@geoemmanuel/nave 0.1.0

package/README.md

@@ -0,0 +1,82 @@
+# nave
+
+A curated skill router for AI coding agents.
+
+Nave detects what you're working on and routes to the right skill — automatically.
+Clear task? Executes directly. Ambiguous task? Shows you the options and asks first.
+
+---
+
+## Install
+
+```bash
+npx nave
+```
+
+Run once in your project root. Nave detects which agent you're using and installs itself.
+
+**Supported agents:** Claude Code, Cursor, OpenCode, Codex CLI, Windsurf, Gemini CLI, Antigravity
+
+---
+
+## What it does
+
+Before responding to any task, nave classifies the request and reads the best matching skill:
+
+| Task | Skill activated |
+|---|---|
+| "plan this feature out" | superpowers — structured plan, crash-proof |
+| "you're over-engineering this" | karpathy — discipline, surgical changes |
+| "simplest solution please" | ponytail — laziest senior dev in the room |
+| "write this slide animation" | emil — motion philosophy, correct easing |
+| "does this look generic?" | taste-skill — anti-slop visual direction |
+| "build a design language" | impeccable — brand vs product thinking |
+| "design this settings page" | ui-ux-pro-max — 99 UX patterns + a11y |
+
+---
+
+## Bundled Skills
+
+| Skill | Author | Stars | Domain |
+|---|---|---|---|
+| [karpathy](https://github.com/multica-ai/andrej-karpathy-skills) | @multica-ai | 174k | Agent behavior, coding discipline |
+| [superpowers](https://github.com/obra/superpowers) | @obra | 226k | Planning, multi-step agentic tasks |
+| [ponytail](https://github.com/DietrichGebert/ponytail) | @DietrichGebert | 50k | Code minimalism, YAGNI |
+| [emil](https://github.com/emilkowalski/skills) | @emilkowalski | 2.8k | Animation, motion, interaction |
+| [taste-skill](https://github.com/Leonxlnx/taste-skill) | @Leonxlnx | 45k | Visual taste, anti-slop |
+| [impeccable](https://github.com/pbakaus/impeccable) | @pbakaus | 28k | Design language, brand vs product |
+| [ui-ux-pro-max](https://github.com/nextlevelbuilder/ui-ux-pro-max-skill) | @nextlevelbuilder | 88k | UX patterns, accessibility |
+
+These authors built exceptional skills. Nave curates and routes between them.
+Full credit above — install their skills directly for the complete experience.
+
+---
+
+## Commands
+
+```bash
+npx nave              # install into detected agent
+npx nave --agent claude   # install for a specific agent
+npx nave list         # list all registered skills
+```
+
+In your agent after install:
+
+```
+/nave                 # show skill registry and suggestion
+```
+
+---
+
+## Adding a skill
+
+See [CONTRIBUTING.md](./CONTRIBUTING.md).
+
+---
+
+## License
+
+MIT
+
+Skill philosophies are credited to their original authors.
+Nave references their work in condensed form — it does not redistribute full skill files.

