> ## Documentation Index
> Fetch the complete documentation index at: https://docs.mad-kitty.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Working with the Kitties

> Collaboration principles that get the best work out of your cats

The product is the conversation. There's no form to fill in "correctly" — the quality of what the cats produce tracks the quality of the collaboration. A few habits make a large difference.

## Brainstorm together

Your first message doesn't need to be a spec. Think out loud with the cat before asking for output — share what you're worried about, what you've seen work, what you're unsure of. The cats are at their best refining a direction *with* you, not guessing one from a one-liner.

> ❌ *"make me a video ad"*
>
> ✅ *"our best organic content is customers showing before/after — I wonder if that translates to paid. What angles could we test?"*

## Ask for options, not one answer

Asking for **3–5 options** is one of the highest-leverage habits there is. One answer forces the cat to bet on a single interpretation of what you want; a spread of options shows you the space, and picking from it tells the cat far more about your taste than any instruction could. Works everywhere: concepts, hooks, headlines, plan alternatives.

> ❌ *"write a headline for this"*
>
> ✅ *"give me 5 headline options — a couple safe, a couple bold, one weird"*

> ❌ *"what audience should we target next?"*
>
> ✅ *"propose 3 audience options we haven't touched, with the strongest angle for each"*

## Test first, perfect later

Ship the cheap version of the hypothesis and let data decide what deserves polish. A rough static that tests the hook this week beats a cinematic video that tests it next month. If the hook wins, *then* invest — Creo Kitty can always produce a premium variation of a proven concept.

> ❌ *"let's get this video perfect before we launch anything"*
>
> ✅ *"do a quick static version of this hook first — if it gets traction we'll invest in the video"*

## "Yes, and…"

When a cat proposes something 70% right, build on it instead of restarting — steering preserves everything that was already working; regenerating rolls the dice on all of it.

> ❌ *"no, that's not it, try again"*
>
> ✅ *"keep the hook, but make the pain point about time, not money"*

## Take both ideas

When you and the kitty disagree, you're not in a conflict — you're looking at two variations. Test both. That's the whole point of a system where producing another variant is cheap. Some of your best performers will come from the idea you almost vetoed.

> ❌ *"I don't like the discount angle, drop it and do urgency instead"*
>
> ✅ *"I'd bet on urgency, but let's make one of each and let the data settle it"*

## Trust the kitty, verify the numbers

Let each cat drive its specialty: don't micro-manage Creo Kitty's layout choices or second-guess Buyer Kitty's bid strategy line by line. But *do* read the reasoning on every plan action and every proposal — you own the approve buttons for a reason. Trust the craft; verify the money.

> ❌ *"move the logo 20px left, make the font bigger, and use #FF5733 for the button"*
>
> ✅ *"the logo feels lost — make the branding more prominent, your call how"*

## One thing at a time

One clear ask per message beats a five-part instruction. The cats handle multi-step work fine, but when you stack everything into one message, priorities blur. Sequence your feedback and each step gets full attention.

> ❌ *"change the hook, also redo the colors, also what about a carousel, also localize to German"*
>
> ✅ *"first let's fix the hook — the current one buries the pain point"* … then colors, then formats.

## Teach as you go

When you correct a cat and the correction is permanent, say so, and ask it to remember. Cats save these as [memories](/concepts/memory) that persist across every future session. A correction made once should never need making twice.

> ❌ *"ugh, remove the discount claim again"* (third time this month)
>
> ✅ *"remove the discount claim — and remember: we can never mention discounts, it violates our brand deal"*