Test Runner Wrapper for Claude Code

Test runners burn through massive amounts of context. Hundreds of lines of output just to say: all green, all good. That’s wasting valuable tokens in your context window.

So I built a wrapper around my test runners. On success, it outputs only a single line. When a test fails, it extracts the relevant information and outputs only what’s actually relevant for the LLM.

I noticed this was a problem months ago. But then I forgot to implement a fix and stuck with my old habits. Now my CLAUDE.md specifies that only the wrapper should be used for running tests. My context window gets spammed significantly less.