Claude: Zero to Hero
A complete, step-by-step learning track for Claude — from “I’ve never typed a prompt” to “I ship production agents.”
Built: August 2026 · Sources: live Anthropic docs (platform.claude.com/docs, code.claude.com/docs, support.claude.com)
How this track is organised
Eight stages, numbered 00 to 07, plus a 99-reference folder you dip into rather than work through. Each file inside a stage is a module, and every module has the same shape:
- You are here — which path it’s on, what plan it needs, how long it takes to read and how long the exercises take, what it assumes
- What you’ll learn — the outcome, in one line
- If you only read one thing — the whole module in plain language, no jargon. Enough to be useful on its own.
- The rest of the module — the actual guide, in depth
- Try it — hands-on exercises you can do immediately, several with worked answers
- Checkpoint — how you know you’ve got it
- Going deeper — links to the official docs for the details this guide compresses
If you’re new and something in The details loses you, the honest advice is to read the summary box, do the exercises, and come back. The box is not a teaser — it’s the load-bearing part.
Modules are numbered in the order you should do them. Nothing later assumes anything you haven’t already read.
learn-claude/
├── 00-start-here/ Orientation. What Claude is, where it lives, what plan you need.
├── 01-foundations/ Chat mastery. Prompting, models, files, artifacts, research.
├── 02-power-user/ No-code power. Projects, Skills, Connectors, Cowork, automation.
├── 03-claude-code/ Agentic coding. CLI, CLAUDE.md, subagents, hooks, MCP, plugins.
├── 04-api/ Building on Claude. Messages API, tools, caching, structured output.
├── 05-agents/ Agent engineering. Agent SDK, context engineering, multi-agent.
├── 06-production/ Shipping. Evals, guardrails, cost, observability, governance.
├── 07-capstones/ Five projects that prove you learned it.
└── 99-reference/ Cheat sheets, glossary, links. Come back to these forever.
Learning paths
You do not have to do all of it. Pick the path that matches where you’re going.
Each module below states its own reading time and exercise time in its You are here line. The totals here are those per-module numbers added up for the modules and capstones in that path’s recipe, not a promise — go at whatever pace makes it stick. For a reading-only pass through a path, add up just the Read figures.
Path A — “I just want to use Claude really well” (21–24 hours)
00-start-here → 01-foundations → 02-power-user → capstone 1 & 2
You’ll never touch code. You’ll end up better at Claude than most engineers. Spread across two or three weeks is more effective than one weekend.
Path B — “I want to code with Claude” (+12–18 hours on top of Path A)
Path A → 03-claude-code → capstone 3
03-claude-code’s modules don’t carry per-module You are here estimates yet, so this delta isn’t backed by a per-module sum the way Path A’s total is — treat it as a rough placeholder pending that follow-up.
Path C — “I want to build products on Claude” (40–60 hours)
Path A (skim) → 03-claude-code → 04-api → 05-agents → 06-production → capstones 4 & 5
Same caveat as Path B: 03-claude-code, 05-agents and 06-production have no per-module estimates yet (only 04-api/00-pricing-and-rate-limits.md does), so this total is not yet a verified sum.
Path D — “I’m technical, skip the basics”
00-start-here/02-the-claude-landscape.md → 01-foundations/02-prompting-fundamentals.md → 01-foundations/03-models-and-modes.md → then jump to 03-claude-code or 04-api.
Full module index
00 · Start here
| # | Module | Why |
|---|---|---|
| 00 | How to use this track | Ground rules, how to practise |
| 01 | What Claude actually is | Mental model: LLM, context, tokens |
| 02 | The Claude landscape | Every surface, one map |
| 03 | Plans, pricing, and limits | What you need to buy, and when |
01 · Foundations — chat mastery
02 · Power user — no code required
| # | Module |
|---|---|
| 01 | Projects |
| 02 | Skills |
| 03 | Connectors and MCP, explained without code |
| 04 | Cowork |
| 05 | Claude inside your apps |
| 06 | Automation: scheduled tasks and routines |
| 07 | Plugins |
| 08 | Advanced prompting patterns |
03 · Claude Code — agentic coding
04 · API — building on Claude
05 · Agents
| # | Module |
|---|---|
| 01 | Agent design principles |
| 02 | The Claude Agent SDK |
| 03 | Context engineering |
| 04 | Multi-agent patterns |
| 05 | Managed Agents |
06 · Production
| # | Module |
|---|---|
| 01 | Evals and testing |
| 02 | Guardrails and safety |
| 03 | Cost and latency optimisation |
| 04 | Observability |
| 05 | Enterprise and governance |
07 · Capstones
Five projects, one per stage, each with a spec and a done-check.
99 · Reference
| File | Use |
|---|---|
| Glossary | Every term in one place |
| Model cheat sheet | Which model, when, what it costs |
| Prompting cheat sheet | Copy-paste patterns |
| Claude Code cheat sheet | Commands, flags, file layout |
| API cheat sheet | Request shapes and parameters |
| Official links | Where to check when this goes stale |
Three rules for getting through this
1. Do the exercises. Reading about prompting improves nothing. The gap between people who are good at Claude and people who aren’t is almost entirely reps.
2. Keep a scratch file. Every time a prompt works surprisingly well, paste it into a notes file. In two weeks that file is worth more than this track.
3. Assume this document is slightly out of date. Claude ships weekly. Every module ends with links to the live docs. When something here disagrees with the docs, the docs win — and the changelogs are listed in 99-reference/official-links.md.
Build & deploy
This site deploys via a GitHub Actions workflow (.github/workflows/pages.yml), not the legacy “build from branch” Pages pipeline. On push to main, the workflow builds the Jekyll site from this repo’s own Gemfile (actions/jekyll-build-pages) and publishes it (actions/deploy-pages). The repo’s Settings → Pages → Build and deployment → Source is set to “GitHub Actions” to match.
Switching off the legacy build means plugins are no longer limited to GitHub’s Pages allowlist — anything installable via the Gemfile is available at build time, and extra build steps (asset hashing, search-index generation, etc.) can be added to the workflow as needed.