Stop Prompting. Start Building Skills.

I produced two Business Requirement Documents in a single day last week. Reviewed them, made adjustments, shipped them. Then I thought: I'm going to be creating more of these. So I told Claude to create a skill for it.

That's the workflow. Do something manually. Do it well. Then turn it into a repeatable, standardised skill so you never have to think about the structure again.

A skill is a markdown file. That's it. It lives in a folder on your machine, and Claude knows to go to that folder when you trigger it.

Inside the markdown file is a workflow: what to ask for, what to produce, what structure to follow, what persona to adopt when reviewing the output. When you invoke the skill, Claude reads the file and follows the steps.

No code. No plugins. No complex setup. A markdown file in a folder.

I don't sit down and architect a skill from scratch. That's backwards.

I do the work first. I produce a report, a presentation, a document — whatever the deliverable is. I get it to a standard I'm happy with. Then I hand Claude the final output and say: "This is the thing I created. This is the structure. Create a skill for it so it's standardised."

Claude looks at what I produced and reverse-engineers the workflow. It captures the structure, the headings, the decision points, the review stages. And now I have a skill I can trigger with a single command.

The key is giving it a real example of the finished product, not a theoretical brief.

My BRD skill isn't just "produce a document." It's a full workflow:

Gather inputs. What domain source material exists? What notes, transcripts, policies?

Draft. Produce the BRD using a defined structure — all the headings, all the sections, in the right order.

BA Manager Review. Claude adopts a reviewer persona and critiques its own work. This is the analyse-review-adjust stage. It's not just generating content — it's stress-testing the output before I even look at it.

Cross-check. Verify that every piece of source material is accounted for in the final document. Nothing dropped. Nothing invented.

This entire workflow lives in one skill file. One trigger. One output.

My workshop report skill is split across multiple sub-skills because it produces 40-80 pages of output. If that was one giant skill, it would consume your entire context window. So it's broken into discrete stages, each handled by a sub-agent.

But the BRD skill? One file. I just want one output to review. I don't need a draft stage and a final stage as separate interactions. So I keep it together.

The rule is simple: if the output is so large it'll choke your context window, split it. If you just want one clean deliverable to review, keep it in one file.

Once you've created the skill, have Claude review it for token efficiency.

I ran this exercise live — told Claude to analyse my BRD skill and find ways to compress it. The plan came back with a 60% reduction. The inline template was 210 lines of full markdown. The model doesn't need a rendered template to produce a document — it needs a 20-line structural outline. That's 1,800-2,000 tokens saved on every single invocation.

If you're on a subscription, that's the difference between staying on your current tier and upgrading. If you're on the API, that's real money on every single call.

Every time I find myself doing something more than twice, it becomes a skill. Presentations. Social media posts. Carousels. Business requirements. Workshop reports. Email searches.

Each skill takes maybe 20 minutes to create. Each one saves hours every time it runs. And because they're just markdown files, they're easy to update, easy to share, and easy to version.

This is how a solo operator scales without hiring. Not by working more hours. By encoding your expertise into reusable workflows that execute at machine speed.

Build the skill once. Run it forever.

Enjoy this post, refer new subscribers and get the Time Flux Method - what I use to help people save 2+ hours a day.