Stats
Actions
Tags
From llmstxt-generator
Guide for understanding and creating llms.txt files - format specification, best practices, and token optimization
How this skill is triggered — by the user, by Claude, or both
Slash command
/llmstxt-generator:llms-format-guideThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
The **llms.txt** file is a standardized Markdown-based format that makes website content optimized and accessible for Large Language Models at inference time. Proposed by Jeremy Howard (Answer.AI) in September 2024, it has become the standard for LLM-consumable documentation.
The llms.txt file is a standardized Markdown-based format that makes website content optimized and accessible for Large Language Models at inference time. Proposed by Jeremy Howard (Answer.AI) in September 2024, it has become the standard for LLM-consumable documentation.
Instead of LLMs struggling to parse HTML or reading massive websites that exceed context windows, llms.txt provides:
# [Project/Site Name]
> [Brief summary: 2-3 sentences with critical context]
[Optional: Additional details without headings - lists, paragraphs]
## [Section Name - e.g., "Documentation"]
- [Link Title](URL): Optional description
- [Another Link](URL): Brief, informative description
## [Another Section]
- [Link](URL): Description
## Optional
- [Link](URL): Can be skipped for shorter LLM context
- [Another Link](URL): Secondary resources
Only the H1 heading is required:
# My Project Name
Everything else is optional but recommended.
1. Blockquote Summary (highly recommended)
> FastHTML is a Python library for building server-rendered hypermedia applications with HTMX, Starlette, and Uvicorn.
2. Introduction Paragraphs (without headings)
# My Project
> [Summary]
This project includes comprehensive API documentation, tutorials, and examples. Compatible with Python 3.8+.
- **Use Case 1**: Description
- **Use Case 2**: Description
3. Section Organization (H2 headings)
## Getting Started
- [Installation](URL): How to install
- [Quick Start](URL): 5-minute intro
## API Reference
- [Core Functions](URL): Main API
4. Link Descriptions (concise, specific)
# Good descriptions
- [Authentication Guide](URL): OAuth2, API keys, session management
- [CLI Reference](URL): Complete command list with examples
# Avoid vague descriptions
- [Docs](URL): Documentation
- [API](URL): API stuff
5. Optional Section (for secondary resources)
## Optional
- [Advanced Features](URL): Beyond basics
- [Troubleshooting](URL): Fix common issues
- [Community Resources](URL): Third-party tutorials
The "Optional" section signals to LLMs and tools that these links can be skipped if context is limited.
# FastCore
> FastCore is a collection of utility functions and types that Fast.ai uses throughout its libraries.
Core utilities for Python development, with no dependencies beyond the Python standard library.
## Installation & Getting Started
- [Installation Guide](https://github.com/fastai/fastcore#install): Simple pip install
- [Quick Tour](https://fastcore.fast.ai/): Overview of main features
## API Documentation
- [Collections](https://fastcore.fast.ai/collections.html.md): Working with lists, dicts, sets
- [Foundation](https://fastcore.fast.ai/foundation.html.md): Core utilities and helpers
- [Functional](https://fastcore.fast.ai/functional.html.md): Functional programming utilities
## Examples
- [Foundation Examples](https://github.com/fastai/fastcore/tree/master/examples): Real-world usage
## Optional
- [API Reference (Full)](https://fastcore.fast.ai/): Comprehensive API docs
- [GitHub Repository](https://github.com/fastai/fastcore): Source code and issues
# Express.js
> Express is a minimal and flexible Node.js web application framework that provides a robust set of features for building web and mobile applications.
Fast, unopinionated web framework for Node.js with powerful routing and middleware.
## Getting Started
- [Installation & Setup](https://expressjs.com/en/starter/installing.html): Get started in minutes
- [Hello World Example](https://expressjs.com/en/starter/hello-world.html): Your first app
- [Basic Routing](https://expressjs.com/en/starter/basic-routing.html): Route basics
## Guide & Documentation
- [Routing Guide](https://expressjs.com/en/guide/routing.html): Advanced routing patterns
- [Middleware](https://expressjs.com/en/guide/using-middleware.html): Using and writing middleware
- [Error Handling](https://expressjs.com/en/guide/error-handling.html): Catching and handling errors
## API Reference
- [Request Object](https://expressjs.com/en/api.html#req): req API
- [Response Object](https://expressjs.com/en/api.html#res): res API
## Optional
- [Template Engines](https://expressjs.com/en/guide/using-template-engines.html): EJS, Pug, etc.
- [Database Integration](https://expressjs.com/en/guide/database-integration.html): Connect to databases
# Claude Code Documentation
> Claude Code is Anthropic's official command-line interface for Claude, the AI assistant. It includes features for file management, Git operations, AI-powered analysis, and plugin development.
Comprehensive documentation for building with Claude through the CLI.
## Quick Start
- [Installation](https://claude.dev/docs/installation): Get started in 5 minutes
- [First Steps](https://claude.dev/docs/first-steps): Your first Claude Code session
## Core Concepts
- [Commands Overview](https://claude.dev/docs/commands): All available commands
- [Tools Reference](https://claude.dev/docs/tools): Built-in tools and capabilities
- [Projects Structure](https://claude.dev/docs/projects): Organizing your work
## Advanced
- [Plugin Development](https://claude.dev/docs/plugins): Create custom plugins
- [MCP Integration](https://claude.dev/docs/mcp): Connect external services
- [Hooks & Automation](https://claude.dev/docs/hooks): Automate workflows
## Examples
- [Plugin Examples](https://github.com/anthropics/claude-code-plugins): Real plugin code
- [Sample Projects](https://github.com/anthropics/claude-code-examples): Tutorial projects
## Optional
- [Troubleshooting](https://claude.dev/docs/troubleshooting): Common issues
- [Community Plugins](https://marketplace.claude-code.dev): Community-built plugins
Good Summary:
FastHTML is a Python library for building server-rendered hypermedia applications with HTMX and Starlette. It's NOT compatible with FastAPI syntax, but enables rapid development of dynamic web apps.
