Weeks 21-22: Final Project — Build Your Own AI Agent

This is it. Two weeks to build an AI agent that solves a real problem for you.

No templates. No step-by-step instructions. You decide what to build, how to build it, and how to make it good.

Requirements

Your final project must have:

2+ custom skills — not copies of what you built before, but new skills that work together
Connected to Telegram — you (and others) can chat with it from a phone
Solves a real problem — something you'd actually use, not a toy demo
Custom personality — SOUL.md and IDENTITY.md that make it yours

Timeline

Week 21: Build

Day 1-2: Pick your project and plan the skills
Day 3-4: Write SOUL.md, IDENTITY.md, and your first skill
Day 5-6: Write your second skill and test everything
Day 7: Fix bugs, polish

Week 22: Polish and Present

Day 1-3: Test with real use, fix issues, improve skills
Day 4-5: Have at least 2 other people try it and give feedback
Day 6: Make final improvements based on feedback
Day 7: Write your final reflection

Getting Started

Pick a project from project_ideas.md or come up with your own
Plan your skills — what should each one do?
Start with templates — use the files in project_template/ as a starting point
Build on your server:

ssh root@YOUR_IP_ADDRESS

Create your project:

# Back up your current workspace (just in case)
cp -r /opt/openclaw/workspace /opt/openclaw/workspace_backup

# Edit your soul
nano /opt/openclaw/workspace/SOUL.md

# Edit your identity
nano /opt/openclaw/workspace/IDENTITY.md

# Create skills
mkdir -p /opt/openclaw/workspace/skills/skill_one
nano /opt/openclaw/workspace/skills/skill_one/SKILL.md

mkdir -p /opt/openclaw/workspace/skills/skill_two
nano /opt/openclaw/workspace/skills/skill_two/SKILL.md

Test as you go:

openclaw chat "test message for skill one"
openclaw chat "test message for skill two"

Deliverables

Your SOUL.md (copy to this repo)
Your IDENTITY.md (copy to this repo)
Your skill files (copy each SKILL.md to this repo)
Screenshots of at least 5 Telegram conversations showing both skills in action
Feedback notes — what did the 2+ people who tested it say?
reflection.md — your final reflection on the course

Grading Yourself

Ask yourself:

Would I actually use this agent? (If yes, you built something good)
Do both skills work reliably? (Test each one 5+ times)
Does the personality feel right? (Not generic, not annoying)
Did other people find it useful or interesting?

Stuck?

You have all the tools:

Re-read your SKILL.md files from previous weeks for patterns
Ask Claude for help: "I'm building an OpenClaw skill for [X]. It should [describe behavior]. Write me a SKILL.md."
Check OpenClaw logs: openclaw logs --tail 50
Re-read the Anthropic blog post from Week 13 for agent design ideas

This is your project. Make it something you're proud of.

README.md