Your First Workflows

You now understand what OpenClaw is, how to install it, how the Gateway works, how skills and memory function, and how to secure your setup. In this final lesson you will build practical workflows, create a custom skill from scratch, learn techniques for effective agent prompting, and explore where to go next.

Workflow 1: Morning Briefing

Set up a heartbeat task that delivers a morning summary every day.

Step 1: Edit your HEARTBEAT.md:

- Every weekday at 8:00 AM, send me a morning briefing that includes:
  - Today's calendar events
  - Unread email count and any flagged messages
  - Weather forecast for my location
  - Top 3 items from my task list

Step 2: Install the required skills:

clawhub install google-calendar
clawhub install email-reader
clawhub install weather-check
clawhub install task-manager

Step 3: Add your location to USER.md:

## Location
- City: London, UK
- Timezone: Europe/London

The agent will now check this list on every heartbeat cycle and send you the briefing when the conditions match.

Workflow 2: GitHub PR Monitor

Have your agent watch a repository and notify you when pull requests need attention.

Step 1: Install the GitHub skill:

clawhub install github

Step 2: Add a heartbeat item:

- Check github.com/my-org/my-repo for open PRs that:
  - Are assigned to me
  - Have been open for more than 24 hours without review
  - Have failing CI checks
  Send a summary via WhatsApp if any PRs match.

Step 3: Make sure your GitHub token is configured:

openclaw config set GITHUB_TOKEN ghp_your_token_here

The agent will proactively notify you about PRs that need your attention, rather than you checking GitHub manually.

Workflow 3: Smart Home Automation

Use OpenClaw as a natural-language interface for your smart home.

clawhub install homeassistant

Now you can send messages like:

• "Turn off all the lights downstairs"
• "Set the thermostat to 21 degrees"
• "Is the garage door open?"
• "Turn on the porch light at sunset"

The agent translates natural language into Home Assistant API calls. Combined with the heartbeat, you can set up automations like "If the temperature drops below 15 degrees, turn on the heating and let me know."

Building a Custom Skill: Meeting Notes

Let's build a skill from scratch that processes meeting transcripts and generates structured notes.

Step 1: Create the skill directory:

mkdir -p ~/.openclaw/skills/meeting-notes

Step 2: Create SKILL.md:

---
name: meeting-notes
description: Process meeting transcripts into structured notes with action items
user-invocable: true
command-dispatch: /meeting-notes
---

When the user provides a meeting transcript (pasted text, uploaded file,
or audio transcription), process it into the following format:

## Meeting Notes: [meeting title or topic]
**Date**: [date]
**Attendees**: [list of participants mentioned]

### Key Decisions
- [Numbered list of decisions made]

### Action Items
| Owner | Task | Deadline |
|-------|------|----------|
| [name] | [task] | [deadline if mentioned, otherwise "TBD"] |

### Discussion Summary
[3-5 bullet point summary of the main topics discussed]

### Open Questions
- [Any unresolved questions or items needing follow-up]

Rules:
- Keep the summary concise (no more than 5 bullets)
- Flag any action items without a clear owner as "UNASSIGNED"
- If no deadline is mentioned, use "TBD"
- Preserve exact quotes for key decisions

Step 3: Test the skill by sending your agent a message:

/meeting-notes

Here's the transcript from today's standup:

Sarah: We finished the authentication module yesterday.
Tom: I'm blocked on the database migration. Need Sarah's help.
Sarah: I can pair with Tom this afternoon.
Manager: Let's ship auth by Friday. Tom, can you have the migration done by Thursday?
Tom: Thursday works if Sarah and I pair today.

The agent will process the transcript and return structured notes with action items.

Effective Agent Prompting

Getting the best results from your OpenClaw agent requires different techniques than prompting a chatbot. Here are key strategies:

Be Specific About the Outcome

Instead of vague requests, describe the exact result you want:

Vague	Specific
"Help me with emails"	"Draft a reply to the email from Sarah declining the meeting and suggesting Thursday instead"
"Check my code"	"Review the PR at github.com/my-org/app/pull/42 for security issues and comment on any findings"
"Plan my week"	"Look at my calendar for next week, identify open slots, and suggest when to schedule the three tasks in my task list"

Use Multi-Step Instructions

The agent excels at sequential tasks. Give it a clear chain:

1. Check my calendar for tomorrow
2. Find any conflicts between meetings
3. Draft reschedule emails for the conflicting ones
4. Send me the drafts for approval before sending

Set Boundaries

Tell the agent what it should NOT do:

Research competitors for our pricing page.
DO NOT visit any URLs that require authentication.
DO NOT make any purchases or sign up for accounts.
Summarize findings in a table comparing features and prices.

Leverage Memory

Remind the agent to use its memory:

Based on what you know about my coding preferences from previous sessions,
review this pull request and flag anything that doesn't match my style.

Useful CLI Commands

Here are commands you will use regularly as you build workflows:

# Check agent status
openclaw status

# View recent logs
openclaw logs --tail 50

# List installed skills
openclaw skills list

# Change the AI model
openclaw models set anthropic/claude-opus-4-6

# Add a new channel
openclaw channels add telegram

# Run a security audit
openclaw security audit

# Reconnect a channel
openclaw channels reconnect whatsapp

# Update OpenClaw
npm update -g openclaw@latest

Next Steps

You now have a solid foundation for using OpenClaw. Here is where to go from here:

• Explore ClawHub at clawhub.com — browse hundreds of community skills
• Read the full docs at docs.openclaw.ai — deep dives into every feature
• Join the community — the OpenClaw Discord server and GitHub Discussions are active
• Build and share skills — publish your custom skills to ClawHub for others to use
• Experiment with models — try different AI providers and compare results for your use cases
• Set up a second channel — connect another platform to see how the same agent works across channels

Key Takeaway

OpenClaw becomes powerful when you combine skills, memory, and the heartbeat into workflows that work for you. Start with simple automations (morning briefings, PR monitoring), build custom skills for your specific needs, and use specific multi-step prompts to get the best results. The entire system is plain text files and npm packages—there is no magic, just well-organized infrastructure for AI agents.