pbakaus_impeccable/AGENTS.md

Impeccable

The vocabulary you didn't know you needed. 1 skill, 22 commands, and curated anti-patterns for impeccable style. Works with Cursor, Claude Code, Gemini CLI, and Codex CLI.

Repository Purpose

Maintain 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.

Architecture: Option A (Feature-Rich Source)

We use a feature-rich source format that gets transformed for each provider:

• Source files (source/): Full metadata with YAML frontmatter, args, descriptions
• Build system (scripts/): Transforms source → provider-specific formats
• Distribution (dist/): Committed output files for 4 providers

Why Option A?

Cursor doesn't support frontmatter or arguments (lowest common denominator). Instead of limiting all providers, we:

1. Author with full metadata in source files
2. Generate full-featured versions for providers that support it (Claude Code, Gemini, Codex)
3. Generate downgraded versions for Cursor (strip frontmatter, rely on appending)

Repository Structure

impeccable/
├── source/                      # EDIT THESE! Single source of truth
│   ├── commands/                # Command definitions with frontmatter
│   │   └── normalize.md
│   └── skills/                  # Skill definitions with frontmatter
│       └── impeccable/
├── dist/                        # Generated outputs (committed for users)
│   ├── cursor/                  # Commands + Agent Skills
│   │   └── .cursor/
│   │       ├── commands/*.md
│   │       └── skills/*/SKILL.md
│   ├── claude-code/             # Full featured
│   │   └── .claude/
│   │       ├── commands/*.md
│   │       └── skills/*/SKILL.md
│   ├── gemini/                  # TOML commands + modular skills
│   │   ├── .gemini/
│   │   │   └── commands/*.toml
│   │   ├── GEMINI.md
│   │   └── GEMINI.*.md
│   └── codex/                   # Custom prompts + Agent Skills
│       └── .codex/
│           ├── prompts/*.md
│           └── skills/*/SKILL.md
├── api/                         # Vercel Functions (production)
│   ├── skills.js                # GET /api/skills
│   ├── commands.js              # GET /api/commands
│   └── download/
│       ├── [type]/[provider]/[id].js   # Individual downloads
│       └── bundle/[provider].js        # Bundle downloads
├── public/                      # Website for impeccable.style
│   ├── index.html               # Main page
│   ├── css/                     # Modular CSS (9 files)
│   │   ├── main.css             # Entry point with imports
│   │   ├── tokens.css           # Design system
│   │   └── ...                  # Component styles
│   └── app.js                   # Vanilla JS
├── server/                      # Bun server (local dev only)
│   ├── index.js                 # Serves website + API routes
│   └── lib/
│       └── api-handlers.js      # Shared API logic (used by both server & functions)
├── scripts/                     # Build system (Bun)
│   ├── build.js                 # Main orchestrator
│   ├── lib/
│   │   ├── utils.js             # Shared utilities
│   │   ├── zip.js               # ZIP generation
│   │   └── transformers/        # Provider-specific transformers
│   │       ├── cursor.js
│   │       ├── claude-code.js
│   │       ├── gemini.js
│   │       └── codex.js
├── README.md                    # End user documentation
├── DEVELOP.md                   # Contributor documentation
└── package.json                 # Bun scripts

Website (impeccable.style)

Tech Stack:

• Vanilla JavaScript (no frameworks)
• Modern CSS with Bun's bundler (nesting, OKLCH colors, @import)
• Local Development: Bun server with native routes (server/index.js)
• Production: Vercel Functions with Bun runtime (/api directory)
• Deployed on Vercel with Bun runtime

Dual Setup:

• /api directory contains individual Vercel Functions for production
• /server directory contains monolithic Bun server for local development
• /server/lib/api-handlers.js contains shared logic used by both
• Zero duplication: API functions and dev server import the same handlers

Design:

• Editorial precision aesthetic
• Cormorant Garamond (display) + Instrument Sans (body)
• OKLCH color space for vibrant, perceptually uniform colors
• Editorial sidebar layout (title left, content right)
• Modular CSS architecture (9 files)

API Endpoints (Vercel Functions):

• / - Homepage (static HTML)
• /api/skills - JSON list of all skills
• /api/commands - JSON list of all commands
• /api/download/[type]/[provider]/[id] - Individual file download
• /api/download/bundle/[provider] - ZIP bundle download

Source File Format

Commands (source/commands/*.md)

---
name: command-name
description: Clear description of what this command does
args:
  - name: argname
    description: Argument description
    required: false
---

Command prompt here. Use {{argname}} placeholders for arguments.

