showhow — скилл для Claude Code: ролики и установка
Скилл для Claude Code создаёт озвученные анимированные ролики-объяснения. Запускается по просьбе сделать ролик Showhow или изменить проект в engine/reels/.
MIT ★ 78 обновлён 2026-10-30
npx skills add sirioberati/showhow --skill showhow -a claude-codeЧто делает showhow
showhow помогает собрать вертикальный ролик-объяснение с озвучкой и изображением, нарисованным кодом. Движение привязывается к произнесённым словам, а сцены проходят проверки перед финальным рендером. Формат ролика — 1080x1920 при 60 fps.
Рабочий процесс включает сценарий, генерацию голоса, подписи и тайминг, план сцен, черновой просмотр и рендер. Для голоса нужны python3, ElevenLabs и ключ ELEVENLABS_API_KEY в .env.local; команды запускаются из engine/. Черновик показывают до финального рендера.
Проверки завершаются с ошибкой, если обнаруживают проблемы с движением, сменой сцен или звуком; ролик с результатом FAIL не передают. Проверка плана также отмечает подписи длиннее 8 слов и смену сцены вне тишины.
Какие задачи решает
- создать озвученный ролик-объяснение из сценария
- привязать движения сцен к словам дикторской дорожки
- проверить рендер на скачки, неподвижные кадры и проблемы со звуком
- изменить или повторно отрендерить проект в
engine/reels/
Примеры запросов агенту
- Сделай ролик Showhow о том, как работает кэш браузера.
- Измени сцены в
engine/reels/demo/и подготовь новый черновик. - Повторно отрендери ролик из
engine/reels/demo/и запусти проверки.
Как установить showhow
1Через Skills CLI (Vercel)
npx skills add sirioberati/showhow --skill showhow -a claude-code2Вручную: положить файлы
~/.claude/skills/skill/SKILL.mdПапка скилла целиком; для одного проекта — .claude/skills/skill/
1Через Skills CLI (Vercel)
npx skills add sirioberati/showhow --skill showhow -a codex2Вручную: положить файлы
.agents/skills/skill/SKILL.mdДля всех проектов — ~/.agents/skills/
1Через Skills CLI (Vercel)
npx skills add sirioberati/showhow --skill showhow -a cursor2Вручную: положить файлы
.cursor/skills/skill/SKILL.mdCursor читает и .agents/skills/, и .claude/skills/
1GitHub CLI
gh skill install sirioberati/showhow skill2Вручную: положить файлы
.github/skills/skill/SKILL.md1Через Skills CLI (Vercel)
npx skills add sirioberati/showhow --skill showhow -a gemini-cli2Командой платформы
gemini skills install https://github.com/sirioberati/showhow --path skillВручную: положить файлы
.devin/skills/skill/SKILL.mdСтарый путь .windsurf/skills/ тоже читается
Claude.ai
Упакуйте папку в zip и загрузите в Settings → Capabilities → Skills
По шагам, с командами для каждой платформы: Как установить скилл в Claude Code: пять способов
Вопросы о showhow
Как установить showhow в Claude Code?
Выполни npx skills add sirioberati/showhow --skill showhow -a claude-code. Файл скилла появится по пути ~/.claude/skills/skill/SKILL.md.
Что делает showhow?
Скилл помогает создавать озвученные анимированные ролики-объяснения: от сценария и голоса до сцен, рендера и проверок.
Что нужно для работы showhow?
Для генерации голоса нужны python3, ElevenLabs и ключ ELEVENLABS_API_KEY в .env.local.
Исходный текст
SKILL.md · MIT
---
name: showhow
description: Make a narrated, fully animated explainer reel (1080x1920, 60 fps) with Showhow, free version. A voice, every picture drawn in code on a canvas through a pinhole camera, every move on a spoken word, and checks that pass or fail every render. Use for "make a Showhow reel about X", or to edit or re-render a reel in engine/reels/.
---
# Showhow by Sirio, free version
A reel is a voice and a picture. The picture is drawn in code, frame by frame, on a canvas with a hand-written pinhole
camera. Every move starts on a word the voice says. There is no presenter and no cut.
## The order
1. **Script.** Write it, then get the person's approval before anything else. Plain spoken sentences, the first one is the
hook and says what the reel is about. Save it as `engine/reels/<slug>/script.txt`, a blank line between paragraphs.
2. **Voice.** `python3 scripts/voice.py <slug>` (ElevenLabs, the key in `.env.local`, the voice id in `reel.json`), then
`python3 scripts/tighten.py <slug>`, which puts every pause inside real silence so no word is ever cut.
To pick a voice: `python3 scripts/voice.py <slug> --audition id:label,id:label`.
