Claude Code Lab
ondrej-svec/claude-code-lab
Aprenda agentic coding com Claude Code através de prática estruturada
Ver no GitHubGuia prático e estruturado para programação com agentes usando Claude Code, do zero até loops compostos. Inclui 10 capítulos de fundações, biblioteca de tópicos avançados, plugin diagnóstico para avaliar sua configuração e skill de atualização automática.
Instalação
npx convez-ai ondrej-svec/claude-code-lab1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Dominar padrões de coding agentic do zero
- Entender context engineering e iteração com Claude
- Diagnosticar e otimizar sua configuração de Claude Code
- Explorar compound engineering (brainstorm → plan → work → review)
Veredito

Recurso educacional sólido e bem arquitetado para quem quer ir além dos basics de Claude Code. O plugin diagnóstico e a skill de freshness automática agregam valor real, mantendo o lab atualizado com novos shipments do Anthropic.
README
A tradução para o português deste README ainda está na fila.
claude-code-lab

→ Live at cc-lab.ondrejsvec.com
A hands-on guide to agentic coding with Claude Code — from install to compound. Bilingual (EN / CS), practice-oriented, with copyable prompts and sample projects you can run through the full loop.
Ten spine chapters teach the path; a small library of depth entries goes further on specific topics; a diagnostic plugin asks how your setup measures against the patterns; a freshness skill keeps the lab honest with whatever Anthropic ships next.
Spine — ten chapters
- Before we start — orientation
- Your first task — first contact
- Teach Claude your project — context
- Iteration and control — craft
- Voice and modalities — voice dictation across surfaces, multimodal context
- The ecosystem — skills, plugins, MCP, hooks
- Compound engineering — brainstorm → plan → work → review → compound
- Where to go next — harness building, autonomy, sandboxing
- Reference — commands, settings, primitives at a glance
- Behind the scenes — stats from building the lab itself
Library — depth on demand
Chapter-affiliated entries that go further than a chapter could without bloating the path. Three entries shipped:
- Context engineering as practice (chapter 4 affinity) — the four moves, three layers of memory,
/compactrecipes, externalize-before-you-forget - Autonomous loops — when and when not (chapter 4 affinity) — the loop that earns its keep vs the Ralph Wiggum loop, pre-flight checklist, harness-as-precondition
- Using
/cc-lab-diagnose(chapter 6 affinity) — what the diagnostic reads, how to interpret an observation, honest limits
Browse at cc-lab.ondrejsvec.com/en/lab/library.
/cc-lab-diagnose — the personal-mirror plugin
A Claude Code plugin that reads your repo (or your personal Claude Code config, or both) and returns three to five evidence-grounded observations — each tied to a specific file, a specific quote, and a specific chapter you can read for the missing context. Cross-platform, zero telemetry, runs locally.
/plugin marketplace add ondrej-svec/claude-code-lab
/plugin install cc-lab@cc-lab
/cc-lab-diagnose project
See the library entry for the full walkthrough.
What's inside this repo
app/— public guide site (Next.js 16, Tailwind 4, Rosé Pine theme, dark/light)content/{en,cs}/— chapter MDX (10 chapters × 2 locales)content/{en,cs}/library/— library entry MDX (3 entries × 2 locales)docs/cc-lab-design-system.md— the lens: voice rules, audience, anti-goals, visual conventions, library architectureplugins/cc-lab/—/cc-lab-diagnosemarketplace plugin (skill + Sonnet judge subagent + bundled chapter knowledge + HTML artifact template)skills/cc-lab-screenshot/— internal screenshot capture skill (computer-use MCP).claude/skills/cc-lab-freshness/— internal lens-aware freshness skill that watches Anthropic's shipping surface and proposes lab-shaped edits as PRs (runs as an Anthropic Routine on a weekly schedule; seescripts/freshness/README.md)samples/python-react/— minimal Python + React sample projectsamples/dotnet-core/— minimal cross-platform .NET Core sample projectskill/— companion Claude Code skill that walks through the lab from your terminal
Running locally
pnpm install
pnpm dev
Open http://localhost:3000. The guide is public and needs no runtime environment variables.
Scripts
pnpm dev # dev server (Turbopack)
pnpm build # production build
pnpm start # run built app
pnpm lint # eslint
pnpm test # vitest unit
pnpm test:e2e # playwright e2e
Visuals
Three production paths, each fitting a different shape (see docs/cc-lab-design-system.md "Hybrid visual strategy" for the full rules):
- Conceptual diagrams (boxes, arrows, flow) — Codex GPT Image 2 via the
image-genskill. Helpers:scripts/visuals/gen-chapter-hero.sh(chapter heroes, ship-locked) andscripts/visuals/gen-library-diagram.sh(library diagrams, no ship). - Static terminal stills —
freezethemed rose-pine-moon viascripts/visuals/freeze.config.json, rendered as<TerminalOutput>MDX. - Animated terminal sessions —
vhs, themed viascripts/visuals/vhs.theme.json, rendered as<TerminalCast>MDX. - Desktop / mobile screenshots —
cc-lab-screenshotskill (computer-use MCP). Manual capture is a documented fallback.
Web index captures regenerate via scripts/capture/:
./scripts/capture/generate-fixtures.py # rebuild CLI ANSI fixtures
./scripts/capture/capture-cli.sh # render CLI PNGs
./scripts/capture/capture-web.sh # render WEB PNGs (needs pnpm dev)
Design
Rosé Pine Dawn (light) / Moon (dark). Manrope body, Space Grotesk display. Lifted from the Harness Lab design system.
Contributing
See CONTRIBUTING.md. Czech corrections especially welcome.
Security
See SECURITY.md — use GitHub private vulnerability reporting, don't file public issues for security bugs.
License
MIT — © 2026 Ondrej Svec.