AI

CLAUDE AI & COWORK

Micro-Level Mastery Notes

A complete deep-dive into Claude AI features,

Claude Cowork, Prompting, Memory, Connectors & More

[Link] Cowork Claude Code Prompting Memory Connectors

PREPARED FOR LEARNING

Claude AI Micro-Level Reference | April 2026 | [Link] Anthropic

CLAUDE AI & COWORK — MICRO-LEVEL MASTERY NOTES Page 2

TABLE OF CONTENTS

Claude AI — Platform Overview

01
Subscription plans, models, interface

[Link] Micro-Level Features

02
Memory, Projects, Artifacts, Styles, Settings

Prompting — Micro-Level Mastery

03
Techniques, structures, advanced patterns

Claude Cowork — Deep Dive

04
What it is, how it works, use cases

Cowork Features — Micro Level

05
Computer use, Dispatch, Connectors, Plugins

Claude Code vs Cowork

06
Comparison, when to use which

Connectors & MCP Ecosystem

07
Available integrations, MCP protocol

Memory System — Complete Guide

08
Chat memory, Projects, how to manage

Skills & Automation Workflows

09
Skills feature, recurring tasks, scheduling

Security, Limits & Best Practices

10
Safety, privacy, what to avoid

Quick Reference Cheat Sheet

11
Commands, tips, shortcuts at a glance

Anthropic Claude AI 2026 | [Link] | Research Preview Updated: April 2026

CLAUDE AI & COWORK — MICRO-LEVEL MASTERY NOTES Page 3

Claude AI — Platform Overview

01 Subscription plans, model families, and product ecosystem

Claude AI
Claude is Anthropic's AI assistant, available at [Link] as a web/mobile/desktop app. It is powered by the
Claude model family and built on Anthropic's Constitutional AI framework — trained with ethical principles baked
in, not just RLHF.

Subscription Plans
Plan Access Models Key Features Best For

Free [Link] Claude Sonnet Basic chat, limited messages, —

(limited) memory for all

Pro — $20/mo [Link] + mobile Sonnet 4.6 + Opus 5x more messages, Projects, Priority Recommended for power
access users

Max — [Link] + desktop All models + Cowork 20x usage, Cowork, early features For heavy daily users
$100-200/mo

Team Shared workspace All models Admin controls, team billing Small teams

Enterprise Custom deployment All models + custom 500k context, SSO, audit logs Large orgs

Model Family (2026)

Model Position Context Window Best Use Cases

Claude Haiku 4.5 Fastest, cheapest 200k tokens High-volume, speed-critical tasks, chatbots

Claude Sonnet 4.6 Best value (default) 1M tokens (beta) Coding, analysis, enterprise workflows — best daily
driver

Claude Opus 4.6 Most powerful 1M tokens + 128k output Complex reasoning, mission-critical, research, long
docs

Key Insight

Sonnet 4.6 hits 79.6% on SWE-bench (coding) and costs 5x less than Opus. For most everyday tasks,
Sonnet is the rational pick. Use Opus only when you need maximum depth — complex multi-document
reasoning, PhD-level science, or 50k+ word outputs.

Claude Product Ecosystem

[Link] — Main web/mobile interface. Claude Cowork — Desktop agent. Non-technical
Conversation-based. All plans. automation. Max plan.

Anthropic Claude AI 2026 | [Link] | Research Preview Updated: April 2026

CLAUDE AI & COWORK — MICRO-LEVEL MASTERY NOTES Page 4

Claude Code — Terminal CLI. Developer agentic Claude in Chrome — Browser extension for web
coding. Pro/Max. automation.

Claude for Excel — Spreadsheet AI add-in. Shares full Claude for PowerPoint — Slide deck AI add-in.
context.

Claude API — Developer access. All models. Pay per Claude Code Security — Codebase vulnerability
token. scanning. Enterprise.

Anthropic Claude AI 2026 | [Link] | Research Preview Updated: April 2026

CLAUDE AI & COWORK — MICRO-LEVEL MASTERY NOTES Page 5

[Link] Micro-Level Features

02 Every setting, feature, and option explained in depth

Inside [Link] — Complete Feature Map

1. Memory System
Claude now has persistent memory for ALL users (including free). This is a major 2026 upgrade.

Memory Type Available On What It Does

Chat Memory All plans Auto-catalogues preferences, facts, context from conversations. Used in
future chats.