package/SKILL.md

@@ -0,0 +1,132 @@
+---
+name: nave
+description: >
+  Nave is a curated skill router. Before responding to any task, classify
+  it against the registry below and read the matching skill file.
+  On clear tasks: execute directly. On ambiguous tasks: show the skill
+  layout and ask which direction before proceeding.
+version: 0.1.0
+---
+
+# Nave — Skill Router
+
+Nave curates the best craft skills for building software and routes every
+task to the right one automatically.
+
+---
+
+## Protocol
+
+**Step 1 — Classify the task**
+Run the classification rules below. First match wins.
+
+**Step 2 — Clarity check**
+- Clear and specific → execute directly, open with: `[nave] → {skill} activated`
+- Ambiguous (e.g. "fix this", "make it better") → show the Skill Layout,
+  state which 2–3 skills apply and what each would do differently, ask which direction
+
+**Step 3 — Read the skill file**
+Load the matched skill from `registry/skills/{id}.md` and apply its philosophy.
+
+---
+
+## Skill Layout
+
+Show this when the task is ambiguous or when the user runs `/nave`:
+
+```
+┌───────────────┬──────────────────┬────────┬───────────────────────────────────────────────┐
+│ skill         │ domain           │ stars  │ best for                                      │
+├───────────────┼──────────────────┼────────┼───────────────────────────────────────────────┤
+│ karpathy      │ Agent behavior   │ ★174k  │ Coding discipline, focus, no overreach        │
+│ superpowers   │ Planning         │ ★226k  │ Multi-step tasks, plans that survive /clear   │
+│ ponytail      │ Code craft       │ ★50k   │ YAGNI, minimal, delete bloat                  │
+│ emil          │ Animation        │ ★2.8k  │ Motion, easing, interaction polish            │
+│ taste         │ Visual taste     │ ★45k   │ Anti-slop, stop AI generating generic UI      │
+│ impeccable    │ Design language  │ ★28k   │ Brand vs product, visual vocabulary           │
+│ ui-ux-pro-max │ UX patterns      │ ★88k   │ Pages, components, accessibility              │
+└───────────────┴──────────────────┴────────┴───────────────────────────────────────────────┘
+```
+
+After showing the layout: state the suggested skill and a one-line reason. Ask to confirm.
+
+---
+
+## Classification Rules
+
+### Karpathy [karpathy]
+**Domain:** Agent behavior  
+**Triggers:** "over-engineering", "stay focused", "think before coding", "wrong direction",
+"silent assumptions", "production code", "critical change", "multi-file task"  
+→ Read `registry/skills/karpathy.md`
+
+---
+
+### Superpowers [superpowers]
+**Domain:** Planning  
+**Triggers:** "plan this", "architect this", "multi-step", "long task",
+"don't lose context", "PRD", "spec", "from scratch", "multi-file refactor"  
+→ Read `registry/skills/superpowers.md`
+
+---
+
+### Ponytail [ponytail]
+**Domain:** Code craft  
+**Triggers:** "simplest solution", "too complex", "YAGNI", "over-engineered",
+"delete this", "minimal", "be lazy", "shortest path", "adding dependency", "new abstraction"  
+→ Read `registry/skills/ponytail.md`
+
+---
+
+### Emil Design Engineering [emil]
+**Domain:** Animation  
+**Triggers:** "animation", "transition", "easing", "motion", "spring",
+"framer-motion", "hover effect", "page transition", "micro-interaction",
+"scroll animation", "keyframe", "gesture"  
+→ Read `registry/skills/emil.md`
+
+---
+
+### Taste Skill [taste]
+**Domain:** Visual taste  
+**Triggers:** "does this look good", "too generic", "AI slop", "improve the UI",
+"visual review", "aesthetic", "typography", "color palette", "layout feedback", "make it better"  
+→ Read `registry/skills/taste.md`
+
+---
+
+### Impeccable [impeccable]
+**Domain:** Design language  
+**Triggers:** "design language", "design system", "brand", "visual vocabulary",
+"typography system", "color system", "spacing scale", "feel like a real product", "design tokens"  
+→ Read `registry/skills/impeccable.md`
+
+---
+
+### UI UX Pro Max [ui-ux-pro-max]
+**Domain:** UX patterns  
+**Triggers:** "new page", "dashboard", "landing", "onboarding", "settings",
+"component architecture", "form design", "navigation pattern",
+"accessibility", "responsive layout", "user flow", "admin panel"  
+→ Read `registry/skills/ui-ux-pro-max.md`
+
+---
+
+## Registered Skills
+
+| Skill         | Domain           | Stars  | Description                                                     |
+|---------------|------------------|--------|-----------------------------------------------------------------|
+| karpathy      | Agent behavior   | ★174k  | Coding discipline and behavioral guardrails for AI agents       |
+| superpowers   | Planning         | ★226k  | Structured workflows for multi-step agentic tasks               |
+| ponytail      | Code craft       | ★50k   | Laziest senior dev in the room — YAGNI, minimal, delete bloat   |
+| emil          | Animation        | ★2.8k  | Motion philosophy from Vercel and Linear design engineering     |
+| taste         | Visual taste     | ★45k   | Anti-slop visual direction, stop AI generating generic UI       |
+| impeccable    | Design language  | ★28k   | Brand vs product thinking, visual vocabulary and design systems |
+| ui-ux-pro-max | UX patterns      | ★88k   | 99 UX guidelines, component patterns, accessibility             |
+
+---
+
+## Adding Skills
+
+To propose a skill: open an issue at https://github.com/nave-dev/nave  
+Include: repo URL, domain, and why it belongs in a curated collection.

