A pull request goes up, and then it sits. The first pass on most PRs is mechanical: does it break the build, do the tests still pass, does the diff do what the description says, any obvious problems. That pass can wait a day for a busy teammate, and the author waits with it.

This is a walkthrough for handing that first pass to Lemon Code, a free coding agent that runs in your browser (there is a Mac app too). No API key needed to start. By the end, every new PR in a repo gets an automatic first-pass review, and a human still makes the real call.

What you are setting up

Lemon Code writes, runs and tests code for you, and it connects to the tools where your work lives: GitHub, Slack, Linear, Jira and 1,500+ other apps, plus MCP servers. You describe what you want in plain words. It shows you what it did: every edited file, every command it ran, the tests that passed.

For this tutorial the outcome is one scheduled automation: read each new pull request, run the checks, and leave review comments on what to fix.

1. Open Lemon Code

Go to lemoncode.ai in your browser and open the app. There is nothing to install and no API key to find.

2. Connect GitHub

Open the Connectors screen, find GitHub, and sign in. That is the whole connection step. You can disconnect at any time from the same screen.

3. Describe the reviewer

Now describe the job in words. Something like:

Watch my-repo for new pull requests. For each one: read the diff, run the test suite, and check the change against the PR description. Leave a review comment for each concrete problem you find, with the file and line. Do not approve or merge anything.

Two things in that prompt matter. It says what "a problem" means concretely (broken tests, diff not matching the description), and it says where the authority stops (no approving, no merging). Write both parts for your own setup.

4. Watch one run before you trust it

Run it on an open PR and read the output. Lemon Code shows the files it looked at, the commands it ran and the results, so you can see whether its comments are real findings or noise. If the comments are off, adjust the wording and run it again. Ten minutes here is what makes the automation worth keeping.

5. Put it on a schedule

Once the comments are good, tell it to run on a schedule: every new PR, or a sweep of open PRs each morning. The first pass is already there when a human opens the review.

What this does not change

A first pass is not a review. The agent catches mechanical problems and saves your team the wait; the design call, the taste, and the merge button stay with people. If a PR matters, a human still reads it.

The details

Lemon Code is free in the browser and on Mac. If you already use Claude Code, Codex, or Gemini CLI, you can bring your own agent and your own subscription or API key instead. Connect and disconnect tools from the Connectors screen whenever you want.

Try it on one repo and one chore: https://lemoncode.ai/home?ref=20b9f909-776a-4f7c-bf3a-3aa5234e1457

If it breaks on your setup, that is worth knowing too.