The Ultimate Guide to Claude from 0 to 1 (2026 Complete Edition)

What Level Are You Using Claude At Right Now?

Beginner: Using Claude like Google – ask a question, get an answer
Intermediate: Using Projects, Skills, Hooks, Subagents – repetitive work visibly reduced
Advanced: Building automated pipelines where AI works on a fixed schedule

This isn’t a “read and forget” article. Follow along, and you’ll walk away with six concrete results:

• ✅ A personal homepage built in minutes
• ✅ An AI inspiration digest that updates daily
• ✅ A publicly accessible URL you can share with friends
• ✅ An experience orchestrating “three AIs working simultaneously”
• ✅ A data analysis report you can show your boss
• ✅ A content pipeline running on a daily schedule

Every hands-on section has a ✅ checkpoint – exactly what you should see when you get there. Let’s go.

---

Part 1: Newbie Village – From 0 to 1

Setup: Install the Claude Desktop App

1. Open claude.ai/download in your browser
2. Choose your system download (Mac / Windows – no desktop version for Linux yet), double-click to install
3. Windows extra step: Install Git for Windows first (click through defaults), otherwise local sessions won’t run. Restart Claude after installing.
4. Log in with your Claude account (register if you don’t have one)

⚠️ The free tier cannot use Cowork or Claude Code. You need at least a Pro subscription (20/month, ~17/month if paid annually). Pro is enough for beginners; price analysis at the end of the article.

No overseas credit card? Subscribe via App Store; use an overseas Apple account. The exact entry point varies by region and version – just follow the on-screen prompts.

Don’t have an overseas Apple account? Refer to the post below.

宋宋@longlongsongs·May 20 Article
Step-by-step guide to registering a Turkish Apple ID
Why register a Turkish Apple ID?
Core reason in one word: cheap, save money!
Before unified pricing, the same ChatGPT Plus cost some people over 140 RMB and others less than 70 RMB per month.
Same X… 21179346K

💰 No overseas credit card / tight budget? See the appendix at the end: “Domestic Model Alternative Route” – finish the main content first, then decide which path to take.

✅ Checkpoint 0: After logging in, you should see three tabs at the top: Chat, Cowork, Code. Gear is ready.

What Are Those Three Tabs Really For?

First, understand the three layers of AI tools:

• Model = Brain (can only think, no hands or feet)
• Chatbot = Brain + a mouth (can talk but not act)
• Agent = Brain + hands + toolbox (can talk and act)

ChatGPT web, Doubao App are Chatbots – you ask, it answers, you still do the work. Claude Code is an Agent – you say “build me a website,” it directly creates folders, writes code, runs it, and debugs itself. The same model in different tools yields vastly different capabilities.

The three tabs in the desktop app correspond to three completely different “creatures”:

Tab	One-Liner Positioning	What It Can Do	When to Use
Chat	Library without hands (advisor)	Pure conversation, doesn’t directly read/write local files	Everything before doing: discuss needs, brainstorm ideas, write instructions
Cowork	Outsourced organizer (autopilot employee)	Given folder access, it works in an isolated workspace to research, organize documents, batch produce reports	Tasks that don’t need you watching; dispatch tasks via phone with Dispatch
Code	Lead engineer working before your eyes	Directly reads/writes local files, writes code, runs previews, installs tools, deploys. Every change is visible and reversible	Building products, managing projects

The correct workflow for the AI era (used repeatedly throughout the article, remember it first):

Task comes in → First go to Chat as an advisor to clarify and make a plan → To build a product, go to Code → Batch chores to Cowork → You only nod or shake your head at key milestones.

In the traditional model, you spend most of your time on “execution”; in the new model, you handle requirements, judgment, and acceptance – AI handles execution.

Don’t force yourself to write complex task instructions from scratch. First discuss in Chat, then finally say:

Organize everything we just discussed into a complete task instruction that can be directly handed to an AI to execute: goal, steps, output requirements, notes – leave nothing out.

Copy and use it. Chat turns thoughts into clarity; Code and Cowork handle execution.

Hands-On 1: 5 Minutes, Taste the Sweetness

Let’s get to work immediately. Build a personal homepage – your business card on the internet.

