Skills are what make an OpenClaw agent useful. Without skills, your agent can hold a conversation. With the right skills, it can manage your email, book appointments, monitor your servers, post to social media, and track your GitHub issues — all without you asking.
There are thousands of skills on ClawHub (clawhub.ai). That number sounds like a good thing. In practice, it creates its own problem: how do you know which ones to install, which to avoid, and in what order?
This is what I learned after making a lot of wrong choices first.
What Skills Actually Are
A skill is a plugin — a piece of code that gives your agent a new capability. Each skill:
- Defines a set of actions your agent can take ("check inbox," "send email," "create calendar event")
- Handles the authentication with external services (OAuth, API keys, tokens)
- Tells your agent how and when to use those actions
When you install a skill, your agent gains access to whatever that skill provides. When you uninstall it, that access goes away.
Skills are installed via the ClawHub CLI:
clawhub install <skill-name>
After installation, you usually need to configure the skill — authenticate with the external service via OAuth, set preferences in your openclaw.json, and specify what the skill should monitor. The skill's ClawHub page shows the setup steps.
Why Skill Selection Matters
Not all skills are equal. ClawHub doesn't curate — it hosts. That means:
Some skills are abandoned. The last update was 18 months ago. They may work fine, or they may break with recent OpenClaw versions. There's no automated compatibility checker.
Some skills conflict. Two email skills installed at the same time fight over the same email endpoint. Two Telegram skills can cause your agent to respond to every message twice. Conflicts aren't always obvious.
Some skills have excessive permissions. A skill that monitors your GitHub notifications shouldn't need write access to your repositories. Read the skill's permissions before installing.
Some skills are low-quality. The registry includes experiments, duplicates, and skills built for one person's specific setup. Quality varies wildly.
The install count on ClawHub is your best signal. High install count + recent update = usually safe. Low install count + old update = install with caution.
Which Skills to Install First
This is what I'd recommend based on actual usage, not theory:
Start here (most reliable, high value):
Telegram — The most installed skill on ClawHub. You probably already have this if you followed the onboard wizard. If not, install it first. Everything else is secondary to having a messaging channel that works.
AgentMail — Email integration. Monitors your inbox, summarizes messages, drafts replies for your review. This is the skill that most immediately changes daily work. High install count, consistently maintained.
Google Calendar — Bidirectional calendar access. Your agent can check your schedule, create events, and manage bookings. Pairs well with email for scheduling workflows.
Second tier (add these once your first three are stable):
GitHub — For anyone who works with code or tracks software projects. Creates issues, monitors pull requests, summarizes repository activity.
Linear — Project management integration. If you use Linear for task tracking, this connects your agent to your project board.
Obsidian — Knowledge management. Your agent can read your notes vault, create new notes, and link information across your notes.
Slack — If your team uses Slack, this connects your agent to channels and direct messages. One of the more complex setups — requires both SLACK_BOT_TOKEN and SLACK_APP_TOKEN in your configuration.
Browser automation — Your agent can navigate websites, fill forms, and extract information. Powerful but use with caution — these skills require more careful permission scoping.
Skills to Approach With Caution
Multi-tool bundles. Skills that combine 6 different integrations in one package often do all of them poorly and create difficult-to-debug conflicts. Install individual skills instead.
File system skills with write access. Be explicit about what directories your agent can write to. An overly permissive file system skill is a significant security risk, especially combined with CVE-2026-25253 or default network exposure.
Unmaintained skills. Last update older than 6 months: install only if you're confident you can debug it yourself when it breaks.
Skills with very low install counts from unknown authors. Not because they're necessarily bad, but because there's no track record.
How to Install and Configure a Skill
The basic pattern:
clawhub install agentmail
This downloads and installs the skill. Then restart the gateway to pick it up:
openclaw gateway restart
For skills that need external service access (like email), you'll typically need to:
- Authorize via OAuth (the skill provides a URL to open in a browser)
- Grant the requested permissions
- Set preferences in your
openclaw.jsonor via the web dashboard athttp://127.0.0.1:18789/
Test it immediately. Send a message to your agent: "What's in my inbox?" If the skill is working, it responds with a summary. If nothing happens or you get an error, check openclaw logs for the skill's error output.
Managing Skills Over Time
Install new skills one at a time. After each install:
- Test that it works
- Test that your other skills still work
- Check resource usage (skills can have memory leaks)
If you're adding multiple skills, wait at least a day between each. This makes it easy to identify which skill caused a problem when one appears.
To remove a skill:
clawhub uninstall <skill-name>
openclaw gateway restart
To check which skills are installed:
clawhub list
Keep your installed skill list short. The more skills your agent has, the more maintenance surface you have. Twenty skills that all work is better than fifty skills where three are broken.
Full OpenClaw tutorial for beginners →
Skip the skill selection headache — Volos curates skills matched to your workflow →
