Use Claude Code and Codex Together

Use Claude Code and OpenAI Codex on the same project in Synara. Plan a change, hand off implementation, review the diff, and verify the result.

You can use Claude Code and OpenAI Codex on the same project in Synara by handing one task between providers, or by giving each provider a separate task and Git worktree. This tutorial uses a sequential handoff: Claude inspects a problem, Codex implements the agreed change, and Claude reviews the resulting diff.

The provider roles are interchangeable. Choose them for your workflow and account access; this example does not depend on a particular model or claim that either provider is always better at a role.

Before you start

  • Download Synara and add a local Git repository.
  • Follow the Claude Code setup guide and Codex setup guide. Confirm both runtimes work with your own accounts.
  • Choose a small change you can verify. The example below uses a search form that incorrectly submits empty queries; adapt the file paths and commands to your project.
  • Inspect existing changes and run the project's normal checks so you know the starting state.

Use one worktree for this sequential workflow. If you want both agents to edit at the same time, use separate worktrees and define independent scopes.

1. Ask Claude Code to inspect the behavior

Create a task, select Claude Code, and confirm its repository and working directory. Ask for an investigation before authorizing edits:

Inspect how this project's search form handles an empty query.
Find the submit handler, request boundary, and relevant existing checks.
Explain whether empty input produces an unnecessary request or error.
Propose the smallest correction if the problem is confirmed.
Do not edit files. Include file references and verification steps.

Read the referenced code and try the current form yourself. If the behavior is already correct, stop or choose another confirmed issue. A plausible agent explanation is not evidence that a fix is needed.

If there is a defect, agree on the intended behavior, such as treating whitespace-only input as empty while preserving normal searches. Record that decision in the task before handing off.

2. Hand the task to Codex

Wait for Claude's turn to finish. Inspect the working tree and preserve any changes already present. Use the task's Hand off control to select Codex when it is available in your installed version.

The task environment continues, and Synara passes context to the receiving provider. Provider-native sessions, hidden reasoning, tools, permissions, and background tasks are not interchangeable. See Provider handoffs for the full boundary.

Give Codex an explicit implementation instruction:

Implement the agreed empty-query fix in this task's current worktree.
First inspect git status, the current diff, and the files identified above.
Verify the diagnosis against the code before changing it.
Preserve existing local changes, normal searches, and unrelated behavior.
Use the existing validation pattern. Add or adjust a focused regression
check if the repository has a suitable test for this behavior.
Run the relevant checks and report the exact commands and results.
Do not commit or push yet.

Answer required approvals and questions as they appear. If Codex discovers that the agreed behavior conflicts with the existing API contract, resolve that question before expanding the change.

3. Review the implementation with Claude Code

Once Codex finishes, inspect the diff yourself. Then hand the task back to Claude Code for a review with a specific scope:

Review the current search-form diff without editing.
Check empty input, whitespace-only input, and ordinary searches.
Verify that validation happens before sending an unnecessary request.
Read the actual test output and report gaps or confirmed defects
with file references. Do not treat the previous summary as proof.

A second provider can identify a missed assumption, but it can also repeat the first provider's mistake. Resolve findings against the actual code and behavior. If a fix is needed, assign one provider to make it and review the updated diff again.

4. Verify the result in the running app

Start the app using the repository's documented command from this task's terminal. Open it in Synara's browser and exercise these cases:

InputExpected result for this example
Empty fieldNo search request is sent; the form follows the agreed empty-input behavior.
Spaces onlyThe form treats the input as empty.
A normal search termA request is sent and the existing results behavior still works.
A term with no matchesThe normal no-results state remains usable.

Where relevant, inspect network requests as well as visible messages. Run the focused regression check and the broader checks required by the repository. Record anything you could not verify.

Read Browser verification for the browser workflow and Your first task for the complete review process.

5. Commit the reviewed change

Inspect git status --short and the complete diff. Confirm that the change contains only the intended fix and verification. Commit the reviewed files, then use the pull-request workflow if the project requires it.

The result should be one understandable change with evidence for its behavior, regardless of how many providers contributed.

When should both agents run in parallel?

Use a handoff when the next step depends on the previous result, such as implementation followed by review. Use parallel tasks when the scopes can proceed independently, such as a client and server built against an agreed contract.

For concurrent editing, give each task its own worktree and designate one owner to integrate both changes. Start with the worked example in Run AI coding agents in parallel.

Common questions

Does Synara include access to Claude Code or Codex?

Synara connects to your installed runtimes and configured accounts. Provider authentication, model availability, billing, and usage limits remain with each provider. Connecting both does not combine their allowances.

Does handing off copy everything from the original provider?

No. The task keeps its environment and Synara passes context, but the receiving provider uses its own session capabilities. Ask it to inspect the files, current diff, and check output before continuing.

Can I reverse the provider roles?

Yes. You can start with Codex and hand the task to Claude Code. Keep the same rules: finish the active turn, inspect the environment, state the next objective, and verify the resulting work.

What if a provider cannot start?

Check its CLI and authentication outside Synara, then consult Provider troubleshooting. Preserve the working tree while correcting the runtime setup.

Last updated on