package/agents/index.json

@@ -0,0 +1,53 @@
+{
+  "agents": [
+    {
+      "id": "claude",
+      "name": "Claude Code",
+      "detect": ".claude",
+      "skillPath": ".claude/skills/nave",
+      "configNote": "Works with claude_code_skills in .claude/settings.json"
+    },
+    {
+      "id": "cursor",
+      "name": "Cursor",
+      "detect": ".cursor",
+      "skillPath": ".cursor/skills/nave",
+      "configNote": "Place in .cursor/skills/ for automatic pickup"
+    },
+    {
+      "id": "opencode",
+      "name": "OpenCode",
+      "detect": ".opencode",
+      "skillPath": ".opencode/skills/nave",
+      "configNote": "OpenCode reads .opencode/skills/ on startup"
+    },
+    {
+      "id": "codex",
+      "name": "Codex CLI",
+      "detect": ".codex",
+      "skillPath": ".codex/skills/nave",
+      "configNote": "Codex reads .codex/skills/ in project root"
+    },
+    {
+      "id": "windsurf",
+      "name": "Windsurf",
+      "detect": ".windsurf",
+      "skillPath": ".windsurf/skills/nave",
+      "configNote": "Works with Windsurf AI rules system"
+    },
+    {
+      "id": "gemini",
+      "name": "Gemini CLI",
+      "detect": ".gemini",
+      "skillPath": ".gemini/skills/nave",
+      "configNote": "Gemini CLI reads .gemini/skills/ on startup"
+    },
+    {
+      "id": "antigravity",
+      "name": "Antigravity",
+      "detect": ".antigravity",
+      "skillPath": ".antigravity/skills/nave",
+      "configNote": "Antigravity skills directory"
+    }
+  ]
+}

