Part 3 of 6

Part 3 — Skills: Deterministic Helpers for Your Workspace

Created: April 19, 2026 | Modified: May 3, 2026

The workspace is set up and its always-on rules are pinned. What's still missing is the work the rules sit around — the procedures you keep retyping the prompt for, every time.

The reply you write twice a week. The brief you fill out for every new piece of work. The check that has to clear before a draft goes anywhere. Same shape every time, even when the topic changes.

Part 3 turns those into named procedures. Each one is a file you named and placed. Each gets a line in the Instructions so Cowork knows when to reach for it.

Two products, same word: Skills

Cowork calls this product feature Skills. That word collides with the Claude Code Skill, which is an on-disk procedure under a completely different mechanism. They are not the same thing. This article is about Cowork Skills, and uses procedure in the prose to keep the distinction clear when both products show up in the same paragraph. The Cowork surface name is Skills; the body of the work is procedures.

When you finish, the workspace has new files inside it: one for each kind of repeatable work you keep doing here. The Instructions has new lines underneath what was already there, each one pointing at one of those new files. Nothing got overwritten. The workspace grew by one branch.

What that buys you:

A short list of the kinds of work you actually repeat — drawn from your own answer, not from a template.
A file per kind of work, named and placed by you, that the workspace reaches for whenever that kind of work comes up.
An Instructions file you can still read end-to-end in under a minute, just with a few more ALWAYS read lines than it had before this run.

What counts as repeatable work

A way of summarizing something that lands on your desk. A way of starting a draft that always opens the same way. A check that runs before something is sent. A formatted view of input data. The kinds of repeatable work that show up in most workspaces are simpler than people expect, and you don't have to come in with a list — the walkthrough surfaces yours from the way you describe the work.

The line that matters: if you can sketch what the output should look like before you start, the work is repeatable. Same fields, same length, same flavour, even when the topic changes. If your one-sentence answer points at work that's actually open-ended, the prompt says so — that work belongs to a focused helper, not a procedure.

If the work you're trying to save is open-ended — different shape every time, no sketchable output — don't force it into a procedure here. Let the prompt send it to the focused-helpers walkthrough instead. Procedures earn their place by being repeatable; everything else belongs in a focused helper.

How to run it

Open the prompts panel and paste the Part 3 prompt. The model re-reads the Instructions file at the root and the slots you've already named. Then it asks you in one sentence what kind of work you find yourself doing here over and over, and walks you through the rest. You name everything. The model proposes shapes; you decide what each piece is called and where it lives — inside one of the slots you already named, at the root, or inside a new slot that earns its place.

Part 3 needs the Part 1 Instructions file at the workspace root. If the prompt finds no Instructions file, or finds one with no Part-1-shaped pointers, it stops and tells you to run Part 1 first. Run Part 1 before this prompt — don't try to run them together.

The always-on rules don't have to be in place. This walkthrough works fine without them. But if they're there, the new procedures get to assume them.

What happens when you paste the prompt

The shape is the same one the earlier walkthroughs use. This time it runs scoped to repeatable work, and you own four decisions inside it: the list, placement, naming, and the append.

It opens by quoting back what's already in place — the layout, the slot names, the files already there — so neither of you is guessing. Then it asks you, in one sentence, what kind of work you find yourself doing here over and over. Short or generic answers get a couple of short multiple-choice follow-ups; that's the only place the prompt pushes back, and it's pushing back on the answer, not on you.

The list. You get a plain-language list of the kinds of repeatable work this workspace seems to want to handle, drawn from your own answer. Each item is a sentence; no filenames yet. You drop one, rename one, add what's missing, or say it looks right. The list does not move forward until you say it does.

Placement. For each item, you decide where it fits — inside one of the slots you already named, at the root, or inside a new slot that earns its place because nothing existing covers it. You stay in charge of the layout. A new slot never gets added unilaterally; if one's needed, you say so, and you say what it's called.

Naming and contents. One item at a time. You name the file. The draft of what's inside gets pulled from the item's purpose plus everything the workspace already knows, and you read it before it lands. Accept it, edit it, or ask for a different angle. Nothing lands until you say go.

The append. When the files are written, you see the exact new lines proposed for the Instructions — one per file, in the same ALWAYS read [your-file] for [what-it-covers] shape. You approve them. They land underneath the existing lines; those stay exactly where they were.

When you're done

The workspace has more files in it than it did before this run, and the Instructions has a few more pointers. The trunk is unchanged at the top. Each new branch leads to a file you can open in a text editor and read.

A practical note for after the run. The first time you actually use one of these procedures on real work, you will probably want to tweak it. That's expected. Open the file, edit the part that didn't quite fit, save it. The next run reads the saved version. The workspace gets sharper the more you use it. The procedures are not meant to be perfect on day one.

Next: Part 4 — Helpers Built for Focused Work extends the same workspace with focused-attention helpers for the work that needs more than a procedure.

Part 3 of 6