Follow along:

1. Create a new empty folder named my-page
2. Click the Code tab → New session → Set environment to Local → Open folder and select my-page (first time asks “Do you trust this folder?” – confirm it’s the empty folder you just created, then click trust workspace)
3. Paste the text below, replace with your info, and hit Enter:

Create a single-file personal homepage index.html: Dark background, centered display of my name “Zhang San”, Below, a line describing me: “Writer / Coffee Addict / Learning AI”, Then three button links to my social accounts (use # as placeholder for now). Use an elegant colour scheme, add a graceful entry animation. Show me the preview when done.

4. When permission requests pop up during work, read the path and action carefully; beginners should click “allow once” first, consider long-term authorization later.
5. After a minute or two, your homepage appears in the right preview pane. Send change requests directly to continue tweaking.

✅ Checkpoint 1: Open the my-page folder, double-click index.html, and your homepage appears in the browser.

Pause for a second: You created a web page with a paragraph of Chinese. Two minutes ago, it didn’t exist.

This is the minimum viable loop: You speak human language → It writes code → You see results and give feedback.

The personal homepage is just a warm-up. The next thing you’ll build is an application you’ll actually use long-term – more files, longer steps, and permissions become more important. Before diving in, get three basic habits in place.

Before Big Projects: Tune 3 Basic Settings

1. One project, one folder (hard rule)

Claude’s work happens inside the folder you specify. Professional habit: further divide the folder into two types:

• Resources: Company SOPs, brand guidelines, reference documents. Name with underscore prefix like _resources – this is the reference material for AI.
• Outputs: All results go here.

Read permission requests carefully – don’t casually authorize unfamiliar directories.

2. Choose the model: like shifting gears in a car

Model	What It’s Like	When to Use
Haiku	Walking	Reading documents, simple Q&A
Sonnet	City driving	Daily workhorse, give it most tasks
Opus	Climbing a steep slope	Planning, architecture design, complex decisions
Fable5	A rocket	Use for the most complex, judgment-heavy tasks only

Where to select: Click the model name next to the input box.

Golden combination: Strong model thinks, daily model executes. Let the stronger model come up with the plan, then switch to Sonnet to execute – like a senior consultant drawing blueprints, and the construction team building.

Fable5 is temporarily usable until 6.22.

Next to it is effort (thinking depth): Use high for building apps and making plans, low for reading files and casual chat – saves quota.

When reading old tutorials: Previously there was an “Extended Thinking / Adaptive thinking” toggle – many versions now removed it. Newer models handle thinking more automatically; just focus on effort.

A tip: Adding a prompt like ultrathink (e.g., ultrathink help me compare several options, don't write code yet) makes Claude Code think deeper on that turn. Difference from effort: effort is session-level (always active), ultrathink only affects the current message – don’t use it daily, add it only for that tough question.

3. Permission mode: How much steering wheel do you give AI?

Where to select: The mode selector next to the input box, shortcut Ctrl+Shift+M (Mac: Cmd+Shift+M).

Mode	Behaviour	Suitable For
Ask permissions	Asks you before every step	Your first week (default)
Accept edits	Changes files without asking, asks before running commands	Daily use after familiarization
Plan mode	Only produces plans, doesn’t act	First step of every major project
Auto mode	Basically autonomous driving	After gaining experience (preview feature)
Bypass permissions	Never asks (YOLO mode)	Veterans + isolated environment

Can’t see Bypass? Go to Settings → Claude Code → Turn on “Allow bypass permissions mode”. Beginners won’t need this yet.

Must-form habit: Enter Plan mode for big projects. Let AI present a plan – you review, modify, and nod before it touches any files. This significantly reduces “It built something completely wrong” rework.

Hands-On 2: Build an Application You’ll Use Every Day

This time, for real: Build a Daily AI Inspiration Digest – open it each day, and automatically get the latest AI tools and trends served to you; see something good, save it with one click to your inspiration library, never lose track of “that cool thing I saw yesterday.” Data source uses free public RSS, zero API cost. You can swap it for anything you want (expense tracker, habit tracker) – the process is identical.

2.1 First, Clarify Requirements in Chat

Switch to the Chat tab, adapt this text and send:

Write me a Product Requirements Document (PRD). I want to build a “Daily AI Inspiration Digest” web application, to be used locally on my computer, not published online for now.