Weak Summary:
FastHTML is a library for web development.
Why? The good summary clarifies what it IS, what it's NOT, and why it matters—exactly what an LLM needs.
Structure sections in the order users encounter them:
Good:
- [Authentication: OAuth2, API Keys, JWT](URL)
- [Error Handling: Try/catch, middleware, recovery](URL)
Poor:
- [Auth](URL)
- [Errors](URL)
Why? Descriptions help LLMs understand what's in each link without fetching it.
llms.txt can link outside your domain:
- [HTMX Documentation](https://htmx.org/): Understand HTMX attributes and capabilities
- [HTTP Status Codes](https://httpwg.org/specs/rfc9110.html): Official HTTP spec reference
If information is in your summary, don't repeat it in linked descriptions:
# Good
> FastHTML uses HTMX and Starlette for server-rendered apps.
## Frameworks
- [HTMX](URL): JavaScript library for dynamic interactions
- [Starlette](URL): Web framework providing routing and middleware
# Avoid
> FastHTML uses HTMX and Starlette.
- [HTMX: JavaScript library from Fast.ai](URL): Understand HTMX
- [Starlette: Web framework used by FastHTML](URL): Server framework
Generate your llms-ctx-full.txt and test:
llms_txt2ctx https://your-site.com/llms.txt -o context.xml
Then ask Claude, ChatGPT questions using that context to verify accuracy.
Descriptions should be concise:
# Good (clear, specific, ~60 chars)
- [Installation](URL): Simple pip install with no dependencies
# Too verbose (150+ chars)
- [Installation Guide](URL): Complete instructions on how to install the package using Python's package manager with detailed troubleshooting steps for common issues
Token efficiency is critical since context windows are limited. Here's how to optimize:
Primary docs should fit in most context windows. Put supplementary content in "Optional":
## Core APIs
- [Request](URL): Request object reference
- [Response](URL): Response object reference
## Optional
- [Legacy Endpoints](URL): Deprecated endpoints (usually not needed)
- [Internal Implementation](URL): Architecture details
Impact: ~40% smaller default context without losing essential info.
Instead of 10 individual endpoint docs:
# Less efficient (many small links)
- [GET /users](URL): Get users
- [POST /users](URL): Create user
- [PUT /users/:id](URL): Update user
- [DELETE /users/:id](URL): Delete user
# More efficient (single comprehensive link)
- [User API Reference](URL): All user endpoints (GET, POST, PUT, DELETE)
Each link should be unique. Remove links to the same content:
# Redundant
- [API Documentation](URL)
- [Complete API Reference](URL)
- [API Docs](URL)
# Better
- [API Reference](URL): All endpoints, authentication, error handling
When possible, link to consolidated docs rather than multiple small pages:
# Efficient (one page summary)
- [Quickstart Guide](URL): Installation, first request, common patterns (5 min read)
# Less efficient (requires multiple fetches)
- [Installation](URL)
- [Your First Request](URL)
- [Configuration](URL)
llms.txt can reference .md or .html.md versions of pages:
# Good (clean markdown)
- [API Reference](https://docs.example.com/api-ref.html.md): Markdown version
# Less efficient (HTML parsing required)
- [API Reference](https://docs.example.com/api-ref.html): HTML version
Project with 50 pages of documentation:
Without llms.txt:
With llms.txt (optimized):
Before publishing your llms.txt, verify:
- [Docs](URL)
- [API](URL)
- [More Info](URL)
Why Bad: LLMs can't tell what's in these links. Use specific titles.
- [Installation Guide](URL): This comprehensive guide covers all the detailed steps required for installing the package on various operating systems including Windows, macOS, and Linux, along with troubleshooting for specific versions and configurations.
Why Bad: Wastes tokens. Keep under 100 characters.
## Documentation
- [Docs](URL)
- [Documentation](URL)
- [Full Docs](URL)
Why Bad: Same content linked 3 ways. Choose one authoritative link.
- [Amazing Quickstart](URL): The fastest way to get up and running!
- [Awesome API Reference](URL): Everything you need to know!
Why Bad: Emotional language doesn't help LLMs. Be factual.
## Advanced Features
- [Internal Implementation](URL): How we built this
- [Legacy APIs](URL): Old endpoints
- [Experiments](URL): Research docs
Why Bad: These aren't essential. Put in "Optional" to let LLMs skip if needed.
The llmstxt-generator plugin uses this format to:
When you run /llmstxt:generate, the Python script creates llms.txt following these best practices automatically using Ollama LLM models.
npx claudepluginhub brunogama/llmstxt-generatorGenerates llms.txt and llms-full.txt from git repo wikis for LLM-readable project docs per llms.txt spec. Resolves repo URL/branch first.
Generates a standards-compliant llms.txt file for any website by crawling the live site or reading the codebase directly. Helps make websites discoverable by AI agents and LLMs.
Validates and generates llms.txt files — the emerging standard for AI guidance on websites. Can check an existing llms.txt or crawl a site to create one from scratch.