What to Tell Claude Code to Test (and What to Skip)

If you're using Claude Code to build apps, you've probably noticed it loves writing tests. Ask it to build a feature and it'll offer to test it. Ask it to fix a bug and it'll suggest adding coverage. Left to its own devices, it will happily generate hundreds of tests for your application. Most of them won't matter. Here's the filter I use before writing any test: if this breaks silently, what happens? If the answer is "nothing for a while" or "a user gets a slightly wrong result," skip it. If the answer is "data gets corrupted," "someone gets charged incorrectly," or "the failure is invisible until it's too late," write the test. That's the whole framework. Everything below is just applying it. One important caveat: this is a strategy for the build phase. If you're shipping an MVP or a side project and you're the only person working on it, this is the right approach. Once you have a team, long-lived systems, and daily deploys, your testing philosophy needs to evolve. But that's a diffe