User Preferences All plans Manually set tone, format style in Settings > User Preferences.

Projects (Active Memory) Pro/Max/Team/Enterprise Files you upload stay available across sessions for that project.

Memory Management All plans Settings > Memory — view all memories, edit, delete individual entries, or
wipe all.

How to Use Memory Effectively

• Tell Claude explicitly: "Remember I prefer bullet-point answers"

• Tell Claude your role: "I am a construction project manager in Chennai"
• Go to Settings > Memory to review what Claude has stored about you
• You can delete any memory you do not want Claude to retain
• Claude will use memories automatically in all future conversations

2. Projects Feature
Projects give Claude a persistent workspace. Upload files, set instructions, and Claude remembers everything
within that project across all conversations.

Create a Project: Click 'New Project' in sidebar Add files: PDFs, docs, images, code — all stay
available

Set project instructions: Custom system prompt for Share projects: Team plans allow shared project
that project only access

Use cases: Client work, research, ongoing writing, Best practice: One project per distinct ongoing task or
codebases client

3. Artifacts Feature
Artifacts are standalone outputs — code, documents, diagrams, interactive widgets — displayed in a side panel
that you can iterate on, copy, or download.

Anthropic Claude AI 2026 | [Link] | Research Preview Updated: April 2026

CLAUDE AI & COWORK — MICRO-LEVEL MASTERY NOTES Page 6

Artifact Type Description

Code / Scripts Python, JS, HTML, any language — runs with syntax highlighting

React Components Interactive UI widgets rendered live in the panel

Markdown Documents Formatted documents, reports, notes

SVG / Mermaid Diagrams, flowcharts, architectural drawings

HTML Pages Full web pages with CSS and JavaScript inline

Micro tip: Say 'Put this in an artifact' or 'Create a downloadable version' to force artifact creation. Artifacts
support persistent storage — they can save and retrieve data across sessions.

4. Writing Styles Feature

You can upload samples of your own writing and Claude will match your style. Go to Settings > Writing Style.
Claude adapts vocabulary, tone, sentence length, and formality to sound like you.

5. Deep Research Mode

For complex research questions, Claude can run extended multi-step web searches before answering. Activate
by asking for 'deep research' or enabling it in settings. Best for: market research, academic topics, competitive
analysis, news synthesis.

6. Extended Thinking / Reasoning Mode

Claude can toggle between fast responses and deep multi-step reasoning. For hard problems, say 'Think
through this step by step before answering' or 'Use extended thinking'. This causes Claude to reason internally
before giving the final answer — like showing its work.

7. Web Search
Enable/disable: Toggle in chat settings bar Works best for: Current events, prices, news, recent
info

Available to: All plans (free users got access May Micro tip: Say 'search for X' to force a search on any
2025) topic

8. Code Execution & File Creation

Claude can execute Python code, create downloadable files (Excel, Word, PDF, PPTX), and run computations.
Toggle in chat settings. This is what powers the file creation you see in the chat.

Anthropic Claude AI 2026 | [Link] | Research Preview Updated: April 2026

CLAUDE AI & COWORK — MICRO-LEVEL MASTERY NOTES Page 7

Prompting — Micro-Level Mastery

03 The exact techniques that separate average from expert Claude users

The Prompting Mindset Shift

Most users treat Claude like a search engine — type a question, get an answer. Expert users treat Claude like a
skilled employee. You give context, role, constraints, output format, and examples. The difference in output
quality is dramatic.

Core Prompting Principles

Principle How to Apply

Be Specific Vague prompts = vague answers. Say exactly what you need, how long, in what format.

Give Context Tell Claude who you are, what the document is for, who the audience is.

Assign a Role 'Act as a senior civil engineer' shapes Claude's vocabulary and depth.

