From Zero to Multi-Agent Engineer with OpenCode¶
Learn OpenCode from first terminal session to confident multi-agent workflows. This 10-week, 40+ hour course is designed for complete beginners who want a practical path into AI-assisted engineering.
By the end, you will be able to:
- Use OpenCode's Plan and Build agents with good judgment.
- Write reusable commands, skills, and custom agents.
- Run multi-agent workflows with clear handoffs.
- Connect MCP servers and plugins safely.
- Evaluate, harden, and version agent workflows for real team use.
- Complete a capstone multi-agent PR-review pipeline.
Start Here¶
Begin with Week 1 - What Is This World?.
Follow the course in order:
- Read the lesson.
- Review the slides.
- Complete each lab before moving on.
- Finish the reflection so the workflow becomes your own.
If you are new to the terminal, that is expected. Week 1 starts there.
What You Need¶
You do not need prior experience with AI coding tools, agent loops, or command-line development.
You will need:
- A computer running macOS, Linux, or Windows with WSL2.
- An internet connection.
- An account with at least one AI model provider, such as Anthropic, OpenAI, or a local Ollama setup.
Free tiers are enough for the first half of the course. Provider setup is covered in Week 2.
Course Map¶
| Phase | Weeks | What you learn |
|---|---|---|
| Foundations | 1-2 | AI and LLM mental models, terminal basics, git basics, OpenCode setup, and your first agent session |
| Core Mastery | 3-6 | Plan vs. Build judgment, custom commands, reusable skills, and custom agents |
| Advanced + Capstone | 7-10 | Multi-agent workflows, MCP and plugins, production safety, evals, and the final capstone |
Weekly Path¶
| Week | Focus | Start |
|---|---|---|
| Week 1 | AI, agents, terminal, and git foundations | Lesson |
| Week 2 | Installing OpenCode and running the first task | Lesson |
| Week 3 | Plan mode, Build mode, and safe execution judgment | Lesson |
| Week 4 | Custom slash commands for repeated workflows | Lesson |
| Week 5 | Skills as reusable on-demand knowledge | Lesson |
| Week 6 | Custom agents and permission boundaries | Lesson |
| Week 7 | Multi-agent workflows and handoff contracts | Lesson |
| Week 8 | MCP servers, plugins, and external capabilities | Lesson |
| Week 9 | Production practices, evals, cost, context, and secrets | Lesson |
| Week 10 | Capstone multi-agent PR-review pipeline | Spec |
How Each Week Works¶
Each week follows the same rhythm:
- Concept: learn one new primitive or judgment pattern.
- Demo: watch the workflow in a realistic setting.
- Lab: practice in a real repo or course exercise.
- Reflection: write down what worked, what failed, and how you will apply it.
The course introduces one major primitive per week so the pieces stack cleanly.
Quick References¶
Keep these open while you work:
- Glossary for course vocabulary.
- OpenCode Cheat Sheet for commands, configuration areas, and safety checks.
- Prompt Patterns for reusable ways to ask agents for help.
For Instructors And Authors¶
Instructors should read PLAN.md first, then use
instructor-guide/pacing.md for pacing, common failure modes, and grading
support.
Authors should treat PLAN.md as the course design contract. Module content
has its own review workflow, so update lesson and lab files intentionally.
Repository Layout¶
modules/week-XX-<name>/ Lessons, slide sources, labs, and reflections
labs/ Sample repos used across multiple labs
instructor-guide/ Pacing notes, failure modes, and rubrics
reference/ Glossary, cheat sheet, and prompt patterns
PLAN.md Course design contract and curriculum rationale
Site Status¶
All course modules are currently marked status: done.
This learner site is published at
opencode-course.pages.dev. GitHub
Actions builds the MkDocs site and deploys it to Cloudflare Pages on pushes to
master. Required Cloudflare deployment secrets are configured in GitHub
Actions.