Skip to content

Your First Session

You've installed OpenCode and oMO. Now let's use them. This walkthrough covers opening a project, asking a question, and seeing the agent pipeline in action.

Open a project

Navigate to any project directory and launch OpenCode:

bash
cd ~/Developer/projects/my-project
opencode

The TUI opens. You'll see a chat interface with a prompt at the bottom and a sidebar showing session history.

TIP

If this is your first time in the project, oMO may prompt you to create a CLAUDE.md or AGENTS.md file. Accept — these files give agents context about your project's stack, conventions, and rules. You can edit them later.

Ask a question

Let's start simple. Type a question about your codebase:

What does the main entry point do?

Press Enter. Here's what happens behind the scenes:

  1. OpenCode receives your message and routes it through oMO's pipeline.
  2. Sisyphus (the executor agent) picks up the task. It reads your project files, finds the entry point, and explains what it does.
  3. The response appears in the TUI.

This is a straightforward query — no planning step needed. Sisyphus handles it directly.

Try a bigger task

Now let's give the pipeline something that benefits from planning. Type:

/start-work

This invokes oMO's /start-work command, which triggers the full pipeline:

  1. Prometheus (the planner) breaks your task into a structured plan with clear steps.
  2. Atlas (the orchestrator) assigns each step to the right agent.
  3. Sisyphus (the executor) carries out the work — reading files, writing code, running commands.
  4. Oracle (the consultant) reviews the output and flags issues.

You'll see the pipeline in action as each agent contributes. The TUI shows which agent is currently active and what it's doing.

Use a skill

oMO ships with skills you can invoke with slash commands. Try one:

/diagnose

The /diagnose skill starts a structured debugging loop: reproduce → minimize → hypothesize → instrument → fix → regression-test. It's designed for when something is broken and you need a disciplined approach to finding the root cause.

Other useful skills for beginners:

CommandWhat it does
/start-workStart a structured work session from a Prometheus plan
/diagnoseDebug a bug with a disciplined loop
/grill-meStress-test your design decisions through interview
/zoom-outGet broader context about how code fits together
/tddTest-driven development with red-green-refactor

See Built-in Skills for the full list.

Project convention files

As you work, you'll notice oMO references two files in your project root:

  • CLAUDE.md: instructions for Claude Code sessions. Covers your stack, coding conventions, and rules.
  • AGENTS.md: instructions for OpenCode agents. Similar content, formatted for the agent pipeline.

These files are how oMO gives agents project context without you re-pasting it every session. You can edit them directly — they're plain Markdown.

A minimal AGENTS.md might look like:

markdown
# My Project

## Stack
- Language: TypeScript
- Framework: Astro
- Runtime: Bun

## Rules
- Always run `bun test` before committing
- Use conventional commits
- Never commit .env files

INFO

For full configuration details, see Config (oh-my-openagent.json).

Session workflow

Here's a typical session flow with oMO:

  1. Open your project: cd into the repo and run opencode.
  2. Describe your task: type what you want to do, or use a slash command like /start-work.
  3. Review the plan: if the pipeline activates, Prometheus presents a plan. Read it. Approve or adjust.
  4. Watch execution: Sisyphus works through the steps. You can intervene at any point.
  5. Verify the result: check the changes, run tests, make sure everything works.
  6. Commit: when you're satisfied, commit the changes. OpenCode can help with commit messages too.

Tips for beginners

  • Be specific: "Add a dark mode toggle to the settings page" works better than "make it look better."
  • Use slash commands: they encode best practices. /diagnose for bugs, /tdd for features, /grill-me for design review.
  • Check the plan before execution: when Prometheus presents a plan, read it. You can ask for changes before Sisyphus starts coding.
  • Iterate: if the first result isn't right, say so. The agent adjusts. You don't need to start a new session.
  • Commit often: after each successful step, commit. This gives you rollback points and keeps diffs small.

Where to go next

  • The oMO Pipeline — understand how Prometheus, Atlas, and Sisyphus divide labor.
  • Skills Ecosystem — explore all available skills and learn to create your own.
  • Playbooks — follow real workflows for starting projects, adding features, and debugging.
  • Reference — deep-dive into configuration, model selection, and agent settings.

Built with VitePress. Contribute on GitHub.