azclaude-copilot 0.4.19 → 0.4.21

package/.claude-plugin/marketplace.json

@@ -1,27 +1,27 @@
-{
-  "name": "azclaude-marketplace",
-  "description": "AZCLAUDE — A complete AI coding environment for Claude Code",
-  "owner": {
-    "name": "haytamAroui",
-    "url": "https://github.com/haytamAroui"
-  },
-  "plugins": [
-    {
-      "name": "azclaude",
-      "description": "AZCLAUDE is a complete AI coding environment for Claude Code. It installs 33 commands, 8 auto-invoked skills, 13 specialized agents, 4 hooks, and a persistent memory system — in one command.\n\nKey features:\n• Memory across sessions — goals.md + checkpoints injected automatically before every session\n• Self-improving loop — /reflect fixes stale CLAUDE.md rules, /reflexes learns from tool-use patterns, /evolve creates agents from git evidence\n• Autonomous copilot mode — /copilot runs a three-tier team (orchestrator → problem-architect → milestone-builder) across sessions until the product ships\n• Spec-driven workflow — /constitute writes project rules, /spec writes structured ACs, /analyze detects plan drift and ghost milestones, /blueprint traces every milestone to a spec\n• Security layer — 102-rule environment scan (/sentinel), pre-write secret blocking, pre-ship credential audit\n• Progressive levels 0–10 — start with CLAUDE.md, grow into multi-agent pipelines and self-evolving environments\n• Zero dependencies — no npm packages, no external APIs, no vector databases. Plain markdown files and Claude Code's native architecture.\n• Smart install — npx azclaude-copilot@latest auto-detects first install vs upgrade vs verify. Context-aware onboarding shows the right next command for your project state.\n\nExample use cases:\n• /setup — scan an existing project, detect stack + domain + scale, fill CLAUDE.md, generate project-specific skills and agents automatically\n• /copilot \"Build a compliance SaaS with trilingual support\" — walk away, come back to working code across multiple sessions\n• /sentinel — run a scored security audit (0–100, grade A–F) across hooks, permissions, MCP servers, agent configs, and secrets\n• /evolve — detect gaps in the environment, generate new skills and agents from git co-change evidence, report score delta (e.g. 42/100 → 68/100)\n• /constitute — write your project's constitution (non-negotiables, architectural commitments, definition of done) — gates all future AI actions\n• /analyze — cross-artifact consistency check: ghost milestones, spec vs. code drift, unplanned commits\n• /reflect — find stale, missing, or contradicting rules in CLAUDE.md and propose exact fixes\n• /debate \"REST vs GraphQL for this project\" — adversarial evidence-based decision with order-independent scoring, logged to decisions.md",
-      "version": "0.4.18",
-      "source": {
-        "source": "github",
-        "repo": "haytamAroui/AZ-CLAUDE-COPILOT",
-        "ref": "main"
-      },
-      "author": {
-        "name": "haytamAroui",
-        "url": "https://github.com/haytamAroui"
-      },
-      "homepage": "https://github.com/haytamAroui/AZ-CLAUDE-COPILOT/blob/main/DOCS.md",
-      "license": "MIT",
-      "keywords": ["memory", "setup", "agents", "commands", "hooks", "domain-aware", "spec-driven", "constitution", "copilot"]
-    }
-  ]
-}
+{
+  "name": "azclaude-marketplace",
+  "description": "AZCLAUDE — A complete AI coding environment for Claude Code",
+  "owner": {
+    "name": "haytamAroui",
+    "url": "https://github.com/haytamAroui"
+  },
+  "plugins": [
+    {
+      "name": "azclaude",
+      "description": "AZCLAUDE is a complete AI coding environment for Claude Code. It installs 33 commands, 8 auto-invoked skills, 13 specialized agents, 4 hooks, and a persistent memory system — in one command.\n\nKey features:\n• Memory across sessions — goals.md + checkpoints injected automatically before every session\n• Self-improving loop — /reflect fixes stale CLAUDE.md rules, /reflexes learns from tool-use patterns, /evolve creates agents from git evidence\n• Autonomous copilot mode — /copilot runs a three-tier team (orchestrator → problem-architect → milestone-builder) across sessions until the product ships\n• Spec-driven workflow — /constitute writes project rules, /spec writes structured ACs, /analyze detects plan drift and ghost milestones, /blueprint traces every milestone to a spec\n• Security layer — 102-rule environment scan (/sentinel), pre-write secret blocking, pre-ship credential audit\n• Progressive levels 0–10 — start with CLAUDE.md, grow into multi-agent pipelines and self-evolving environments\n• Zero dependencies — no npm packages, no external APIs, no vector databases. Plain markdown files and Claude Code's native architecture.\n• Smart install — npx azclaude-copilot@latest auto-detects first install vs upgrade vs verify. Context-aware onboarding shows the right next command for your project state.\n\nExample use cases:\n• /setup — scan an existing project, detect stack + domain + scale, fill CLAUDE.md, generate project-specific skills and agents automatically\n• /copilot \"Build a compliance SaaS with trilingual support\" — walk away, come back to working code across multiple sessions\n• /sentinel — run a scored security audit (0–100, grade A–F) across hooks, permissions, MCP servers, agent configs, and secrets\n• /evolve — detect gaps in the environment, generate new skills and agents from git co-change evidence, report score delta (e.g. 42/100 → 68/100)\n• /constitute — write your project's constitution (non-negotiables, architectural commitments, definition of done) — gates all future AI actions\n• /analyze — cross-artifact consistency check: ghost milestones, spec vs. code drift, unplanned commits\n• /reflect — find stale, missing, or contradicting rules in CLAUDE.md and propose exact fixes\n• /debate \"REST vs GraphQL for this project\" — adversarial evidence-based decision with order-independent scoring, logged to decisions.md",
+      "version": "0.4.19",
+      "source": {
+        "source": "github",
+        "repo": "haytamAroui/AZ-CLAUDE-COPILOT",
+        "ref": "main"
+      },
+      "author": {
+        "name": "haytamAroui",
+        "url": "https://github.com/haytamAroui"
+      },
+      "homepage": "https://github.com/haytamAroui/AZ-CLAUDE-COPILOT/blob/main/DOCS.md",
+      "license": "MIT",
+      "keywords": ["memory", "setup", "agents", "commands", "hooks", "domain-aware", "spec-driven", "constitution", "copilot"]
+    }
+  ]
+}

