/create-llms-for-skills

description: Generate comprehensive llms.txt from URLs using agent-browser
argument-hint: <urls> [requirements]

Create llms.txt from URLs

Use agent-browser CLI to access target URLs, extract content, and generate comprehensive llms.txt files.

Arguments: $ARGUMENTS

• First argument(s): urls (required) - one or more URLs, space-separated
• Last argument: requirements (optional) - additional requirements or instructions (if the last argument is not a URL)

---

Tool Priority

1. **agent-browser CLI** (preferred) - Full browser automation 2. **WebFetch** (fallback) - If agent-browser unavailable

**DO NOT use:**

• Claude in Chrome MCP
• Direct Fetch without user confirmation

---

Instructions

1. Parse Arguments

From `$ARGUMENTS`, parse:

• Identify all URLs (starting with http:// or https://)
• Remaining content serves as additional requirements

2. Use agent-browser CLI

agent-browser is a **command-line tool** with specific subcommands:

# Step 1: Open the page
agent-browser open "https://docs.rs/{crate}/latest/{crate}/"

# Step 2: Extract content using CSS selectors
agent-browser get text ".docblock"              # Main documentation
agent-browser get text ".module-item"           # Module list
agent-browser get text ".item-decl"             # Type declarations
agent-browser get text "pre.rust"               # Code examples

# Step 3: Close browser
agent-browser close

**Common selectors for docs.rs:**

| Selector | Content | |----------|---------| | `.docblock` | Main documentation text | | `.module-item` | Module/item list | | `.item-decl` | Function/struct declarations | | `pre.rust` | Code examples | | `.feature-flag` | Feature flags | | `#reexports` | Re-exports section |

**For multiple pages**, repeat open/get/close for each submodule.

3. Content Extraction Strategy

For Rust crate documentation (docs.rs):

1. Main crate page → Overview, re-exports, modules list
2. Each major module → Public items, examples
3. Important types → Methods, trait implementations
4. Examples section → Complete runnable code

**Extraction focus**:

• Core concepts and principles
• API function signatures and parameter descriptions
• Code examples (complete and runnable)
• Configuration options and best practices
• Common patterns and use cases
• Feature flags and cargo features

4. Generate llms.txt

Consolidate all content and generate in the following format:

# {Crate Name}

> {One-line description from crate docs}

**Version:** {version} | **docs.rs:** {url}

---

## Overview

{Detailed explanation of core concepts from crate-level docs}

## Modules

### {module_name}

{Module description}

#### Key Types

| Type | Description |
|------|-------------|
| `TypeName` | Brief description |

#### Key Functions

```rust
// Function signature with doc comment
pub fn function_name(param: Type) -> ReturnType

Code Examples

// Complete code example from docs
use crate_name::...;

fn main() {
    // Example code
}

---

Feature Flags

| Feature | Description | Default | |---------|-------------|---------| | `feature_name` | What it enables | yes/no |

---

Common Patterns

Pattern 1: {Name}

// Pattern code

Pattern 2: {Name}

// Pattern code

### 5. Save Output

```bash
# Generate timestamp
timestamp=$(date +%Y%m%d%H%M)

# Determine crate name from URL
# e.g., https://docs.rs/tokio/latest/tokio/ → tokio

# Save location
~/tmp/${timestamp}-{crate_name}-llms.txt

Inform the user of the file path after output is complete.

---

Fallback: WebFetch

If agent-browser is not available:

1. Use WebFetch to get main page content
2. Parse the response for key sections
3. May need multiple WebFetch calls for subpages
4. Inform user that content may be incomplete

---

Quality Requirements

• [ ] Comprehensive content: Include actual API descriptions and code examples
• [ ] Clear sources: Mark source URL for each section
• [ ] Complete structure: Maintain the hierarchy of the original documentation
• [ ] Usable code: Example code should be complete and runnable
• [ ] Consistent format: Use consistent Markdown formatting
• [ ] Feature flags: Document all cargo features

---

Workflow Integration

This command is the first step in the Skills creation workflow:

1. **create-llms-for-skills** (this command) → Generate llms.txt 2. **create-skills-via-llms** → Create skills based on llms.txt

---

Example Usage

# Generate llms.txt for tokio
/create-llms-for-skills https://docs.rs/tokio/latest/tokio/

# Generate for multiple URLs
/create-llms-for-skills https://docs.rs/serde/latest/serde/ https://serde.rs/

# With additional requirements
/create-llms-for-skills https://docs.rs/axum/latest/axum/ "Focus on routing and extractors"