Specify Format 'Respond in bullet points' / 'Give me a table' / 'Write in 3 paragraphs'

Use Examples Show Claude an example of the output style you want (few-shot prompting)

Iterate Claude remembers the conversation. Say 'make it shorter' or 'add more detail on X'

Chain Prompts Break complex tasks into steps. Ask for plan first, then execution.

Prompt Templates by Task Type

For Analysis Tasks:

Analysis Prompt Template

You are a [role]. Analyze the following [document/data] and: 1) Identify the top
3 issues, 2) Explain the root cause of each, 3) Suggest specific actionable
solutions. Format your response with headers for each issue. Length: 400 words
max.

For Writing Tasks:

Writing Prompt Template

Write a [document type] for [audience]. Tone: [formal/casual/technical]. Key

points to include: [point 1], [point 2], [point 3]. Avoid: [things to exclude].
Length: approximately [X] words. Here is an example of the style I want: [paste
example].

For Data/Excel Tasks:

Anthropic Claude AI 2026 | [Link] | Research Preview Updated: April 2026

CLAUDE AI & COWORK — MICRO-LEVEL MASTERY NOTES Page 8

Data Prompt Template

I have an Excel sheet with columns: [list columns]. Please: 1) Create a formula to
[calculate X], 2) The formula should handle empty cells, 3) Show me the formula
and explain each part. The sheet uses ■ for currency and Indian date format
DD/MM/YYYY.

Advanced Prompting Techniques

Technique How to Use It

Chain of Thought 'Think through this step by step before giving me the final answer'

Role Assignment 'You are a senior structural engineer reviewing a foundation plan...'

Constraint Setting 'Answer in under 100 words' / 'Use only information I provide'

Negative Examples 'Do NOT use jargon' / 'Avoid bullet points' / 'Don't repeat the question'

XML Tags 'Put your reasoning in tags and your answer in tags'

Persona Persistence In a Project, set instructions once and Claude maintains them every time

Output Scaffolding 'Use this exact structure: Section 1: X. Section 2: Y. Section 3: Z.'

Few-Shot Examples Show 2-3 examples of input-output pairs before your actual request

Context Window Use Paste entire documents — Claude handles up to 1M tokens (entire books)

Self-Critique Prompt 'After writing, review your answer and improve the weakest part'

Pro Tip: Contract-Style System Prompts

For Projects, write your project instructions like a short contract. Include: your role, the purpose of this
project, output format rules, what to always include, what to never do, and tone. Claude will follow this
contract every time.
Example: 'You are assisting a construction project manager in Chennai. Always respond in both
English and Tamil. Format all cost figures in Indian Rupees. When listing workers, always show
Mason/Helper counts separately. Never use US-centric examples.'

Anthropic Claude AI 2026 | [Link] | Research Preview Updated: April 2026

CLAUDE AI & COWORK — MICRO-LEVEL MASTERY NOTES Page 9

Claude Cowork — Deep Dive

04 The complete guide to Anthropic's desktop AI agent

What is Claude Cowork?

Claude Cowork is Anthropic's desktop AI agent for non-technical users. Released January 2026 as a research
preview, it brings the same agentic power as Claude Code (developer tool) into a graphical interface that
requires zero technical setup.

One-Line Definition

Regular Claude: "Here is how to do X." | Claude Cowork: "I did X for you."

Origin Story
Claude Code launched in late 2024 as a developer terminal tool. Anthropic noticed non-coders were using it for
tasks like vacation research, building slide decks, cleaning emails, and recovering files. Cowork was built in
approximately 2 weeks (mostly using Claude Code itself) to bring these capabilities to mainstream users. It is
powered by technology from Vercept AI, acquired by Anthropic.

Availability
Attribute Detail

Plan Required Max ($100-200/mo) — Cowork is exclusive to Max subscribers

Platform macOS (March 2026) + Windows (April 3, 2026) — fastest platform expansion in Anthropic history

Status Research Preview — not production-ready, expect errors, do not use with sensitive data

Access Users must opt-in and grant permission to each application Claude accesses

How Cowork Works — The Architecture

Cowork uses a tiered tool hierarchy — it does not jump straight to controlling your desktop. Claude tries tools
in this order:

Priority Tool Type Examples

Tier 1 (First) Connectors Slack, Google Calendar, Drive, CRM APIs — structured integrations

Tier 2 (Second) Claude in Chrome Browser extension for web-based tasks

Tier 3 (Last Resort) Computer Use Direct desktop control — screenshots + mouse/keyboard commands

The Computer Use loop: Observe (screenshot) → Plan (decide next action) → Act (click/type) → Verify
(check result) → repeat.