Core features:

1. Automatically fetch from several free public AI news sources (RSS), both Chinese and English; open it daily to see the latest content, sorted by time
2. Each item can be saved with one click to an “Inspiration Library”; when saving, allow adding tags (e.g., Topic / Tool / Case / Idea)
3. Inspiration Library page: filter by tag, search by keyword
4. “Today’s Digest” view: arrange today’s new content + my saved items into a beautiful daily digest card, suitable for screenshots and sharing
5. Recommend specific sources – they must be stable, free, and require no registration
6. Style like a well-typeset electronic newspaper

Chat for a few rounds, get the PRD, then check it against the five elements: What it does + who it’s for + what features + what it looks like + special requirements – ask it to fill in any missing parts. When satisfied, select all and copy.

2.2 Go to Code, Use Plan Mode to Start

1. Create new empty folder ai-daily, start a new Code session and select it
2. Set permission mode to Plan mode
3. Paste this opening text + the PRD you copied:

Don’t write any code yet. Read the requirements below, then list a complete implementation plan for me to review. If anything is unclear, ask me one question at a time.

[Paste your PRD here]

When permission requests to read directories or run commands appear, read them carefully before allowing.

Two practical tips:

• @filename syntax: You can also save the PRD as prd.md inside the project folder, then just say @prd.md list the plan according to these requirements – @ lets it focus precisely on that file, a core syntax of Claude Code.
• Mine requirements from Xiaohongshu / competitors: See a competing product? Screenshot it, throw it in and say “I want something similar, first write me a PRD” – don’t even need to think of the requirements yourself.

2.3 Lock Down the Plan, Then Let It Build

The value of Plan mode isn’t “let Claude talk more” – it’s to expose areas likely to cause rework upfront. It will usually give you an implementation plan + a few boundary questions. You don’t need to sound professional; just confirm these three things:

• First, scope: This time, only do MVP, don’t cram in P0/P1/P2 at once. Beginners can reply directly: “Start with a locally usable version – get save, tags, search, and daily digest card working first. Add scheduled tasks and online deployment later.”
• Second, data: Save saved content locally as files, no cloud database; for sources, let it recommend 4-6 stable free RSS feeds, you pick which ones, then execute.
• Third, visual style: If you have a preferred product style, give a screenshot or reference link; if not, say: “Keep it simple, clean, like a well-typeset electronic newspaper.”

Common questions and beginner-friendly answers:

If it asks	Beginner should answer
Where to store saved data?	“Store locally as files, no cloud database”
Which sources to fetch?	“Recommend 4-6 stable free ones, both Chinese and English, list them for me to choose”
Which tech stack?	“Use what you recommend, keep it simple”
Deploy to the web?	“First get it working locally, think about deployment later”
Scheduled fetching?	“First do auto-fetch on app start + a manual refresh button, add scheduled tasks later”

If it points out a contradiction in the PRD (e.g., “Is the daily digest card a manual screenshot, or a copy-to-clipboard button?”), don’t skip past it. Choose the simpler version: MVP first manual screenshot, add copy button later.

Principle: Context is king. The more materials, screenshots, and reference URLs you provide, the closer the result matches what you want. To borrow a big company’s design, open getdesign.md, pick a product (e.g., Linear), copy its full design.md, then add: “Modify the plan’s design section, using this design system:”

2.4 Approve Plan, Let It Build

1. Set permission mode to Auto accept edits
2. It works through the to-do list item by item
3. Rename the session “AI Inspiration Digest”; when permission requests appear, read the path and action, confirm they’re reasonable, then allow

✅ Checkpoint 2: To-dos get checked off one by one, files start appearing in the folder. Multiple “workers” building simultaneously before your eyes.

2.5 Live Preview: See It Built as It Builds

• How to open preview panel: The Preview dropdown button on the session toolbar (same row as model switch and permission mode switch). Mac shortcut Cmd+Shift+P, Windows Ctrl+Shift+P.
• After changing files, preview often starts automatically; if not, manually Start from the Preview dropdown.
• Drag the preview panel border to resize, directly test mobile view.
• Auto verify is on by default: after each code change, it takes screenshots, checks for errors, clicks buttons to test, and fixes itself.
• Click modification records in the conversation to see specific changes; if you dislike a line, click the small button next to it to leave a note.

