Code review

Two other models read the diff before you do.

I'd rather have Codex read Claude's diff than have Claude grade itself. So Claude makes the change, Codex writes the tests and runs them, and agy, Google's Gemini agent, reads the diff for whatever both of them missed. Their reports come back to the session you started in.

Free and open sourceRuns on Mac and LinuxYour own Claude and Codex logins
Overvibe's window with the sidebar and a Claude Code session, as agy's review of the diff arrives as a message
Recorded on a demo machine: agy's review arrives in the Claude session that asked for it.

The model that wrote the code is the worst one to review it

Ask Claude to review its own change and you get a nice explanation of why it's fine. Whatever edge case it didn't think of while writing, it also won't think of while reviewing.

One agent on its own
  • The author reviews its own diff
  • Tests get written last, or never
  • You paste the diff into a second chat yourself
  • Three opinions in three tabs, and you merge them in your head
Agents side by side in Overvibe
  • Codex and agy read it with their own training
  • Codex writes tests while Claude is still at it
  • One command starts each reviewer, attached to Claude
  • Every report shows up in Claude's session as a message

How a review goes

  1. 1Claude makes the change

    Use Claude Code the way you already do, in your repo.

  2. 2It hands out the checks

    Codex gets the tests and agy gets a read-only review. Both sit under Claude in the sidebar.

  3. 3The reports come back

    Each helper sends its result with overvibe send, and its row goes grey once it's done.

  4. 4You read one update

    Claude pulls the findings together and fixes the ones you agree with.

Behind it, two commands. Claude runs these on its own when you ask for a review. You can type them yourself too. Read the CLI reference

$ overvibe new --attach --tool codex --cwd . --name "api: codex tests" \
    --prompt "Write bun tests for POST /invoices and run them. When done, run: overvibe send \"$OVERVIBE_SESSION\" \"<result>\""
$ overvibe new --attach --tool agy --cwd . --name "api: agy review" \
    --prompt "Review the diff in src/routes for bugs. Do not edit. When done, run: overvibe send \"$OVERVIBE_SESSION\" \"<findings>\""

Questions

Will agy edit my files?

Only if the prompt lets it. For a review, tell it to leave the files alone, like the command above does.

Which model should write the tests?

Any of them. I'd start with Codex, since it runs the suite in its own terminal and sends the numbers back.

Do I need all three CLIs installed?

No. Claude Code plus either Codex or agy already gets you a second opinion.

Other ways to pair agents up

Put a second reviewer on your next diff

Install Overvibe, start Claude in your repo, and ask it to bring Codex and agy in on the change.

The Overvibe app: a coordinator with its Codex and Gemini (agy) helpers in the sidebar, each with its status, and one session waiting on you