Add ProPR to a codebase you already have: the first week

You already have a product, a repository and a way of shipping. ProPR joins that without changing it: the work arrives as pull requests, your checks run, and you press merge. The first week below starts with one small issue and ends with a planned feature.

Overview

Example project. Ledgerly is a small invoicing web app with three years of history, a test suite, and a backlog of forty open issues. Two people work on it part time. The project is an illustration written to show the steps.

A codebase with history that added ProPR: mcptest.io, a developer web app, has merged 200 agent-written pull requests this way since July 2025.

Keep the stack you have

There is nothing to pick. ProPR works on the repository as it is. The one thing worth doing on day one is telling the agent how to install and test the project, so it runs your tests the way your pipeline does. That is a short setup script in the repository; the repository best practices show the file, and the agent writes it for you if you make that the first issue.

Day one: add the repository and run one small issue

In ProPR, open Repositories, choose Add Repository, and enter the repository name. Pick the smallest item in the backlog: a confusing empty-state message, a date shown in the wrong format. Add the AI label to that GitHub issue, or start it as a New Task from the ProPR dashboard. Twenty minutes later there is a pull request. Read the summary, look at the diff, watch your own checks go green, and merge it like any other.

@youopened an issue

Empty invoices page says "No data"

When a new customer has no invoices yet, the page shows "No data". Show "No invoices yet" with a button that opens the new-invoice form. Keep the existing styles.

Day two: get a second opinion on an open pull request

Pick a pull request a teammate wrote and type /review as a comment. ProPR reads the diff and posts numbered findings: merge blockers, suggestions, and a score. Add a model name after the command to hear from a different model than the one that wrote the code. Nothing changes until you ask; /fix with the numbers you agree with applies those and no others. The second-model review playbook shows one on a public pull request.

A follow-up run on the pull request: read the PR comment about the phone layout, move the summary below the form, run the checkout tests
A follow-up run started from a pull request comment.

Day three: fix a bug with a test

Take a bug from the backlog and ask for the test first.

@youopened an issue

Invoice total ignores a discount applied after tax

Write a failing test that reproduces this with the example from customer report #318, then fix it. Acceptance: the new test passes, every existing test passes, and the PR description shows the before and after totals.

The pull request shows the test, the fix and the numbers. If your pipeline fails on it, switch on auto follow-up on failed CI for the repository and the next time a check goes red the agent pushes a fix commit to the same branch with the failure explained. The CI fix playbook shows the loop.

Day four: ask for changes in your own words

Not every pull request is right the first time, and the comment thread is where you fix that. Comment on a line of the diff or in the conversation, and a new commit arrives on the same branch with a reply that links to the undo. Keep the habit from day one: say what you want, read what came back, merge when it is right. The follow-up playbook shows the exchange in full.

@youcommented on the pull request

Good fix. Move the discount logic into the existing InvoiceTotals helper instead of a new file, and add one more test for a 100% discount.

Day five: plan a feature

Now a feature from the backlog that would take a week by hand. Open Planner Studio, pick the repository, and describe it. ProPR already knows the codebase from indexing it, shows you which files it will read, and drafts a plan you refine in chat until each task is a pull request you would happily review on its own.

Add recurring invoices: a customer can be set to receive the same invoice monthly or quarterly. Reuse the existing invoice form and email sending. Keep the schema change, the scheduling job, the form changes and the email changes as separate tasks, with tests for each.

A plan for that feature looks like this.

  1. Schema: recurrence settings on an invoiceDone when: a migration adds the recurrence fields, the model validates monthly and quarterly, and tests cover the validation. No user-facing change.
  2. Scheduled job that creates the next invoiceDone when: a daily job creates the next invoice for each due recurrence, never creates duplicates, and tests cover month-end dates.
  3. Recurrence options on the invoice formDone when: the form shows a "repeat" choice with a preview of the next three dates, and the screen matches the existing form styles at desktop and phone widths.
  4. Email for generated invoicesDone when: a generated invoice sends the existing invoice email with a line explaining it is recurring, and a test checks the email content.

Approve the plan and ProPR files the issues. Choose Execute as Epic PR so they run in order on one branch, each with its own pull request, and the feature lands as one epic pull request you merge when it is complete.

Context Scope at 60%, between Focused, Expanded and Full Scan: a dependency-graph scan with a budget of up to 588k tokens
The files the planner will read, ranked by relevance, before any token is spent.

After the first week

The backlog is now a queue. Label the clear issues and let them become pull requests while you review. Let ProPR scan the repository for improvements and turn the ones you keep into a plan. Put a weekly automation on the repository that reports what changed and what looks risky. When a bigger outcome has no clear task list yet, hand it to a goal and steer it from its draft pull request.

Good to know

  • The agent works in a clean Linux container with Node.js and Python and your repository's setup script. It never sees your production servers or databases; it runs your tests against the fixtures the suite already has.
  • Your branch protection rules, required checks and reviewers all apply. ProPR opens pull requests; it merges nothing unless you turn that on for a plan.
  • Each pull request records the model, the prompt and the cost, so you see what each change took from your AI subscription.

Start your first week.

Everything above happens in a browser. ProPR Hosted gives you your own instance, run by the ProPR team; the live demo lets you click through plans and pull requests first.

On your own server, self-host ProPR, free and open source.

All Build something walkthroughs