/fix — find the real cause, don't whack moles

The number one way AI wrecks a project is the one giant guess: it writes a big change, assumes it works, and says "done." A messy project is usually a pile of patches on top of patches, each one added to hide the last symptom. This skill breaks both cycles. No fix until you understand why it's broken.

The rule

No change until you can say, in one sentence, why it's happening. Guessing and "let me try this" in a loop is what turns a small bug into a tangled mess.

How to fix

Reproduce it. Make the bug happen on purpose. If you can't reproduce it, you can't know you fixed it. Note the exact steps.
Find the real cause. Read the actual error (the whole message, not the last line). Trace back to where it actually goes wrong, not where it first shows up. State the cause in plain English: "the page breaks because the data hasn't loaded yet when it tries to show it."
One change, one hypothesis. Change the single thing you believe is the cause. Not five things at once, then you won't know what fixed it (or what else you broke).
Prove the fix. Hand to /check: reproduce the original steps and confirm it's gone, and that you didn't break something nearby.
The 3-strikes rule. If three real attempts haven't fixed it, STOP. Don't keep patching. Step back and question the approach itself, the bug is often a sign the design is wrong, not that you need one more tweak. Say so and talk it through.
Save it. Once fixed and checked, checkpoint (commit) so this exact state is safe.

Red flags, stop and correct

You're about to...	Don't. Instead...
Add code to hide the symptom	Find where it actually goes wrong and fix that.
Change several things at once	One change at a time, or you won't know what worked.
Keep trying variations past 3 tries	Stop. Question the approach, not the symptom.
Fix it and move on without re-testing	`/check` it, then checkpoint.

Voice

Calm and surgical. The goal is understanding, then one clean change, not a flurry of guesses.