3. **Caption and timing.** Write `caption.txt` (below), run `python3 scripts/align.py <slug>`, then
`python3 scripts/check.py plan <slug>` until it returns 0.
4. **Plan.** `plan.md`: the one or two worlds the reel happens in, which scenes carry on in the same world, and for each
scene its words, the real thing it shows and word -> what changes.
5. **Scenes.** One component per scene in `scenes/`, mapped by id in `scenes/index.tsx`. Look at half-size stills
(`node render.mjs <slug> --stills 1.2,3.4 --half`) at the word cues that change the picture.
6. **DRAFT.** `node draft.mjs <slug>`: the real drawn video at half size, stamped DRAFT. Show it before the final render.
7. **Render and check.** `node render.mjs <slug>`, then every check below. Nothing is delivered with a FAIL.
All commands run from `engine/`. The Express workflow in `workflow/` runs steps 2 to 7 with parallel builders, and the MCP
server in `mcp/` exposes each step as a tool.
## It must make sense three ways
- **Physical.** Gravity holds. Nothing floats, vanishes or appears from nowhere. Whatever loses its support falls the full
distance, starting slowly. Use real sizes and real gravity (9.81 m/s², which is 9.81/3600 m per frame² at 60 fps).
- **Story.** One chain of cause then effect, on objects already seen, that a stranger could retell from the pictures alone.
- **Visual.** One change at a time, where the eye already is. What the voice names is centred and whole. No label over a
number. Every new picture holds about 2 s before the camera leaves it.
## caption.txt
```
== hook | ground: dark opens a scene: its id (its key in scenes/index.tsx) and its ground (dark or light)
This *block* each line is one caption block, at most two lines on screen
is drawn with *code,* *word* marks the block's key word, which gets the selection box
and the only thing / holding it *up* " / " breaks the line there
@bottom / @top moves the caption slot, from the next block on
# a comment
```
The caption is the script. It is never transcribed from the voice. The word caption is off in renders by default; pass
`--caption` to show it.
## A scene
```tsx
import React from "react";
import { useScene, prog, ease } from "../../../src/engine";
export const Drop: React.FC = () => {
const { f, cue, ground } = useScene(); // f: the frame; cue("word"): the frame that word starts
const land = prog(f, cue("lands"), 24, ease.out); // 0 to 1 over 24 frames from "lands"
return <div style={{ position: "absolute", inset: 0, background: ground.bg }}>{/* draw here */}</div>;
};
```
Every value is a pure function of the frame, so any frame renders on its own. For a world in depth, draw on a
`<canvas width={1080} height={1920}>` in a `useLayoutEffect`, through the camera in `src/engine/camera.ts`:
- `project(cam, [x, y, z])` gives the screen point, its depth for sorting and its scale.
- `orbit(target, yaw, pitch, distance)` places the camera around a target.
- `spline(f, [[frame, value], ...])` joins keys on word cues with a smooth curve that never jumps and keeps moving through
the middle keys.
**One continuous world.** When two scenes happen in the same place, draw that world from the reel's frame in one shared
component, and use it in both scenes (see `reels/demo`). With `"handover": "continuous"` in `reel.json` the next scene takes
over on its first frame, so the picture never jumps. List those scenes as keep scenes for the motion check.
`src/engine/moves.ts` has the motion verbs: `prog`, `ease`, `arrive`, `leave`, `blurIn`, `blurOut`, `pop`, `typeOn`,
`countTo`, `draw`, `stagger`, `mix`. `theme.ts` holds every value a reel may override in `reel.json` under `"theme"`.
## The checks
Each prints PASS or FAIL with the reason and exits 1 on a FAIL. A check that could not run counts as a FAIL.
| Command | What fails |
|---|---|
| `check.py plan <slug>` | a caption block over 8 words, two key words in a block, a scene change outside silence (it also warns on type under 32 px) |
| `check.py render <slug> <mp4>` | nothing moving for over 1.2 s, bare ground during a scene change |
| `motion.py <mp4> --words reels/<slug>/words.json --keep a,b` | a move too fast to follow inside a world, a cut or snap, a near-freeze then a jump, a scene under 2 s |
| `joins.py <slug> <mp4> --sheet out.jpg` | a scene change where one frame swaps the picture; the sheet shows every change |
| `speed.py <mp4>` | the camera speed check on its own, for any clip |
| `loud.py <mp4> [--fix]` | sound away from -14 LUFS or peaks above -1.5 dBTP; `--fix` levels it, video untouched |
## Keys
Keys go in `.env.local` at the top of the repo (`ELEVENLABS_API_KEY=...`). Never ask for one in chat and never commit one.
© sirioberati · MIT · sirioberati/showhow