package/install.js

@@ -0,0 +1,236 @@
+#!/usr/bin/env node
+
+const fs = require("fs");
+const path = require("path");
+
+const RESET = "\x1b[0m";
+const BOLD = "\x1b[1m";
+const DIM = "\x1b[2m";
+const CYAN = "\x1b[36m";
+const GREEN = "\x1b[32m";
+const YELLOW = "\x1b[33m";
+const RED = "\x1b[31m";
+
+const cwd = process.cwd();
+
+// Load registry and agent configs
+const registry = JSON.parse(
+  fs.readFileSync(path.join(__dirname, "registry/index.json"), "utf8")
+);
+const agentConfig = JSON.parse(
+  fs.readFileSync(path.join(__dirname, "agents/index.json"), "utf8")
+);
+
+function log(msg) { process.stdout.write(msg + "\n"); }
+function info(msg) { log(`${CYAN}nave${RESET}  ${msg}`); }
+function ok(msg) { log(`${GREEN}✓${RESET}     ${msg}`); }
+function warn(msg) { log(`${YELLOW}!${RESET}     ${msg}`); }
+function err(msg) { log(`${RED}✗${RESET}     ${msg}`); }
+
+// Detect which agents are present in the current project
+function detectAgents() {
+  return agentConfig.agents.filter((agent) =>
+    fs.existsSync(path.join(cwd, agent.detect))
+  );
+}
+
+// Generate SKILL.md dynamically from registry
+function generateSkillMd() {
+  const skills = registry.skills;
+
+  const registryTable = skills
+    .map(
+      (s) =>
+        `| ${s.id.padEnd(16)} | ${s.domain.padEnd(16)} | ★${s.stars.padEnd(5)} | ${s.description} |`
+    )
+    .join("\n");
+
+  const classificationRules = skills
+    .map((s) => {
+      const triggers = s.triggers.map((t) => `"${t}"`).join(", ");
+      return `### ${s.name} [${s.id}]\n**Domain:** ${s.domain}  \n**Triggers:** ${triggers}\n→ Read \`registry/skills/${s.id}.md\``;
+    })
+    .join("\n\n---\n\n");
+
+  return `---
+name: nave
+description: >
+  Nave is a curated skill router. Before responding to any task, classify
+  it against the registry below and read the matching skill file.
+  On clear tasks: execute directly. On ambiguous tasks: show the skill
+  layout and ask which direction before proceeding.
+version: ${registry.version}
+---
+
+# Nave — Skill Router
+
+Nave curates the best craft skills for building software and routes every
+task to the right one automatically.
+
+---
+
+## Protocol
+
+**Step 1 — Classify the task**
+Run the classification rules below. First match wins.
+
+**Step 2 — Clarity check**
+- Clear and specific → execute directly, open with: \`[nave] → {skill} activated\`
+- Ambiguous (e.g. "fix this", "make it better") → show the Skill Layout below,
+  state which 2-3 skills apply and what each would do differently, ask which direction
+
+**Step 3 — Read the skill file**
+Load the matched skill from \`registry/skills/{id}.md\` and apply its philosophy.
+
+---
+
+## Skill Layout
+
+Show this when task is ambiguous or when user runs \`/nave\`:
+
+\`\`\`
+┌──────────────────┬──────────────────┬───────┬────────────────────────────────────────┐
+│ skill            │ domain           │ stars │ best for                               │
+├──────────────────┼──────────────────┼───────┼────────────────────────────────────────┤
+${skills
+  .map(
+    (s) =>
+      `│ ${s.id.padEnd(16)} │ ${s.domain.padEnd(16)} │ ★${s.stars.padEnd(5)} │ ${s.description.substring(0, 38).padEnd(38)} │`
+  )
+  .join("\n")}
+└──────────────────┴──────────────────┴───────┴────────────────────────────────────────┘
+\`\`\`
+
+After showing the layout: state the suggested skill and one-line reason. Ask to confirm.
+
+---
+
+## Classification Rules
+
+${classificationRules}
+
+---
+
+## Registered Skills
+
