zerotoken-openclaw

What This Skill Does

The zerotoken-openclaw skill provides an interface to the ZeroToken MCP server, designed specifically to facilitate low-token, reusable browser automation within the OpenClaw environment. By leveraging trajectory recording and script playback, this skill enables agents to convert complex, multi-step browser interactions into lightweight, executable sequences. This is particularly effective for recurring or scheduled tasks, minimizing the context overhead that typical AI-driven browser navigation usually incurs.

Installation

To integrate this skill, follow these steps:

1. Ensure you have the necessary environment for OpenClaw/MCPorter installed.
2. Execute the install command: clawhub install openclaw/skills/skills/amos144/zerotoken.
3. Configure your openclaw.json to enable persistent connectivity. Because OpenClaw creates isolated processes, you must use the streamable-http mode to maintain browser state. Launch the ZeroToken HTTP service in the background (e.g., zerotoken-mcp-http) and point the mcpServers.zerotoken configuration in openclaw.json to the local URL (typically http://localhost:8000/mcp).
4. If dependencies are missing, ensure playwright is installed in your target environment using playwright install chromium.

Use Cases

• Scheduled Monitoring: Automate daily price checks or status reporting on websites that require multiple login and navigation steps.
• Data Extraction Pipelines: Use recorded trajectories to build stable data scraping tasks that don't need to re-think common navigation logic every time.
• Regression Testing: Run standardized browser journeys repeatedly to ensure UI elements are functioning as expected without manually guiding the agent through the steps each time.

Example Prompts

1. "Record my steps for logging into the dashboard, navigating to the reports tab, and downloading the CSV. Save this as a reusable task named 'daily-report'."
2. "Execute the 'daily-report' browser task I saved yesterday and summarize the data it retrieves."
3. "Open the browser, navigate to the news portal, and extract the headlines using the existing trajectory script for that site."

Tips & Limitations

• Performance: Always prefer recording a trajectory for repetitive tasks to save on token costs significantly compared to natural language steering.
• State Management: Remember that browser_init is a mandatory prerequisite. If your session disconnects, the skill will require a fresh init to restore browser context.
• Limitations: This skill is not suitable for pages that require high-level human intuition or complex fuzzy decision-making. If the UI changes drastically, the recorded trajectory may fail; in such cases, you will need to re-record the interaction sequence.