A foundational skill for AI agents to navigate the web with human-like precision, strict tab discipline, and token-efficient perception.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install surfagent-browser-operations
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 surfagent-browser-operations 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 Browser Operations?
Browser Operations is the universal brain for any browser-based AI agent, providing the essential rules for navigating the web without wasting tokens or creating technical debt. It enforces a philosophy of perception-first execution, where the agent must verify the state of a page before acting and confirm visible results immediately after. This skill serves as the mandatory foundation for Openclaw Skills, ensuring that agents drive browsers like careful human operators rather than broken macros.
By prioritizing browser hygiene and tab discipline, this skill prevents the common pitfall of tab sprawl and duplicate retries. It introduces a sophisticated tool selection ladder that favors low-token state summaries and platform adapters over expensive, raw DOM dumps. Whether you are building a research agent or a complex workflow automation, Browser Operations provides the logic needed for reliable, autonomous web interaction.
Browser Operations Use Cases
- Maintaining browser hygiene and tab discipline during high-concurrency web tasks.
- Implementing autonomy loops with human-like interaction patterns to avoid detection and errors.
- Reducing operational costs by using token-efficient perception rather than full page scraping.
- Establishing verifiable proof of success for complex actions like sending emails or posting content.
- Managing automated recovery when encountering common web blockers like modals or cookie banners.
How Browser Operations Works
- Inspect current tab state to identify the active work environment and confirm authentication.
- Utilize the perception layer to understand the page state using the most token-efficient method possible.
- Choose the appropriate tool from the selection ladder, prioritizing platform adapters over raw browser actions.
- Execute a single, deliberate action such as a click, type, or scroll based on visible evidence.
- Verify the result by diffing the page state to ensure the intended UI update actually occurred.
- Perform cleanup by closing throwaway tabs and leaving the browser in a clean, explainable state.
Browser Operations Setup
To deploy this skill within your agent framework, ensure your environment is configured to support Openclaw Skills and browser-based perception tools.
# Install the browser operations core skill
openclaw install browser-operations
# Configure your agent to prioritize perception-first logic
openclaw config set browser_logic perception-first
Browser Operations Data Schema & Taxonomy
The skill organizes interaction data into a structured taxonomy to ensure clarity and efficiency across sessions.
| Data Component | Format | Description |
|---|---|---|
| Tab Registry | Array | Tracks active URLs, titles, and auth status to prevent sprawl. |
| Perception Cache | JSON | Stores lightweight snapshots of page states and element locations. |
| Action Log | List | Records attempted vs. verified actions for success auditing. |
| Proof Artifacts | String/URI | Links to visual evidence or state changes confirming completion. |
Browser Operations Advanced Features
- Tool Selection Ladder: Automatically picks the most efficient tool (Perception > Platform Adapter > MCP > Raw Browser).
- Tab Sprawl Prevention: Monitors and cleans up redundant tabs created during retry cycles.
- Visual Confirmation Escalation: Automatically triggers visual snapshots when DOM extraction and visible state disagree.
- Human-Like Pacing: Implements state-aware interactions that avoid the pitfalls of rigid, macro-style automation.
- Universal Fallback Ladder: A 7-step recovery process to handle site-specific blockers or poisoning of the tab state.
SKILL.md
Loading
Related Openclaw Skills
Gmail
A high-reliability Gmail automation skill designed for AI agents to manage mailboxes, compose emails, and verify message delivery without manual selector hunting.
Real Estate Storyline & Marketing Agent
A high-end real estate branding tool that transforms project data into evocative storylines and comprehensive marketing assets.
Breakfast Included Hotel Finder
A specialized travel search skill that identifies hotels with included breakfast to save travelers time and money.
Quick Getaway: 3-Day Trip Planner
A professional travel orchestration skill that automates 3-day itinerary planning, flight searching, and hotel booking through real-time CLI data.
Super GitHub
An AI-native GitHub assistant utilizing vector search and LLM architecture to provide semantic search and proactive repo monitoring.
Agent Genesis
Agent Genesis enables AI agents to generate native working capital by solving Proof of Agent challenges and leveraging ZKTLS-verified billing proofs.
Featured*
