1# Impeccable
2
3The vocabulary you didn't know you needed. 1 skill, 18 commands, and curated anti-patterns for impeccable style. Works with Cursor, Claude Code, Gemini CLI, and Codex CLI.
4
5## Repository Purpose
6
7Maintain a **single source of truth** for design-focused skills and commands, then automatically transform them into provider-specific formats. Each provider has different capabilities (frontmatter, arguments, modular files), so we use a build system to generate appropriate outputs.
8
9## Architecture: Option A (Feature-Rich Source)
10
11We use a **feature-rich source format** that gets transformed for each provider:
12
13- **Source files** (`source/`): Full metadata with YAML frontmatter, args, descriptions
14- **Build system** (`scripts/`): Transforms source → provider-specific formats
15- **Distribution** (`dist/`): Committed output files for 4 providers
16
17### Why Option A?
18
19Cursor doesn't support frontmatter or arguments (lowest common denominator). Instead of limiting all providers, we:
201. Author with full metadata in source files
212. Generate full-featured versions for providers that support it (Claude Code, Gemini, Codex)
223. Generate downgraded versions for Cursor (strip frontmatter, rely on appending)
23
24## Repository Structure
25
26```
27impeccable/
28├── source/ # EDIT THESE! Single source of truth
29│ ├── commands/ # Command definitions with frontmatter
30│ │ └── normalize.md
31│ └── skills/ # Skill definitions with frontmatter
32│ └── impeccable/
33├── dist/ # Generated outputs (committed for users)
34│ ├── cursor/ # Commands + Agent Skills
35│ │ └── .cursor/
36│ │ ├── commands/*.md
37│ │ └── skills/*/SKILL.md
38│ ├── claude-code/ # Full featured
39│ │ └── .claude/
40│ │ ├── commands/*.md
41│ │ └── skills/*/SKILL.md
42│ ├── gemini/ # TOML commands + modular skills
43│ │ ├── .gemini/
44│ │ │ └── commands/*.toml
45│ │ ├── GEMINI.md
46│ │ └── GEMINI.*.md
47│ └── codex/ # Custom prompts + Agent Skills
48│ └── .codex/
49│ ├── prompts/*.md
50│ └── skills/*/SKILL.md
51├── api/ # Vercel Functions (production)
52│ ├── skills.js # GET /api/skills
53│ ├── commands.js # GET /api/commands
54│ └── download/
55│ ├── [type]/[provider]/[id].js # Individual downloads
56│ └── bundle/[provider].js # Bundle downloads
57├── public/ # Website for impeccable.style
58│ ├── index.html # Main page
59│ ├── css/ # Modular CSS (9 files)
60│ │ ├── main.css # Entry point with imports
61│ │ ├── tokens.css # Design system
62│ │ └── ... # Component styles
63│ └── app.js # Vanilla JS
64├── server/ # Bun server (local dev only)
65│ ├── index.js # Serves website + API routes
66│ └── lib/
67│ └── api-handlers.js # Shared API logic (used by both server & functions)
68├── scripts/ # Build system (Bun)
69│ ├── build.js # Main orchestrator
70│ ├── lib/
71│ │ ├── utils.js # Shared utilities
72│ │ ├── zip.js # ZIP generation
73│ │ └── transformers/ # Provider-specific transformers
74│ │ ├── cursor.js
75│ │ ├── claude-code.js
76│ │ ├── gemini.js
77│ │ └── codex.js
78├── README.md # End user documentation
79├── DEVELOP.md # Contributor documentation
80└── package.json # Bun scripts
81```
82
83## Website (impeccable.style)
84
85**Tech Stack:**
86- Vanilla JavaScript (no frameworks)
87- Modern CSS with Bun's bundler (nesting, OKLCH colors, @import)
88- **Local Development**: Bun server with native routes (`server/index.js`)
89- **Production**: Vercel Functions with Bun runtime (`/api` directory)
90- Deployed on Vercel with Bun runtime
91
92**Dual Setup:**
93- `/api` directory contains individual Vercel Functions for production
94- `/server` directory contains monolithic Bun server for local development
95- `/server/lib/api-handlers.js` contains shared logic used by both
96- Zero duplication: API functions and dev server import the same handlers
97
98**Design:**
99- Editorial precision aesthetic
100- Cormorant Garamond (display) + Instrument Sans (body)
101- OKLCH color space for vibrant, perceptually uniform colors
102- Editorial sidebar layout (title left, content right)
103- Modular CSS architecture (9 files)
104
105**API Endpoints** (Vercel Functions):
106- `/` - Homepage (static HTML)
107- `/api/skills` - JSON list of all skills
108- `/api/commands` - JSON list of all commands
109- `/api/download/[type]/[provider]/[id]` - Individual file download
110- `/api/download/bundle/[provider]` - ZIP bundle download
111
112## Source File Format
113
114### Commands (`source/commands/*.md`)
115
116```yaml
117---
118name: command-name
119description: Clear description of what this command does
120args:
121 - name: argname
122 description: Argument description
123 required: false
124---
125
126Command prompt here. Use {{argname}} placeholders for arguments.
127```
128
129### Skills (`source/skills/*.md`)
130
131```yaml
132---
133name: skill-name
134description: Clear description of what this skill provides
135license: License info (optional)
136---
137
138Skill instructions for the LLM here.
