AI Guides › Step-by-step guides
By Nigel Guy · 6 min read
Most people assemble a content workflow by renting it: a template pack, a prompt bundle, a "content calendar" file that someone else designed around their own business. It looks finished on day one and fits badly by day ten, because none of it knows your voice, your audience or your habits. The fix is not a better template. It is three small things you build yourself, once, in Claude.
The rule: build the three layers separately. A Project holds what Claude should know, a Skill holds how a repeatable job is done, and an Artifact is a small tool you use. Mix them and you get one giant prompt that nobody maintains.
A note on scope. This guide was commissioned from a social-media post about someone's personal "content stack" of three self-built tools. We could not verify which three tools that author built, so we have not described them. What follows is the durable, checkable version: three builds you can make today with features Anthropic documents, using the exact menus as they stand at time of writing.
You need a Claude account. At time of writing, Projects, Skills and Artifacts are all available on the Free plan, though Free accounts are limited to five Projects, and some extras (such as recording a skill on Mac) need Pro or Max. Pro is listed at $20 a month, or $200 a year; that is about £15 a month at time of writing, but check the £ price at checkout because exchange rates and VAT move it.
Skills need "code execution" turned on, and Artifacts need "Cloud code execution and file creation" enabled in Settings. If a build below does not appear, check that setting first. Have ready: three to five pieces of your own writing you are proud of, a one-paragraph description of your audience, and one content job you repeat every week.
Use this to write the instructions, in a normal chat, before pasting the result into the Project.
You are an editor who specialises in turning loose notes into a clear house style guide.
Context: I write [CONTENT_TYPE, e.g. a weekly newsletter] for [AUDIENCE]. My goal is [GOAL, e.g. to get readers to reply]. My three samples are pasted below.
Task:
1. Read the samples and describe my voice in no more than eight bullet points (sentence length, tone, words I overuse, words I avoid).
2. Write a project instruction of under 300 words that tells an assistant how to write as me, what to ask me before drafting, and what never to do.
3. List anything you had to guess, so I can correct it.
Rules: do not invent facts about me or my business. If the samples are too few or too similar to judge, say so and ask for more instead of padding. Before answering, check that every bullet is supported by something in the samples.
SAMPLES:
[PASTE_SAMPLES]
Fill in the content type, audience, goal and samples. Keep the saved instructions short; put long reference material in the knowledge area instead.
A Skill is a folder with a SKILL.md file. Per Anthropic's help page, it needs front matter with a name (up to 64 characters) and a description (up to 200 characters) saying what it does and when to use it, followed by plain instructions.
SKILL.md inside a folder named for the skill.You are a technical writer who creates reusable instruction files for AI assistants.
Write a SKILL.md for this single job: [JOB_DESCRIPTION].
Inputs I will always give: [INPUTS]. A good result looks like: [WHAT_GOOD_LOOKS_LIKE]. Output format: [FORMAT, e.g. headline plus 300-word draft].
Requirements:
- Start with YAML front matter: a name (max 64 characters) and a description (max 200 characters) stating what it does and when to use it.
- Then give numbered steps, a short example of input and output, and a "do not" list.
- Instruct the assistant to ask for any missing input rather than guess.
- Cover one workflow only; if my description contains two jobs, tell me and split them.
Check the character limits before you answer and show the counts.
Fill in the job, inputs, what good looks like and the format. Keep one skill per job; a skill that tries to do everything gets triggered at the wrong moments.
Artifacts are self-contained things Claude makes in a side panel: documents, dashboards or small interactive tools. Anthropic says Claude creates one when the content is substantial and self-contained, typically over 15 lines. You can edit by asking in chat, store them in the Artifacts sidebar tab, and share them publicly or with named people.
You are a front-end developer building a small single-page tool for a non-technical user.
Build an interactive tool that does this: [TOOL_PURPOSE, e.g. a content planner with a column for each weekday and a status for each idea].
Fields I need: [FIELDS]. It should run entirely in the browser, work on a phone, and need no sign-up.
Constraints: plain, readable layout; no external services; no invented sample data other than clearly labelled placeholders. Ask me about anything unclear before building. When done, list what the tool cannot do.
Fill in the purpose and fields. Test it with real entries before relying on it. Anthropic notes stored artifact data is text only and capped at 20 MB per artifact, so this is a notebook, not a database.
description so it names the situation.