Kern

1# Kern
2 
3A Claude Code plugin that catches AI-generated design defaults before they ship.
4 
5Kern routes every design through a persona system, picks a varied subset of anti-patterns per run (so two runs never produce the same critique), scores output on a 0-100 sameness scale, and fixes copy, color, layout, and interaction before the code lands.
6 
7Built for Shadcn/Radix/Next.js stacks. Runs a team of specialist agents that plan, design, implement, and review collaboratively.
8 
9## The Problem
10 
11AI design tools (v0, Lovable, Bolt.new, Cursor Composer) converge on the same visual vocabulary. Purple/blue gradient, centered hero, Inter font, three identical feature cards, "Get Started" CTA. A developer tool and a consumer app look like the same product built on the same template.
12 
13Kern fixes this with persona-first generation and a sameness score that quantifies how close a design is to the AI default.
14 
15## Installation
16 
17Add the raymondware marketplace, then install Kern:
18 
19```
20/plugin marketplace add raymondware/kern
21/plugin install kern@raymondware
22```
23 
24That's it. Restart Claude Code and the `/kern:*` commands are available.
25 
26## Quick Start
27 
28```bash
29# Generate a component from scratch
30/kern:design hero section for a CI/CD observability platform
31 
32# Audit existing code for AI-default patterns
33/kern:audit
34# (paste TSX when prompted)
35 
36# Make an existing design less AI-looking
37/kern:differentiate
38 
39# Generate design specs without code
40/kern:plan pricing page for a developer API product
41```
42 
43## Commands
44 
45| Command | Does |
46|---|---|
47| `/kern:design <desc>` | Full pipeline: persona detection, specialists, implementation, review |
48| `/kern:audit` | Anti-pattern scan with sameness score against existing code |
49| `/kern:differentiate` | Identify which AI tool fingerprints are present and apply targeted fixes |
50| `/kern:plan <desc>` | Design specs only (typography, color, layout, motion) -- no code |
51| `/kern:review` | Full review pass: critic ensemble, copy editor, accessibility audit |
52| `/kern:compare <a> <b>` | Side-by-side comparison with similarity score |
53| `/kern:polish` | Rewrite a component against Kern's standards, show diff |
54| `/kern:copy` | Microcopy audit: button labels, errors, empty states, CTAs |
55| `/kern:research` | Scan design communities for new anti-patterns |
56| `/kern:draw <desc>` | Run the anti-pattern selector standalone (rotation verification) |
57 
58## How It Works
59 
60### 1. Persona Detection
61 
62Kern picks a persona from your description, then loads that persona's rules for font, color, layout, and copy. Five personas:
63 
64| Persona | Register |
65|---|---|
66| `developer-tool` | Restrained, dense, function-first. Linear/Vercel register. |
67| `consumer-saas` | Warm, spacious, approachable. Things 3 / Cron register. |
68| `creative-tool` | Bold, editorial, personality-forward. Figma / iA Writer register. |
69| `b2b-enterprise` | Conservative, trust-building, data-forward. Stripe / Ramp register. |
70| `e-commerce` | Product-first, conversion-focused. Allbirds / Ridge register. |
71 
72### 2. Variation Rule
73 
74Every run picks a different subset of anti-patterns. Kern reads the audit log (`state/draws.jsonl`) and excludes the previous run's subset. Two runs on the same prompt produce different critiques. No template drift.
75 
76### 3. Anti-Pattern Selection
77 
78Each run draws 12-18 patterns from a pool of 69, weighted by surface (landing page, dashboard, pricing), persona, industry, and audience. The selection is logged to `state/draws.jsonl` before any agent runs.
79 
80### 4. Parallel Critics + Synthesizer
81 
82Four critics run in parallel (design, hierarchy, interaction, microcopy), plus a copy editor and accessibility auditor. The synthesizer merges their output, removes duplicates, and computes a consensus sameness score.
83 
84### 5. The Sameness Score
85 
86Every design gets a 0-100 score on AI-default similarity:
87 
88| Score | Meaning |
89|---|---|
90| 0-20 | Distinctive. Clear decisions made. Ship. |
91| 21-40 | Has character. Some defaults showing. |
92| 41-60 | Generic. Could be any AI-generated SaaS site. |
93| 61-80 | Default AI aesthetic. Multiple tool fingerprints present. |
94| 81-100 | Template unmodified. Do not ship. |
95 
96Gate threshold: **40** for Kern-produced output, **60** for external audits.
97 
98## Anti-Pattern Library
99 
100Kern ships with 69 anti-patterns across three base files and a research-sourced directory:
101 
102| File | Patterns |
103|---|---|
104| `anti-patterns/visual.md` | Gradient slop, nested card soup, Inter default, centered glowing hero, AI palettes, drop shadows, inverted dark mode, neon without context |
105| `anti-patterns/copy.md` | Marketing verbs, hedging, filler openers, self-descriptors, em-dashes, passive voice, vague headlines, broken microcopy |
106| `anti-patterns/interaction.md` | Dishonest CTAs, loading as marketing, motivational empty states, bouncy interactions, modal fatigue, toast spam, dark patterns |
107| `anti-patterns/sourced-from-research/` | Community-sourced patterns with citations |
108 
109## Reference Files
110 
111| File | Purpose |
112|---|---|
113| `skills/kern/references/fonts.md` | Font stacks by context: display, body, mono, serif |
114| `skills/kern/references/tokens.md` | Spacing, radius, type scale, color philosophy |
115| `skills/kern/references/exemplars.md` | Sites that get it right and why |
116| `skills/kern/references/dribbble-refs.md` | Reference shots by category |
117| `skills/kern/references/ai-fingerprints.md` | Tool-specific fingerprints for `/kern:differentiate` |
118| `skills/kern/references/personas/` | Five persona files with font, color, layout, copy rules |
119 
120## Agent Team
121 
122Kern runs a team of specialized agents:
123 
124### Design Specialists (Haiku)
125`typography-specialist` `color-specialist` `layout-specialist` `motion-specialist`
126 
127### Development (Sonnet)
128`component-architect` `component-implementer` `style-implementer` `accessibility-implementer`
129 
130### Critics (parallel ensemble)
131`design-critic` `hierarchy-critic` `interaction-critic` `microcopy-critic` `copy-editor` `accessibility-auditor` `responsive-validator`
132 
133### Orchestration (Opus 4.8)
134`conductor` `anti-pattern-selector` `critique-synthesizer`
135 
136## Contributing
137 
138See `CLAUDE.md` for the contributor guide. Agents live flat in `agents/`. Anti-patterns in `anti-patterns/`. Commands in `commands/`.
139 
140## License
141 
142MIT. See `LICENSE`.