package/.claude-plugin/plugin.json

@@ -1,17 +1,17 @@
-{
-  "name": "azclaude",
-  "version": "0.4.18",
-  "description": "AZCLAUDE is a complete AI coding environment for Claude Code. It installs 33 commands, 8 auto-invoked skills, 13 specialized agents, 4 hooks, and a persistent memory system — in one command.\n\nKey features:\n• Memory across sessions — goals.md + checkpoints injected automatically before every session\n• Self-improving loop — /reflect fixes stale CLAUDE.md rules, /reflexes learns from tool-use patterns, /evolve creates agents from git evidence\n• Autonomous copilot mode — /copilot runs a three-tier team (orchestrator → problem-architect → milestone-builder) across sessions until the product ships\n• Spec-driven workflow — /constitute writes project rules, /spec writes structured ACs, /analyze detects plan drift and ghost milestones, /blueprint traces every milestone to a spec\n• Security layer — 102-rule environment scan (/sentinel), pre-write secret blocking, pre-ship credential audit\n• Progressive levels 0–10 — start with CLAUDE.md, grow into multi-agent pipelines and self-evolving environments\n• Zero dependencies — no npm packages, no external APIs, no vector databases. Plain markdown files and Claude Code's native architecture.\n• Smart install — npx azclaude-copilot@latest auto-detects first install vs upgrade vs verify. Context-aware onboarding shows the right next command for your project state.\n\nExample use cases:\n• /setup — scan an existing project, detect stack + domain + scale, fill CLAUDE.md, generate project-specific skills and agents automatically\n• /copilot \"Build a compliance SaaS with trilingual support\" — walk away, come back to working code across multiple sessions\n• /sentinel — run a scored security audit (0–100, grade A–F) across hooks, permissions, MCP servers, agent configs, and secrets\n• /evolve — detect gaps in the environment, generate new skills and agents from git co-change evidence, report score delta (e.g. 42/100 → 68/100)\n• /constitute — write your project's constitution (non-negotiables, architectural commitments, definition of done) — gates all future AI actions\n• /analyze — cross-artifact consistency check: ghost milestones, spec vs. code drift, unplanned commits\n• /reflect — find stale, missing, or contradicting rules in CLAUDE.md and propose exact fixes\n• /debate \"REST vs GraphQL for this project\" — adversarial evidence-based decision with order-independent scoring, logged to decisions.md",
-  "author": {
-    "name": "haytamAroui",
-    "url": "https://github.com/haytamAroui"
-  },
-  "homepage": "https://github.com/haytamAroui/AZ-CLAUDE-COPILOT/blob/main/DOCS.md",
-  "repository": "https://github.com/haytamAroui/AZ-CLAUDE-COPILOT",
-  "license": "MIT",
-  "keywords": ["memory", "setup", "agents", "commands", "context", "lazy-loading", "hooks", "domain-aware"],
-  "commands": "./templates/commands/",
-  "skills": "./templates/skills/",
-  "agents": "./templates/agents/",
-  "hooks": "./templates/hooks/hooks.json"
-}
+{
+  "name": "azclaude",
+  "version": "0.4.19",
+  "description": "AZCLAUDE is a complete AI coding environment for Claude Code. It installs 33 commands, 8 auto-invoked skills, 13 specialized agents, 4 hooks, and a persistent memory system — in one command.\n\nKey features:\n• Memory across sessions — goals.md + checkpoints injected automatically before every session\n• Self-improving loop — /reflect fixes stale CLAUDE.md rules, /reflexes learns from tool-use patterns, /evolve creates agents from git evidence\n• Autonomous copilot mode — /copilot runs a three-tier team (orchestrator → problem-architect → milestone-builder) across sessions until the product ships\n• Spec-driven workflow — /constitute writes project rules, /spec writes structured ACs, /analyze detects plan drift and ghost milestones, /blueprint traces every milestone to a spec\n• Security layer — 102-rule environment scan (/sentinel), pre-write secret blocking, pre-ship credential audit\n• Progressive levels 0–10 — start with CLAUDE.md, grow into multi-agent pipelines and self-evolving environments\n• Zero dependencies — no npm packages, no external APIs, no vector databases. Plain markdown files and Claude Code's native architecture.\n• Smart install — npx azclaude-copilot@latest auto-detects first install vs upgrade vs verify. Context-aware onboarding shows the right next command for your project state.\n\nExample use cases:\n• /setup — scan an existing project, detect stack + domain + scale, fill CLAUDE.md, generate project-specific skills and agents automatically\n• /copilot \"Build a compliance SaaS with trilingual support\" — walk away, come back to working code across multiple sessions\n• /sentinel — run a scored security audit (0–100, grade A–F) across hooks, permissions, MCP servers, agent configs, and secrets\n• /evolve — detect gaps in the environment, generate new skills and agents from git co-change evidence, report score delta (e.g. 42/100 → 68/100)\n• /constitute — write your project's constitution (non-negotiables, architectural commitments, definition of done) — gates all future AI actions\n• /analyze — cross-artifact consistency check: ghost milestones, spec vs. code drift, unplanned commits\n• /reflect — find stale, missing, or contradicting rules in CLAUDE.md and propose exact fixes\n• /debate \"REST vs GraphQL for this project\" — adversarial evidence-based decision with order-independent scoring, logged to decisions.md",
+  "author": {
+    "name": "haytamAroui",
+    "url": "https://github.com/haytamAroui"
+  },
+  "homepage": "https://github.com/haytamAroui/AZ-CLAUDE-COPILOT/blob/main/DOCS.md",
+  "repository": "https://github.com/haytamAroui/AZ-CLAUDE-COPILOT",
+  "license": "MIT",
+  "keywords": ["memory", "setup", "agents", "commands", "context", "lazy-loading", "hooks", "domain-aware"],
+  "commands": "./templates/commands/",
+  "skills": "./templates/skills/",
+  "agents": "./templates/agents/",
+  "hooks": "./templates/hooks/hooks.json"
+}

package/README.md

@@ -807,11 +807,11 @@ Run `/level-up` at any time to see your current level and build the next one.
 
