Learn Claude

Set up once

Use one account. Set defaults. Save one repeated workflow.

Sign in once
Use the same account on the devices and apps you already use.
Open Claude · Download Claude
Set personal instructions

Tell Claude your default tone, detail level, and when it should ask first.

Instructions guide
Help me write personal instructions for Claude. Ask me four questions, one at a time, about: - Tone and formality I prefer. - Detail level I want by default. - When you should ask before answering. - Topics or formats to avoid. After I answer, propose a short, edit-ready instructions block I can paste into Claude personalization. Keep it under 120 words.
Turn on Memory and search
Memory and chat search make follow-ups smarter because Claude can build on prior context.
Memory guide
Add one Connector
Pick the app you use most, review access, and test with non-sensitive material.
Directory
Save your first Project
Choose one repeated workflow and add its instructions, files, and examples.
Project guide

Use one loop everywhere

Context -> ask -> check -> improve -> save.

1

Context

Goal, audience, source material, examples, rules.

2

Ask

Request the exact output: email, table, slide outline, draft, design, or workflow.

3

Check

Look for missing facts, assumptions, weak wording, and wrong format.

4

Improve

Fix one weakness at a time instead of restarting.

5

Save

Turn repeated work into a Project, template, Artifact, or agentic workflow.

Context bundle

Use this context bundle before you answer.

Goal:
[WHAT I WANT DONE]

Source of truth:
[FILE, NOTE, WEBPAGE, MESSAGE, SCREENSHOT, FOLDER, OR APP]

Rules:
- Preserve: [WHAT SHOULD NOT CHANGE]
- Ask first before: [WHAT NEEDS MY APPROVAL]
- Avoid: [WHAT I DO NOT WANT]

Before finalizing, list anything uncertain, missing, or important for me to verify.

Fix vague prompts fast

Meeting note

Vague: "Summarize this meeting note."

Why weak: no audience, length, or source rules.

Fixed prompt

Summarize this meeting note for:
[AUDIENCE]

Source of truth:
[PASTE NOTE]

Goal:
Help me leave with decisions, owners, and deadlines.

Return:
- 3 decisions.
- 3 owners.
- 3 deadlines.
- Anything unclear before finalizing.

Code change

Vague: "Refactor this file."

Why weak: no scope, check command, or smallest-change rule.

Fixed prompt

Help me improve this file:
[FILE OR PATH]

Before editing:
- Explain the relevant part first.
- Propose the smallest useful change.
- Avoid unrelated cleanup.

After editing:
- Run the existing check command if one exists.
- Summarize what changed.
- Tell me what to test.

Turn a rough request into a strong prompt

Type normally. The builder adds context, output rules, and checks.

What do you want Claude to help you do?

Choose the Claude surface

Choose the output

Optimized prompt

Model fit Use Sonnet 4.6 for most everyday work. Switch only when the task clearly needs a different fit.

Haiku 4.5 Sonnet 4.6 Opus 4.7

Use the right Claude surface

Filter by task, then choose the simplest surface that fits.

Chat

Drafts, uploads, images

Quick questions, emails, file summaries, decisions, and image analysis.

Upload files Vision guide

Research

Multi-source answers

Several searches, connected context, and citations you can check.

Use Research When to use

Projects

Reusable workspaces

Saved files, examples, instructions, and role for repeated work.

Set up Project Instructions

Artifacts

Editable outputs

Documents, charts, prototypes, small apps, and visual previews.

Artifact docs

Claude Design

Visual drafts

One-pagers, prototypes, slides, and design-system-aware visuals.

Design guide

Word

Document review

Review, rewrite, and improve selected text while preserving structure.

Set up in Word Guide

PowerPoint

Deck building

Storyline, slide structure, clearer messages, and better visuals.

Set up in PowerPoint Guide

Excel

Workbook help

Formula explanations, table checks, analysis, and doc/deck handoff.

Set up in Excel Guide

Chrome

Website tasks

Reading, comparing, navigating, and low-risk browser workflows.