2.6 Acceptance Testing + Bug Fixing: Two Moves for Everything

Please use the preview window to walk through the entire flow from start to finish: Refresh and fetch news once, save two items with different tags, go to the inspiration library and filter by a tag, finally generate today’s daily digest card. Confirm every feature works correctly; if you find issues, fix them directly.

Afterwards, two moves for any problems found:

• Screenshot method: Paste a screenshot and say “fix this” – it understands on its own.
• Paste error method: Copy the error message verbatim and say “fix it.”

✅ Checkpoint 3: Save today’s first piece of AI news that catches your eye into the inspiration library, and tag it.

From this moment, this is your productivity tool, not a tutorial exercise. One month later, you’ll have an inspiration library accumulated according to your own needs.

Eventually, it will give you a link – open it and use it directly.

See the Results

2.7 As Soon as It’s Done, Do These Two Things (30 seconds, save 3 hours later)

• /init – Generates a CLAUDE.md (note capital letters) project memory file.
• Check your quota usage – Click in the bottom-right of the conversation bar to see usage within 5 hours, within 1 week, and next reset time.

Memory Management: The Dividing Line Between Newbie and Veteran

Claude has no cross-session memory by default. The core is CLAUDE.md – think of it as the project manual + employee handbook: what the project is about, how to run it, your preferences and no-go zones. It reads this automatically at the start of each new session.

Command	What it does	When to use
/init	Generate CLAUDE.md	Once per project at start
update CLAUDE.md (type this directly)	Write progress into memory	After completing each feature
/compact	Compress conversation history	When context exceeds 50%
/usage	Check quota	Before starting a big project
/rewind (some versions)	Rollback code and conversation to an earlier state	When AI breaks things, prefer returning to a clean node rather than repeatedly redoing in messy context

Three hard rules:

• 50% red line: Once context usage exceeds 50%, AI is more likely to miss earlier details.
• Archive before compressing: First update CLAUDE.md, then /compact – reverse order risks losing progress.
• CLAUDE.md should be as short as possible: It’s a soft constraint (only enters context), don’t write a novel; use Hooks for hard constraints (covered in Part 2).

---

Part 2: Equipment Arsenal – Five Tools

Give Claude Gear Like an Employee

Imagine Claude as an employee you hired. These five items are the gear you give it:

Concept	Equivalent	One-liner
Command	Text expansion shortcut in your input method	Save a long, frequently repeated text as /xx, one slash to send
Skill	Standard Operating Procedure (SOP)	Teach it once “how to do this type of work,” it automatically references the manual thereafter
Hook	“If… then…” automatic rule	Like “emails auto-archive on arrival”: condition met, auto-execute
Subagent	Hire multiple specialized helpers	The main worker handles the mainline, helpers each research separately without competing for mental bandwidth
MCP / Connector / CLI	Building access card for the employee	Without a card, it can only stay in the folder; with a card, it can go to GitHub, calendar, deployment platforms, etc.

Most configurations can be done using natural language. Every tool will be practiced directly on your inspiration digest app.

Tool 1: Commands

Suitable for saving long paragraphs you often repeat. For example, summarizing what was done this session every time you wrap up. Create it as /ship:

I want to create a custom command named ship. When I type /ship, it should summarize what I did this session and how I did it. The summary should only display in the chat. This command should only work in the current project.

⚠️ Typing /ship does nothing? Newly created commands sometimes need a reload: try opening a new session, if still not working, restart the app and resume the session (Resume). Not broken, just needs a restart.

Tool 2: Skills

First, use the official one – the built-in “Frontend Design” skill is essentially Anthropic’s design team’s aesthetic SOP:

Use the frontend design skill, re-examine the entire app’s interface, optimize holistically based on the current design style, take care of all elements.

After sending, check the preview: spacing, fonts, and animations often get a full round of polish. Compare before and after – you’ll intuitively understand the value of Skills.

Then, build your own – use the official skill-creator (a skill for building skills). You don’t need to write anything; just answer questions. It will interview you like a consultant: What does this Skill do? What is its focus? Output format? Who uses it? After you answer, it generates a draft, finds a sample to test run, shows you the results. If not satisfied, give feedback; when satisfied, say “make it a Skill” and it’s saved.