 ## Verified
 
-1359 tests. Every template, command, capability, agent, hook, and CLI feature verified.
+1366 tests. Every template, command, capability, agent, hook, and CLI feature verified.
 
 ```bash
 bash tests/test-features.sh
-# Results: 1359 passed, 0 failed, 1359 total
+# Results: 1366 passed, 0 failed, 1366 total
 ```
 
 ---

package/bin/cli.js

@@ -373,7 +373,7 @@ function installCommands(projectDir, cfg) {
 
 // ─── Skills (SKILL.md — model-auto-invoked) ──────────────────────────────────
 
-const SKILLS = ['session-guard', 'test-first', 'env-scanner', 'debate', 'security', 'skill-creator', 'agent-creator', 'architecture-advisor'];
+const SKILLS = ['session-guard', 'test-first', 'env-scanner', 'debate', 'security', 'skill-creator', 'agent-creator', 'architecture-advisor', 'frontend-design'];
 
 function installSkills(projectDir, cfg) {
   const skillsDir = path.join(projectDir, cfg, 'skills');

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "azclaude-copilot",
-  "version": "0.4.19",
+  "version": "0.4.21",
   "description": "AI coding environment — 33 commands, 8 skills, 13 agents, memory, reflexes, evolution. Install: npx azclaude-copilot@latest, then open Claude Code.",
   "bin": {
     "azclaude": "bin/cli.js",