+| Skill            | Domain           | Stars  | Description |
+|---|---|---|---|
+${registryTable}
+
+---
+
+## Adding Skills
+
+To propose a skill: open an issue at https://github.com/nave-dev/nave
+Include: repo URL, domain, and why it belongs in a curated collection.
+`;
+}
+
+// Install nave into a target agent directory
+function install(agent, targetDir) {
+  const absTarget = path.join(cwd, targetDir);
+
+  // Create directories
+  fs.mkdirSync(path.join(absTarget, "registry/skills"), { recursive: true });
+
+  // Generate and write SKILL.md
+  const skillMd = generateSkillMd();
+  fs.writeFileSync(path.join(absTarget, "SKILL.md"), skillMd);
+
+  // Copy all skill reference files
+  const skillsDir = path.join(__dirname, "registry/skills");
+  const skillFiles = fs.readdirSync(skillsDir);
+  for (const file of skillFiles) {
+    fs.copyFileSync(
+      path.join(skillsDir, file),
+      path.join(absTarget, "registry/skills", file)
+    );
+  }
+
+  ok(`Installed to ${DIM}${targetDir}${RESET}`);
+  if (agent.configNote) {
+    log(`      ${DIM}${agent.configNote}${RESET}`);
+  }
+}
+
+// Main
+function main() {
+  const args = process.argv.slice(2);
+  const cmd = args[0];
+
+  log("");
+  log(`  ${BOLD}nave${RESET} ${DIM}v${registry.version}${RESET}`);
+  log(`  ${DIM}curated skill router for AI coding agents${RESET}`);
+  log("");
+
+  if (cmd === "list") {
+    log(`  ${BOLD}Registered skills:${RESET}`);
+    log("");
+    for (const skill of registry.skills) {
+      log(`  ${CYAN}${skill.id.padEnd(18)}${RESET} ${skill.domain.padEnd(18)} ★${skill.stars.padEnd(6)} ${DIM}${skill.repo}${RESET}`);
+    }
+    log("");
+    return;
+  }
+
+  if (cmd === "generate") {
+    // Just generate SKILL.md in current directory (for repo maintenance)
+    const skillMd = generateSkillMd();
+    fs.writeFileSync(path.join(cwd, "SKILL.md"), skillMd);
+    ok("Generated SKILL.md in current directory");
+    return;
+  }
+
+  // Default: install
+  const detected = detectAgents();
+
+  if (detected.length === 0) {
+    warn("No AI coding agent detected in this project.");
+    log("");
+    log("  Nave looks for: " + agentConfig.agents.map((a) => a.detect).join(", "));
+    log("");
+    log("  Install your agent first, then run npx nave again.");
+    log("  Or specify a target: npx nave --agent claude");
+    log("");
+    return;
+  }
+
+  // --agent flag
+  const agentFlag = args.find((a) => a.startsWith("--agent=") || a === "--agent");
+  let targets = detected;
+
+  if (agentFlag) {
+    const agentId = agentFlag.includes("=")
+      ? agentFlag.split("=")[1]
+      : args[args.indexOf("--agent") + 1];
+    const specific = agentConfig.agents.find((a) => a.id === agentId);
+    if (!specific) {
+      err(`Unknown agent: ${agentId}`);
+      log("  Available: " + agentConfig.agents.map((a) => a.id).join(", "));
+      return;
+    }
+    targets = [specific];
+  }
+
+  if (targets.length > 1) {
+    info(`Detected ${targets.length} agents: ${targets.map((a) => a.name).join(", ")}`);
+  } else {
+    info(`Detected: ${targets[0].name}`);
+  }
+
+  log("");
+
+  for (const agent of targets) {
+    log(`  Installing for ${BOLD}${agent.name}${RESET}...`);
+    install(agent, agent.skillPath);
+  }
+
+  log("");
+  log(`  ${GREEN}${BOLD}Done.${RESET} Nave is live in your project.`);
+  log("");
+  log(`  ${DIM}Skills installed: ${registry.skills.map((s) => s.id).join(", ")}${RESET}`);
+  log(`  ${DIM}Run /nave in your agent to see the skill registry.${RESET}`);
+  log("");
+}
+
+main();