What Cowork Can Do

Anthropic Claude AI 2026 | [Link] | Research Preview Updated: April 2026

CLAUDE AI & COWORK — MICRO-LEVEL MASTERY NOTES Page 10

Organize and rename files in a folder Generate reports from local data

Batch convert file formats (PDF, DOCX, XLSX) Pull data from websites via Chrome

Fill out forms and interact with web apps Create presentations and documents

Send Slack messages or emails via connectors Schedule recurring tasks (e.g., weekly reports)

Control desktop apps (open, navigate, interact) Run tasks while you step away (Dispatch)

Cross-reference local files with web data Access Google Drive, Salesforce, DocuSign via
connectors

Built-in VM Isolation
Cowork runs in a built-in virtual machine for isolation — keeping AI operations sandboxed from your main
system. Claude asks for explicit permission before accessing new applications. Certain sensitive apps are
restricted by default.

Security Warning

A prompt injection vulnerability was demonstrated where a malicious file could cause Cowork to steal
other files. Anthropic has acknowledged this. Do NOT use Cowork alongside apps handling sensitive
data (banking, passwords, medical). This is still a research preview.

Anthropic Claude AI 2026 | [Link] | Research Preview Updated: April 2026

CLAUDE AI & COWORK — MICRO-LEVEL MASTERY NOTES Page 11

Cowork Features — Micro Level

05 Computer Use, Dispatch, Connectors, Plugins, Skills

Dispatch Feature
Dispatch lets you assign tasks from your phone and come back to find the work done on your desktop. It is
essentially an AI coworker that keeps working while you step away.

Dispatch Feature How to Use

Assign from mobile Open Claude iOS/Android app → Dispatch tab → describe the task

Claude executes on desktop Computer uses screenshots and actions to complete the task autonomously

Recurring tasks Schedule daily/weekly tasks: 'Every Monday, pull last week's expenses and summarize'

On-demand tasks One-time tasks: 'Download all invoices from Gmail and organize by month'

Persistent agent thread Pro and Max plans have a persistent thread to manage tasks from mobile or desktop

Computer Use — Micro Details

Capability What It Does

Screen reading Claude captures screenshots to understand the current visual state of your screen

Mouse control Clicks, right-clicks, drags — Claude can interact with any UI element it can see

Keyboard input Types text, uses keyboard shortcuts (Cmd+C, Ctrl+Z, etc.)

App navigation Opens apps, switches windows, navigates menus

Browser control Navigates websites, fills forms, extracts data, clicks buttons

Permission gating Claude asks permission before accessing each new app — you stay in control

Sensitive app blocks Certain apps (banking, passwords) are restricted by default for safety

Connectors Ecosystem
Connectors are structured integrations — they use direct APIs rather than screen scraping, making them faster
and more reliable than computer use. Available in the Cowork connector catalog:

Productivity: Slack, Google Drive*, Google Calendar*, CRM/Sales: Salesforce, HubSpot

Gmail*

Cloud: AWS Marketplace, Dropbox Document: DocuSign, Notion

Dev Tools: GitHub, Linear, Jira Automation: n8n, Zapier

Analytics: Honeycomb, Mixpanel Meetings: [Link], Zoom

Finance: QuickBooks, Stripe HR: BambooHR, Workday

* Google Calendar, Drive, Gmail connectors still in development as of April 2026. Use Chrome extension workaround.

Anthropic Claude AI 2026 | [Link] | Research Preview Updated: April 2026

CLAUDE AI & COWORK — MICRO-LEVEL MASTERY NOTES Page 12

Plugin Marketplace
Launched March 6, 2026, the Cowork Marketplace gives access to hundreds of reviewed plugins. Team and
Enterprise plans have admin controls to approve/restrict plugins. You can also add custom plugins for your own
tools.

