I added three short lessons to Automation Skills: comparing the full cost of an AI task, setting up a fair model test, and checking an AI report’s conclusion. Each has a simple example you can use on your own work. Open the Automation Skills course to try them.

If you want to compare AI models on work you actually need to finish, this kit gives you the prompts and a simple way to record the results. Inside: - The twelve video slides, with a short explanation under each one. - Six real prompts for films, a game, websites and data reports. - The meteorite data file and the cost and API-time results. - A blank comparison sheet and an extra website result where Opus cost less. Start with pack.pdf, then open START-HERE.md inside the file kit. Pick one task and give both models the same prompt and input files. Full walkthrough: https://youtu.be/cio-vM6DkEs What task would you use to compare two models?

Three short lessons are up in the Automation Skills course. Each one stands on its own, so start with whichever sounds like your problem: • When to start a new AI chat instead of carrying on in the old one • How to write a CLAUDE.md that does not eat your usage limit • How to make Claude Code's /compact keep the part you actually care about All three come from the same idea: an AI model has no memory, so the whole conversation is sent again every time you press Enter, and that is what your limit is actually being spent on. If any step in them does not make sense, say so underneath and I will rewrite it. That is more useful to me than a thank you.

This is everything from the video about what actually eats your usage limit: the ten slides with a written explanation under each one, the four prompts, and the one-page guide to writing a prompt properly. It is for you if you keep running out of limit in the middle of something and have never been told what is spending it. WHAT IS ATTACHED • The complete pack (PDF) — the ten slides explained so they make sense on their own, the four prompts with the words you need to change marked, a page on the four commands, and two pieces that are not in the video • The one-page prompt guide (PDF) — meant to be kept open beside you while you write a prompt • A starter CLAUDE file — a block you can copy into the file Claude Code reads at the start of every session in a project • All four prompts as plain text — easier to copy from than a PDF START HERE Start with the starter CLAUDE file, because it is the one that keeps paying. Open a folder you actually work in, make a new file there called CLAUDE, with the normal .md extension, paste the block in, replace the parts in angle brackets, and delete the lines that do not apply to you. Then open Claude Code in that folder and type /context. The number it shows against your memory files is what that file costs you on every single message. There is nothing to install and nothing to sign up for. Every command in here is already inside Claude Code. ONE QUESTION Full walkthrough, with all of it running live: https://youtu.be/bvaHxURbMK0 What is the point in your day where you notice the limit? Reply with it and I will tell you which of the seven fixes I would point at it first.

This is everything used to build a ten page website five different ways in Claude Code: one session, parallel sessions, subagents, agent teams, and a dynamic workflow. It is for you if you have only ever used one chat and want to know what the other four are, what they cost, and which one your own job actually needs. What is attached: - pack.pdf, the whole thing in one document. Nine diagrams with an explanation under each, all twelve prompts with the words you need to change marked, and one page that is not in the video - brief.md, the complete website brief. This is the file every prompt reads - prompts.md, the twelve prompts as plain text, easier to copy from than a PDF - site-check.md and site-check.js, the two files Claude wrote when it was asked to save a job as a reusable command. A real one, so you can see what a saved command is - finished-site.zip, the ten pages that came out the other end Start with brief.md. Make a new empty folder anywhere on your computer, put brief.md inside it, open that folder in Claude Code, and paste prompt 1 from the pack. That is the whole setup. There is nothing to install. If you have never done any of this before, do the first four prompts and stop. They build a working page in one conversation, and they are the part everything else is measured against. One question, and I would genuinely like to know: what is the job you keep doing by hand that you think might be too big for one chat? Reply with it and I will tell you which of the five I would point at it.