Set up Chrome

Connectors

Your apps and files

Trusted services Claude can search, use, or show as interactive app cards.

Set up connectors Interactive apps Directory

Mac + Windows

Desktop app

Daily work, app handoff, Cowork, and Mac quick entry when available.

Download app Desktop guide Mac quick entry

Claude Cowork

Longer desktop jobs

Selected files, folders, apps, progress visibility, and handoff.

Cowork guide

Plugins

Ready-made Cowork setups

Skills, connectors, and subagents bundled for a role or workflow.

Set up plugins Directory

iOS + Android

Mobile work

Voice, quick capture, messages, calendar help, and app handoff.

iOS guide Android guide Dictation

Claude Code

Repository work

Inspect first, make the smallest useful edit, then run checks.

Open Claude Code Guide

Chat or Project?

Chat for one-time work. Project when you would reuse the same background, files, or instructions.

Search or Research?

Search for one fact. Research when Claude needs several sources, citations, and synthesis.

Artifact or Design?

Artifact for an editable output. Design when the visual result itself needs polish.

Connector or Plugin?

Connector adds one trusted app. Plugin bundles a Cowork workflow with tools and skills.

Answer or thinking?

Answer for simple work. Thinking when reasoning, math, code, or tradeoffs need a careful path and final check.

Desktop or Mobile?

Desktop for longer work with files and apps. Mobile for voice, capture, messages, and quick review.

Cowork or Code?

Cowork for desktop files and apps. Code for repositories, tests, reviews, and pull requests.

Claude terms in plain English

Instructions

Saved preferences for tone, detail, format, and when Claude should ask first.

Project

A focused workspace for repeated work with the same files and instructions.

Artifact

An editable output such as a page, chart, document, prototype, or small app.

Research

A deeper mode where Claude searches across sources and returns citations you can check.

Context engineering

Giving Claude the goal, source material, examples, rules, and checks before it works.

Extended thinking

A deeper reasoning mode for hard problems. Use it when the answer needs a careful path, not just a quick draft.

Long context

Giving Claude a lot of source material at once. Ask it to map the sources before answering.

Connector

A controlled connection between Claude and another app or service.

Plugin

A Cowork add-on that can include skills, connectors, and subagents for a role or workflow.

Agentic workflow

A repeatable job with inputs, allowed actions, checks, and a handoff.

MCP

A standard way to connect Claude Code or agents to tools and data sources.

Hook

A command that can run before or after a Claude Code action, often for checks.

Set these three once

1

Skills

Skills teach Claude repeatable procedures for specialized work.

Try: browse the directory and add one Skill for a workflow you repeat.

Skills guide

2

Memory

Memory helps Claude carry useful context into future chats.

Try: ask Claude what it remembers, then correct anything stale.

Memory guide

3

Personal instructions

Instructions and styles set your default tone, depth, and output shape.

Try: set a concise default and add exceptions for high-stakes work.

Instructions guide

Skills, connectors, and plugins directory Styles guide

Repair weak answers instead of starting over

Pick the problem and copy the recovery prompt.

What went wrong?

Too vague

Use when: Claude sounds reasonable but you still cannot act on the answer.

Reliability check

Review your last answer before I rely on it.

Source of truth:
[PASTE OR NAME THE FILE, NOTE, WEBPAGE, OR MESSAGE]

For each important claim, label it as:
- Confirmed from the source of truth.
- Reasonable inference.
- Needs checking.

Then correct the answer if needed and give me a short review checklist.

When to trust, when to verify

Trust without extra checking

Tone shaping, summaries of material you provided, reformatting, brainstorming, and facts you can sanity-check at a glance.

Verify before relying

Numbers, names, dates, citations, code that touches data, medical / legal / financial decisions, and anything you will send or publish.

Verify with the source

Claims about specific people, organizations, products, or events; quotes; statistics; and anything outside your own material.

Use stronger patterns

Show advanced patterns

1

Artifact iteration

Ask Claude to revise the output itself: add a filter, simplify a page, make a dashboard, or add a toggle.

Try: "Keep the content, make it easier to scan, and add one useful control."

