Skip to main content

5. OpenClaw Skills

Action Step

Complete this before moving on.

Ask your OpenClaw agent to create a simple skill — for example, "Create a skill that summarizes any text I give you into exactly 3 bullet points." Then test it by giving it a paragraph to process.

Comment in Slack

Post your answer in your onboarding channel.

Why do you think you sometimes need to explicitly tell your agent to follow a skill's instructions "to the letter" instead of it just doing it automatically?

Training Guide

You built skills in Section 3. Now let's see how skills work in OpenClaw — same concept, slightly different mechanics.

Skills Are Still Just Markdown Files​

  • In Claude Code: you created .md files with instructions and loaded them via copy-path or /skill
  • In OpenClaw: same thing — skills are markdown files with instructions
  • They live in the agent's workspace on the server

It's the same Neo-in-the-Matrix concept — you're uploading knowledge to its brain.

How to Create a Skill for Your Agent​

Option 1: Tell your agent in Slack to create a skill — "Create a skill that processes meeting transcripts and extracts action items with owners and due dates." It'll write the skill file on its server.

Option 2: Write the skill yourself and have the tech team add it to your agent's workspace.

Option 3: Copy a skill from the team's Skills and SOPs repo.

The Adherence Nuance​

Key insight: sometimes you have to explicitly tell your agent to invoke the skill. Sometimes you have to tell it to follow the instructions to the letter.

Why? Because OpenClaw decides on its own whether to use a skill — unlike Claude Code where you manually load it.

Tips for better adherence:

  • Be explicit: "Use the transcript-processing skill to process this"
  • Be firm: "Follow the skill instructions exactly — don't improvise"
  • As models get smarter, this gets better — but for now, be direct

Same prompting principles from Section 2 — specific numbered steps, important instructions first, show examples. It applies even more here because you can't see what the agent is doing step-by-step.

ClawdHub — The Skill Registry​

ClawdHub is a community marketplace — 13,000+ skills. Think of it like an app store for your agent's abilities. Weather, email summaries, Salesforce queries, calendar management, etc.

At LeanScale, the tech team manages which skills get installed — you don't install random ones yourself. Why: security — not all community skills are safe.

Invoking Skills via Slash Commands​

Skills with slash command access can be triggered with /skill <name> in chat. Example: /skill transcript-processor then paste the transcript. Same concept as slash commands in Claude Code — just in Slack instead of the terminal.