You commented TOOLKIT, so here is the setup for all five. One reusable Skill, a scheduled task, a published artifact, a plugin, and the Chrome extension. Exact clicks, exact plan tiers, and the one mistake that breaks each one.
Part one set Claude up to know you. Part two sets Claude up to do work.
Part one covered the five that make Claude useful on day one: get the desktop app, fill out your personal instructions, connect your tools, learn to switch models, and create one Project. Do those first. Everything below assumes they are done.
These five turn Claude from something you talk to into something that runs. They are also the five people quit on, because every article explains what they are and never says where to click.
So this is the setup, not the explainer. Each one gets the exact path, the plan you need, your first real run, and the mistake that makes it fail. Where I already wrote the deep guide, I link it.
Do This Before Anything Else
Go to Settings, then Capabilities, then turn on "Code execution and file creation." One toggle, and it unlocks both Skills and Artifacts. If you go looking for Skills later and the option is greyed out or missing, this is why.
A Skill is a set of instructions Claude loads when the job matches. Write it once, stop re-explaining yourself. The full breakdown, plus how the skill-creator interview works, is in this guide. Here is only how to get one live.
What plan you need: Pro and up. Anthropic's help center also lists Free, but their developer docs say Pro and above for uploading your own, so assume you need Pro.
The route that needs no terminal. Anthropic's own tutorial. You build the Skill by talking to Claude.
Where Skills live: Customize, then Skills in the left sidebar. Direct link is claude.ai/settings/capabilities. For pre-built ones: Customize, Skills, the "+" button, then "Browse skills." Honest note, Anthropic keeps moving this menu and their own pages disagree. If Skills are not under Customize, check Settings, Capabilities.
If You Build The File Yourself
The file is SKILL.md, uppercase. A header with two required fields, name and description, then instructions and examples below. Name rules: max 64 characters, lowercase letters, numbers and hyphens only, and it cannot contain "anthropic" or "claude." In a zip, the folder inside must be named the same as the skill. Failed uploads are almost always one of three things: folder name does not match, no SKILL.md inside, or bad characters in the name.
If you use Claude Code: no menu, files only. Personal skills go in ~/.claude/skills/your-skill-name/SKILL.md, project skills in .claude/skills/your-skill-name/SKILL.md. Run one by typing /your-skill-name. Important: the folder name is what you type, not the name inside the file. Match them and save yourself an hour.
Your First Real Run
Ask for the thing, then watch the thinking
Do not test with "use my skill." Ask for the work like you normally would. "Draft the quarterly review for the Nolan account."
Then watch Claude's thinking for "Using [skill name]." That is your proof it fired. No line, no Skill, and you are just getting regular Claude with extra steps.
The Mistake That Makes It Fail
A vague description. Claude decides whether to load a Skill by reading it, so "helps with docs" never triggers and "creates quarterly business reviews from client data and past reports" does. Write it for the moment it should fire. Second: Skills do not sync. One uploaded to claude.ai does not exist in Claude Code or the API, and the reverse is true too. They are also private to your account, so you cannot hand one to a teammate.
This is the one that makes Claude work while you are asleep. It is called Scheduled tasks.
Read This Before You Go Looking
Scheduled tasks live inside Claude Cowork, not regular Chat. If you have been hunting in your normal chat window, that is why. Cowork is in beta. The pricing page says all paid plans, but it is rolling out starting with Max, so on Pro you may not see it yet. That is not you doing it wrong.
The exact path:
Shortcut: type /schedule in any Cowork session. More on schedule syntax and running unattended in this guide and this one.
Your frequency options are: hourly, daily, weekly, on weekdays, or manually. "Manually" means it only runs when you press the button, which is more useful than it sounds.
The part that surprises people: scheduled tasks run remotely. Your computer does not need to be awake and the app does not need to be open. That is different from what a lot of older posts still say. Close the laptop. It still runs.
The Scheduled page shows upcoming and past runs, and lets you edit, pause, resume, delete, or run a task now.
Your First Real Run
A daily briefing at 7am
Use Anthropic's own example: "Summarize Slack messages, emails, or calendar events from the past 24 hours." Set it to daily. That is the task that earns its keep in week one.
The other one worth building is a weekly report: "Compile data from Google Drive, spreadsheets, or connected tools into a formatted summary." Pick the day you actually need it, which is the morning before your team meeting, not Friday afternoon.
The Mistake That Makes It Fail
Writing the prompt like a note to yourself. Nobody is there to answer follow-up questions, so anything vague comes back useless. "Check my email" gives you nothing. "Summarize emails from the last 24 hours, group by sender, flag anything needing a reply today, under 200 words" gives you something you will read. Second: your tools have to already be connected. Test the prompt by hand once before you schedule it.
Building artifacts is covered in this guide and this one, so the build part here is short. The half nobody covers is publishing, which turns a thing on your screen into a link you send a client.
Plans, and read this one carefully:
| Plan | Can build artifacts | Can publish publicly |
|---|---|---|
| Free | Yes | Yes |
| Pro | Yes | Yes |
| Max | Yes | Yes |
| Team | Yes | No, sharing stays inside your organization |
| Enterprise | Yes | No, sharing stays inside your organization |
Yes, that is backwards. A Free account can publish, a company Enterprise account cannot. Also: artifacts are not on mobile. Do this one at a desktop.
To build one: click "Artifacts" in your sidebar, which signals what you are after. Describe the problem you want solved, not the code you want written. Let Claude ask clarifying questions. Then say "Can you create this for me?" It builds in a window right of the chat. Claude makes an artifact when content is significant and self-contained, usually over 15 lines, so tiny asks stay in the chat.
