Free fiche
Nine modules on what actually drives your usage up. Every concept explained, then shown on my own setup: my files, my skills, my automations, and two months of real measurements.
230
sessions measured, July and August
19×
average re-reads of what goes in
0
lines of code needed to apply it
What happens when you hit Enter
The packet resent on every message, the cache and its single condition, the three layers. Including the exact one-hour cache rule that most articles get backwards.
Five habits
Why /compact costs more than /clear. Why switching models mid-session gets billed. And the measured proof that unplugging your external tools barely saves anything now.
The four places to store knowledge
CLAUDE.md, memory, skills, notes. What each costs and when. With my full stack, file by file, and one of my own instructions pointing at a folder that vanished months ago.
Building your own skills
The two ways the same file fires, the header that decides whether Claude finds it, and the naming trap that hid the most useful measuring tool from me.
Hooks
When they fire, what the program can answer, and one of mine in full. Plus three traps I hit, including the one that lied to Claude without anyone noticing.
Delegating to an agent
Why its forty pages never land in your conversation, and why delegating a short task costs more than doing it. With the measuring trap that doubles a bill.
With nobody watching
Running Claude on a schedule. The three rules of an automation that lasts, and the order that keeps you from dropping it after a week.
Measuring
The three built-in tools. And the table holding the least intuitive conclusion: cutting the bill and cutting the footprint are not the same moves.
Fourteen days, one move a day
The plan. The first three days change nothing, they measure. Every move takes a few minutes.
One email, one link, and it's yours. You also get Umami, my monthly newsletter on what I build and what it costs.
Is it really free?
Yes. You leave an email, you get the link. Nothing to pay, nothing to install.
Do I need to code?
No. Everything in the fiche is asked of Claude in plain language. The file excerpts are there so you recognise your own, not so you write them.
Where do the numbers come from?
From my own usage, 1 July to 24 August 2026, 230 sessions, measured continuously. The configuration excerpts are my real files, shortened for readability but never rewritten.
Does it apply to Claude in general, or only Claude Code?
The cache and context mechanics apply everywhere. Skills, hooks and agents are specific to Claude Code.