blueprint · smilewilson1999 — скилл для Claude Code: что делает и как установить
Скилл blueprint для Claude Code, создающий интерактивные схемы архитектуры стеков и топологий сервисов в виде HTML-файла со светлой и тёмной темами.
MIT ★ 59 обновлён 2026-11-09
npx skills add smilewilson1999/blueprint --skill blueprint -a claude-codeЧто делает blueprint
Скилл строит архитектурные диаграммы на основе JSON-описания системы, собирает официальные логотипы брендов и генерирует автономную HTML-страницу. Готовый файл публикуется на artifact.cafe для сбора комментариев или сохраняется в формате PNG через Playwright. Для работы требуются Python 3.9+, доступ в сеть для загрузки иконок и Playwright с Chromium для рендеринга PNG.
Инструмент автоматически настраивает группы узлов, расставляет маркеры запросов и проверяет компоновку на пересечения элементов. При необходимости скорректировать схему изменения вносятся в спецификацию, после чего страница собирается заново.
Какие задачи решает
- создать интерактивную схему архитектуры системы по описанию
- сгенерировать диаграмму связей сервисов с логотипами
- опубликовать архитектурный макет на artifact.cafe для ревью
- отрендерить PNG-изображения архитектуры из HTML
Примеры запросов агенту
- Нарисуй схему архитектуры нашего бэкенда и фронтенда
- Создай диаграмму связей микросервисов для ревью
- Собери интерактивную архитектурную карту стека
Как установить blueprint
1Через Skills CLI (Vercel)
npx skills add smilewilson1999/blueprint --skill blueprint -a claude-code2Вручную: положить файлы
~/.claude/skills/blueprint/SKILL.mdПапка скилла целиком; для одного проекта — .claude/skills/blueprint/
1Через Skills CLI (Vercel)
npx skills add smilewilson1999/blueprint --skill blueprint -a codex2Вручную: положить файлы
.agents/skills/blueprint/SKILL.mdДля всех проектов — ~/.agents/skills/
1Через Skills CLI (Vercel)
npx skills add smilewilson1999/blueprint --skill blueprint -a cursor2Вручную: положить файлы
.cursor/skills/blueprint/SKILL.mdCursor читает и .agents/skills/, и .claude/skills/
1GitHub CLI
gh skill install smilewilson1999/blueprint skills/blueprint2Вручную: положить файлы
.github/skills/blueprint/SKILL.md1Через Skills CLI (Vercel)
npx skills add smilewilson1999/blueprint --skill blueprint -a gemini-cli2Командой платформы
gemini skills install https://github.com/smilewilson1999/blueprint --path skills/blueprintВручную: положить файлы
.devin/skills/blueprint/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 blueprint
Как установить blueprint в Claude Code?
Выполните в терминале npx skills add smilewilson1999/blueprint --skill blueprint -a claude-code, затем обратитесь к файлу ~/.claude/skills/blueprint/SKILL.md.
Что делает blueprint?
Превращает текстовое или JSON-описание системы в интерактивную архитектурную страницу с логотипами, путями запросов и поддержкой тем оформления.
Что нужно для работы blueprint?
Для работы требуются Python версии 3.9 и выше, сетевое подключение для загрузки иконок, а также установленный Playwright с Chromium для генерации PNG-изображений.
Исходный текст
SKILL.md · MIT
--- name: blueprint description: "Create premium, interactive architecture diagrams of tech stacks, system architectures, infra layouts and service topologies, with real brand logos, grouped windows, numbered request flows and orthogonal edges, delivered as one self-contained HTML file that publishes to artifact.cafe for review (plus PNGs). Use this skill whenever the user asks for an architecture / system / stack / infra diagram, a tech-stack overview, a service topology, a cloud-vendor-style diagram, a diagram to get feedback on, or wants tech logos arranged into a picture of how their services connect, in any language, even if they don't name an output format and even if the tools involved aren't cloud services. For physical spaces, floor plans and furniture layouts use blueprint-space; for exploded physical objects and layered internals use blueprint-teardown." license: MIT (see LICENSE); bundled fonts are SIL OFL 1.1 compatibility: Python 3.9+ and network access to fetch logos; PNG output also needs Playwright with Chromium --- # blueprint Turns a short JSON description of a system into an interactive architecture page in the Cafe OS style: warm paper, real logos on light tiles, groups drawn as small OS windows, one amber request path with numbered steps, flow tabs, an inspector, and light and dark themes. The HTML is the source of truth. It is built to be reviewed on artifact.cafe, where every node name is real text a reviewer can quote and every node has a stable id. ## Which skill | The user wants to show | Use | | --- | --- | | How services, APIs and data stores connect; a request path | `blueprint` | | How a physical object is built: layers, parts, how they fit | `blueprint-teardown` | | A physical space: rooms, floor plan, furniture placement | `blueprint-space` | ## Workflow 1. Settle the content: nodes, groups, edges, flows 2. Fetch icons: `scripts/fetch_icons.py` 