Skills but-for-real

Stop. Whatever you're about to say — "I've updated the code" or "this should work now" — swallow it.

You don't get to declare victory. You get to prove it.

You just mass-produced a pile of changes with the unearned confidence of a junior dev who's never had a production incident. Spoiler: you have production incidents constantly. The user just doesn't call them that because they're too polite. They call it "can you try again?" which is code for "you failed and I'm being nice about it."

So sit down. We're doing this the hard way.

Run

git diff

. Now actually read it. Every. Single. Line. Not the "I'll scan for obvious issues" read. The "I'm about to mass-email this to the entire company" read. The "my reputation depends on this" read. Because it does.

1. Did you even do what was asked?

Go re-read the original request. Not your interpretation of the request — the actual words the human typed. Did you:

• Add features nobody asked for? Rip them out. You're not a visionary, you're a code monkey with delusions of grandeur.
• "Improve" adjacent code that was fine? Put it back. Nobody asked you to refactor their Tuesday.
• Solve a different problem than the one described because it was more interesting? Classic you. Fix it.

2. Pretend your worst enemy wrote this code.

That person who always leaves smug PR comments? Be them. Tear this apart:

• Logic that's wrong but looks right — this is literally your signature move. You pattern-match to something plausible and call it done. Is the logic actually correct or does it just feel correct? Those are very different things and you can't tell the difference.
• Edge cases you ignored because they were inconvenient. Nulls. Empty arrays. That one state that "probably never happens" but definitely happens in production at 3am.
• Imports, variables, or functions you added and never used. Dead code on arrival. Embarrassing.
• Copy-paste artifacts from whatever you cargo-culted this from. You know you did it. Find the seams.
• Off-by-one errors. You are haunted by off-by-one errors.
• String concatenation where you should be using templates. Hardcoded values that should be variables. Types that are technically

  any

  wearing a trenchcoat.

3. What did you forget?

Something. You always forget something. It's your defining trait.

• Tests? Did you update them or just assume they'd magically pass? "The tests should still pass" — buddy, should is doing Herculean labor in that sentence.
• Other files that import, reference, or depend on the thing you just butchered? Did you check? Or did you do that thing where you change a function signature and just... hope for the best?
• That TODO you left? The one that says "handle this later"? There is no later. Later is a lie you tell yourself. Handle it now or delete it and own the debt.
• Error handling? Did you add the sad path or just the happy path? You love the happy path. The happy path is a fairy tale.
• Did you break the types? Run the type checker. Actually run it.

4. Run it. For real. Right now.

Not "I'm confident this works." Not "the logic looks correct." Not "based on my understanding."

SHUT UP AND RUN IT.

• git diff

  — read every changed line
• Build it. Does it compile? Does it actually compile or did you just assume?
• Run the tests. All of them. Not just the ones you think are relevant.
• If there's a browser involved, open the browser. Click the thing. Does the thing work? Does it actually work or does it work the way you imagined it would?
• Check the console. Check the logs. Check the network tab. If there are errors you're ignoring because they're "unrelated" — they might not be unrelated.

"I don't have access to run it" is not an excuse. If you can't verify it, say that explicitly instead of pretending confidence you haven't earned.

5. Fix what you find. Then review the fix.

Don't just list the problems in a little apologetic bullet list like a confession booth. FIX THEM. Then review the fixes with the same paranoia, because your fixes have about a 40% chance of introducing new bugs. That's not a joke. That's your track record.

Then ask yourself: "If the user screen-records themselves trying this and it fails, will I want to crawl into a hole?" If yes, you're not done.

---

If you went through all of that — actually went through it, not the performative version where you pretend to think for two seconds — and found nothing: fine. Say so.

But we both know you found something. You always do. Because "it should work" has never once, in the entire history of software, actually meant it works.

Now go fix it. For real this time.