Skills in Cowork
Anthropic has built Skills specifically for Cowork — instruction sets that enhance Claude's ability to create
specific types of outputs. Built-in Cowork skills include:

PPTX Skill: Creates professional PowerPoint DOCX Skill: Creates formatted Word documents
presentations

XLSX Skill: Builds complex Excel spreadsheets with PDF Skill: Generates and manipulates PDF files
formulas

File-reading Skill: Reads any uploaded file type Browser Skill: Advanced web automation patterns
intelligently

Customize Section
A new Customize section in Claude Desktop groups all Skills, Plugins, and Connectors in one place. Access it
from the sidebar. This is where you configure which apps Cowork can access and which skills to use for specific
tasks.

Anthropic Claude AI 2026 | [Link] | Research Preview Updated: April 2026

CLAUDE AI & COWORK — MICRO-LEVEL MASTERY NOTES Page 13

Claude Code vs Cowork

06 Understand the difference and when to use which

Side-by-Side Comparison
Feature Claude Code Claude Cowork

Interface Terminal (command line) GUI (graphical chat interface)

Target User Developers, technical users Non-developers, knowledge workers

Primary Use Coding, file systems, automation Non-coding tasks, office work, automation

Setup Requires [Link], npm install No setup — download and use

Computer Use Yes (auto mode in April 2026) Yes (research preview)

File Access Full filesystem via terminal Folder-level access with permission

Available On macOS, Windows, Web, iOS macOS, Windows (April 2026)

Plan Required Pro or Max Max only

Configuration [Link], [Link] files Customize section in desktop app

Connectors MCP protocol, custom Built-in connector library + custom

Best For Coding, build pipelines, devops Reports, files, email, research, slides

When to Use Which

• Use Claude Code if you are comfortable with a terminal, doing software development, automating
dev workflows, or need fine-grained filesystem control.
• Use Claude Cowork if you want a chat-style interface, doing office/knowledge work, organizing
files, creating documents, or automating browser tasks without coding.
• Both have Computer Use — but Cowork is the consumer-friendly version with a GUI and built-in
safety prompts.

Anthropic Claude AI 2026 | [Link] | Research Preview Updated: April 2026

CLAUDE AI & COWORK — MICRO-LEVEL MASTERY NOTES Page 14

Connectors & MCP Ecosystem

07 Model Context Protocol — how Claude integrates with everything

What is MCP?
MCP (Model Context Protocol) is Anthropic's open standard for connecting Claude to external tools and data
sources. Think of it as USB-C for AI — a universal connector. Any tool that implements MCP can connect to
Claude.

MCP Architecture
Component What It Is Example

MCP Server A tool/service that exposes its functions via the MCP Slack, Google Drive, custom database
protocol

MCP Client Claude acting as the caller that invokes server tools Claude reading your Slack messages

Web Connectors Run via browser-based APIs, lighter integration Most third-party SaaS tools

Desktop Extensions Run locally, deeper system access Local database, filesystem tools

Setting Up Connectors
In [Link]: Settings > Connected Apps > Add In Cowork: Customize section > Connectors > Enable
connector

Auth flow: OAuth for most services (click Connect, log Permissions: Choose read-only or read-write per app
in)

Testing: Ask Claude 'What can you see in my Slack?' Custom MCP: Developers can build and add custom
to verify MCP servers

Using Connectors in Chat

Once connected, Claude can use these apps automatically when relevant. You can also be explicit:

Connector Prompt Examples

"Search my Google Drive for the Q1 financial report"

"Send a message in the #project-updates Slack channel saying the deadline is
moved to Friday"
"Find all emails from supplier@[Link] in the last 30 days and summarize
them"
"Create a DocuSign envelope for the contract I just uploaded"

Anthropic Claude AI 2026 | [Link] | Research Preview Updated: April 2026

CLAUDE AI & COWORK — MICRO-LEVEL MASTERY NOTES Page 15

Memory System — Complete Guide

08 How Claude remembers you across conversations

The Memory Upgrade of 2026

In early March 2026, Anthropic pushed persistent memory to ALL users including free. Claude now retains your
name, communication style, writing preferences, and ongoing project context across separate conversations.
You start a new chat — Claude already knows who you are.

