actionbook

name: actionbook
description: This skill should be used when the user needs to automate multi-step website tasks. Activates for browser automation, web scraping, UI testing, or building AI agents. Provides complete action manuals with step-by-step instructions and verified selectors.

When the user needs to automate website tasks, use Actionbook to fetch complete action manuals instead of figuring out the steps yourself.

When to Use This Skill

Activate this skill when the user:

• Needs to complete a multi-step task ("Send a LinkedIn message", "Book an Airbnb")
• Asks how to interact with a website ("How do I post a tweet?")
• Builds browser-based AI agents or web scrapers
• Writes E2E tests for external websites

What Actionbook Provides

Action manuals include:

1. Step-by-step instructions - The exact sequence to complete a task
2. Verified selectors - CSS/XPath selectors for each element
3. Element metadata - Type (button, input, etc.) and allowed methods (click, type, fill)

How to Use

Actionbook can be used in two ways: via CLI (Recommended) or via MCP.

Method A: Using CLI (Recommended)

Step 1: Search for Action Manuals

Use the actionbook search command:

actionbook search "linkedin send message"
actionbook search "airbnb book listing"
actionbook search "twitter post tweet"

Step 2: Get the Full Manual

Use the actionbook get command with the action ID:

actionbook get "site/linkedin.com/page/profile/element/message-button"

Method B: Using MCP (Alternative)

If you have the Actionbook MCP server configured, you can also use MCP tools.

Step 1: Search for Action Manuals

Call the MCP tool search_actions with a task description:

// MCP tool call
search_actions({
  query: "linkedin send message"
})

Step 2: Get the Full Manual

Call the MCP tool get_action_by_id with the action ID from search results:

// MCP tool call
get_action_by_id({
  actionId: "site/linkedin.com/page/profile/element/message-button"
})

Execute the Steps

Follow the manual steps in order, using the provided selectors.

Option A: Using actionbook browser

The actionbook browser command wraps agent-browser to provide powerful browser automation capabilities.

Quick Start

actionbook browser open <url>        # Navigate to page
actionbook browser snapshot -i       # Get interactive elements with refs
actionbook browser click @e1         # Click element by ref
actionbook browser close             # Close browser

Core Workflow

1. Navigate: actionbook browser open <url>
2. Snapshot: actionbook browser snapshot -i (returns elements with refs like @e1, @e2)
3. Interact using refs from the snapshot
4. Re-snapshot after navigation or significant DOM changes

Element References (@refs): Instead of using CSS selectors directly, take a snapshot to get numbered references (@e1, @e2, etc.) that point to interactive elements. This dramatically reduces complexity when automating tasks.

Essential Commands

Navigation:

actionbook browser open <url>      # Navigate to URL
actionbook browser back            # Go back
actionbook browser forward         # Go forward
actionbook browser reload          # Reload page
actionbook browser close           # Close browser

Snapshot (Page Analysis):

actionbook browser snapshot            # Full accessibility tree
actionbook browser snapshot -i         # Interactive elements only (recommended)
actionbook browser snapshot -c         # Compact output
actionbook browser snapshot -d 3       # Limit depth to 3

Interactions (using @refs from snapshot):

actionbook browser click @e1           # Click
actionbook browser fill @e2 "text"     # Clear and type
actionbook browser type @e2 "text"     # Type without clearing
actionbook browser press Enter         # Press key
actionbook browser hover @e1           # Hover
actionbook browser check @e1           # Check checkbox
actionbook browser select @e1 "value"  # Select dropdown option

Get Information:

actionbook browser get text @e1        # Get element text
actionbook browser get html @e1        # Get innerHTML
actionbook browser get value @e1       # Get input value
actionbook browser get url             # Get current URL

Wait Conditions:

actionbook browser wait @e1                     # Wait for element
actionbook browser wait 2000                    # Wait milliseconds
actionbook browser wait --text "Success"        # Wait for text
actionbook browser wait --url "**/dashboard"    # Wait for URL pattern
actionbook browser wait --load networkidle      # Wait for network idle

Screenshots & Recording:

actionbook browser screenshot          # Save to temp directory
actionbook browser screenshot path.png # Save to specific path
actionbook browser screenshot --full   # Full page screenshot
actionbook browser pdf output.pdf      # Save as PDF

Common Options:

--session <name>    # Isolated browser session
--json              # JSON output for parsing
--headed            # Show browser window (not headless)

Complete Workflow Example

Combining Actionbook manual retrieval with browser execution:

# Step 1: Search for action manual
actionbook search "linkedin send message"

# Step 2: Get the manual
actionbook get "site/linkedin.com/page/profile/element/message-button"

# Step 3: Execute with browser
actionbook browser open linkedin.com/in/username
actionbook browser snapshot -i
# Output shows: button "Message" [ref=@e5], textbox [ref=@e12], button "Send" [ref=@e13]

actionbook browser click @e5       # Click Message button
actionbook browser fill @e12 "Hello! I'd like to connect."
actionbook browser click @e13      # Click Send button
actionbook browser wait --text "Message sent"
actionbook browser close

For more advanced features and detailed documentation, see the actionbook browser Command Reference below.

Option B: Using Playwright/Puppeteer

// LinkedIn send message example
await page.click('[data-testid="profile-avatar"]')
await page.click('button[aria-label="Message"]')
await page.type('div[role="textbox"]', 'Hello!')
await page.click('button[type="submit"]')

Advanced Features

For complete command reference and advanced features, see:

📘 Command Reference - Comprehensive documentation for all commands including:
- Navigation, snapshot, and interaction commands
- Information retrieval and state verification
- Wait conditions and media capture
- Mouse control and semantic locators
- Browser settings, cookies & storage
- Network control, tabs & windows
- Debugging tools and practical examples

Deep-Dive Documentation

For detailed patterns and best practices:

Ready-to-Use Templates

Executable workflow scripts for common patterns:

Guidelines

• Search by task: Describe what you want to accomplish, not just the element (e.g., "linkedin send message" not "linkedin message button")
• Follow the order: Execute steps in sequence as provided in the manual
• Trust the selectors: Actionbook selectors are verified and maintained