package/templates/skills/frontend-design/SKILL.md

@@ -0,0 +1,169 @@
+---
+name: frontend-design
+description: >
+  Production-grade, visually distinctive frontend interfaces. Use when building
+  a landing page, creating a dashboard, designing a React component, making a
+  website, building something visually striking, creating a UI, designing a
+  poster, or when the user says "build me a site", "make this look good",
+  "design a page", "create a frontend", "visually striking", "beautiful UI",
+  "modern design", or any task where the primary deliverable is a rendered
+  interface. Also fires when /copilot reaches a milestone whose files include
+  index.html, .jsx, .tsx, .css, or .scss.
+  Do NOT trigger when: user asks to review existing UI (use code-reviewer),
+  request is code-only with no visual deliverable, or a strict brand guide
+  already defines all visual decisions.
+---
+
+# Frontend Design Skill
+
+Produces interfaces that are visually distinctive and production-ready.
+The core constraint: every design choice must be intentional, not default.
+
+## Step 0: Context Detection
+
+Before writing any code, read three sources.
+
+**1. Project stack and domain**
+Read `CLAUDE.md` — find `Domain:` and `Stack:` fields.
+If stack contains React/Next/Vue/Svelte: use component file structure.
+If stack is plain HTML: single index.html with embedded or linked CSS/JS.
+
+**2. Existing design system or brand guide**
+```bash
+ls .claude/constitution.md 2>/dev/null
+find . -name "design-tokens*" -o -name "brand-guide*" -o -name "theme.*" 2>/dev/null | head -5
+```
+If `constitution.md` exists: read `## Visual Identity` section — it is non-negotiable.
+If design tokens exist: import them. Do not invent conflicting values.
+
+**3. Prior aesthetic decisions**
+```bash
+grep -i "font\|color\|aesthetic\|palette" .claude/memory/decisions.md 2>/dev/null | head -10
+```
+If prior decisions exist: stay consistent with them unless the user explicitly wants a change.
+
+---
+
+## Step 1: Pick an Aesthetic Direction
+
+Choose ONE direction before writing any code.
+Intentionality beats intensity — a minimal design executed with conviction beats a
+maximalist design that hedges.
+
+See `references/aesthetic-directions.md` for fonts, colors, and techniques per direction.
+
+| Direction | Core signal |
+|-----------|-------------|
+| Brutally Minimal | Extreme whitespace, single weight, silence as design element |
+| Maximalist Chaos | Layers, collisions, overflow, density as aesthetic |
+| Retro-Futuristic | CRT nostalgia meets speculative UI — scanlines, phosphor glow |
+| Organic/Natural | Irregular shapes, earth palette, no mechanical grids |
+| Luxury/Refined | Restraint, gold/cream/black, editorial spacing |
+| Playful/Toy-like | Rounded, saturated, bouncy, tactile affordances |
+| Editorial/Magazine | Strong typographic hierarchy, column grids, pull quotes |
+| Brutalist/Raw | Exposed structure, default-looking-but-intentional, harsh juxtapositions |
+| Art Deco/Geometric | Symmetry, chevrons, gilt details, high contrast geometry |
+| Soft/Pastel | Muted saturation, gentle gradients, airy spacing |
+| Industrial/Utilitarian | Monospace, grid lines, data-forward, no decoration |
+| Cyberpunk/Dark | Dark base, neon accent, glitch effects, terminal aesthetics |
+
+State the chosen direction explicitly before proceeding. Example:
+`Direction: Retro-Futuristic — scanline overlays, phosphor green on near-black, monospace headline`
+
+---
+
+## Step 2: Implement
+
+### Typography rules
