Master Claude Cowork: Automate Your Work with AI

Covers merging Chat/Cowork into one tab, connecting a folder with read/write permission, running your first prompt, and reading the Progress/Context panels.

Reading vs. editing a folder, renaming/deleting files, and the Manual/Automatic/Skip approval settings.

Scanning a messy folder and having Claude suggest categories and naming before touching anything.

Difference between account-wide instructions and folder/task-specific instructions, plus letting Claude write the CLAUDE.md for you

Creating a standard chat project, attaching files, instructions, and memory.

How a Cowork project links to a real folder on your computer, saves outputs directly, and differs from a chat project.

What connectors are, browsing the directory, connecting Gmail, and Always Allow / Needs Approval / Blocked settings.

Full example: scanning Gmail proposals, researching companies, and generating a Google Doc — then scheduling it weekly.

Installing Claude for Chrome, chatting with open tabs, controlling the browser, and chaining a multi-source research task into a PDF.

The "Teach Claude" feature, recording a workflow into a shortcut, and its limits — blocked sites, context limits.

Building a Gmail-based project that analyzes your sent emails to create a personal writing-style profile, then using a repeatable instruction file so every drafted reply sounds like you.

Creating a "pimp my inbox" instruction that scans recent emails, sorts them into labeled categories, and reports back — without ever deleting or sending anything.

Three more Gmail workflows: downloading attachments into organized Drive folders via the Chrome extension, building a spending tracker from receipts, and labeling unread promotional emails for review.

Three more Gmail workflows: downloading attachments into organized Drive folders via the Chrome extension, building a spending tracker from receipts, and labeling unread promotional emails for review.

Letting Claude tidy a messy Downloads folder into categorized subfolders with an index file, then a deeper example reading invoices and statements to generate monthly expense reports in Sheets and Excel.

Using the pptx skill to turn a company's style guide, overview, and financials into a fully branded PowerPoint deck that follows exact colors, fonts, and structure.

Converting a source document into a clickable flashcard artifact plus an Anki-ready export, then leveling it up with difficulty tags, distractors, and mixed question formats.

An introduction to Claude's artifacts — what they render, chat vs. Cowork creation, building a landing page and a quiz from scratch, editing versions, and publishing a shareable link.

Going beyond static artifacts to build a "live" employee directory tool that edits a real spreadsheet on your computer directly from an interactive interface.

Restricting Claude to only the links and documents you provide, so it never fills gaps with outside knowledge, then testing that it correctly refuses anything outside those sources.

Interviewing Claude to build a personal reference profile, then using it with the Chrome extension to fill out real forms accurately, flagging anything it can't confidently answer.

Understanding Claude Skills as the onboarding manual for an already-expert model — step-by-step instructions, tone, and context that work identically across Chat, Cowork, and Claude Code.

Browsing and installing skills from Anthropic's built-in directory, plus community marketplaces like SkillHub, Skillsmp, and Smithery — and how to bring any downloaded skill into Claude.

Breaking down a SKILL.md file's required structure — the always-loaded metadata, the step-by-step markdown instructions, and the bundled scripts, references, and assets loaded only when needed.

Testing how a skill's description determines whether Claude loads it automatically, comparing a plain request against an explicit skill call, and using a skill that runs local scripts.

Using Anthropic's built-in skill-creator skill and a five-part template to turn a plain-language idea into a working, installable skill.

Converting an earlier voice-profile-plus-instructions workflow into one proper skill, using an expanded nine-part template with a good example and things to avoid.

Combining the web-research, pdf, and social skills into a single daily-AI-brief skill that researches, writes, compiles a PDF, and hands off to another skill automatically.

Understanding plugins as a complete starter kit — skills, connectors, sub-agents, and hooks bundled for a whole role or workflow — and how they differ from a single skill.

Expanding an existing plugin with a new skill, a bundled style guide, and a frozen copy of another skill — versioning the plugin as it grows.

Building a real hooks.json entry that runs one skill automatically after another finishes, based on a condition like output length — true conditional automation, not just prompting.

Understanding how a sub-agent runs a step on a separate, constrained chat instance — with its own model and rules — to keep the main conversation's context window light.

Setting up a recurring task by chatting with Claude or filling in the form directly, understanding "runs on this computer," and reviewing a scheduled task's run history.

Understanding prompt-injection risk from files, skills, and plugins pulled from the web, the three approval levels, folder sandboxing, and best practices like duplicating sensitive files before you work on them.