Guides to AI coding agents on GitHub.
Start with your first pull request, even with no GitHub experience. Pick a walkthrough for the thing you want to build, read how seven real projects were built the same way, then the questions engineers ask before they let an agent near a repository.
Overview
Start here
New to GitHub? Your first AI pull request, start to finish.
From a new GitHub account to a merged change, with every new word explained.
How to review a pull request when you didn't write the code
The two-minute habit that keeps every AI-written change under your control.
The words behind an AI-built change, in plain language
Repository, issue, branch, pull request, review, merge and ten more, each with where you meet it in ProPR.
Build something
Six worked examples, each from the first prompt to the pull requests you review. Four start from an empty repository; two bring the code you already have. The five-step pattern they share is one page.
Build a website or landing page from an empty repository
A landing page on Astro and Cloudflare, built one reviewed pull request at a time.
Build a business dashboard or internal tool from an empty repository
An internal tool on Next.js and Supabase: sign-in, data, charts, one pull request per piece.
Build a browser game from an empty repository
A 2D game on Phaser, each level, menu and sound landing as its own reviewed change.
Build a mobile app from an empty repository
An Expo app you test on your own phone, built screen by screen in reviewed pull requests.
Add ProPR to a codebase you already have: the first week
From one small issue to a planned feature in a week, without changing how the code ships.
Take over or modernise an older codebase: tests first, then refactors, then features
Pin the behaviour with tests, refactor in small reviewed steps, then add the features that waited.
Case studies
Seven real projects built this way, with counts from GitHub as of 11 November 2026. All 7 case studies →
- Game Claude Cope 458 merged pull requests, 412 written by agents
- Web app mcptest.io 200 agent-written pull requests merged over sixteen months
- Commercial product dig.vin 185 merged pull requests between 17 July and 11 September 2026
- Developer tool Agent Tank 56 merged pull requests, 46 written by agents
- Business productPrivate project A messaging-assistant platform 770+ merged pull requests, 650+ written by agents
- Mobile appPrivate project A ten-year-old mobile app 110+ agent-written pull requests merged
- Marketing site propr.dev, this website 113 merged pull requests, 76 opened by agents
Questions engineers ask
How to get an AI coding agent to open pull requests from GitHub issues
Label an issue, get a linked pull request back, and review it like any other.
Run Claude Code on GitHub issues and pull requests
Your Claude subscription, a model label on the issue, and a pull request back.
Run AI coding agents on your own server
What self-hosting keeps on your infrastructure, and what it takes to run.
Keep a human approval step on every AI-written change
Plan approval, PR review, manual merge, and what the agent is never allowed to do.
See what each AI-written pull request costs
Per-task usage, per-model comparison, and subscriptions versus API keys.
Fix failing CI runs with an AI agent, in the same pull request
A red check becomes a fix commit on the same branch, explained in a comment.
Walk one change end to end next.
The playbooks follow single changes from the first prompt to the merged pull request, and the quickstart gets a stack running in two commands.