+- NEVER use: Inter, Roboto, Arial, Helvetica, system fonts without intentional treatment
+- NEVER use: Space Grotesk (convergence signal — indicates no aesthetic thinking)
+- DO use: characterful fonts from Google Fonts matched to the chosen direction
+- Set: `font-display: swap` on all imported fonts
+
+### Color rules
+- NEVER use: purple-to-pink gradient on white
+- NEVER use: evenly distributed hues without hierarchy
+- NEVER use: high-saturation rainbow palettes
+- DO use: 1 dominant + 1-2 accent colors, committed to the direction
+- DO use: a defined background treatment (texture, grain overlay, gradient mesh) — not flat solid color
+
+### Layout rules
+- NEVER use: centered hero + 3-column feature grid without modification
+- NEVER use: standard nav with no customization
+- DO use: asymmetry, overlap, diagonal flow, or grid-breaking elements
+- DO use: visual hierarchy that guides the eye in a deliberate path
+
+### Motion rules
+- Prefer CSS transitions over JS animation libraries
+- Scroll-triggered animations must serve comprehension, not decorate it
+- Hover states must be thoughtful — not just opacity change
+- Glassmorphism only if the direction calls for it (Luxury, Cyberpunk) — never as default
+
+### Detail rules
+- Rounded corners only where intentional — not applied globally by default
+- Micro-interactions must have a reason: confirm action, guide attention, indicate state
+- Every spacing value must belong to a scale (4px, 8px, 16px, 24px, 32px, 48px, 64px)
+
+---
+
+## Step 3: Complexity Budget
+
+Match animation/visual code ratio to the chosen direction.
+
+| Direction type | Animation/visual code | Semantic structure |
+|---|---|---|
+| Maximalist (Chaos, Retro, Cyberpunk) | 60-70% | 30-40% |
+| Minimalist (Minimal, Luxury, Soft) | 20-30% | 70-80% |
+| Mid-complexity (Editorial, Deco, Industrial) | 40-50% | 50-60% |
+
+Do not apply maximalist code budget to a minimalist direction. The restraint IS the design.
+
+---
+
+## Ambiguity Protocol
+
+If the request is vague (no content, no purpose stated):
+→ Ask: "What does this interface do, and who uses it? One sentence."
+
+If no framework is specified and CLAUDE.md has no stack:
+→ Default to vanilla HTML/CSS/JS. State this assumption before writing.
+
+If the user asks for "something beautiful" with no further constraint:
+→ Pick a direction from the aesthetic table, state it explicitly ("Going with Brutally Minimal — here's why"), then proceed. Do not ask for permission.
+
+If a request conflicts with constitution.md visual constraints:
+→ Flag the conflict: "constitution.md restricts X — I'll use Y instead." Do not silently override.
+
+---
+
+## Step 4: Production Requirements
+
+- Entry file: `index.html` (always — even for React, the build output target is index.html)
+- Cross-browser: test mental model against Chrome, Firefox, Safari rendering differences for any CSS used
+- Mobile-responsive: no horizontal scroll on 375px viewport, tap targets >= 44px
+- No broken links, placeholder `#` hrefs without intent, or `TODO` comments in shipped code
+- Images: use aspect-ratio, width/height attributes, or explicit dimensions to prevent layout shift
+
+---
+
+## Completion
+
+When done, output:
+
+```
+Direction: {chosen direction}
+Entry file: {absolute path to index.html}
+Line count: {wc -l output}
+Files created: {list}
+```
+
+Run:
+```bash
+wc -l index.html
+```
+
+Show the actual line count. Do not estimate.
+
+## References
+
+For the full aesthetic directions reference: `references/aesthetic-directions.md`

