Mastering OpenClaw's Built-in Agent Tools

OpenClaw empowers AI agents with a robust suite of built-in tools designed to replace the older openclaw-* skill patterns. These native tools are type-safe, eliminate shelling, and provide a direct, efficient interface for agent operations. Understanding and leveraging these tools is fundamental to building powerful automated workflows.

Why Native Tools?

The shift to native tools represents a significant evolution in agent capability. Unlike the older skill-based approach, which often required wrapping CLI commands in YAML or JSON, native tools are first-class citizens within the OpenClaw ecosystem. This provides several key advantages:

• Type Safety & Reliability: Tools have explicit, well-defined schemas, reducing errors from malformed commands.
• Performance & Efficiency: Direct integration eliminates the overhead of parsing and executing external scripts.
• Structured Interaction: The tool framework handles the conversation loop (calling, returning results) seamlessly, allowing agents to focus on the task logic.

Core Tool Categories

OpenClaw's tools are organized into logical groups, which can be used as shorthands in permission policies.

File System (group:fs)

These tools manage file operations within the agent's workspace.

• read: Retrieve the contents of a file.
• write: Create or overwrite a file.
• edit: Make precise, text-based edits to a file by replacing an exact string.
• apply_patch: Apply multi-hunk code edits, ideal for complex refactors.

Runtime & Execution (group:runtime)

These tools execute code and manage processes.

• exec: Run shell commands. Can run synchronously or in the background.
• process: Manage background exec sessions (log, poll, kill).

Web Interaction (group:web)

For interacting with online content.

• web_search: Perform web searches using the Brave Search API (requires a key).
• web_fetch: Extract readable content from a URL, converting HTML to markdown or plain text.

UI Automation (group:ui)

For interacting with graphical user interfaces, either in a browser or on a desktop.

• browser: Control a dedicated, managed browser instance for automation tasks like form filling or data scraping.
• canvas: Drive the UI of a paired node application (e.g., a macOS app) for deep desktop automation.

Agent & Session Management (group:sessions)

For orchestrating multi-agent workflows.

• sessions_list / sessions_history: Discover and inspect other agent sessions.
• sessions_spawn: Spawn a dedicated sub-agent to handle a specific task.
• sessions_send: Send a message to another session.
• session_status: Check the status and model of the current session.

Automation & Control (group:automation)

For managing scheduled tasks and the gateway itself.

• cron: Create, list, and manage recurring cron jobs.
• gateway: Apply config changes or restart the OpenClaw gateway service.

Communication (group:messaging)

For interacting with various chat channels.

• message: Send messages, reactions, and perform channel actions on platforms like WhatsApp, Discord, and Telegram.

Advanced Tool Management

OpenClaw provides sophisticated mechanisms to configure and restrict tool access, ensuring security and appropriate resource use.

Tool Profiles and Policies

You can define base tool allowances with profiles (minimal, coding, messaging, full) and then use allow/deny lists to refine them. This allows you to set different tool access levels for different agents. For example, a support agent might have a messaging profile with access to the slack tool added via an allow list.

Per-Provider Restrictions

The tools.byProvider configuration lets you apply tool restrictions based on the LLM provider or a specific model. This is useful if a particular model struggles with certain tools or if you want to limit expensive operations on higher-tier models.

Loop Detection

The loop-detection tool guards against infinite or inefficient tool-call loops. It can be configured to warn or break a loop after a certain number of repeated, non-progressing calls, helping to maintain system stability.

Conclusion

Mastering OpenClaw's native tools is the key to unlocking its full potential. By moving away from clunky skill wrappers to a structured, type-safe tool system, OpenClaw provides a powerful and reliable foundation for building sophisticated, automated assistants. Start by exploring the core tool categories, leverage profiles for security, and use sessions_spawn to create complex, multi-step workflows that get real work done.