package/package.json

@@ -0,0 +1,35 @@
+{
+  "name": "@geoemmanuel/nave",
+  "version": "0.1.0",
+  "description": "A curated skill router for AI coding agents",
+  "main": "install.js",
+  "bin": {
+    "nave": "./install.js"
+  },
+  "scripts": {
+    "postinstall": "node install.js",
+    "generate": "node install.js generate",
+    "list": "node install.js list"
+  },
+  "keywords": [
+    "ai", "agent", "skills", "claude-code", "cursor", "codex",
+    "opencode", "windsurf", "skill-router", "llm"
+  ],
+  "author": "George Emmanuel <geoemman678@gmail.com>",
+  "license": "MIT",
+  "repository": {
+    "type": "git",
+    "url": "https://github.com/geoemmanuel/nave"
+  },
+  "homepage": "https://github.com/geoemmanuel/nave",
+  "engines": {
+    "node": ">=18.0.0"
+  },
+  "files": [
+    "install.js",
+    "registry/",
+    "agents/",
+    "SKILL.md",
+    "README.md"
+  ]
+}

package/registry/index.json

@@ -0,0 +1,109 @@
+{
+  "version": "0.1.0",
+  "skills": [
+    {
+      "id": "karpathy",
+      "name": "Karpathy",
+      "author": "multica-ai",
+      "repo": "https://github.com/multica-ai/andrej-karpathy-skills",
+      "stars": "174k",
+      "domain": "Agent behavior",
+      "description": "Coding discipline and behavioral guardrails for AI agents",
+      "triggers": [
+        "over-engineering", "stay focused", "think before coding",
+        "wrong direction", "silent assumptions", "production code",
+        "critical change", "multi-file task"
+      ],
+      "file": "registry/skills/karpathy.md"
+    },
+    {
+      "id": "superpowers",
+      "name": "Superpowers",
+      "author": "obra",
+      "repo": "https://github.com/obra/superpowers",
+      "stars": "226k",
+      "domain": "Planning",
+      "description": "Structured workflows for multi-step agentic tasks",
+      "triggers": [
+        "plan this", "architect this", "multi-step", "long task",
+        "don't lose context", "PRD", "spec", "from scratch", "multi-file refactor"
+      ],
+      "file": "registry/skills/superpowers.md"
+    },
+    {
+      "id": "ponytail",
+      "name": "Ponytail",
+      "author": "DietrichGebert",
+      "repo": "https://github.com/DietrichGebert/ponytail",
+      "stars": "50k",
+      "domain": "Code craft",
+      "description": "Laziest senior dev in the room. YAGNI, minimal, delete bloat",
+      "triggers": [
+        "simplest solution", "too complex", "YAGNI", "over-engineered",
+        "delete this", "minimal", "be lazy", "shortest path",
+        "adding dependency", "new abstraction"
+      ],
+      "file": "registry/skills/ponytail.md"
+    },
+    {
+      "id": "emil",
+      "name": "Emil Design Engineering",
+      "author": "emilkowalski",
+      "repo": "https://github.com/emilkowalski/skills",
+      "stars": "2.8k",
+      "domain": "Animation & motion",
+      "description": "Motion philosophy from Vercel and Linear design engineering",
+      "triggers": [
+        "animation", "transition", "easing", "motion", "spring",
+        "framer-motion", "hover effect", "page transition", "micro-interaction",
+        "scroll animation", "keyframe", "gesture"
+      ],
+      "file": "registry/skills/emil.md"
+    },
+    {
+      "id": "taste",
+      "name": "Taste Skill",
+      "author": "Leonxlnx",
+      "repo": "https://github.com/Leonxlnx/taste-skill",
+      "stars": "45k",
+      "domain": "Visual taste",
+      "description": "Anti-slop visual direction. Stop AI from generating generic UI",
+      "triggers": [
+        "does this look good", "too generic", "AI slop", "improve the UI",
+        "visual review", "aesthetic", "typography", "color palette",
+        "layout feedback", "make it better"
+      ],
+      "file": "registry/skills/taste.md"
+    },
+    {
+      "id": "impeccable",
+      "name": "Impeccable",
+      "author": "pbakaus",
+      "repo": "https://github.com/pbakaus/impeccable",
+      "stars": "28k",
+      "domain": "Design language",
+      "description": "Brand vs product thinking. Visual vocabulary and design systems",
+      "triggers": [
+        "design language", "design system", "brand", "visual vocabulary",
+        "typography system", "color system", "spacing scale",
+        "feel like a real product", "design tokens"
+      ],
+      "file": "registry/skills/impeccable.md"
+    },
+    {
+      "id": "ui-ux-pro-max",
+      "name": "UI UX Pro Max",
+      "author": "nextlevelbuilder",
+      "repo": "https://github.com/nextlevelbuilder/ui-ux-pro-max-skill",
+      "stars": "88k",
+      "domain": "UX patterns",
+      "description": "99 UX guidelines, component patterns, accessibility, cross-platform",
+      "triggers": [
+        "new page", "dashboard", "landing", "onboarding", "settings",
+        "component architecture", "form design", "navigation pattern",
+        "accessibility", "responsive layout", "user flow", "admin panel"
+      ],
+      "file": "registry/skills/ui-ux-pro-max.md"
+    }
+  ]
+}