Skills (source/skills/*.md)

---
name: skill-name
description: Clear description of what this skill provides
license: License info (optional)
---

Skill instructions for the LLM here.

Build System

Uses Bun for fast builds. Modular architecture:

• utils.js: Shared functions (parseFrontmatter, readSourceFiles, writeFile, etc.)
• Transformer pattern: Each provider has one focused file
• Registry: transformers/index.js exports all transformers
• Main script: build.js orchestrates everything (~50 lines)

Run: bun run build

Provider Transformations

1. Cursor (Agent Skills Standard)

• Commands: Body only → dist/cursor/.cursor/commands/*.md (no frontmatter support)
• Skills: Agent Skills standard → dist/cursor/.cursor/skills/{name}/SKILL.md
  • Full YAML frontmatter with name/description
  • Reference files in skill subdirectories
• Installation: Extract ZIP into your project root, creates .cursor/ folder
• Note: Agent Skills require Cursor nightly channel

2. Claude Code (Full Featured)

• Commands: Full YAML frontmatter → dist/claude-code/.claude/commands/*.md
• Skills: Full YAML frontmatter → dist/claude-code/.claude/skills/{name}/SKILL.md
• Preserves: All metadata, all args
• Format: Matches Anthropic Skills spec
• Installation: Extract ZIP into your project root, creates .claude/ folder

3. Gemini CLI (Full Featured)

• Commands: TOML format → dist/gemini/.gemini/commands/*.toml
  • Uses description and prompt keys
  • Transforms {{argname}} → {{args}} (Gemini uses single args string)
• Skills: Modular with imports → dist/gemini/GEMINI.{name}.md (root level)
  • Main GEMINI.md uses @./GEMINI.{name}.md import syntax
  • Gemini automatically loads imported files
• Installation: Extract ZIP into your project root, creates .gemini/ folder + skill files

4. Codex CLI (Full Featured)

• Commands: Custom prompt format → dist/codex/.codex/prompts/*.md
  • Uses description and argument-hint in frontmatter
  • Transforms {{argname}} → $ARGNAME (uppercase variables)
  • Invoked as /prompts:<name>
• Skills: Agent Skills standard → dist/codex/.codex/skills/{name}/SKILL.md
  • Same SKILL.md format as Claude Code with YAML frontmatter
  • Reference files in skill subdirectories
• Installation: Extract ZIP into your project root, creates .codex/ folder

Key Design Decisions

Why commit dist/?

End users can copy files directly without needing build tools.

Why separate transformers?

• Each provider ~30-85 lines, easy to understand
• Can modify one without affecting others
• Easy to add new providers

Why Bun?

• Much faster than Node.js (2-4x)
• All-in-one toolkit (runtime + package manager)
• Zero config, TypeScript native
• Node.js compatible (works with existing code)

Why modular skills for Gemini/Codex?

• Better context management (load only what's needed)
• Cleaner file organization
• Gemini: Uses native @file.md import feature
• Codex: Uses routing pattern with AGENTS.md guide

Why vanilla JS for website?

• No build complexity
• Bun handles everything natively
• Modern features (ES6+, CSS nesting, OKLCH colors)
• Fast, lean, maintainable

Adding New Content

1. Create source file in source/commands/ or source/skills/
2. Add frontmatter with name, description, args (for commands) or license (for skills)
3. Write body with instructions/prompt
4. Build: bun run build
5. Test with your provider
6. Commit both source and dist files

Important Notes

• Source is truth: Always edit source/, never edit dist/ directly
• Test across providers: Changes affect 4 different outputs
• Argument handling: Write prompts that work with both placeholders and appending
• Cursor limitations: No frontmatter/args, so design for graceful degradation

Documentation

• README.md: End user guide (installation, usage, quick dev setup)
• DEVELOP.md: Contributor guide (architecture, build system, adding content)
• This file (AGENTS.md): Context for AI assistants and new developers

Provider Documentation Links

• Agent Skills Specification - Open standard
• Cursor Commands
• Cursor Rules
• Cursor Skills
• Claude Code Slash Commands
• Anthropic Skills
• Gemini CLI Custom Commands
• Gemini CLI GEMINI.md
• Codex CLI Slash Commands
• Codex CLI Skills