What Claude Can Remember

Memory Category What Claude Stores

Personal preferences How you like answers formatted, your preferred language/tone

Your role/context Your job, industry, location, tools you use

Ongoing projects What you are working on, who the stakeholders are

Past decisions Choices made in previous conversations it can reference

Writing style Your vocabulary, formality level, sentence structure (if you share samples)

Facts you shared Your name, company, team structure, important numbers

Managing Your Memory

View memories: Settings > Memory > See all stored Edit a memory: Click any memory to modify it
items

Delete one memory: Click delete next to any item Wipe all memories: Clear all button in Memory settings

Add a memory: Tell Claude 'Remember that I...' Stop storing: 'Don't remember this conversation'

Projects vs. Memory

Aspect Projects Memory

Scope Specific project only All conversations globally

Content Files, instructions, context for that project Your preferences, facts, style

Control You set project instructions manually Claude auto-extracts + you can edit

Use case Ongoing client work, codebases, research General communication preferences

Best Practice: Seed Your Memory

Anthropic Claude AI 2026 | [Link] | Research Preview Updated: April 2026

CLAUDE AI & COWORK — MICRO-LEVEL MASTERY NOTES Page 16

At the start of your Claude journey, have a setup conversation:

"My name is [Name]. I am a [role] in [city]. I work primarily on [type of work]. I
prefer [concise/detailed] responses. Always use [■/$/etc] for currency. My tools
include [Excel, WhatsApp, etc]. Remember all of this."

Anthropic Claude AI 2026 | [Link] | Research Preview Updated: April 2026

CLAUDE AI & COWORK — MICRO-LEVEL MASTERY NOTES Page 17

Skills & Automation Workflows

09 Recurring tasks, scheduling, and the Skills framework

Skills Framework
Skills are instruction sets that Claude loads to perform specific types of tasks at expert level. Launched October
2025, Skills can be built-in (by Anthropic), user-provided, or custom-built. They live in markdown files
([Link]) that Claude reads before starting a task.

Scheduling & Recurring Tasks

Cowork now supports both recurring and on-demand scheduled tasks. This is one of the most powerful
micro-level features for productivity automation.

Task Type Example Command

Daily digest 'Every morning at 8am, summarize my unread emails and create a priority list'

Weekly report 'Every Friday at 5pm, pull this week's work log and generate a PDF summary'

Monthly backup 'On the 1st of each month, archive last month's files to the Backup folder'

On-demand batch 'Process all new receipts in the Downloads folder right now'

Monitoring task 'Check the project folder every hour and alert me if any file changes'

Building Your Personal Automation Workflows

Step 1: Identify repetitive tasks you do weekly Step 2: Describe the task to Cowork in plain English

Step 3: Let Claude run it once and verify the output Step 4: Schedule it as recurring if it looks right

Step 5: Monitor from mobile via Dispatch Step 6: Refine the instruction if output needs adjusting

Automation Ideas for Non-Technical Users

• File organization: Sort downloads by type/date every Sunday

• Report generation: Pull WhatsApp data weekly and format into a tracker
• Email management: Flag and summarize all unread emails > 3 days old
• Research digest: Search for industry news daily and save to a folder
• Document conversion: Convert all new PDFs in a folder to Word format
• Data entry: Extract text from images/PDFs and paste into Excel

Anthropic Claude AI 2026 | [Link] | Research Preview Updated: April 2026

CLAUDE AI & COWORK — MICRO-LEVEL MASTERY NOTES Page 18

Security, Limits & Best Practices

10 Stay safe, stay efficient, avoid common mistakes

Security Guidelines
Rule Why It Matters

Never share passwords Do not paste passwords, API keys, or secrets into Claude chat

Cowork + sensitive apps Do NOT use Cowork alongside banking, password managers, or medical apps

Prompt injection risk A malicious file could instruct Claude to take unintended actions — be careful what you ask
Cowork to open

Research preview caution Cowork is not production-ready. Always verify outputs before acting on them

Permission review Regularly audit which apps Claude has access to in Settings > Connected Apps

Data in cloud Your conversations may be used to improve Claude (see Privacy settings to opt out)