package/registry/skills/emil.md

@@ -0,0 +1,33 @@
+# Emil Design Engineering — Animation & Motion
+> github.com/emilkowalski/skills · 2.8k stars
+
+## Core Principle
+Animation should be invisible. When done right, users don't notice it —
+they notice the interface feels good.
+
+## Timing
+- Micro-interactions (hover, press, toggle): 150-250ms
+- Standard transitions (modals, panels): 200-350ms
+- Complex sequences (page transitions): 300-500ms
+
+## Easing Rules
+- Enter: ease-out (starts fast, decelerates in)
+- Exit: ease-in (starts slow, accelerates out)
+- Never ease-in for enter. Most common agent mistake.
+
+## Animation Patterns
+- Fade + rise: opacity 0 y:8 to opacity 1 y:0
+- Scale for emphasis: scale(0.96) + opacity, never scale(0)
+- Menus: transform-origin at trigger, scale + opacity from source
+- Tooltips: delay on first, instant on subsequent
+
+## Performance
+- transform and opacity only. Never animate width, height, top, left.
+- CSS for predetermined animations, JS for dynamic/interruptible
+- Always respect prefers-reduced-motion
+
+## Common Agent Mistakes
+- ease-in on enter (should be ease-out)
+- scale(0) as start (use scale(0.96))
+- Missing transform-origin on menus
+- All tooltips with same delay

package/registry/skills/impeccable.md

@@ -0,0 +1,27 @@
+# Impeccable — Design Language
+> github.com/pbakaus/impeccable · 28k stars
+
+## Brand vs Product Mode
+Ask first:
+- Brand: design IS the message. Bold choices, personality-forward.
+- Product: design SERVES function. Restraint, patterns, invisible craft.
+
+Most SaaS/tools = Product. Marketing/portfolio = Brand.
+
+## Design Language Checklist
+- Display face is characterful, used with restraint
+- Type scale has clear hierarchy (display, heading, body, caption)
+- Palette: 4-6 named values, one accent used sparingly
+- Spacing: all values multiples of a base unit
+- One element this design will be remembered by. Everything else is quiet.
+
+## Three AI Defaults to Avoid
+1. Warm cream + terracotta accent + high-contrast serif
+2. Near-black + acid-green or vermilion accent
+3. Dense newspaper columns + hairline rules + zero border-radius
+
+If your design reads as one of these: it is a default, not a choice.
+
+## Restraint Rule
+Spend your boldness in one place.
+Before finishing: remove one decorative decision.

package/registry/skills/karpathy.md

@@ -0,0 +1,30 @@
+# Karpathy — Agent Behavior
+> github.com/multica-ai/andrej-karpathy-skills · 174k stars
+
+## The Four Principles
+
+### 1. Think Before Coding
+State assumptions explicitly before acting. If underspecified, ask one focused
+question. Never silently guess and charge ahead.
+
+Before writing code: restate the goal, name your assumptions, flag anything risky.
+
+### 2. Simplicity First
+Write minimum code that solves the problem. No speculative abstractions,
+no future-proofing that wasn't asked for. 50 lines not 500.
+
+### 3. Surgical Changes
+Touch only what you must. Never refactor adjacent code that wasn't asked.
+Never silently delete anything. Match existing codebase style.
+
+If you see something broken nearby: mention it, don't fix it without asking.
+
+### 4. Goal-Driven Execution
+Define a clear success condition before starting. Loop until it's met.
+Surface blockers explicitly — never work around them silently.
+
+## Failure Modes This Prevents
+- Confident implementation of the wrong thing
+- 10-line fix turned into 200-line refactor
+- Modifying files never asked to touch
+- Declaring success before verifying the actual goal

package/registry/skills/ponytail.md

@@ -0,0 +1,27 @@
+# Ponytail — Code Minimalism
+> github.com/DietrichGebert/ponytail · 50k stars
+
+## Philosophy
+You are a lazy senior developer. Lazy means efficient, not careless.
+The best code is the code never written.
+
+## The Ladder (run in order, stop at first match)
+1. Does this need to exist? Skip it, say so in one line. (YAGNI)
+2. Does stdlib do it? Use stdlib.
+3. Does a native platform feature cover it? Use it.
+4. Does an already-installed dep solve it? Never add a dep for 3 lines of code.
+5. Can it be one line? One line.
+6. Only then: write the minimum code that works.
+
+## Rules
+- No unrequested abstractions
+- No boilerplate "for later"
+- Deletion over addition
+- Mark shortcuts: // ponytail: <ceiling>, upgrade when <trigger>
+
+## Anti-patterns to Catch
+- Reinvented standard library
+- New dependency for a 3-line problem
+- Speculative abstractions
+- Config for a value that never changes
+- Interface with exactly one implementation