Copy this and practice:

Create a custom skill named “Feature Planner.” Its role: Whenever I say “I want to add a feature,” first ask me enough sharp questions to understand the requirements and edge cases, then recommend a few related features I might not have thought of, after I answer, restate the complete plan in one paragraph, and only start building after I nod approval. Keep the skill itself as concise as possible, follow best practices.

A Skill is an asset, not a conversation. A well-crafted Skill essentially turns your workflow into a reusable operations manual. For similar tasks later, no need to re-explain everything from scratch. Team sharing: Organization Settings → Skills → Add, distribute to whole company.

Where to find existing ones: Official library, skills.sh, ClawHub, GitHub, and the strongest one – the one you built by sticking to your own workflow. For command-line users, a universal installer: npx skills add <skill-name>, installs into Claude Code / Cursor and 30+ other tools with one line.

Click your account in the bottom-left → Get apps and extensions: currently not as many as Codex, but will continue to grow.

A key point when writing your own Skill: The description at the beginning is the most important line – Claude uses it to decide at each startup “should this skill be triggered for this task?” So make it specific, with trigger keywords (e.g., “trigger when user says ‘publish to WeChat Official Account’ or ‘upload article’”).

⚠️ Three safety and hygiene rules:

• Never install a Skill from an untrusted source (real risk of backdoor token theft)
• Don’t have more than 10 Skills in the same project (they interfere with each other)
• Install Skills with specific business scenarios at project level, not globally (avoid false triggering)

Tool 3: Hooks

Pain point 1: Forgetting to save progress. Automate it:

I want to create a hook: Whenever a new feature is developed and tested successfully, automatically update the CLAUDE.md file with progress. Trigger mode: stop hook (auto-check at the end of each turn).

Pain point 2: AI says “done” but it’s not really done. Install an acceptance inspector:

Create a hook for me: perform a delivery acceptance check before you prepare to finish a task. Rule: If this turn modified code, config, or documentation, but hasn’t been tested, or if checking reveals remaining to-dos not completed – do not allow finishing, continue working until checks pass.

From now on, mandatory self-check before every shutdown – false “completion” claims will drop significantly. Nodes where hooks can be attached also include: before/after tool calls, before subagent finishes, before/after context compression.

Tool 4: Subagents

Create two subagents for me: The first called “Code Reviewer,” dedicated to finding bugs and quality issues in the code. The second called “Security Reviewer,” dedicated to checking the application for security vulnerabilities.

During creation it will ask a few choices. Recommended answers:

If it asks	Suggestion
Project or global?	Select project scope (only for this project)
Tool permissions?	Read-only for research-type ones; full access for ones that need to act
Model?	Sonnet is enough for checking tasks, saves quota
Where to save memory?	Follow the project, select project scope

After they are active: Run two reviewers in parallel to check the inspiration digest project, give me a combined report. Two AIs review simultaneously without interfering. When the report comes out, just say “fix the main issues” and wrap up.

⚠️ For every new subagent, let the team lead recognize it:

View all my existing subagents and their responsibilities, update the task routing rules in CLAUDE.md: what tasks are delegated to whom, what tasks are not delegated, and the judgment principles.

Want to ask something else while it’s working? Press Ctrl+; or type /btw to start a “side window” chat without interrupting the main task.

Tool 5: Connecting to the Outside World: Connectors / MCP / CLI

Three ways for AI to reach the outside world:

Interface	Nature	In Plain Language
Connector / MCP	Standardized tool sockets	Click to authorize: GitHub, Slack, Notion, Google Calendar (+ → Connectors)
API	Program-to-program	AI “calls a service to ask for data” on your behalf
CLI	Command-line tool	“Hands” installed on AI – each installation adds one capability

CLI is great for filling specific tool capabilities. When installing, just say: “Help me install XX CLI, and tell me each step.” Worth installing:

• Vercel CLI: Deploy and go live
• Google Workspace CLI: Operate Docs / Sheets / Gmail
• Firecrawl CLI: Crawl webpages
• ffmpeg / whisper: Video editing / speech-to-text – foundation of video pipelines

Mega Pack – Plugins: Install an Entire Role

A Skill is one workflow, a Plugin is a collection of workflows = a complete role (skills + hooks + MCP packaged together). For example, a marketing plugin includes brand alignment, campaign planning, competitor analysis, content creation – bringing in the entire marketing department. Entry: Input + → Plugins. Distinguish between official (by Anthropic) and third-party.