2

Reason, answer, check

For complex work, ask Claude to outline the steps, answer, then check against your rules.

Try: "Before answering, outline the steps. Then give the answer. End with what I should verify."

3

Meta-prompting

When the task is fuzzy, ask Claude to interview you before writing the final prompt.

Try: "Ask me the five questions needed to write the best prompt for this task."

Prompt architect

I want Claude to help me with this task:
[DESCRIBE THE TASK]

Do not answer yet.

First, ask me the five questions that would help you write the strongest prompt.
After I answer, create:
1. A clear reusable prompt.
2. A shorter quick-use version.
3. The best Claude surface for this workflow.
4. A checklist for judging the result.

Build agentic workflows safely

Start manually. Automate only after the goal, inputs, allowed actions, checks, and handoff are clear.

Show agent workflow builder

Beginner

Personal workflow

Use a Project, saved prompt, source files, and a review checklist.

Good for: meeting prep, document review, inbox drafting, research synthesis, Office work.

Builder

Repository workflow

Use Claude Code when files, tests, command-line checks, or pull requests matter.

Add later: custom commands, subagents, hooks, and MCP tools when they solve a repeated problem.

Developer

Hosted agent

Use the Agent SDK when the agent must run inside an app or service.

Require: narrow tools, logging, monitoring, and review before high-impact actions.

1

Job

What the agent helps with.

2

Trigger

When the workflow starts.

3

Inputs

Files, notes, sites, issues, or messages.

4

Actions

What it may do and what needs approval.

5

Checks

Facts, tests, errors, and review points.

6

Handoff

What you review before trusting it.

Agent workflow brief

Help me design an agentic workflow with Claude.

Workflow goal:
[WHAT THE AGENT SHOULD HELP WITH]

Trigger:
[WHEN IT STARTS]

Inputs it may use:
[FILES, NOTES, WEBSITES, ISSUES, MESSAGES, OR DATA]

Allowed actions:
[WHAT IT MAY DO]

Actions that need my approval:
[WHAT IT MUST ASK BEFORE DOING]

Return the simplest version, the safety checks, the handoff, and what would require Claude Code, GitHub Actions, or the Agent SDK later.

Agent SDK Hosting Tool use MCP GitHub Actions

Copy only what you need

The task picker and Prompt Lab cover most work. Use these for reusable workflows.

Research and connected apps

Help me set up Claude for this workflow.

Workflow:
[WHAT I WANT TO DO REPEATEDLY]

Sources or apps involved:
[FILES, DRIVE, EMAIL, CALENDAR, SLACK, LINEAR, DOCS, OR OTHER]

Research needed:
[WHAT CLAUDE MAY NEED TO INVESTIGATE]

Recommend:
1. Whether to use Chat, Project, Research, a connector, or a Cowork plugin.
2. The first connector or plugin to set up, if any.
3. The permissions I should review before connecting it.
4. A safe first test using non-sensitive material.
5. The exact prompt to use after setup.

Chrome or Cowork safety

Browser and desktop task

Help me run this task safely.

Goal:
[WHAT I WANT DONE]

Material Claude may use:
[FILES, FOLDERS, APPS, WEBSITES, OR CONNECTORS]

Rules:
- Explain your approach before acting.
- Ask before changing, sending, deleting, posting, sharing, or opening sensitive material.
- Stop if access, uncertainty, or risk is unclear.

Start by telling me the safest first action.

Claude Code

Help me with this codebase in plain English.

Task:
[DESCRIBE THE CHANGE]

Before editing:
- Explain the relevant part of the project.
- Identify the files likely involved.
- Tell me the smallest useful change.

While editing, avoid unrelated cleanup and run the relevant check if one exists.
Afterward, summarize what changed and what I should test.

What's new digest

I want to keep up with Claude without getting overwhelmed.

I will paste notes from official Claude release notes, documentation, or announcements.

For each change, tell me:
1. What changed in plain English.
2. Which everyday workflow it helps.
3. Whether a beginner should try it now, save it for later, or ignore it.
4. One safe test task I can use to learn it.

