Refactor compiler breakage across a huge codebase with Claude
Claude helped a team clean up widespread compiler failures after they tightened settings in a 50-project mono-repo.
- What they did
- On a mono-repo with roughly 50 projects and more than 1M lines, the team switched on stricter compiler checks and got hit with tens of thousands of errors. Claude then sorted the failures into repeatable patterns and rewrote broad sets of code to match, using the team’s custom skills or new scripts when that made the job safer. They kept structural or behavior-changing work out of the automated passes and limited bulk refactors to performance and cosmetic fixes.
- How it went
- The agent updated over 200K lines, which the author described as effectively out of reach for a person working inside an IDE. That cleanup also cleared the way for a move to dotnet 10 LTS and made later dependency upgrades easier. No runtime or wall-clock figures were reported.
- Tips
- Use bulk automation only for refactors that do not change behavior; save logic changes for separate, tightly reviewed sessions. It helps to classify issues first so the agent can handle one safe pattern at a time.
Try it yourself with Claude
In [repo or path], I want to enable [compiler flag or warning level] and deal with the errors it exposes. Build the project, tally the failures by pattern, and start with the most common safe category. For that category, generate a reusable script or skill that updates the code consistently across the repo, then run it and inspect the new build output. Continue with the next safe category after I approve the plan. Do not batch any change that could alter behavior, and pause for my approval before any payment, sending, deletion, submission, or other irreversible action. Show a concise change summary before each round.
Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.
Get 3 agent uses like this every morning
With the day's top AI news. Free, and one click to unsubscribe.
Related agent uses
Plan work in a PRD, then assign the build
A developer asks a stronger model for a plan, turns it into tasks, and routes implementation to lower-cost agents.
Build an anonymous board for AI hopes and worries
They used Claude Code to create a site where people can post anonymously, with one model sorting entries and another finding themes.
Use Copilot to scaffold a full-stack app
They let GitHub Copilot start a new full-stack app, then kept fixing setup issues until the React interface was styled and usable.
Use Claude as an async coding partner
They paired with Claude and Jules while coding locally, then copied changes back and forth instead of sharing a live editor.
Use Claude to build a Lean runtime for web targets
A developer had Claude help create Lasm, a Lean 4 runtime that can run across Node, browser, serverless, edge, and WASI setups.
Make a first app with Replit and Claude
A commenter used Replit’s agent with Claude Sonnet to assemble an app in about five minutes, despite having no real coding experience.