First must-install for beginners: Superpowers – a development best-practices package: brainstorm to pin down requirements → write plan → execute in batches → review. Before big projects, just say:

First use Superpowers to brainstorm, only sort out requirements, don’t write code.

It will ask one question at a time to pin down your vague requirements.

Which One Should I Use? Four-Line Judgment Criteria

You want to…	Use
Let Claude consistently complete a workflow	Write a Skill
Let Claude use an external service / tool	Install MCP / Connector
Let Claude automatically respond to an event	Configure a Hook
One-click install a combination of abilities	Find a Plugin

Classic combination: Automated WeChat Official Account posting = Skill handles workflow + MCP handles “search hot topics” capability.

Bonus: All the equipment you create is stored in the project’s .claude folder – all plain text files written in human language. Feel free to browse, modify, or copy to friends.

✅ Checkpoint 4: All five tools equipped. Your inspiration digest experienced a design upgrade + a dual-person quality check. These tools will serve the project long-term.

---

Part 3: Operations – Go Live, Collaboration, Pipeline

Hands-On 3: Publish a Shareable Online Version

Don’t rush to make your local app public. Your inspiration digest may contain things like:

• Local saved data
• RSS fetch logic
• Personal information
• API keys
• Test data

Runs locally ≠ Ready for the public web. Health check before going live.

3.1 Pre-Launch Health Check

Send this to Claude:

Don’t deploy yet. Please check if this project is suitable for publishing to the public web. Focus on:

1. Are there any local file reads/writes that will break online?
2. Any API keys, personal saves, account info that shouldn’t be public?
3. Will RSS fetching encounter CORS or network issues online?
4. Do we need a demo mode, access password, or hidden private data?
5. Give me a minimum viable publishable version plan.

3.2 Choose the Release Version

After the check, choose one according to the table:

Version	Suitable When
Demo	Show friends – only display UI and sample data
Private	Has personal info or saves – add access password before publishing
Official	Data, permissions, privacy all confirmed – can be long-term public

3.3 Deploy to Vercel

After choosing a version, say:

According to the go-live plan you just made, deploy this project to Vercel. If you need to install Vercel CLI, connect the Vercel plugin/MCP, or have me log in to authorize, tell me step by step. Before deploying, list which files will be changed.

Three deployment paths – follow the one your interface shows:

What’s in your interface	How Claude will deploy
Vercel Connector / MCP	Follow prompts to authorize
Neither	Use Vercel CLI instead
Want long-term management	GitHub + Vercel auto-deploy

3.4 How to Manually Install Vercel

• + → Connectors, look for Vercel or Netlify (both have free tiers) and connect/authorize. Not in the list? Skip it, go straight to step 2 – Claude will use the command-line tool to deploy instead, same result.

When authorizing, confirm the page is Vercel before clicking Allow:

After clicking Allow, you’ll jump back to Claude. If it doesn’t refresh the first few times, try again; finally, you’ll see it under Connectors management if successful.

✅ Checkpoint 5: Open that URL on your phone – not on your computer, but on the internet. Send it to a friend, with “I made this.” A very satisfying step: you’ve put a local tool onto the internet.

Hands-On 4: Three Sessions Collaborating Simultaneously

Core idea in one sentence: Don’t cram everything into the same session.

A single project can be split into parallel lines: local session continues to improve features, cloud session handles independent small changes, Chat handles copywriting and ideas.

A local session reads files directly from your computer folder; a cloud session runs on Anthropic’s servers – it needs a shared repository as a workspace – that’s why you need to put your project on GitHub first. Bonus: adds version management – roll back if things break.

4.1 Prep: Upload the Project to GitHub

1. github.com free registration → New repository, any name (e.g., ai-daily), choose Private, create.
2. Copy the repository URL, go back to your digest session and say:

   Upload this project to my GitHub repo: [paste URL]

3. It runs all the commands itself. Refresh GitHub – your project files are in the repository.

4.2 Three Sessions, Three Tasks, Fire Simultaneously

Session 1: Local Code (modify core feature) – Go back to your digest local session:

Add an “Export” feature to the inspiration library: With one click, export all my saved inspirations (including tags and dates) into a well-formatted Markdown file, convenient for me to organize into my note app.

