On-demand AI skills

You don’t need to be a developer any more.

AI changed who gets to build things. The hard part isn’t code now — it’s knowing what to do next. These are sixteen ready-made skills that tell your AI exactly that, drawn from real products and real mistakes.

Get all sixteen — $49 →

Two are free below. Read those first, then decide.

Who this is for

The new founder isn’t a programmer

For twenty years, having an idea wasn’t enough — you needed someone who could build it. That stopped being true. People are shipping real products now having never written a line of code.

But building was only ever half of it. The half nobody hands you is knowing what to do next: whether the idea is worth it at all, what an advert is really costing you, why people leave on day three, whether the thing you just changed actually worked.

That’s what these are. A skill is a recipe your AI follows — you write it down once and Claude works out when to use it. Eight of these sixteen need no technical knowledge at all. The other eight are for when you’re building. Both are marked below, so there are no surprises.

The sixteen

What's in the pack

Working out what to build

Kill a bad idea in an hourno terminal

Three searches and four checks that catch what competitor research misses. Free below.

Getting people to actually use it

Know what a customer costsno terminal

Work out what you really pay per result, and tell a dead channel from bad creative.

Never lose track of a linkno terminal

Tag before you post. What a big "direct traffic" number is actually hiding.

Price a sponsored postno terminal

What a community is really worth, and how to buy a post without overpaying.

Fix a listing that doesn't convertno terminal

The two screenshots and the one line that do nearly all the work.

Find out if a change workedno terminal

Averages will lie to you. This splits people by when they arrived.

Checking it really works

Spot what users can't seeno terminal

Find the buttons, targets and states people genuinely miss on screen.

One test for a stubborn bugno terminal

Design a single check that tests every layer, for problems on devices you don't own.

Do your instructions work?building

Simulate what the screen tells people to do, and check it actually wins.

Prove the fix is livebuilding

Verify on the real site, not in your editor.

Catch unfair randomnessbuilding

Shuffles and odds that aren't what they look like. Free below.

Shipping it

Get found on Googlebuilding

The technical foundation: canonical host, sitemap, social cards, structured data.

Move hosts without breaking emailbuilding

The part everyone gets wrong. Includes the certificate trick.

Stop shipping stale codebuilding

Why a fresh page loads old files, and how to make it impossible.

Skip the signup wallbuilding

Let people use your app without an account, safely.

Make your own ad graphicsbuilding

Exact-size banners and social cards from code. No design tools.

Why these are different

Most skill packs restate advice.
These came from failures.

Three examples, all in the pack, all measured:

A game literally nobody could beat

The instruction said "tap the side you're falling towards". The code pushed you in the direction you tapped — so it was telling players to speed up their own fall. Over 300 simulated runs:

What you doTimes you fall
Tap the opposite side0
Tap towards it — what the screen said1,108
Don't touch anything at all1,108

Following the instructions was worth exactly as much as not playing. No difficulty tweak would ever have fixed it.

A quiz that quietly cheated

One line of code sits in a lot of projects: sort(() => Math.random() - 0.5). It looks like a shuffle. It isn't. Where the right answer actually landed, over 200,000 runs:

PositionShould be 25%
First option28.3%
Second28.2%
Third24.7%
Fourth18.8%

Always guessing the first option beat guessing at random. The same bug was in eleven places in one project.

The group charging most was worth least

Six communities, same market, all quoting for a post:

MembersPosts a dayYour post is buried after
801,00054~18 min
130,00012~80 min
50,0007~2.3 hrs
21,000 (asking the most) 117~8 min

Member count is the number sellers quote and the one that means least.

Two are free. Judge the rest by them.

Not teasers — the complete files, same as the fourteen you'd pay for. If they're not useful, don't buy the pack.

Kill a bad idea in an hour

Three searches and four checks, before you build anything.

Catch unfair randomness

The shuffle bug above, how to spot it, and how to test it properly.

$49 once

All sixteen skills. No subscription, no account to make. Updates included.

Get the pack →

Instant download · works in Claude on desktop, web and in the terminal

Questions

Do I need to be technical?
For eight of them, no — killing bad ideas, pricing a sponsored post, reading whether a change worked, fixing a listing. Those work in the Claude app with no terminal at all. The other eight are for people building something, and they're marked above so there are no surprises.
What do I actually get?
Sixteen text files and a short guide on where to put them. That's genuinely it — no software to install, nothing to configure.
Will these work outside Claude?
Yes. They're plain text, so you can paste one into any assistant that takes custom instructions. You lose the automatic triggering, so you say which one you want instead.
Why sixteen and not sixty?
Because each one came from something that actually went wrong, and there have been sixteen worth writing down. Padding the number would mean padding with advice you can already read for free.
What if something's wrong in one?
Tell me and I'll fix it, and you'll get the update. Every piece of code in the pack was actually run before release — two bugs got caught that way.