A lightweight and token-efficient browser automation tool that allows AI agents to control web pages using stable accessibility references.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install pinchtab-helper
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install pinchtab-helper using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is PinchTab Browser Control?
PinchTab Browser Control is a specialized tool designed to optimize how AI agents interact with web environments. By leveraging accessibility trees instead of heavy image-based processing, it reduces token consumption to just 800 tokens per page, making it 5 to 13 times more cost-effective than screenshot-based methods. This integration for Openclaw Skills provides a robust framework for agents to navigate, click, and input data with high precision and minimal latency.
Built for developers who need reliable web interaction, it supports multiple browser instances and uses stable element references (refs) to ensure that automation scripts don't break when page layouts shift slightly. It is an essential component for any developer building complex web-reliant workflows within the Openclaw Skills ecosystem.
PinchTab Browser Control Use Cases
- Automated web research and data gathering for AI analysis.
- Executing multi-step form submissions and transactional workflows.
- Headless browser testing and UI verification via AI agents.
- Efficient navigation of complex web applications with limited token windows.
How PinchTab Browser Control Works
- The PinchTab service is initialized locally or via a Docker container on port 9867.
- The AI agent sends a navigation command to open a specific URL.
- The skill generates a snapshot of the page, mapping interactive elements to short, stable reference IDs (e.g., e5).
- The agent performs actions like click, fill, or press by targeting these specific reference IDs.
- State changes are verified through subsequent snapshots or visual screenshots to ensure task completion.
PinchTab Browser Control Setup
To start using this skill within your Openclaw Skills environment, first install the PinchTab CLI:
# Install via Shell (macOS/Linux)
curl -fsSL https://pinchtab.com/install.sh | bash
# Or install via npm
npm install -g pinchtab
Launch the background service:
pinchtab &
Ensure your agent has access to the local endpoint at http://localhost:9867.
PinchTab Browser Control Data Schema & Taxonomy
PinchTab organizes its interaction data using a reference-based system to minimize context overhead:
| Component | Type | Description |
|---|---|---|
| Snapshot | Text | An accessibility-tree representation of the DOM with interactive element IDs. |
| Element Ref | String | A unique identifier (e.g., e1, e2) used to target elements for actions. |
| Instance ID | UUID | Identifies a specific browser session in multi-instance setups. |
| Action Payload | JSON | Contains the command type (click, fill, etc.) and target reference. |
PinchTab Browser Control Advanced Features
- Multi-instance support allowing the agent to manage several browser sessions concurrently.
- Token-optimized snapshots providing high-fidelity page context at a fraction of the cost of vision models.
- Bridge Token support for secure remote access to the browser control service.
- Custom automation profiles to maintain separate cookies, history, and session states for different tasks.
SKILL.md
Loading
Related Openclaw Skills
Docupilot
A document automation integration for generating and managing documents from templates within AI agent workflows.
Cardinal
A powerful project management skill that allows AI agents to organize tasks, track progress, and manage resources within Cardinal.
Memory Defrag
A maintenance skill for AI agents designed to reorganize bloated memory files, remove redundancies, and optimize knowledge retrieval.
WeChat MP Publisher
A comprehensive toolset for AI agents to programmatically manage and publish articles to WeChat Official Accounts.
3-Layer Token Compressor
A local preprocessing utility that leverages Ollama models to compress prompts and summarize history before they reach expensive paid LLM APIs.
Skill Shield
An advanced security management system for OpenClaw extensions that provides automated risk scanning and granular policy control.
Featured*
