Dev.to AI 🤖 Ai 👁 0 📖 1 min read

sub-agents ship features that work individually but still break your app

There's a gap in how we test software built by multiple coding agents. Each agent gets assigned a feature. It builds it, runs its tests, and reports that everything works. But those tests usually stop at the boundaries

There's a gap in how we test software built by multiple coding agents.

Each agent gets assigned a feature. It builds it, runs its tests, and reports that everything works. But those tests usually stop at the boundaries of that feature.

The problem is that customers don't use features in isolation. They move between pages, combine workflows, and expect everything to work together.

That's where things break.

A checkout can work. Pricing can work. But when a customer actually checks out, the wrong price gets displayed. Both agents might have passed their tests, yet the application is still broken.

I built Kintsu to catch exactly these kinds of problems.

Kintsu uses Kane CLI to run integration tests in a real Chrome browser. It checks how features behave together, collects screenshots and logs when something fails, and identifies which coding agent is responsible for the problem.

It doesn't stop at reporting bugs either. Kintsu sends the responsible agent a repair ticket, gives it the failure evidence, and reruns the same test after the fix. It retries up to three times before handing the issue to a human.

There's also a live graph showing which agents and features are connected, and whether their integration tests have passed or failed.

The idea behind Kintsu is straightforward:

Testing whether individual features work isn't enough. You need to test whether the application actually works when those features come together.

And when something breaks, the system should know where to send the problem.

GitHub: https://github.com/fozagtx/kintsu. https://www.testmuai.com/register/

📰 Read the original article on Dev.to AI

Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.