break-fix

Installation
SKILL.md

Break-fix — adversarial e2e bug hunt, then root-cause and fix

Two jobs, in order. Break it: drive the running app until it does something it shouldn't. Fix it: for each real defect, find the actual cause, encode it as a failing e2e test, fix it, and watch the test go green.

The regression test is not paperwork at the end — it's how you prove the bug was real and that the fix addresses it. A test written after a fix, that has never once failed, guards nothing.

Why polite testing finds nothing

The instinct when handed an app is to use it correctly: fill the form with plausible data, click the buttons in order, admire the success toast. That path is the one the developers walked a hundred times while building it. It is already green.

Bugs live where the implied contract of the UI is violated — where a user did something the code never imagined. So your job is to be a bad user, deliberately and systematically. Not random flailing (that finds little and wastes tokens), but targeted violation of the assumptions you can see the code making.

Two things make a hunt productive rather than performative:

  • Read the code first, then attack what it assumes. A hunt informed by five minutes in the validation layer finds ten times what blind clicking finds. When you see parseInt(req.query.page) with no guard, you now know exactly what to type in the URL. When a form validates on the client, go around it.
  • Look where the user can't see. A screen that renders fine while the console throws Unhandled promise rejection and the network tab shows a 500 is a bug that shipped. Most of what you'll find lives in the console, the network panel, and the server's stderr — not the viewport.

Rules of engagement

Installs
10
Repository
alamops/skills
GitHub Stars
53
First Seen
Aug 28, 2026