Five tracks, module 0 open
Five agents. One method. Your own terminal.
تعلّم الوكيل بيديك: اكتب الأمر، وشاهد ما يحدث.
Learn by typing. Every command here is copied from the agent's own docs, and every lesson ends with you running it on your machine.
Step 1 / 5
Every task is a session
Sessions run side by side, each with its own chat and project folder.
Step 2 / 5
Say what you want
Type the task and press Enter. You can correct Claude while it works.
Step 3 / 5
Decide how much it does alone
The permission mode, the model and the machine it runs on are all one click away.
Step 4 / 5
Extend it
Skills, plugins and connectors add abilities. Slash commands run them.
Step 5 / 5
Review before you keep it
The diff shows every change. Claude can review its own code first.
I'll read the project first, then add an Arabic page with a language switch.
lang="ar" and dir="rtl", and a font with Arabic letters.Read 2 files
● Read index.html
● Read styles.css
Plan
- Add
ar/index.htmlwithdir="rtl" - Add a language switch to both pages
- Check both pages at 390 px wide
● Edited index.html +48 −2
● Run npm run build
Checked by the auto-mode classifier
Reads the main thread; answers here without derailing it.
index.html - <nav> + <nav> + <a href="/ar/" hreflang="ar">العربية</a> ar/index.html new + <html lang="ar" dir="rtl"> + <h1>مخبز الحي</h1>
Claude: the Arabic link needs lang="ar" so screen readers switch voice.
~/bakery-site $ npm run build [your build output appears here]
No background tasks running.
Chat
Everyday conversations with Claude. Software work lives in the Code tab; longer jobs and Dispatch live in Cowork.
Message Dispatch a task. It decides how to handle it.
- Google Calendar
- Slack
- GitHub
- Linear
- Notion
- /code-review Bundled
- /design Bundled
- /loop Bundled
- /summarize-changes Personal · ~/.claude/skills/summarize-changes/SKILL.md
- /deploy Project · .claude/skills/deploy/SKILL.md
- commit-commands claude-plugins-official
- A plugin bundles skills, agents, hooks and MCP servers as one unit.
Desktop app
Manage or disconnect your connectors here.
Continue this session from claude.ai/code or the Claude app on your phone. Claude keeps running on this machine, so your files stay here, and the conversation stays in sync on every device.
Bilingual bakery page
$ cd bakery-site && claude
Type a task, / for commands, Shift+Tab to change mode.
Discover Installed Marketplaces
› commit-commands@claude-plugins-official
[more plugins from the marketplaces you add]
⏸ manual mode on
Neighbourhood Bakery
Fresh bread from 6 am
Sourdough, date rolls and Friday specials, a short walk from home.
Order for pickupمخبز الحي
خبز طازج من السادسة صباحًا
اطلب للاستلامIllustrative layout. Claude Design is an Anthropic Labs research preview (April 2026) that turns prompts into designs, prototypes and decks; its real interface differs.
Simulation for learning. Labels follow Anthropic's Claude Code docs, checked 2026-09-30; the real apps can differ.
Sessions each keep their own chat and folder
+ New session (Ctrl+N) runs tasks in parallel
Worktree: an isolated copy, so sessions never collide
Prompt box: Enter sends; type again to redirect it
Transcript: every read, edit and command it runs
Transcript view: Normal, Thinking or Verbose (Ctrl+O)
Permission mode: Manual, Accept edits, Plan, Auto, Bypass
Where it runs: Local, Cloud, SSH or WSL
Model: next to Send, switch it any time
+ menu: Connectors, Slash commands, Plugins
Customize: skills, plugins and connectors in one place
Settings: computer use, worktrees, the bypass switch
Panes: diff, terminal, browser, tasks. Drag to arrange
Review code: Claude comments on its own diff
Views opens more panes
The path
Ten modules, from install to ship
- 0Install and first runBeginner · 20 minOpen
- 1The desktop app, hands onBeginner · 25 minOpen
- 2Permission modesBeginner · 15 minOpen
- 3Slash commandsBeginner · 20 minOpen
- 4CLAUDE.md and memoryIntermediate · 40 minPlanned
- 5SkillsIntermediate · 45 minPlanned
- 6Plugins and MCPIntermediate · 45 minPlanned
- 7Hooks and subagentsAdvanced · 60 minPlanned
- 8Remote Control, Cowork and DispatchAdvanced · 30 minPlanned
- 9Ship it: where to hostBeginner · 15 minOpen
Find your level
Five questions. Then we tell you where to start.
Q1 / 5
Have you installed an AI coding agent before?
Q2 / 5
Do you know what Plan and Accept edits modes do?
Q3 / 5
Have you written a CLAUDE.md or AGENTS.md?
Q4 / 5
Have you created a skill or installed a plugin?
Q5 / 5
Have you run sessions in parallel or from your phone?
The five tracks
FIG. 3 · 5- 01Claude CodeCoding agent: terminal, IDE, desktop app, web
- 02Codex CLICoding agent that runs locally in the terminal
- 03Hermes AgentSelf-improving agent with a learning loop
- 04OpenClawPersonal AI assistant that lives in your chat apps
- 05DeepSeekModel provider; runs inside Claude Code through an Anthropic-format API
Built something?
Where should the app your agent built live?
Answer four questions, or paste your file list. The tool names the shape of your app and the host that fits it.
Open the tool