Three posts, your voice
Everything you've filed, cashed in at once
Where: Claude Code · Folder: Create a new folder
called
blogStart a new Claude Code session
before you begin.
You have four things now: a voice on file (1.3), a look on file (1.4), a skillSkillA reusable bundle of know-how that Claude loads on demand. Lives as a folder of markdown on your machine; kicks in when the conversation matches what the skill is for. you wrote (2.1), and two you borrowed (3.3).
This exercise spends all of them in one go. Three posts, written in your voice, on a site in your design, live on the internet. Twenty-five minutes, and almost none of it is typing.
Step 1 — Write the posts
Prompt:
Using the blog-post skill, write three short posts — 300 words each, no more:
1. AI slop: why more output is not more value
2. AI tone: why everything now sounds the same
3. AI and cognitive load: what it takes off our plate, and what it quietly adds
Use my person.md so they sound like me, not like LinkedIn. Have an opinion in
each one. No "in today's fast-paced world", no bullet-point conclusions.
Read one. If it doesn't sound like you, say so — "the second one is too polished, I don't write in threes" — and let it fix all three.
Step 2 — Build the site
Prompt:
Now use the html-visual skill to build a simple blog site for these three
posts: an index page listing them and a page for each. Style it from my
design.md. Keep it to plain HTML and CSS — no framework, no build step.
Step 3 — Ship it
Prompt:
Deploy this to Vercel with the Vercel CLI so the link opens for anyone without
a login, then give me the URL.
Send it to someone. That's the exercise.
Stuck?Debrief — what should have happened
- Three posts that sound like you wrote them on a good day.
- A site you didn't design, in a style you defined once, four exercises ago.
- A public URL.
Count the briefs you didn't write. No tone-of-voice document for the writer, no style guide for the designer, no ticket for the person who deploys it. Four small files did that work, and they'll do it again tomorrow for something completely different.
This is the real argument for contextContextEverything Claude can see right now — the conversation so far, the files it has read, the tool results, the system instructions. A big text buffer the model reads end-to-end before every reply. files, and it only lands once you've watched them compound. The files are the leverage. The prompts are just the trigger.
Stuck?FAQ for this exercise
Q: The posts still sound generic. A: Your person.md is probably too thin on how you write — most people describe their job and stop. Add three lines about sentence length, what you never say, and something you're annoyed about. Then re-run. Voice comes from constraints, not adjectives.
Q: It ignored my design.md. A: Say so directly: "reread design.md and restyle — use the actual hex values." Skills carry their own opinions about styling, and sometimes they win. Being explicit settles it.
Q: Can I write about something real instead? A: Please do. Swap the three topics for three things you actually have an opinion about at work. The exercise is the machinery, not the subject.
Q: VercelVercelA hosting service. You give it a folder of code; it gives you back a public URL anyone in the world can open — turns your Snake game into a link you can text someone in seconds. is asking me to log in again. A: Same flow as Exercise 3.2 — check the browser tab it opened. If it stalls, ask Claude to run vercel login and walk you through it.
OptionalGo deeper
Point the same machinery at something with a deadline.
Read person.md and design.md. Take the last thing I wrote in Slack that I
had to think about — I'll paste it — and turn it into a short post for our
internal blog, in my voice, on a page styled from design.md.
Most people's best writing already exists, buried in messages they wrote once and never reused. This is a two-minute path from "I said something good in a thread" to "it's published."