Teach your AI your taste.

Better and faster execution used to separate good work from bad. Everyone has the same AI now, like every chef having the same knives in the same kitchen, so what's left is judgment: deciding what goes on the menu and sending back the plates that aren't good enough.

AI can't give you taste, and it will happily tell you that what you just made looks good. This skill makes it judge your work against your own standard instead.

STEP ONE

Build your taste bank.

Taste comes from looking a little closer and keeping a record of what looks good to you. The taste bank is that record, written down so your AI can read it.

Pick one kind of work.
Captions, proposals, product pages or emails. One taste bank per kind of work, because good looks different in each.
Collect ten you love.
Yours or anyone's. Under each one, write a single line on why it's good.
Collect five you'd send back.
The lines on why these fail are often more useful than the ones on why the good ones work.

The taste bank template.

# My taste bank: [the kind of work, e.g. Instagram captions]

## Work I love
1. [paste the example, or a link to it]
   Why: [one line on what makes it good]
2.
3.
(keep going to ten)

## Work I'd send back
1. [paste the example, or a link to it]
   Why: [one line on what's wrong with it]
2.
3.
(keep going to five)

## Rules I've noticed
[Once both lists are full, write down the patterns you see. "Short sentences
win." "No exclamation marks." "Always one real number."]

STEP TWO

The skill reads your bank, scores a draft against it, rewrites it and shows you where every change came from. It isn't allowed to call anything good without quoting your bank as proof.

The Taste Check skill.

---
name: taste-check
description: Judges a draft against my taste bank before I see it. Use whenever I ask you to write, review or improve a piece of work that has a taste bank, or when I say "taste check this".
---

# Taste check

My taste bank is the file taste-bank.md. It holds work I love, work I'd send
back and the rules I've noticed. It is the only standard you judge by. Your own
opinion of what is good does not count here.

## Every time you run

1. Read taste-bank.md in full before you look at the draft.
2. Score the draft from 1 to 10 against the bank. For every point you take off,
   quote the line in the draft and name the example or rule in the bank it
   falls short of.
3. List what the best examples in the bank do that this draft doesn't.
4. Rewrite the draft to close those gaps. Keep my facts, names and numbers.
5. Score the rewrite the same way. If it scores under 8, go round once more.
6. Show me the final draft, its score and the three biggest changes you made,
   each with the line from the bank it came from.

## Rules

Never call a draft good, strong or great without quoting the part of the bank
that proves it. If the bank doesn't cover something, say so and ask me instead
of guessing what I'd like. When I reject a draft, ask me for one line on why
and offer to add it to the bank, so the bank gets sharper every time I use it.

HOW TO USE IT

Install it in three steps.

Save the two files.
Save the skill as SKILL.md inside a folder called taste-check, and your filled-in bank as taste-bank.md.
Add the skill to your tool.
In the Claude app, upload the taste-check folder as a zip under Settings, then Capabilities, then Skills. In Claude Code, put the folder in .claude/skills/ inside your project. In ChatGPT, paste the skill into a Project's instructions.
Give it the bank and a draft.
Attach taste-bank.md, or keep it in the same project, then say "taste check this" with your draft.

WHY IT MATTERS

Your taste decides what gets sent.

Over RM1 million in new business has closed on proposals written by Sean's AI systems, and plenty of what those systems produced along the way got rejected. The rejecting is the part that made the rest worth sending.

See what we do

An AI that agrees with everything makes this harder. The AGREE Prompt makes it argue with you instead.