OpenClaw Skill Complete Guide: Teach AI Repeatable Workflows

What Is a Skill? In One Sentence

Skill = A fixed workflow you teach AI to follow.

It’s like training a new colleague on a task: do A first, then B, then C. A Skill writes this process down so OpenClaw’s Agent can run it repeatedly.

---

Why Do You Need Skills?

You might be thinking: “I can just tell ChatGPT directly, why bother writing a Skill?”

Great question. Here’s the difference:

Direct Chat	Using a Skill
Describe from scratch every time	Write once, use forever
Inconsistent results (prompt may vary)	Consistent results (fixed workflow)
Can’t connect to other tools	Can connect to Google Drive, Notion, etc.
Only handles single steps	Can orchestrate multi-step workflows

An Analogy

• Direct chat = Verbally telling an assistant each time “look up XX, organize into a table, then email the boss”
• Skill = Writing an SOP for the assistant — from now on, just say “run the weekly report”

---

Basic Structure of a Skill

A Skill consists of the following parts:

# 📄 my-first-skill.yaml
name: "Weekly News Digest"
description: "Search for the latest news on a specified topic, organize into a summary, save to Google Drive"

# Trigger method
trigger:
  type: manual  # manual = manual trigger | schedule = scheduled trigger
  # schedule: "0 9 * * 1"  # Every Monday at 9 AM (cron format)

# Input parameters
inputs:
  - name: topic
    type: string
    description: "Topic to search for"
    default: "AI Technology Trends"

# Steps
steps:
  - id: search
    action: web_search
    params:
      query: "{{topic}} latest news this week"
      max_results: 10

  - id: summarize
    action: llm_generate
    params:
      prompt: |
        Please organize the following search results into a summary:
        {{search.results}}
        
        Format requirements:
        - One paragraph per news item
        - Include title, source, and key summary
        - End with a "Weekly Trend Summary"

  - id: save
    action: google_drive_create
    params:
      title: "Weekly_Report_{{topic}}_{{date}}"
      content: "{{summarize.output}}"
      folder: "Weekly Reports"

Looks like a lot? Don’t worry, let’s break it down step by step.

---

Breaking Down Each Part of a Skill

1. Basic Info (name / description)

name: "Weekly News Digest"
description: "Search for the latest news on a specified topic, organize into a summary, save to Google Drive"

• name: The Skill’s name — used when you tell the Agent “run XX”
• description: Describes what the Skill does — the Agent references this to decide when to use it

2. Trigger Method (trigger)

trigger:
  type: manual

Trigger Type	Description	Example
manual	Manual trigger	You say “run weekly report”
schedule	Scheduled execution	Runs automatically every Monday at 9 AM
event	Event-driven trigger	Runs when a specific Email is received

3. Input Parameters (inputs)

inputs:
  - name: topic
    type: string
    description: "Topic to search for"
    default: "AI Technology Trends"

Parameters make your Skill flexible. The same Skill can run with different topics:

• “Run Weekly News Digest with AI Technology Trends”
• “Run Weekly News Digest with Blockchain”

4. Steps (steps)

This is the core of the Skill. Each step contains:

• id: Step identifier (subsequent steps can reference it)
• action: The action to execute
• params: Parameters for the action

Steps pass data between each other using {{step_id.output}} — like passing a baton in a relay race.

---

Common Actions List

Action	Description	Use Case
web_search	Web search	Gather data
llm_generate	Call LLM	Summarize, translate, analyze
google_drive_create	Create Google Drive file	Save results
google_drive_read	Read Google Drive file	Read existing data
gmail_send	Send Email	Deliver reports
gmail_read	Read Email	Fetch new messages
notion_create	Create Notion page	Note management
notion_query	Query Notion database	Data retrieval
http_request	HTTP request	Call any API

---

Your First Skill: Email Summary Bot

Let’s start with a practical and simple Skill:

Requirements

Every morning, organize unread Emails into a summary so you know what’s important in 30 seconds.

Complete Skill

name: "Email Morning Summary"
description: "Read unread Emails and organize into today's highlights"

trigger:
  type: schedule
  schedule: "0 8 * * *"  # Every day at 8 AM

steps:
  - id: fetch_emails
    action: gmail_read
    params:
      filter: "is:unread"
      max_results: 20

  - id: summarize
    action: llm_generate
    params:
      prompt: |
        Here are today's unread Emails ({{fetch_emails.count}} total):
        
        {{fetch_emails.results}}
        
        Please organize into the following format:
        
        ## 🔴 Needs Immediate Action
        (Emails that need a reply or action today)
        
        ## 🟡 Needs Attention
        (Important but not urgent)
        
        ## 🟢 Acknowledged
        (Notifications, no action needed)
        
        One-line summary per Email.

  - id: notify
    action: gmail_send
    params:
      to: "me"
      subject: "📬 Today's Email Summary ({{date}})"
      body: "{{summarize.output}}"

What Does This Skill Do?

1. Runs automatically at 8 AM every day
2. Reads all unread Emails (up to 20)
3. Uses LLM to classify and organize into “Needs Action / Needs Attention / Acknowledged”
4. Sends the summary to yourself

Set it up once, and every morning the first thing in your inbox is today’s summary.

---

Advanced Techniques

1. Conditional Logic

steps:
  - id: check
    action: llm_generate
    params:
      prompt: "Does this Email contain urgent keywords? Answer yes or no"
  
  - id: alert
    action: gmail_send
    condition: "{{check.output}} == 'yes'"
    params:
      to: "me"
      subject: "⚠️ Urgent Email Alert"
      body: "You received a potentially urgent Email"

2. Loop Processing

steps:
  - id: process_each
    action: llm_generate
    loop: "{{fetch_emails.results}}"
    params:
      prompt: "Translate the following Email content to English: {{item.body}}"

3. Error Handling

steps:
  - id: risky_step
    action: http_request
    params:
      url: "https://api.example.com/data"
    on_error:
      action: gmail_send
      params:
        to: "me"
        subject: "Skill Execution Failure Notification"
        body: "Step risky_step failed: {{error.message}}"

---

Common Skill Examples

📰 Industry News Digest

Search → Summarize → Save to Notion → Email notification

📧 Auto-Classify Customer Emails

Read Emails → LLM classifies → Apply Gmail labels

📊 Competitor Monitoring

Search competitor news → Compare with last week → Generate report

📝 Meeting Notes Organizer

Read recording/transcript → Extract key points → List action items → Save to Notion

---

FAQ

Q: Can I share Skills with others?

Yes! A Skill is just a YAML file — you can share it directly. The OpenClaw community also has a Skill Marketplace where people share their Skills.

Q: What if a Skill fails during execution?

OpenClaw logs every execution. You can see the execution status and error messages for each step in the admin interface.

Q: What’s the maximum number of steps in a Skill?

There’s no technical limit, but we recommend keeping it under 10 steps. If the workflow is too complex, consider splitting it into multiple Skills and chaining them with a “main Skill.”

---

Next Steps

• 🤖 Learn about Agent: Your AI Double — let the Agent auto-select and combine Skills
• 🧠 Understand Soul: Give Your Agent Memory and Personality
• 💬 Prompt Engineering: Write Better Skill Prompts
• 🔗 MCP Protocol: Connect Skills to External Tools
• 📚 Build a Knowledge Management System with Skills
• ☁️ Haven’t deployed yet? See the Cloud Deployment Guide