LLM Documentation Generator

Generate structured, AI-readable documentation following the llms.txt standard with granular files organized by domain.

Output Structure

llm-docs/
├── llm.txt                    # Main index (~1-2 KB)
├── llm.version.txt            # Metadata and sync info (~0.3 KB)
└── llm.{domain}.txt           # Domain-specific files (~3-50 KB each)

Workflow

Phase 1: Language Selection

Ask user preferred language before starting:

¿En qué idioma prefieres la documentación? / What language do you prefer?
- Español
- English
- Bilingual (technical terms in English, explanations in Spanish)

Phase 2: Project Analysis

Identify project type and data sources:

Detect structured data sources:

• JSON metadata files (component docs, OpenAPI specs)
• JSDoc/GoDoc comments
• TypeScript definitions
• Configuration files (package.json, go.mod)
• Existing documentation (README, docs/)

Phase 3: Domain Planning

Based on project type, plan which llm.{domain}.txt files to create:

Frontend/UI: See references/frontend-example.md

• tokens, utilities, styles, brands
• components-atoms, components-molecules, components-organisms

CLI Tools: See references/cli-example.md

• commands, core, gateway, deployment, resources, testing, usage

APIs: See references/api-example.md

• endpoints, models, auth, errors, examples

Libraries: See references/library-example.md

• api, internals, patterns, examples

Phase 4: Implementation Decision

Choose approach based on data availability:

Generator script benefits:

• Auto-updates when code changes
• DRY principle: single source of truth
• Consistent formatting
• Add npm/make script: generate:llms

Phase 5: File Generation

llm.version.txt (always first)

# {Project} LLM Documentation

- **Version**: {semantic version}
- **Last Updated**: {YYYY-MM-DD}
- **Documentation Version**: 1.0.0
- **Files**: {count} domain files
- **Total Size**: ~{X} KB

llm.txt (main index)

# {Project} - LLM Documentation

## Project Metadata
- **Name**: {project name}
- **Type**: {frontend|cli|api|library}
- **Language**: {primary language}
- **Purpose**: {one-line description}

## Quick Reference
- **Key Modules**: {list main areas}
- **Patterns**: {architectural patterns used}
- **Dependencies**: {key dependencies}

## Documentation Structure

### {Domain 1}
#### llm.{domain1}.txt
- **Focus**: {what this file covers}
- **Use when**: {scenarios to read this file}

### {Domain 2}
...

## Reading Guide

1. Start with `llm.version.txt` for metadata
2. Read `llm.{primary-domain}.txt` for core concepts
3. Reference other files as needed

llm.{domain}.txt (domain files)

Each domain file follows this structure:

# {Domain} - {Project}

## Overview
{2-3 sentences explaining this domain}

## {Section 1}

| Name | Type | Description |
|------|------|-------------|
| ... | ... | ... |

## {Section 2}

### {Subsection}
{Content with code examples}

## Related Files
- `llm.{related}.txt` - {why related}

Best Practices

1. File size: Keep each file under 50 KB for optimal LLM context usage
2. Cross-references: Link between files with clear "when to read" guidance
3. Tables: Use markdown tables for properties, tokens, parameters
4. Code examples: Include practical, copy-pasteable examples
5. Hierarchy: Use consistent heading levels (H1 for title, H2 for sections, H3 for subsections)

Generator Script Pattern

When creating a generator script:

// Structure
const config = { COMPONENTS_DIR, OUTPUT_DIR, ... };

// Utilities
function readFile(path) { ... }
function writeOutput(filename, content) { ... }

// Extractors (one per data source)
function extractComponents() { ... }
function extractTokens() { ... }

// Generators (one per output file)
function generateIndex() { ... }
function generateVersion() { ... }
function generateDomain() { ... }

// Main
function main() {
  // Extract all data
  // Generate all files
  // Log summary
}

// Export for testing
module.exports = { extractors, generators };

// Run if main
if (require.main === module) main();

Add to package.json:

{
  "scripts": {
    "generate:llms": "node build-scripts/create-llms-docs.js"
  }
}

Ejemplo Completo de Output

Proyecto: CLI de Deployment

Después de analizar un CLI tool, el skill genera:

llm-docs/llm.version.txt

# DeployCLI LLM Documentation

- **Version**: 2.1.0
- **Last Updated**: 2025-12-15
- **Documentation Version**: 1.0.0
- **Files**: 4 domain files
- **Total Size**: ~35 KB

llm-docs/llm.txt

# DeployCLI - LLM Documentation

## Project Metadata
- **Name**: deploy-cli
- **Type**: CLI Tool
- **Language**: TypeScript
- **Purpose**: Deploy applications to multiple cloud providers

## Quick Reference
- **Key Modules**: commands, providers, config
- **Patterns**: Command pattern, Provider abstraction
- **Dependencies**: commander, chalk, ora

## Documentation Structure

### Commands
#### llm.commands.txt
- **Focus**: All CLI commands and subcommands
- **Use when**: Need to understand available commands and flags

### Providers
#### llm.providers.txt
- **Focus**: Cloud provider integrations (AWS, GCP, Vercel)
- **Use when**: Adding or modifying provider support

### Configuration
#### llm.config.txt
- **Focus**: Config file format and options
- **Use when**: Understanding how users configure the CLI

llm-docs/llm.commands.txt

# Commands - DeployCLI

## Overview
DeployCLI exposes 5 main commands for deployment management.

## Commands

| Command | Description | Flags |
|---------|-------------|-------|
| `deploy` | Deploy to target provider | `--provider`, `--env`, `--dry-run` |
| `rollback` | Revert to previous deployment | `--version`, `--force` |
| `status` | Check deployment status | `--watch`, `--json` |
| `config` | Manage configuration | `--init`, `--validate` |
| `logs` | Stream deployment logs | `--follow`, `--since` |

## deploy

Main deployment command.

### Usage
\`\`\`bash
deploy-cli deploy --provider aws --env production
\`\`\`

### Flags
- `--provider, -p`: Target provider (aws, gcp, vercel)
- `--env, -e`: Environment (development, staging, production)
- `--dry-run`: Simulate without deploying
- `--config, -c`: Path to config file

## Related Files
- `llm.providers.txt` - Provider-specific deployment details
- `llm.config.txt` - Configuration options for deployments

Uso del Output

Una vez generados, los archivos pueden ser:

1. Incluidos en prompts de AI:

   @llm-docs/llm.commands.txt How do I deploy to staging?
   

2. Referenciados en CLAUDE.md:

   ## LLM Documentation
   Ver `llm-docs/` para documentación optimizada para AI.
   

3. Mantenidos automáticamente:

   npm run generate:llms  # Regenerar después de cambios