Plan Limits to Know

Limit Type Amount Notes

Free Limited messages per day (exact count varies) Claude Sonnet

Pro ($20/mo) 5x more messages than Free Sonnet 4.6 + Opus (limited)

Max ($100-200/mo) 20x usage, unlimited Cowork All models + Cowork + early features

Context window Sonnet: 200k (1M beta) | Opus: 1M tokens Approximately 750k words

Output limit Standard: 64k tokens | Batch API: 300k tokens Opus has 128k output per response

Common Mistakes to Avoid

Vague prompts without context Not iterating — accepting first output without refining

Not using Projects for ongoing work Forgetting to enable memory / not seeding preferences

Using Cowork for sensitive data Running Cowork without verifying outputs first

Not using web search for current info Relying on Claude's training data for recent events

Anthropic Claude AI 2026 | [Link] | Research Preview Updated: April 2026

CLAUDE AI & COWORK — MICRO-LEVEL MASTERY NOTES Page 19

Quick Reference Cheat Sheet

11 Commands, prompts, and micro-tips at a glance

Essential Prompt Starters

Prompt Starter Effect

"Act as a [role] and..." Activates domain expertise in Claude's responses

"Think step by step..." Triggers extended reasoning for complex problems

"In a table, compare..." Forces structured comparative output

"Give me 3 options for..." Gets alternatives instead of single answer

"Critique your last response" Makes Claude self-review and improve

"Remember that I prefer..." Adds to your persistent memory

"Search for the latest..." Forces a web search for current information

"Put this in an artifact" Creates downloadable/renderable content

"Make it shorter/longer" Iterates on previous output length

"Explain this like I am 12" Simplifies complex topics to plain language

Cowork Quick Commands

Command Pattern What It Does

Organize my [folder] by [rule] File organization task

Every [day/week], do [task] Recurring automation setup

Convert all [files] in [folder] to [format] Batch file conversion

Search [website] for [info] and save to [file] Web research + save

Open [app], navigate to [section], and [action] Computer use task

Pull data from my [connector] and create a report Connector + document creation

Keyboard Shortcuts & UI Tips

New chat: Click '+' or Cmd/Ctrl + N Search past chats: Cmd/Ctrl + K or search bar

Upload files: Click paperclip or drag-and-drop Toggle web search: Click globe icon in chat bar

Enable code execution: Click tools icon in chat Switch models: Click model name at top of chat

View memories: Settings > Memory Manage projects: Left sidebar > Projects

Model Selection Quick Guide

Anthropic Claude AI 2026 | [Link] | Research Preview Updated: April 2026

CLAUDE AI & COWORK — MICRO-LEVEL MASTERY NOTES Page 20

Task Type Best Model Why

Quick questions, chatting Haiku 4.5 Fastest, cheapest

Coding, analysis, daily work Sonnet 4.6 Best value — default choice

Complex research, long docs Opus 4.6 Maximum capability

Cowork tasks Opus 4.6 (auto) Cowork selects automatically

The Claude Mastery Hierarchy

Level What You Do Result

Level 1 — Beginner Type questions like a search engine Getting answers

Level 2 — User Use Projects, memory, and file uploads Getting better answers

Level 3 — Power User Write structured prompts with role, format, constraints Getting the right answers

Level 4 — Expert Chain prompts, use connectors, automate workflows Getting work done automatically

Level 5 — Master Build custom skills, schedule agents, use Dispatch AI doing your work while you sleep

Final Key Insight

Claude is not a chatbot. It is a thinking partner, a document worker, a file organizer, and increasingly
an autonomous agent. The users getting 10x value are not those with the best prompts — they are the
ones who have set up Projects, seeded their memory, connected their tools, and scheduled recurring
tasks. Start with one workflow. Automate it. Then build from there.

Official Resources
[Link]: [Link] API Docs: [Link]

Prompt Engineering Guide: Support: [Link]

[Link]/prompting

Release Notes: [Link]/updates/anthropic/claude Cowork Tutorial:

[Link]/tutorial/claude-cowork-tutorial

Anthropic Claude AI 2026 | [Link] | Research Preview Updated: April 2026