AI Guides › Step-by-step guides
By Nigel Guy · 8 min read
Most people fix a disappointing Claude with longer prompts. The prompts improve and the routine doesn't: blank chat, same background pasted in, answer copied into another app, dull part done by hand. The problem isn't the prompt. Nothing you do in Claude carries over or reaches past the chat box.
The rule: switch on five things in a fixed order. First the task you repeat, then the task that should run without you, then the output you need to send, then a ready-made toolkit, and last the browser. Each step depends on the one before it, so skipping ahead hands more power to something you haven't tested.
| You need | Notes |
|---|---|
| A paid Claude plan for Steps 2, 4 and 5 | Scheduled tasks, plugins and Claude in Chrome are all on Pro, Max, Team and Enterprise only. Skills (Step 1) also work on Free. Sharing an artifact by link (Step 3) needs Pro or above. |
| The price | At time of writing Anthropic's pricing page lists Pro at US$20 a month, or US$17 a month billed annually ($200 paid up front), with tax excluded. UK buyers are charged in pounds at checkout. I couldn't confirm the exact £ figure on an official Anthropic page, so check your own checkout screen before you pay. |
| Claude on the web or Claude Desktop | You can't change sharing settings in the iOS or Android apps. |
| Google Chrome, for Step 5 only | Other Chromium browsers and mobile aren't supported. |
| One task you did at least twice last week | Name it before you start. |
Check which version you have. Anthropic is gradually rolling out a "new Claude experience" to Pro and Max. If your message box no longer shows separate Chat and Cowork options, you have it, and Step 2 is slightly different for you. That step tells you where.
A skill is a folder holding a SKILL.md file: a name, a description and the steps. Claude reads the description to decide when to load it, so a vague description means a skill that never fires.
If you use Claude Code, skip the zip. A personal skill lives at ~/.claude/skills/<skill-name>/SKILL.md and you can call it with /skill-name.
Fill in the task and how you do it today:
You are helping me turn a task I repeat into a reusable Claude skill.
The task: [TASK_I_REPEAT]
How I do it now, roughly: [MY_CURRENT_STEPS_OR_"NOT SURE"]
Files or formats involved: [FILES_AND_FORMATS]
Work in this order:
1. Before you draft anything, ask me: what starts this task, what a finished
result looks like, and one mistake that would make the result unusable.
Wait for my answers.
2. Write a SKILL.md. Keep the name under 64 characters. Keep the description
under 200 characters and make it name the situation that should trigger
the skill, using words I would actually type.
3. In the body, list the steps I really take in the order I take them, the
inputs each step needs, and the output format, with one short example.
4. List anything you assumed and couldn't confirm from my answers.
5. Show me the folder structure I should zip.
Rules: don't invent steps I didn't describe. Where something is unclear,
ask me rather than guess. Before you reply, check that the description
would make sense to someone who had never seen this conversation.
Test it today on real work. In a new chat, ask for the task in your normal words without naming the skill. If Claude ignores it, rewrite the description, not the body.
Schedule a task only after its skill has worked by hand. Each run is its own Cowork session with your skills, connectors and plugins, and it runs in the cloud, so your computer can be off.
/schedule inside any Cowork task.For the first fortnight, choose an approval mode that asks you before Claude acts. Stick to tasks that draft things or summarise them.
Text you paste into an email is stuck in your chat. Ask Claude to put the result in an artifact, such as a page, tracker or short report, and share that.
Viewers need a Claude account (legacy chat artifacts published with Publish are the exception). An artifact that pulls from connected apps uses the viewer's connections, not yours, and can't use Anyone with the link. To stop sharing, set it back to Only you.
A plugin bundles skills, connectors and sub-agents into one package. In Claude, open Customize in the left sidebar, then the Plugins tab, then Discover. Pick the one plugin that matches your job, such as marketing, finance or operations, and click Add. In classic Cowork, open the Cowork tab before you open Customize.
Plugins are saved to your account, so they follow you into chat, Cowork and Claude Code. Type / in the message box to see the skills a plugin added. Hooks and sub-agents only run in Cowork and Claude Code, so they appear greyed out in chat.
Use one plugin for a week before adding a second. If one of its skills overlaps with yours from Step 1, turn one off.
This is the step where Claude stops answering and starts clicking. That's why it comes last.
The side panel starts in Automatically approve mode. Claude carries on working, checks each action and stops to ask when it thinks something needs your approval. For your first week, switch to a mode that asks before acting. The side panel remembers the mode you choose. Start with one read-only job on a site you know well.
| Test | Pass looks like |
|---|---|
| Ask for your repeated task in a new chat, in everyday words | Claude uses your skill without you naming it |
| Open Scheduled after the first run | The run has a finished output and nothing was sent without you |
| Open your shared link in a private window, signed in as someone else | It opens with the access level you chose |
| Type / in the message box | Your plugin's skills are listed |
| Ask Claude in Chrome to read one page | It reads and reports back without changing anything |