Your First Agent Loop in 10 Minutes

In this tutorial, you will create and run your first multi-agent coding loop in Zenborg. You will watch two models plan a feature, review each other’s work, implement the code in an isolated Git worktree, and verify the result.

What You Need Before You Start

Zenborg installed and running on macOS or Windows.
At least one connected CLI agent (such as Claude Code or Gemini CLI).
A Git repository selected as your current project.

Ordered Screens Overview

During this tutorial, you will navigate through four screens in sequence:

Screen 1: Loops Dashboard — Creating a new loop.
Screen 2: Loop Configuration Panel — Setting roles and scope.
Screen 3: Active Stage Timeline — Monitoring worktree execution.
Screen 4: Review and Diff View — Inspecting final changes.

Step-by-Step Instructions

Step 1: Open the Loops View

Launch Zenborg and click on the Loops tab in the main navigation.
Ensure your target repository appears in the project selector dropdown.

Step 2: Define Your Goal

Scroll to the Create New Loop form.
In the Task Scope text area, enter a concrete development task:
```
Add a footer with the current year to index.html
```

Step 3: Assign Agent Roles

In the Manager Agent dropdown, select antigravity (or claude). The Manager Agent drafts the plan, synthesizes review feedback, and conducts the final code review.
In the Worker Agent dropdown, select browser with gemini (or a CLI agent like gemini). The Worker Agent executes the two plan reviews, writes the code in the worktree, and implements any fix rounds.
Set Max Correction Rounds to 4.
Leave "Allow outside project folder" unchecked to keep edits confined to your repository.

Step 4: Start the Loop

Click the button labeled “Start Orchestration Loop”.
Zenborg creates an isolated temporary Git worktree for the loop.

Step 5: Observe the Pipeline Stages

Watch the timeline update in real time:

plan_draft: Manager Agent analyzes your repository structure and writes a step-by-step plan.
plan_review_1 & plan_review_2: Worker Agent sessions check the plan in parallel for missing edge cases.
plan_synthesis: Manager Agent merges the feedback into a final specification.
execution: Worker Agent opens the worktree, modifies the target files, and runs tests.
review: Manager Agent inspects the Git diff.

Step 6: Review and Merge

Once all stages complete, scroll to the Worktree section.
Click Inspect Diff to review the commit.
When satisfied, click Merge to Main Branch.

[Visual]	[Spoken]
00:00 - 00:12 Close-up of Zenborg desktop app opening to the Loops view. Mouse clicks into the scope text area.	“Welcome to Zenborg. In this quick video, we’ll run an automated coding loop where multiple AI models plan, code, and review a feature together.”
00:12 - 00:25 Typing: “Add an accessible dark mode toggle button to the navbar” into `phase-scope-input`.	“First, we type our task description. We want an accessible dark mode toggle with persistent local storage settings.”
00:25 - 00:40 Selecting Manager Agent in `#planner-agent-input`, then selecting Worker Agent in `#worker-agent-input`. Clicking `"Start Orchestration Loop"`.	“Next, we assign our team: our Manager Agent will plan the architecture, and our Worker Agent will implement the code. Let’s click Start Orchestration Loop.”
00:40 - 00:58 Screen shows stage progress: `plan_draft` completes, parallel review cards light up green, then `execution` starts.	“Notice that Zenborg created an isolated Git worktree. Your working branch is never modified directly. Claude drafts the plan, two reviewers critique it in parallel, and Gemini writes the code.”
00:58 - 01:15 Review stage completes with a green checkmark. User clicks on the Git diff viewer, revealing clean changes.	“The final code review passes with zero issues. You can inspect the clean diff, run your tests, and merge the branch with complete confidence. That’s Zenborg in under ten minutes.”