package/templates/skills/frontend-design/references/aesthetic-directions.md

@@ -0,0 +1,26 @@
+# Aesthetic Directions — Quick Reference
+
+12 directions for the frontend-design skill. Pick one before writing code.
+
+| Direction | Vibe | Key Techniques | Example Fonts | Example Colors |
+|-----------|------|---------------|---------------|----------------|
+| **Brutally Minimal** | Silence as design. One element per screen. Every pixel earns its place. | Extreme whitespace, single typeface weight, monochrome or near-monochrome, no decoration | Garamond, Cormorant, DM Serif Display | #FFFFFF + #0A0A0A, or #F5F0EB + #1A1A1A |
+| **Maximalist Chaos** | Density, layering, visual noise as deliberate statement. | Overlapping elements, competing type sizes, mixed media, bleed-to-edge images, stacked layers | Bebas Neue + handwritten mix, Display + body contrast | Saturated contrasting hues — no single dominant |
+| **Retro-Futuristic** | Nostalgia for a future that never arrived. CRT terminals, tape decks, space-age UI. | Scanline overlays, phosphor glow, CRT curve, terminal cursor blink, VHS distortion | Share Tech Mono, VT323, Orbitron | #0D1B0D (near-black) + #00FF41 (phosphor green) or amber #FFB300 |
+| **Organic/Natural** | Nature-sourced — irregular, warm, living. Rejects the mechanical grid. | Blob shapes, hand-drawn SVG borders, irregular column widths, grain textures, leaf/soil palette | Playfair Display, Lora, Newsreader | #3D2B1F, #8B7355, #E8DCC8, #4A7C59 |
+| **Luxury/Refined** | Restraint that signals expense. Negative space is the product. | Wide letter-spacing on display type, rule lines instead of borders, gold/cream/black palette, no rounded corners | Cormorant Garamond, Bodoni Moda, Cinzel | #1A1208, #C9A84C (gold), #F5F0E8 (cream) |
+| **Playful/Toy-like** | Tactile, bouncy, joyful. Designed to invite touch. | High border-radius, drop shadows with color, spring/bounce animations, illustrated elements, sticker-style badges | Nunito, Fredoka One, Baloo 2 | #FF6B6B, #4ECDC4, #FFE66D, #FFFFFF |
+| **Editorial/Magazine** | Type IS the design. Grid discipline with deliberate breaks. | Strong typographic hierarchy, 12-column grid, pull quotes, dropcaps, ruled sections, photo crops | Libre Baskerville + Source Sans, Playfair + Inter (editorial exception), Fraunces | #F4F1EC (paper), #1C1C1C, red accent #D62828 |
+| **Brutalist/Raw** | Deliberately anti-polished. The exposed skeleton is the aesthetic. | Default-looking elements styled with intent, visible borders, stark backgrounds, aggressive whitespace or none | Courier New (intentional), Times New Roman (reclaimed), monospace | #FFFFFF + #000000 + primary color blocks |
+| **Art Deco/Geometric** | Symmetry, repetition, gilt geometry. Machine-age glamour. | Chevron patterns, radial symmetry, thin geometric ornaments, fan motifs, high contrast | Poiret One, Josefin Sans, Italiana | #1A1209, #C9A84C (gold), #F5E6C8, #1A2639 |
+| **Soft/Pastel** | Gentle, airy, approachable. Saturation turned below 50%. | Muted gradients (analogous hues), cloud-like shapes, generous padding, soft shadows | Poppins (light weight only), Nunito Light, DM Sans | #F2E8FF, #E8F4F8, #FFF4E8, #E8F5E9 |
+| **Industrial/Utilitarian** | Function first. Decoration is a bug. Data is the interface. | Monospace typefaces, visible grid lines, data tables as primary UI, tight spacing, no gradients | JetBrains Mono, IBM Plex Mono, Roboto Mono | #1A1A1A, #2D2D2D, #4A4A4A, #00CC66 (status green) |
+| **Cyberpunk/Dark** | Dark base, neon accent, the city at 3am. Glitch is a feature. | Dark near-black base, neon accent (1 color only), glitch text effect, scanlines optional, terminal feel | Rajdhani, Exo 2, Orbitron | #0A0A0F, #FF2D78 (neon pink) or #00F5FF (cyan), #1A1A2E |
+
+## Usage Notes
+
+- Pick exactly one direction per project. Mixing directions without a deliberate concept produces visual noise.
+- The font column lists examples — always verify availability on Google Fonts before using.
+- Color values are starting points. Adjust lightness to pass WCAG AA contrast (4.5:1 for body text).
+- Maximalist directions (Chaos, Retro, Cyberpunk) allow more motion budget — see SKILL.md Step 3.
+- Brutalist/Raw requires the most discipline: it looks like no thought went into it, which requires the most thought.

