Before you start
Install Diri and the Claude Code and Codex CLIs. Sign in to each agent using its own setup flow, then open a local Git repository with at least one commit.
Diri uses your existing tools and accounts. If an installed agent is missing, open Settings → Agents and refresh discovery.
1. Split the work
Choose tasks that can be reviewed independently. In a small task-list project, one agent can implement filtering while the other adds a summary function.
Add filterTasks(tasks, status) in a new module, with tests for all, open, and done.
Add summarizeTasks(tasks) in a new module, with tests for total, open, and done counts.
Specify the files each task owns. Agree on shared data shapes first. Two agents changing the same interface will still need coordination.
2. Start each agent in a worktree
- Open New Agent and select your project.
- Choose Claude Code. Switch Current folder to New worktree.
- Give the session a clear title, enter the first task, and launch.
- Repeat for Codex, selecting the same project and another New worktree.
Each agent gets a separate checkout and branch. Their edits stay apart while they run. Worktrees share the Git repository; they do not isolate ports, databases, or external services.
This workflow uses a local repository. For remote work, start with an existing project directory on the host.
3. Follow progress
Use the sidebar to switch between sessions and see which one is working or needs input. Read each agent’s request before approving a command. If one task discovers a change to the shared interface, pass that decision to the other agent.
You can close and reopen Diri while local sessions keep running. This does not keep the computer awake or preserve processes across a reboot.
4. Review and combine
Open each session’s Changes pane. Check the diff, ask the agent to run the relevant tests, and review the result. Keep each task in its own commit or pull request.
Merge one branch, bring the other up to date, and run the tests again with both changes present. Separate worktrees prevent agents from overwriting each other’s files during the task; the combined result still needs review.
For setup and shortcuts, see the getting-started guide.