Borrowed Shelf, Handmade Seal

Here is the least flattering fact about everything I’ve shown you across six posts. Of all the leverage in these two codebases — the grilling that drags a real spec out of a vague ask, the domain model that makes everyone say it the same way twice, the door designed before the room, the test that refuses to mock the engine — almost none of it is mine. Nearly every skill doing that work came straight off the shelf, pinned by hash to one man’s public repo. Exactly one was made here, by hand. I could dress that up, and I won’t, because the asymmetry is the most honest thing in the series and it’s also the reason this post exists: working this way genuinely changed how I build, and the lion’s share of the credit is Matt Pocock’s. What follows is the last fundamental, and it isn’t write every skill yourself. It’s knowing which judgment is worth encoding by hand — and being straight about how much of the rest you simply borrow.

Start with the failure, because it’s one you’ve felt even if you never named it. Every project you open, you start cold. You explain to the AI what good looks like here — how you want the spec dragged out, what the words are allowed to mean, where the seam honestly falls — and it’s genuinely useful, and then the session ends and all of it evaporates. Next repo, next week, you say it again, slightly differently, and again after that. Your judgment is real and hard-won and it does not travel; it lives behind your eyes and gets re-typed forever. Kent Beck has a line Matt reaches for in the talk: invest in the design of the system every day. The usual reading is about the code. But the thing you re-explain every session is a design too — the design of how the work gets done — and a skill is just where you invest in that once instead of every morning. Encode the judgment, and it stops evaporating when the session closes. That is the entire mechanism. The rest of this post is only watching it pay out.

So watch it pay out — and notice it isn’t a number. How many skills sit in the lock file was never the point; where they come from is. Every one is pinned by hash to a single public repo — Matt Pocock’s mattpocock/skills — the way you’d pin any dependency. That shelf is maintained, just not by me: he prunes the skills that stopped earning their place and adds the ones he’s still working out, and each time I re-pin I inherit that curation wholesale — the improvements are his, I only pull them down. The ones I reach for most aren’t exotic — grill-with-docs, to-prd, to-issues, the plain pipeline that turns a vague idea into a spec into a stack of issues an agent can pick up, and writing-great-skills, the one that tells a skill worth keeping from a skill worth deleting. All his. Not one of them mine. The same shelf is load-bearing in tinkforge too — the PDF tool from part three, TypeScript compiled to WebAssembly, nothing like this blog — because judgment pinned from a public repo doesn’t care what it’s steering.

And the tightest proof is the one you’re standing inside. This post was shaped by writing-shape — the skill that grows an article block by block, argument before prose, this paragraph included. The series about the fundamentals was written by the fundamentals: every skill the posts were about, turned around and pointed back at the posts themselves. That isn’t a flourish; it’s the commit history.

Now the other half — the one skill that was never Matt’s. Scroll to the bottom of this post, past the sign-off, and you’ll find a small hand-drawn glyph and a line about some Kerala artifact: a fishing net, a peppercorn, a serpent grove. One has closed every post in this series, a different one each time. That seal is the only thing in the entire setup I made myself. It’s in no lock file, it came from no upstream, I wrote it here by hand — because the taste it encodes was too particular to borrow. A fresh typographic motif no earlier post has used; never an image; one Kerala domain per post; a rhyme with the theme only where the rhyme was already there, never forced. Nobody else was ever going to want that rule. It was mine. So I banked it, and now the AI draws the seal while I only decide whether it’s right — which means the byline of a series about encoding judgment is itself a piece of encoded judgment. The stamp at the foot of this post, when you reach it, was placed by the one skill here that answers to no one but me.

Put the two halves together — the borrowed shelf, the handmade seal — and you get the fundamental they add up to. It is not write every skill yourself; almost everything here is borrowed and one piece is mine, and pretending otherwise would be its own dishonesty. Most of your judgment, it turns out, someone has already encoded better than you would, and the move is to take it. The skill is discernment — knowing which taste is particular enough to hand-build, like my one seal; which to pull off the shelf; and, hardest of all, how to tell a good shelf-skill from a bad one. Matt has a name for the moment that last problem arrives: we had tutorial hell, then framework hell, and now, he says, skill hell — the pile is enormous and the new literacy is separating the good ones from the crud. And here the tail goes all the way into the mouth, because that literacy is itself a skill. Matt wrote it — writing-great-skills — and both my repos pin it. I’ve encoded all of this into a new skill in my repo called writing great skills, he says; use this skill to either improve your skills or write great new ones. A skill for judging and making skills, shipped as a skill, sitting in the same lock file as everything it judges. The thing that decides what to encode is itself encoded — and that judgment about judgment is the part most worth banking of all.

So that’s the capstone, and with it the series. The failure was the one hiding under all the others: your hard-won judgment doesn’t travel — you start every project cold, re-explaining your taste to the AI until the session ends and it evaporates. The fundamental is Kent Beck’s, invest in the design of the system every day, read one turn further out — invest in the process, once, so it stops being re-typed every morning. The skill is the skill of making skills, write-a-skill and writing-great-skills, the one that turns a hard-won lesson into a thing that compounds instead of fading. And the artifact is the plainest in the series: this very post, drafted block by block by writing-shape — a skill the series is about — and about to be signed by the one skill that lives in no lock file, the only thing here I made by hand.

Look back down the road one last time. Six posts, and each opened on a failure so old it predates all of this: no one knows what they want, so you grill it out of them; they will say the one idea six different ways, so you name it once and hold everyone to the name; the rooms get deep, so you hang small doors; the AI outruns your reading, so you let a test do the looking. Every fundamental was twenty years old. What was new was only where I put them — because each time I learned one, I stopped trusting myself to remember it and handed it to a skill. Grill me. Say it the same way twice. Design the door. Six posts, six skills, the same quiet move at the end of every one; and that move, not any single fundamental, was what the series was really about. The AI is the tactical programmer, the sergeant on the ground — the strategist is you, and it was always going to be you. The three words that surfaced under every post, and I keep it in a skill, were never a footnote. They were the thesis, said a little more plainly each time until there was nothing left to reveal.

So there is no next time; this is where the road ends. But the skills were never mine to keep. Matt’s are public — mattpocock/skills, the same shelf both my repos pin — and if you take one thing from seven posts, take the shelf, and then go looking for the one rule too particular for it to hold. The judgment worth encoding by hand is the judgment nobody else was ever going to have. Mine draws a small Kerala seal. Yours will be something else entirely. Go encode it.

The whole way of working across these seven posts is Matt Pocock’s — the talk that started it, the shelf of skills the AI actually runs, and more of my working day than I can neatly credit. His courses live at AI Hero, and the skills are open at mattpocock/skills. The fundamental is Kent Beck’s — invest in the design of the system every day — and the framing that you are the strategist while the AI works the ground is Matt’s, from the talk. The recursion, a skill that teaches you to write and judge skills, is his too, from a second talk — “The Missing Manual: How to Write Great Skills” — on navigating “skill hell” with a shared rubric for building skills worth pinning. The recursion — a series drafted by the very skills it documents — and the one seal that lives in no lock file are this project’s own; the rest of the shelf is his, and I only pinned it. The single hand-made skill is mine — and even it stands on his write-a-skill.

║▤▤▤║ Kasavu — the gold-bordered handloom cloth of Kerala, its fine zari border set onto the loom once and then reproduced down the entire length of the cloth, one throw of the shuttle after another.