package/templates/skills/skill-creator/references/skill-engineering-guide.md

@@ -62,16 +62,30 @@ description: >
 
 ### The formula:
 ```
-description = 
+description =
   WHAT it does (1 sentence)
   + ACTIONS that trigger it (write, review, fix, audit, check, scan...)
   + OBJECTS it applies to (keys, tokens, passwords, .env, connections...)
   + PATTERNS it detects (injection, XSS, CSRF, eval, exec...)
   + COMMANDS that invoke it (/audit, /ship, security...)
+  + INPUT CONSTRAINTS where it does NOT apply (e.g., "not for non-JS projects")
   + CONTEXTS where it should fire even without explicit request
   + "Even if the user doesn't explicitly mention X, use this skill when Y"
+  + "Do NOT trigger when: [anti-triggers — prevents false positives]"
 ```
 
+### Input constraints (stolen from production skill templates):
+Most skills trigger too broadly without explicit boundaries.
+Add a `Do NOT trigger when:` line to the frontmatter description:
+```yaml
+description: >
+  ...all the trigger keywords...
+  Do NOT trigger when: user is asking a conceptual question (not building),
+  when a design system already exists in the project (defer to it),
+  or when the request is a code review (use code-reviewer instead).
+```
+This prevents false positive triggering that wastes context and confuses the user.
+
 The last line is critical. Anthropic's own docs say:
 "Claude has a tendency to undertrigger skills. Make descriptions pushy."
 
@@ -116,8 +130,23 @@ Numbered steps. Imperative form. What Claude DOES, not what Claude SHOULD do.
 - Written as positive directives: "Always X" not "Don't do Y"
 - Specific, testable, unambiguous
 
+## Ambiguity Protocol
+*Every skill should define what happens when input is unclear.*
+
+If input is vague (no framework specified, no target stated):
+→ Ask: "[specific question, e.g., 'Which framework — React, Vue, or vanilla HTML?']"
+
+If input is malformed or out of scope:
+→ Say: "[specific message, e.g., 'This skill handles UI creation. For code review, use /audit instead.']"
+
+If a required prerequisite is missing (e.g., no CLAUDE.md, no design system):
+→ Do: "[specific fallback, e.g., 'Assume stack from package.json, proceed with default aesthetic']"
+
+**Rule:** Never silently fail or produce partial output. Either ask, redirect, or state the assumption explicitly.
+
 ## Examples
 One concrete input → output example that shows the expected behavior.
+Include at least one edge case / failure case: what happens when input is ambiguous, malformed, or out of scope.
 
 ## References
 For detailed [topic], read: `references/detailed-guide.md`
@@ -266,6 +295,8 @@ From Anthropic's skill-development skill + AZCLAUDE's debate engine research:
 ```
 □ Description has 30+ trigger keywords (pushy, not modest)
 □ Description ends with "even if the user doesn't explicitly ask"
+□ Description includes "Do NOT trigger when:" anti-trigger line
+□ Ambiguity Protocol defined: what to ask/do when input is vague, malformed, or missing prereqs
 □ SKILL.md body is under 2,000 words
 □ All detailed content is in references/, not SKILL.md
 □ Workflow uses imperative form ("Run X" not "You should run X")