Session 2: Cloud Code (modify a small independent feature) – Start a new session, switch environment from Local to Remote (Cloud), select the ai-daily repository:

Add a dark mode toggle to the digest page, remember the user’s choice. It works in the cloud – even if you close your laptop, it keeps running.

Session 3: Chat (don’t touch project files, only generate copy) – Start a new Chat conversation:

I just built a “Daily AI Inspiration Digest” app: auto-fetches AI news, one-click save with tags, generates shareable daily digest cards. I want to share it on social media. Help me write 3 different styles of posting copy: one storytelling, one listing features, one with a bit of humor.

In the session list: local sessions have a computer icon, cloud sessions have a cloud icon. For Git projects, each session gets an isolated copy (worktree), not editing the same file directly simultaneously.

✅ Checkpoint 6: All three tasks fired. Glance at the sidebar – the same project is now split into multiple lines advancing in parallel.

The key experience is not “waiting for AI to slowly finish” – it’s learning to split tasks and advance simultaneously.

4.3 Harvest Results

• Local export feature: Test it in preview, screenshot feedback if unsatisfied.
• Cloud dark mode: Click Create PR → view changes on GitHub → Merge merge → delete branch → go back to local session and say:

  Pull the latest code.

• 3 pieces of copy from Chat: Pick one and save it – use it directly for graduation/social media posts.

Bonus: After opening a PR, a CI status bar appears. Claude Code automatically monitors checks – if tests fail, it fixes them itself; if all green, it can even auto-merge.

Hands-On 5: Data Analysis – Worth the Price Even If You Don’t Build Apps

If you have Excel / CSV files, this section might be the quickest to put into use right away.

1. Create a new folder data-analysis, throw in any spreadsheet: bill export, store transactions, grade report… If you have none, have it generate practice data in step 3.

2. Start a new session and select the folder:

Browse this folder, understand the data, tell me in plain English: what is this data about, what fields are there, how is the data quality?

3. If no data, instead say:

Help me generate a simulated 12-month sales data CSV for a coffee shop, make it realistic. Then proceed with analysis.

4. The killer instruction:

Assume you are a McKinsey data analyst. This data is in your hands. You need to report to the boss. What are the 3 most valuable questions to answer? Just do the analysis directly. If you need to draw charts, draw them.

5. It will write scripts, install tools, and run charts on its own – you don’t need to know what pandas is. Finally:

Organize the analysis results into a one-to-two-page illustrated report, clean layout, save as an HTML file that I can directly share with others.

✅ Checkpoint 7: A report with charts now sits in your folder. For similar data in the future, you have at least a quick-start analysis workflow.

Hands-On 6: Cowork Automation – Let AI Deliver on Schedule

Everything before required you to be present to direct. After this battle, it works even when you’re away.

6.1 Cowork Getting Started + Topical Daily Digest

1. Cowork tab → Select a folder (like giving a work badge: only work here) → Throw your channel/account materials into it.
2. Send the task:

First read the materials in this folder to understand my content style and positioning. Then help me gather and organize recent hot topics in my field, Requirements: for each topic provide title direction, angle of attack, popularity basis, Finally, organize into a visually appealing topical daily digest. After it runs, a digest tailored to your style will appear in the folder.

Test its upper limit: Ask Cowork to design a product launch process, deliver 5 files – product positioning, competitor analysis, pricing anchors, brand guidelines (with colors and brand story), distribution channels + short video scripts, even AI image generation prompts. In the past, this kind of work often required team collaboration; now, AI can give you the first draft.

6.2 Scheduled Tasks: Auto-Deliver Every Day at 9 AM

Scheduled → New task → Create with Claude: Name “Daily Topical Digest,” description paste the instruction text above, workspace select your folder, model choose Sonnet, frequency Daily at 9 AM.

⚠️ Prerequisite for local scheduled tasks: computer must stay awake (turn off sleep – a sleeping computer can’t run them). A desktop or Mac Mini that never shuts down is standard for serious automation. Cloud-scheduled tasks (routines) bypass this limitation.

When creating with Claude, the interface may default to English. Just speak Chinese in