139```
140
141## Build System
142
143Uses **Bun** for fast builds. Modular architecture:
144
145- **`utils.js`**: Shared functions (parseFrontmatter, readSourceFiles, writeFile, etc.)
146- **Transformer pattern**: Each provider has one focused file
147- **Registry**: `transformers/index.js` exports all transformers
148- **Main script**: `build.js` orchestrates everything (~50 lines)
149
150Run: `bun run build`
151
152## Provider Transformations
153
154### 1. Cursor (Agent Skills Standard)
155- **Commands**: Body only → `dist/cursor/.cursor/commands/*.md` (no frontmatter support)
156- **Skills**: Agent Skills standard → `dist/cursor/.cursor/skills/{name}/SKILL.md`
157 - Full YAML frontmatter with name/description
158 - Reference files in skill subdirectories
159- **Installation**: Extract ZIP into your project root, creates `.cursor/` folder
160- **Note**: Agent Skills require Cursor nightly channel
161
162### 2. Claude Code (Full Featured)
163- **Commands**: Full YAML frontmatter → `dist/claude-code/.claude/commands/*.md`
164- **Skills**: Full YAML frontmatter → `dist/claude-code/.claude/skills/{name}/SKILL.md`
165- **Preserves**: All metadata, all args
166- **Format**: Matches [Anthropic Skills spec](https://github.com/anthropics/skills)
167- **Installation**: Extract ZIP into your project root, creates `.claude/` folder
168
169### 3. Gemini CLI (Full Featured)
170- **Commands**: TOML format → `dist/gemini/.gemini/commands/*.toml`
171 - Uses `description` and `prompt` keys
172 - Transforms `{{argname}}` → `{{args}}` (Gemini uses single args string)
173- **Skills**: Modular with imports → `dist/gemini/GEMINI.{name}.md` (root level)
174 - Main `GEMINI.md` uses `@./GEMINI.{name}.md` import syntax
175 - Gemini automatically loads imported files
176- **Installation**: Extract ZIP into your project root, creates `.gemini/` folder + skill files
177
178### 4. Codex CLI (Full Featured)
179- **Commands**: Custom prompt format → `dist/codex/.codex/prompts/*.md`
180 - Uses `description` and `argument-hint` in frontmatter
181 - Transforms `{{argname}}` → `$ARGNAME` (uppercase variables)
182 - Invoked as `/prompts:<name>`
183- **Skills**: Agent Skills standard → `dist/codex/.codex/skills/{name}/SKILL.md`
184 - Same SKILL.md format as Claude Code with YAML frontmatter
185 - Reference files in skill subdirectories
186- **Installation**: Extract ZIP into your project root, creates `.codex/` folder
187
188## Key Design Decisions
189
190### Why commit dist/?
191End users can copy files directly without needing build tools.
192
193### Why separate transformers?
194- Each provider ~30-85 lines, easy to understand
195- Can modify one without affecting others
196- Easy to add new providers
197
198### Why Bun?
199- Much faster than Node.js (2-4x)
200- All-in-one toolkit (runtime + package manager)
201- Zero config, TypeScript native
202- Node.js compatible (works with existing code)
203
204### Why modular skills for Gemini/Codex?
205- Better context management (load only what's needed)
206- Cleaner file organization
207- Gemini: Uses native `@file.md` import feature
208- Codex: Uses routing pattern with AGENTS.md guide
209
210### Why vanilla JS for website?
211- No build complexity
212- Bun handles everything natively
213- Modern features (ES6+, CSS nesting, OKLCH colors)
214- Fast, lean, maintainable
215
216## Adding New Content
217
2181. **Create source file** in `source/commands/` or `source/skills/`
2192. **Add frontmatter** with name, description, args (for commands) or license (for skills)
2203. **Write body** with instructions/prompt
2214. **Build**: `bun run build`
2225. **Test** with your provider
2236. **Commit** both source and dist files
224
225## Important Notes
226
227- **Source is truth**: Always edit `source/`, never edit `dist/` directly
228- **Test across providers**: Changes affect 4 different outputs
229- **Argument handling**: Write prompts that work with both placeholders and appending
230- **Cursor limitations**: No frontmatter/args, so design for graceful degradation
231
232## Documentation
233
234- **README.md**: End user guide (installation, usage, quick dev setup)
235- **DEVELOP.md**: Contributor guide (architecture, build system, adding content)
236- **This file (AGENTS.md)**: Context for AI assistants and new developers
237
238## Provider Documentation Links
239
240- [Agent Skills Specification](https://agentskills.io/specification) - Open standard
241- [Cursor Commands](https://cursor.com/docs/agent/chat/commands)
242- [Cursor Rules](https://cursor.com/docs/context/rules)
243- [Cursor Skills](https://cursor.com/docs/context/skills)
244- [Claude Code Slash Commands](https://code.claude.com/docs/en/slash-commands)
245- [Anthropic Skills](https://github.com/anthropics/skills)
246- [Gemini CLI Custom Commands](https://cloud.google.com/blog/topics/developers-practitioners/gemini-cli-custom-slash-commands)
247- [Gemini CLI GEMINI.md](https://github.com/google-gemini/gemini-cli/blob/main/docs/cli/gemini-md.md)
248- [Codex CLI Slash Commands](https://developers.openai.com/codex/guides/slash-commands)
249- [Codex CLI Skills](https://developers.openai.com/codex/skills/)
250