The best beginner tool set is the smallest one needed for your task. A summarization request may need no external tool. A research task needs source access; a file task needs access to the correct workspace; a messaging task needs a configured channel and deliberate permission to send.
Independent selection guide. References reviewed on October 2, 2026 against the official tools overview. This is a capability checklist, not a ranking of tested marketplace packages.
Start with a working assistant
openclaw gateway status
openclaw config validate
openclaw dashboard
Confirm a normal chat reply before adding a plugin, skill, or external account. Use the local quickstart if setup is unfinished.
The earlier YAML tool definitions on this page did not describe the current Gateway configuration. Do not copy generic file_write, http_request, email-password paths, or config.yaml examples into OpenClaw without checking the installed schema and integration docs.
Choose capabilities by the outcome
| Task | Capability to consider | Result you should verify |
|---|---|---|
| Summarize supplied text | Model response, no external tool required | Findings match the supplied text |
| Review a workspace document | File reading with the intended path access | The correct file was read |
| Save a draft | File writing limited to a disposable folder | Actual file contents and path |
| Research a question | Supported web/search integration | Each important source supports the answer |
| Navigate a website | Browser tools appropriate to the task | Target page and intended action |
| Run a local command | Command execution under an explicit policy | Command result and its effects |
| Work with a messaging account | Official channel or supported integration | Correct account, recipients, and access rules |
These are capability descriptions rather than a universal list of installed tool IDs. Availability depends on the version, agent, plugins, tool policy, and platform.
Begin with text and one workspace task
Try the first assistant exercise using text you provide. Next, if you need file output, use a disposable workspace folder and inspect the saved result. This sequence separates answer quality from filesystem permissions.
Check the active workspace:
openclaw config get agents.defaults.workspace
Use the official workspace documentation before changing instruction files or paths. A denied file operation may be a policy or mount issue, not a reason to grant access to the entire home directory.
Add web access with a source check
Review the official web tools reference for provider setup and the tools supported by your release. Confirm any required account or API credential privately.
Ask for source links and distinguish facts from inference. Open the sources behind the claims that matter. If a page cannot be read or a provider is unavailable, the assistant should state that instead of inventing a citation.
Browser automation is a separate capability from a simple web fetch. Start with reading and checking information before allowing account changes, submissions, or purchases.
Treat shell access as a separate permission
A shell command can affect files, processes, and network services. Use the official execution-tool reference and inspect the policy for the actual agent.
For sandboxed execution, check the effective rules:
openclaw sandbox explain
openclaw sandbox list
Read the sandbox guide for the difference between tool policy and filesystem isolation. A sandbox does not make every command harmless or put all plugins inside the same boundary.
Distinguish tools, skills, and channels
Tools provide actions. Skills supply instructions and supporting resources that help an agent use capabilities. Channels connect conversations to messaging services. Installing a skill is not proof that a tool is enabled, that an account is authorized, or that a channel can send messages.
Use the official skills reference and our skill selection guide when a repeated task needs reusable instructions. Check source, author, requested access, and updates before installation.
Keep credentials and costs visible to the operator
Do not place real keys in a prompt, public example, screenshot, or a shell-history command. Follow the secrets documentation for supported credential handling.
External tools can have usage charges and data-sharing implications. Review each provider's controls and our API key guide before leaving the capability active for scheduled work.
FAQ
Should I enable every available tool?
Enable what the task needs and test the result. More capabilities also introduce more account setup, permissions, dependencies, and failure modes.
Is summarization a separate tool?
An assistant can summarize supplied text as a model task. External document access or parsing may require tools depending on the source format.
How do I fix an unavailable tool?
Inspect the relevant integration documentation, agent policy, credentials, and installed version. Do not invent a tool name or replace the runtime config with a generic YAML list.