To publish it:
Whoever you send it to needs no Claude account to open it or use it. Two more buttons. "Customize" lets other Claude users remix it into their own copy while yours stays untouched. "Get embed code" appears after publishing so you can drop it into your own site, and it needs an "Allowed domains" list, comma separated.
The Mistake That Makes It Fail
This one is permanent. Unpublishing is one way. In Anthropic's words, once you unpublish an artifact, you cannot publish that same artifact again, and it permanently deletes all associated storage data. No undo. Unpublish to clean it up and want the link back, you build a brand new artifact and send everyone a new URL.
Your First Real Run
Build the thing you keep doing in a spreadsheet
A pricing calculator for your services. An intake form for new clients. A quiz that sorts people into the right package. Anything you rebuild by hand every time.
Build it, publish it, send the link to one person and watch them use it. That loop, build to link to someone else's screen, is the point. Most people build ten artifacts and send none.
Anthropic's own line is the clearest one: "Each plugin bundles skills, connectors, and sub-agents into a single package, so you get a ready-to-go setup from your first conversation."
If those three words blur together, here is the difference in one table. Keep it.
| Thing | What it is | Think of it as |
|---|---|---|
| Skill | One capability, one set of instructions | A single recipe |
| Connector | A link to an outside service | The pipe to your data |
| Plugin | Skills plus connectors plus sub-agents, bundled | The whole kit for a role |
What plan you need: any paid plan. Pro, Max, Team or Enterprise. Not on Free.
To install one in the app: open "Customize" in the left sidebar, select the "Plugins" tab, click "Browse plugins," then "Install." The Knowledge Work marketplace is added by default, so there is something to browse on day one. You can also add Anthropic's Financial Services, Legal, and Life Sciences marketplaces. To add one from GitHub: Customize, Plugins, Personal plugins, "+", "Add marketplace," "Add from a repository," then paste the repo or git URL. Full catalog at claude.com/plugins.
In Claude Code: type /plugin for a four-tab panel: Discover, Installed, Marketplaces, Errors. Install with /plugin install name@claude-plugins-official. Add a marketplace with /plugin marketplace add owner/repo. Install mid-session and you must run /reload-plugins or it sits there doing nothing.
Picking the right plugin for your job is covered in the plugins guide, and the ten I actually use are in this one.
Anthropic's Own Warning
Straight from their docs: "Anthropic doesn't control what MCP servers, files, or other software are included in plugins and can't verify that they work as intended." Translation: a plugin is someone else's code running with your access. Install from Anthropic's own marketplaces first. Pulling from a random GitHub repo, check who made it before connecting it to your email.
Your First Real Run
Install one, then immediately give it a real job
Install one plugin that matches what you did last Tuesday. Then, in a new conversation, hand it a real task from your week. Not a test question.
Judge it on one thing: did it change the output, or just sit there? A plugin that does not visibly change how Claude answers is not helping. Uninstall it, try another.
The Mistake That Makes It Fail
Installing eight in one afternoon. Every plugin adds instructions and tools to the pile, and when they overlap Claude gets less sure what you want, not more. Install one. Use it a week. Then add the second. And in Claude Code, if a fresh plugin seems dead, you forgot /reload-plugins.
This is the one almost nobody has installed, and it is the one that changes your day the most.
It is called Claude in Chrome. In the Chrome Web Store the listing is titled just "Claude," published by Anthropic. Check that publisher name before you install anything.
What you need: it is in beta, there is no waitlist, and it works on all paid plans: Pro, Max, Team and Enterprise. It is Chrome only. Not Edge, not Brave, not Arc, not mobile.
The install, start to finish:
What it can actually do, which is far more than "chat about this page":
Now the safety part, and do not skim it. This is an AI with hands on your browser. Anthropic built real limits in, and you should know where they are.
Where The Line Actually Sits
You remain accountable. For anything Claude publishes or sends for you, and for any purchase or financial transaction it makes. Guardrails reduce risk, they do not move responsibility. Anthropic's own advice is the rule: avoid financial transactions, password management, or anything involving sensitive personal data.
One note: browser interactions are more compute-intensive than regular chats, so the extension eats more of your limit. Save it for tasks worth the spend.
Your First Real Run
The form you have filled out forty times
Pick one of these three. Fill out an application you have typed the same answers into before. Enter information into your CRM after a call, the job everyone puts off until Friday. Or book something, a table, an appointment, a flight search.
Start with the CRM one. It is repetitive, the data is already in front of you, and you will know in ninety seconds whether this changes your week. Watch the first run all the way through. You are learning what it does well before you trust it with anything.
The Mistake That Makes It Fail
Being vague about which tab. It is a browser agent, so it needs to know where it is working. "Update the client record" is a coin flip. "On the tab with the Salesforce contact page open, update the notes field with the summary from our call" works. Name the page, the field, and what goes in it. Second: do not hand it a task on a page you are not watching. And if it stops to ask you something, that is the safety system working, not a bug.
Five things is a lot for one sitting, so do not do one sitting. Here is the order that gets you a win fastest.
One a day for a week and you are done. That puts you past where almost anyone gets with this tool, because most people stop at typing questions in a box.
If You Only Do One Thing
Install the Chrome extension and use it on one real task today. No plan confusion, no beta rollout to wait on, no file format to get right. Add to Chrome, pin it, and hand it the boring thing you were going to do anyway.
© 2026 Mariah Brunner. All rights reserved.