Academies Skills Workshop 🔨 Makers · Module 02

Skills that speak differently.

a seven-step walkthrough · audience awareness · about twenty-five minutes

⚠ The temptation

Pick a "voice" from a list and apply it.

But voices aren't templates — they're a specific person you're imitating. Pick one human you actually know whose voice you can hear, and write the Skill in their voice — not in "a friendly voice."

Step 1 of 7
Step 01 · The problem

The same fact can be right or wrong depending on who's listening.

Your Sprouts Skill has facts. Facts are true no matter who's reading. But the way you say a fact changes everything — and your Skill sounds ridiculous if it uses the same voice for your teacher and your best friend.

Wrong for the teacher

To the school principal

"Yo, Buddy literally ate my homework, I'm not even kidding, he's done it like three times this year. 🤦"

✓ Right for the teacher

To the school principal

"Unfortunately, my dog Buddy chewed up my homework last night — it's not the first time this has happened. I can bring the remains if you'd like to see them."

Same fact. Same dog. Same chewed homework. Different human on the other end — so the sentence has to change. That's what this module is about.

A Skill with only one voice is like a piece of music that only plays in one key. It works, but it's rigid. Real Skills — the ones you'll keep using for years — have voice profiles built in.

Step 02 · Pick the raw material

Pick one real fact you could work with.

A base fact is the underlying truth — the thing that doesn't change no matter who's listening. We're going to take one base fact and rewrite it three different ways.

🐾 A fact about my pet
👨‍👩‍👧 About someone in my family
🎨 About my hobby
🏫 About something at school
📍 About a place I love
🕐 About my routine
Got it — a fact about . You'll get to write it in step 5.

Pick the domain your fact belongs in, not the specific fact yet. You'll write the fact itself later.

Step 03 · Watch the voice change

Same fact. Three voices.

Click the audience buttons below and watch the sentence change. Same underlying truth — but the voice shifts completely. Try all three on each switcher.

Base fact

My dog Buddy has a small scar on his left ear from running into a doorframe when he was a puppy.

→ said to Mom

You know that scar on Buddy's ear? He still has it from the door incident. He'll never learn, that dog.

Base fact

My grandmother adds a small piece of orange peel to almost every soup she makes.

→ said to Mom

Did you notice Grandma's still putting orange peel in everything? I watched her do it yesterday. She thinks we don't notice.

Base fact

I can land a kickflip about 1 in every 8 tries. It took me 4 months to learn.

→ said to Mom

I've been working on kickflips for four months. I can land them about one out of eight times now — I know it doesn't sound like a lot but that's real progress for me.

Look at what changed between the three voices. The facts didn't — Buddy still has the scar, Grandma still uses orange peel, the kickflip success rate is still 1-in-8. What changed was tone, slang, level of detail, and what's taken for granted. That's a voice profile.

Step 04 · The actual code

This is how a Skill stores voice profiles.

Here's a real Skill about Grandma's cooking — but with a new section: voice_profiles. Each profile teaches Claude how to adapt the facts for a specific kind of listener.

grandmas-cooking.skill.yaml
# A Skill about Grandma's cooking — with voice profiles

name: "Grandma's Cooking"
description: "How my grandmother actually cooks, adapted to the listener"

facts:
  - "Uses cold water for dough, never warm"
  - "Adds a small piece of orange peel to most soups"
  - "Refuses store-bought wrappers"
  - "Three pinches of salt, not four"

voice_profiles:

  default:
    tone: "warm, observational, quietly proud"
    vocabulary: "plain, no slang, no jargon"

  to_family:
    tone: "familiar, slightly teasing, uses shared references"
    vocabulary: "casual, assumes the listener has eaten Grandma's food before"
    example: "Did you notice she's STILL putting orange peel in everything?"

  to_teacher:
    tone: "descriptive, observational, slightly formal"
    vocabulary: "explanatory, assumes the listener has never met Grandma"
    example: "My grandmother has a distinctive cooking signature — she adds a small piece of orange peel to most soups."

  to_friend:
    tone: "enthusiastic, exaggerated, playful"
    vocabulary: "slang okay, emphasis words okay, can use 'literally' and 'obsessed'"
    example: "My grandma is literally obsessed with putting orange peel in every soup."

when_to_use: "Any question about Grandma's cooking — pick the voice profile based on who's asking"

The facts don't duplicate — they're written once. The voice profiles only describe how to say them, not what to say. That's the whole trick. You can add more profiles later (to_sibling, to_neighbor, to_someone_who_cooks_professionally) without rewriting a single fact.

Step 05 · Your turn

Write your base fact — then adapt it.

One real fact about your . Then three versions — one for Mom, one for a teacher, one for a friend. As you type, the full YAML builds itself underneath.

my-voice-skill.yaml
# Fill in the form above and watch this update.
name: ...
facts:
  - ...
voice_profiles:
  ...

If two of your voices sound almost the same, that's useful information — it means you haven't noticed a difference yet. Try harder to imagine the different listeners.

Step 06 · Spot the mismatch

Can you tell which voice belongs to which audience?

Two rounds. In each round, both sentences say the same fact — but only one fits the listener. Pick the right one, then tap again to see why.

You're writing a scholarship essay about a family member. Which version belongs?

You're texting your closest friend about something your brother did. Which version fits?

The "correct" voice isn't about being formal or casual in general — it's about matching the actual person you're talking to. A Skill that can do this automatically saves you from re-explaining your family to every new audience. It also teaches Claude something rare: that voice itself is a form of respect.

Step 07 · You did it
🔊

You just taught your Skill to have voices.

It now knows there's a difference between a grandma you're talking about and a grandma you're talking to. And between a friend and a principal. That's audience awareness.

What you just learned

  • Facts don't change, but the way you say them does.
  • A voice profile is a set of rules for how to say a fact to a specific kind of listener.
  • You don't duplicate the facts — you only describe the tone and vocabulary for each audience.
  • The "right" voice isn't about formal vs casual. It's about matching the actual person.
  • Voice itself is a form of respect — and now your Skill knows that.

In Module 03, you'll learn the last Makers pattern: Skills that remember what happened last time. A Skill that knows you already asked about Grandma's dough yesterday is a Skill that doesn't waste your time.

★ Before you call it done

Three questions. Same three. Every time.

These are the same three questions for every module in Kindling. They are how you check whether AI did the part it should and you did the part only you could. Tap each one to mark it true.

★ ★ ★

This is yours. Ship it.