test-loop
Test loop — run everything, then drive it to green
Two jobs, in order. Run: execute the whole suite — every layer, e2e included — and report exactly what passed and failed. Fix: when something failed, find out why, fix the cause, and run everything again until it's green or you hit a wall that needs a human.
The rule that makes this loop worth trusting: green has to mean the software works. A suite made green by skipping the failing test, deleting it, or loosening its assertion until it matches the wrong output is worse than a red one — it's a red suite that lies.
Where this sits
This is Phases 7 and 8 of the implement delivery loop, packaged to run on its own. It typically follows write-tests, and it also accepts must-fix findings from a code review (e.g. the code-review skill) as items to fix before the final green run.
1. Run-only or to-green?
Read what was asked:
- "Run the tests" / "what's failing?" → run and report, then offer the fix loop. Nobody asked you to edit code yet.
- "Get it green" / "fix the failing tests" / "CI is red, sort it out" → run, then loop.