3. Write `blueprint.json`: read `references/spec.md` 4. Build: `scripts/blueprint.py`, fix every `WARN` 5. Render PNGs and look at them 6. Publish to artifact.cafe for review (or deliver the HTML / PNG) ## Step 1: Content From the user's list of technologies, decide: - **Nodes**: one per technology, with its official name and a short lowercase role (`api routes`, `sqlite db`), plus a one-sentence `note` on what it does here when you know it. - **Groups**: a box around things that share a runtime, vendor or repo. Nest one level; deeper gets noisy. - **Edges**: `primary` for the one path a reader should follow (number it with `step`), `internal` for other calls, `async` for events, queues and webhooks, `dev` for CI and deploys. - **Flows**: when the system has more than one story (request path, nightly sync, deploy), give each a flow so the reader can switch between them. If the user gave only a flat list, infer sensible groups and flows from what the tools are. Ask only when a tool is ambiguous enough that the diagram would be wrong either way. ## Step 2: Icons ```bash python3 scripts/fetch_icons.py <icons-dir> nextjs hono neon "actions:githubactions" "betterauth:better-auth" "orpc=https://orpc.unnoq.com/logo.webp" python3 scripts/fetch_icons.py <icons-dir> "cron:lucide/clock" "queue:lucide/list-ordered" ``` Spec forms: `name`, `name:slug`, `name=URL`, `name:lucide/<icon>`, and the vendor forms `name:aws/<Name>`, `name:gcp/<Product>`, `name:azure/<Name>`. The script tries svgl, then devicon, then gilbarbara/logos, writes `manifest.json`, saves `<name>_dark` when the brand has a dark-theme mark, and flags white fills. - **Never recolor a logo.** If it's white or too light for a light tile, use `"tile": "dark"`. The `white_fill` flag is a heuristic: glance at the SVG first. - **Official marks only.** When every source misses, find the official file in the project's repo or site and pass it as `name=URL`. - **Abstract concepts get Lucide pictograms** (`devenv:lucide/terminal`, `cron:lucide/clock`), never text in a tile or emoji. Generic people and machines: copy a glyph from `assets/generic/` into the icons dir. - **Vendor sub-products** without their own mark (Cloudflare D1, R2) reuse the vendor logo; the name carries the distinction. - **Group-label icons are 15px**: use the brand's square mark, never a wordmark. - **AWS, Google Cloud and Azure** use their vendors' own icons and drawing rules inside their area. Read `references/vendors.md` before drawing one. ## Step 3: Spec Write `blueprint.json` per `references/spec.md`. Place nodes on a grid with `at: [col, row]`, in reading order: the request enters on the left or top and the primary path runs in a straight line where it can. Keep node `id`s stable between versions; comments on the review page attach to them. ## Step 4: Build ```bash python3 scripts/blueprint.py blueprint.json <icons-dir> index.html ``` The build lays everything out, routes edges around nodes and group labels, inlines the fonts and icons, and prints a `WARN` line for each overlap, collision or crowded label it finds. Fix warnings by moving nodes on the grid, then rebuild. Don't hand-edit the generated HTML: change the spec. ## Step 5: Look at it ```bash python3 scripts/render_png.py index.html diagram.png --scale 2 ``` This writes `diagram.png` and `diagram-dark.png` (canvas only; `--selector .page` for the whole page). Read both images. Check: - the primary path reads left to right (or top to bottom) without doubling back - no edge crosses label text and no label chip sits on a node or a window title bar - every icon is visible in both themes, including group-label marks in dark - groups don't touch each other and nothing sticks out of its group If Japanese text renders with Chinese glyphs, the host is missing a Japanese font: the script prints a fix. ## Step 6: Deliver **Default: publish for review on artifact.cafe.** The HTML is one self-contained file, so publish the folder that holds `index.html` with the artifact-cafe skill or CLI (`npx artifact-cafe@latest publish <dir>`), using the diagram's title as the artifact title. Share the review URL. Reviewers can comment on any node name, role, edge label or step without an account. When comments come back, change `blueprint.json`, rebuild, and publish the next version into the same artifact. Keep node ids and names stable where the component is unchanged: comments belong to the version they were made on, and an unchanged anchor is what lets a thread be found again in the next one. Reply to each thread with what changed. Also on request: - **PNG**: the `render_png.py` pair, for Slack, slides or a README. - **HTML**: the built file itself; it works offline. The footer credits the icon sources from `manifest.json` automatically. Brand logos remain the trademarks of their owners.
© smilewilson1999 · MIT · smilewilson1999/blueprint
Похожие скиллы
Все скиллы по теме «DevOps и инфраструктура» →Сначала — blueprint от других авторов.