:typescript-design-ui |
Commits to a distinctive aesthetic POV, generates OKLCH token system, produces production-grade TS + React + Tailwind v4 code |
| Review | /typescript-ui:typescript-review-ui |
Dispatches 2-3 specialists in parallel for design quality, anti-AI aesthetic, accessibility, and TS type safety |
| Optimize | /typescript-ui:typescript-optimize-ui |
Audits Core Web Vitals (LCP/INP/CLS), bundle size, rendering, with quantified metric impact per finding |
| Improve | /typescript-ui:typescript-improve-ui |
Full 4-specialist sweep (design + anti-AI + perf + TS), deduplicated unified report ordered by impact |
All reviews are read-only. Findings are advisory. You pick which to apply.
git clone https://github.com/TheMizeGuy/typescript-ui-public.git ~/.claude/plugins/typescript-uiThen add to your Claude Code settings:
{
"enabledPlugins": {
"typescript-ui": true
}
}Restart Claude Code.
claude --plugin-dir /path/to/typescript-ui-publicgit clone https://github.com/TheMizeGuy/typescript-ui-public.git /tmp/typescript-ui-public
mkdir -p ~/.claude/plugins/cache/typescript-ui/1.0.0
cp -R /tmp/typescript-ui-public/{.claude-plugin,agents,skills,references,README.md,ARCHITECTURE.md} \
~/.claude/plugins/cache/typescript-ui/1.0.0/After restarting Claude Code, type any of these -- if the skill triggers, it's installed:
/typescript-ui:typescript-design-ui
/typescript-ui:typescript-review-ui
/typescript-ui:typescript-optimize-ui
/typescript-ui:typescript-improve-ui
design a settings page for a developer tool -- dark theme, dense, keyboard-first
The plugin triggers typescript-design-ui, dispatches the design architect, and produces:
- POV statement (e.g. "Tactical Operator -- monospace, functional, single accent")
- Full OKLCH token system (
@themeblock) - Production-grade TS + React components
- Taste audit results
/typescript-ui:typescript-review-ui src/components/
Dispatches design reviewer + anti-slop auditor in parallel. Returns severity-tagged findings.
/typescript-ui:typescript-optimize-ui all
Dispatches the perf engineer. Each finding has quantified impact (e.g. "+1200ms LCP").
/typescript-ui:typescript-improve-ui src/
Dispatches the team lead who coordinates all 4 specialists, deduplicates findings, and presents a unified plan.
| Agent | Color | Focus |
|---|---|---|
ui-design-architect |
Cyan | Greenfield design, POV commitment, token systems, production code |
ui-design-reviewer |
Blue | Visual quality, UX, accessibility, POV coherence (18 lenses) |
ui-perf-engineer |
Yellow | Core Web Vitals, bundle, rendering, React perf (12 angles) |
ui-typescript-engineer |
Purple | TS6 strictness, component typing, state safety (10 angles) |
ui-anti-slop-auditor |
Red | 108 AI-generated aesthetic tell detection |
ui-team-lead |
Green | Multi-agent orchestration, finding dedup, unified report |
All agents run on Opus. All review agents are read-only.
26 self-contained reference files (8,600+ lines) organized into 6 domains. Agents read these before every review -- no external dependencies required.
| Domain | Files | Covers |
|---|---|---|
| typescript/ | 4 | TS6 essentials, component typing, state typing, branded primitives |
| design/ | 6 | OKLCH color, typography, spacing, motion, Tailwind v4, shadcn customization |
| aesthetic/ | 4 | 108 anti-AI tells, POV discovery, 12 distinctive system case studies, taste checklist |
| performance/ | 5 | Core Web Vitals, React 19 perf, CSS perf, bundle/loading, measurement |
| accessibility/ | 4 | WCAG 2.2, keyboard/focus, reduced motion, screen reader |
| architecture/ | 3 | Component patterns, state architecture, styling architecture |
All review/optimize/improve skills accept the same scope argument:
| Argument | Resolves to |
|---|---|
(empty) or diff |
Uncommitted + staged changes |
staged |
Only staged files |
pr |
Diff vs main/master |
<file> |
Single file |
<directory> |
All UI files in dir |
all |
Entire project |
Files filtered to: .tsx, .jsx, .css, .scss. Excludes node_modules, dist, build, .next, out, coverage.
Every agent produces findings in the same structure:
### [SEVERITY] [Category]: one-line title
File: path/to/file.tsx:42-58
Issue: plain-English explanation
Why it matters: concrete consequence
Current code:
(5-15 line extract)
Suggested rework:
(concrete rewrite, applicable verbatim)
Reference: references/<domain>/<file>.md #section
| Tag | Meaning |
|---|---|
| CRITICAL | Blocks ship -- a11y violation, shipped AI-default aesthetic, CWV threshold breach, type unsoundness |
| HIGH | Fix before launch -- failing contrast, broken keyboard nav, significant perf cost |
| MEDIUM | Quality cost -- weak composition, missing exhaustiveness, suboptimal motion |
| LOW | Polish -- missing JSDoc, could use newer feature |
| NIT | Preference -- included sparingly |
The plugin works standalone with its 26 reference files. These optional integrations enhance it:
| Integration | What it adds |
|---|---|
| Context7 | Live library docs for Tailwind v4, React 19, Motion, TanStack Query |
- Claude Code with Opus model support (agents require Opus)
- For the
typescript-improve-uiskill: practical concurrency limit is ~4 agents
MIT
See ARCHITECTURE.md for the internal map including agent-to-skill dispatch, reference-to-agent reading assignments, and cross-reference format.