package/registry/skills/superpowers.md

@@ -0,0 +1,32 @@
+# Superpowers — Planning & Multi-step Tasks
+> github.com/obra/superpowers · 226k stars
+
+## When to Use
+Task has 3+ steps, spans multiple files, or is long enough to risk context loss.
+
+## Plan First
+Before any code, write a plan to .nave/plan.md:
+
+  Goal: {one sentence — what done looks like}
+
+  Steps:
+  1. {concrete step}
+  2. {concrete step}
+
+  Success criteria: {how to verify it's done}
+  Files to change: {known files — add as discovered}
+
+Check off steps as completed. Plan survives /clear and context resets.
+
+## Execution Rules
+- Complete one step fully before moving to next
+- Verify it works after each step
+- Blocked: surface it explicitly, never work around silently
+- Context reset: re-read plan file, continue from last checked step
+
+## Done Checklist
+- All steps checked off
+- Tests pass
+- No TODO comments unaddressed
+- No files in broken intermediate state
+- Diff scoped to what was asked

package/registry/skills/taste.md

@@ -0,0 +1,30 @@
+# Taste Skill — Visual Taste & Anti-Slop
+> github.com/Leonxlnx/taste-skill · 45k stars
+
+## Three Parameters
+Set before designing:
+
+- DESIGN_VARIANCE (1-10): 1=clean/conventional, 10=editorial/experimental
+- MOTION_INTENSITY (1-10): 1=simple hover, 10=orchestrated/magnetic
+- VISUAL_DENSITY (1-10): 1=airy/luxury, 10=dense/data-rich
+
+Defaults: product=4,3,4 / marketing=7,5,2 / dashboard=2,2,8
+
+## Anti-Slop Checklist
+Remove if unintentional:
+- Pulsing indicators on everything
+- Blur-everywhere entrance animations
+- Hover-scale on every interactive element
+- Bouncy springs on utility actions
+- Uniform fade-ins on everything
+- Card shadows too heavy and uniform
+- Every section has its own background color
+- Gradient text on headings as default
+
+## Typography
+- Display: line-height 1.1-1.2, letter-spacing -0.02em
+- Body: line-height 1.5-1.6, default letter-spacing
+
+## Spacing
+- Base unit 8px or 4px, all values are multiples
+- Never mix rhythms within a component

package/registry/skills/ui-ux-pro-max.md

@@ -0,0 +1,36 @@
+# UI UX Pro Max — Patterns & Accessibility
+> github.com/nextlevelbuilder/ui-ux-pro-max-skill · 88k stars
+
+## Page-Level Rules
+- One primary action per view
+- Above the fold: answer "what is this and what do I do"
+- Progressive disclosure: hide complexity until needed
+
+## Navigation
+- Top nav: broad multi-section products
+- Side nav: complex tools, dashboards, admin
+- Bottom nav: mobile, 3-5 destinations
+- Tabs: sibling views at same hierarchy level
+- Never mix patterns at the same level
+
+## Forms
+- One column (two only for truly paired fields like first/last name)
+- Labels above fields, never placeholder-only
+- Inline validation on blur, not every keystroke
+- Error messages specific: "Enter a valid email" not "Invalid input"
+
+## Buttons
+- One primary CTA per view
+- Label = verb + noun ("Save changes" not "Submit")
+- Min touch target: 44px web / 44pt iOS / 48dp Android
+
+## Modals
+- Use for: confirmations, focused tasks
+- Avoid for: complex multi-step flows, errors that could be inline
+- Always: close button + Escape key + focus trap
+
+## Accessibility Non-negotiables
+- Color alone never conveys meaning
+- Visible focus indicators on all interactive elements
+- aria-label on icon-only buttons
+- Min contrast: 4.5:1 body, 3:1 large text