Focus only on practical use.

What changed and what to try

Updated Sunday and Wednesday at 11:00 PM Pacific from official sources.

May 6, 2026

Claude Code feels smoother in real terminals

What changed: Claude Code shipped a batch of reliability and terminal UX fixes—better pasting, less flaky resume/scroll behavior, and safer shutdown when sessions are interrupted.

Try it: update Claude Code, paste a screenshot, and run a short task; then use --resume to pick up where you left off.

Read Claude Code changelog

May 5, 2026

Ready-to-run finance agents arrive as plugins

What changed: Anthropic released ten agent templates for common financial-services workflows (for example: pitchbooks, KYC screening, and month-end close). They ship as plugins for Claude Cowork and Claude Code, with accompanying cookbooks for managed agents.

Try it: pick one template, run it on a non-sensitive sample document, then save your best prompt + checklist as a reusable workflow.

Read announcement

Apr 28, 2026

Creative work connects to more tools

What changed: connectors for Adobe Creative Cloud, Blender, Ableton, Autodesk Fusion, SketchUp, Splice, Affinity, and Resolume help Claude support design, 3D, audio, and production tasks.

Try it: ask Claude to walk you through one feature you have been avoiding in a creative tool, then turn it into a repeatable workflow.

Read announcement

Apr 17, 2026

Claude Design creates polished visual work

What changed: Claude Design helps create designs, prototypes, slides, one-pagers, and other visual drafts.

Try it: ask for a one-page visual summary, then ask Claude to improve clarity before polish.

Read release notes

Older updates

Apr 16, 2026

Opus 4.7 improves complex work

What changed: Opus 4.7 improves software engineering, long-running coding, and vision with images up to about 2,576 pixels on the long edge.

Try it: paste a dense screenshot of a chart, table, or dashboard and ask Claude to extract the data into a clean table.

Read launch note

Apr 9, 2026

Claude Cowork expands desktop help

What changed: Cowork became generally available on macOS and Windows through the Claude Desktop app.

Try it: start with a read-only folder review and approve changes only after Claude explains them.

Read release notes

Mar 25, 2026

Connectors can show interactive apps

What changed: some connectors now show live app cards and fullscreen interfaces inside Claude.

Try it: connect one trusted app, ask a real question, then use the interactive card before asking Claude to summarize the result.

Read guide

Last reviewed against official Claude and Anthropic sources on May 7, 2026.

Official courses and documentation

Use Anthropic courses for structured learning, then source docs for setup and verification.

Anthropic Academy hub Claude 101 Introduction to Claude Cowork AI Fluency Prompt engineering tutorial

Show source links

Learn Claude

Pick a Claude task

Email

Set up once

Sign in once

Set personal instructions

Turn on Memory and search

Add one Connector

Save your first Project

Use one loop everywhere

Context

Ask

Check

Improve

Save

Context bundle

Fix vague prompts fast

Meeting note

Fixed prompt

Code change

Fixed prompt

Turn a rough request into a strong prompt

Optimized prompt

Use the right Claude surface

Drafts, uploads, images

Multi-source answers

Reusable workspaces

Editable outputs

Visual drafts

Document review

Deck building

Workbook help

Website tasks

Your apps and files

Desktop app

Longer desktop jobs

Ready-made Cowork setups

Mobile work

Repository work

Chat or Project?

Search or Research?

Artifact or Design?

Connector or Plugin?

Answer or thinking?

Desktop or Mobile?

Cowork or Code?

Instructions

Project

Artifact

Research

Context engineering

Extended thinking

Long context

Connector

Plugin

Agentic workflow

MCP

Hook

Set these three once

Skills

Memory

Personal instructions

Repair weak answers instead of starting over

Too vague

Reliability check

When to trust, when to verify

Trust without extra checking

Verify before relying

Verify with the source

Use stronger patterns

Artifact iteration

Reason, answer, check

Meta-prompting

Prompt architect

Build agentic workflows safely

Personal workflow

Repository workflow

Hosted agent

Job

Trigger