Cómo usar
- Copia el comando y revisa el paquete antes de instalarlo en tu entorno de agente.
- Consulta las herramientas, credenciales y entornos necesarios en el documento original.
- Describe las entradas y el resultado deseado; contrasta la salida con las fuentes.
npx skills add https://github.com/atmanity/atmee-toolkit --skill plan-avatar-appAntes de empezar
- La instalación no incluye suscripción al modelo, cuota API ni permisos sobre tus archivos.
- Las estrellas e instalaciones reflejan actividad, no una prueba independiente de calidad o seguridad.
Contenido del repositorio de origen con títulos, ejemplos, código, tablas, enlaces e imágenes preservados.
Plan an app around an Atmee avatar
You turn a rough idea ("a pharmacy counter assistant") into a brief: one page that says who the app is for, the one job it does, how the conversation flows, who the avatar is, what it does through tools and what the page shows, which API limits bite, and how the app will be proven live. Nothing is built here; create-avatar and build-avatar-app build from the brief.
This is the Session API path: Atmee runs the conversation, the developer's app surrounds it. While the person is still asking what Atmee is, what it costs or which path fits (or wants only a face for their own voice agent), use atmee-product first.
The grill
Once per session, first run the update check in the atmee-setup skill (its step 0), and go on whatever it says.
Ask in rounds until every line of the brief is settled. The question bank is references/questions.md, ordered by round.
- Number the questions and give your recommended answer with each, so
agreeing settles it. Ask through your question tool when you have one, the way create-avatar's "How to ask the creator" says; without one, a numbered list with a ➡️ recommendation under each.
- Ask only the frontier: two to four questions whose prerequisites are
settled. Tools wait until the job and the avatar's role are known.
- Look facts up instead of asking: the limits (
build-avatar-app,
"Current limits"), the plan's session length and price (get_account), the avatars they already have (list_avatars), their repo and stack.
- Meet a limit head on: when an answer runs into one, name it and offer
the workaround, or write it down as an accepted gap. A brief that hides a limit fails in the first live run.
- Anything never asked keeps your recommendation; list those defaults with
the brief, so the person sees what they accepted.
Inspiration
When you can reach them (they are not public yet), the Atmee demo apps, https://github.com/Atmanity/atmee-ai-demos, are nineteen apps on the Avatar API: a recruiter on an agenda, a trainer with a quiz, a companion that watches the screen, a game narrator, a livestream host, and fourteen tutors, advisors and games under apps/. Each apps/<app>/README.md gives its use case, architecture, how it uses the API and the limits it hit; docs/PORTING.md and docs/API-GAPS.md hold what the live runs taught. Read them for ideas and patterns that worked, as inspiration: the brief fits this person's app, which may look like none of them.
The brief
# <Product name>: <one line>
User and job: who opens it, where, and the one thing they leave with
Flow: front door -> ... -> the done moment; with an agenda, its steps,
each with the condition (about what was said) that ends it
Avatar: role, name, persona in two lines, greeting, language,
guardrails; a new avatar, or the id of one that fits
Tools: | name | notify or await | when the avatar calls it | what the page does or returns |
Screen vs speech: what the page shows as the truth (scores, prices, verdicts,
safety lines) and what the avatar only says
App to avatar: the state (what is true now), notes, events (what happened), when each is sent
Transcript: a chat panel or captions, one of them; the user's own words in it
Sessions: one per visitor, or one host for many watching; how many
avatars; the length cap; who pays
Limits: each one that bites, with its workaround or "accepted gap"
Live test: the run that proves it: what to say or type, the tool calls
that must fire, the end state, the session ended
Deploy: where it runs, how it is gated, the key's scope and spend capDone when every line holds an answer the person gave or accepted and the Limits line has walked every item of build-avatar-app's "Design around these" that the app touches. Show the brief with the defaults, ask "Build from this brief?", and on a yes save it in their repo (for example docs/brief.md).
Then
- `create-avatar`, with the Avatar line as its input. The line's name
and language start step 1 (the person still brings a photo and a voice, or generates them), and the rest becomes step 3's character. Make a purpose-built avatar: it greets and introduces itself as it was made, and its persona outweighs tool instructions (a chess player refuses to show slides). Pick a seed whose mouth reads clearly: a seed is one captured talking clip (the server's "candidate" and "take" mean a seed, and the API's seed field is an informational number, not the clip).
- `build-avatar-app`, from the brief. Its "Prove it live" runs the
brief's Live test line and its "Document and ship" the Deploy line; a platform gap found on the way goes back into the brief's Limits.

