Workflow

Claudette is designed around a simple loop: prompt, review, iterate, merge. Here’s how it works in practice.

The Core Loop

1. Prompt — describe what you want the agent to build or fix
2. Review — inspect the diff to see what changed
3. Iterate — send follow-up prompts to refine the output
4. Merge — integrate the finished work into your main branch

Each workspace is an isolated git worktree with its own branch, so you can run this loop on multiple tasks simultaneously without conflicts.

Working with Multiple Agents

The real power of Claudette is running several agents in parallel:

1. Create multiple workspaces for a single repo — each gets its own branch and agent
2. Assign different tasks to each workspace (e.g., one for a new feature, one for a bug fix, one for tests)
3. Switch between workspaces with ⌘/Ctrl + Shift + [ and ⌘/Ctrl + Shift + ]
4. Review and merge each workspace independently

Since each workspace has its own worktree, agents never step on each other’s work.

Using the Terminal

The integrated terminal (⌘/Ctrl + `` “) runs inside each workspace’s worktree directory. Use it to:

• Run tests after the agent makes changes
• Start a dev server to manually verify the output
• Run any shell commands in the workspace’s isolated environment

Tips

• Be specific in your prompts — the more context you give, the better the output
• Use plan mode (Shift + Tab) for complex tasks — the agent will outline its approach before writing code
• Set up a setup script in repo settings so each new workspace starts with dependencies installed
• Use @-mentions to reference specific files in your prompt (type @ and start typing a filename)