I was on my third week of asking Cowork to draft the same weekly cashflow forecast email. Same recipients. Same four-bullet structure. Same tolerance for variance commentary. Every Monday I was retyping the rules. Every Monday the output drifted a little. Sometimes too long. Sometimes too analytical. Once, an extra section snuck in that I had to delete by hand.
I had been treating Claude like a smart intern who keeps forgetting where the supply closet is. That is a waste of both of us.
A Skill is a small file that tells the model exactly how to handle a specific kind of task. Once written, it loads automatically, and the output snaps to your standard every single time.
From the field, Q1 2026
A Skill fixes this. In plain terms, a Skill is a small file that tells the model exactly how to handle a specific kind of task. The format. The recipients. The rules. The edge cases. The "do this, never do that." Once written, the Skill loads automatically whenever I describe the task in natural language, and the output snaps to my standard every single time.
Three rules I follow before I build one.
Not every task deserves a Skill. The bar is simple.
- It has to repeat. If I will do this task at least monthly, it qualifies. Below that, the cost of writing the Skill is higher than the time it saves.
- It has to have a "right answer" I can describe. If I cannot write down the gold-standard output and the rules that produce it, neither can the model. Building a Skill forces me to clarify my own standard, which is itself a useful exercise.
- The cost of a bad version has to be real. Anywhere consistency matters more than creativity (board updates, lender reporting, monthly reporting packages, AP coding logic, SOPs) is Skill territory. Pure thinking work is not.
The test that decides whether the Skill is real.
Before I trust a new Skill, I run a small evaluation. I write three different ways a colleague might ask for the same task, then have the model produce the output twice for each: once with the Skill loaded and once without. Six outputs, side by side.
The results are almost always stark. Without the Skill, the model defaults to long, analytical, multi-section writing. With the Skill, the outputs converge on the format I actually want. The variance collapses.
What I have built so far, and what it cost.
In the last quarter I have converted the following into Skills: the monthly reporting email, the weekly cashflow forecast note, the monthly close SOP, the vendor credit application autofill, our standardized letterhead and slide template, and the quarterly business review structure. Each one took between thirty minutes and two hours to write and test.
Conservatively, those Skills now produce something I would have otherwise drafted or supervised at least four times a month. The hourly math works out, but the math is not actually the point.
The compounding part nobody talks about.
Here is what surprised me. Each Skill makes the next one easier to build, because the patterns rhyme. The structure of the cashflow Skill informed the structure of the reporting Skill. The reporting Skill informed the QBR Skill. My library is becoming a private operating system for how the finance function expresses itself in writing.
Even better, the Skills are inheritable. The next finance hire does not need to memorize my preferences. They invoke the Skill and the output meets the standard.
That is the shift. We stopped buying software and started writing it, in a form lightweight enough that a CFO can author it on a Saturday morning. The leverage is enormous and the moat is small, but it is the right size moat for our function. Standards expressed as code. Judgment retained by humans. Production handled by the agent workforce we are quietly assembling.
Where to start tomorrow.
Pick the next recurring task on your calendar. Write down the gold-standard version of the output and the five rules that govern it. That single document is the first draft of your first Skill. Build it. Test it against itself with and without. Then move to the next one.
A year from now your library will be the most valuable asset on your laptop. None of it will be on the balance sheet. All of it will be on the income statement.