Here's how Week 4 goes
Week 3 gave one lane of your work a place to live; this week you write down a method you repeat, so Claude runs it the same way every time.
Place or method?
A Project is the kitchen and a Skill is the recipe: one holds a lane of your work, the other holds a procedure you repeat.
Week 3 ended on the line: a Project is a place, a Skill is a method. Now you have both halves, and they stack. Sort these five; a couple are closer calls than they look.
Everything Claude should know about our GEAR UP program.
How we write the weekly partner update, step by step.
The board memo format with our headings and sign-off.
Background docs for the fall report cycle.
Our checklist for turning event notes into a recap email.
They stack: run a method inside a place and the window gets both.
Anatomy of a Skill
Write a Skill the way you'd train a new teammate: when to use it, what to do, one sample of good output, and what never happens.
Skills live with you: create one under Skills in your own Claude settings and Claude loads it when the task calls for it. The same four parts also work as a saved text block you paste in first; step 3 below shows where to keep it.
Write one
Claude writes the first draft of your Skill; you supply one finished example and then test what it hands back.
- Pick a task you've done three or more times the same way: the weekly update, the meeting summary, the recap email.
- Have Claude draft the Skill from your work, using the prompt below. You supply one raw input and the finished version you made from it.
- Save it where it lives: under Skills in your own Claude settings, or in a doc titled "Skill · [task name]" that you paste in at the start of each run.
- Test the draft on fresh input the Skill has never seen.
- Tighten the one step that came out wrong. It's usually something you do on autopilot and never thought to name.
- Run it once more. A method that works twice in a row is ready to share. (Once your team relies on it, the regression check keeps every later edit honest.)
Claude writes a better method from one worked example than from any description of the task.
Watch how I do this task, then write the instructions you'd give yourself to repeat it: when to use, steps, one example, guardrails. Here's my input and my finished version: [paste both]
This one turns event notes into a recap email. Fresh notes go in; the house format comes out. The event details below are invented for the demo.
Team: about 30 scholars came out for yesterday's welcome session at our partner campus, and the financial aid office joined for the back half. One follow-up before Friday: three scholars asked about DCTAG paperwork and we owe them a clear answer. Full notes are in the shared folder.
[Your name]
New notes went in and the recap came out on format, because you wrote the format down once.
Why DC CAP cares
One written method pays off twice: once on your own plate, and again the day a teammate loads it.
Brand content, from everyone
DC CAP set an FY27 goal: every staff member produces external brand content independently, using DC CAP templates plus AI. A Skill that carries the template is how you get there; write it once and the format shows up right every time.
The first pattern your team inherits
A method in your head helps you; a method written down helps whoever loads it next. Share yours with a teammate this week, or bring it to office hours. We are stronger when learning together.
What you can do now
Writing a method clearly enough to hand off is Description and Delegation in one move, and you just did both.
This is your own check that the week landed. It stays in your browser — nobody else sees it.
Under the hood
Skip it and lose nothing. Three questions people usually ask about skills.
What actually loads when a Skill runs?
When a Skill triggers, the method enters the window: trigger, steps, example, guardrails. (At larger scales Claude keeps only a Skill’s name and description in view and loads the body when the trigger fires; the effect is the same.) Claude reads it the way it reads everything else you load, which is why the writing matters: a vague step gets a vague result, and a missing guardrail never gets applied. A skill also takes space in the window, so keep it as short as the method allows.
Why does one example beat three sentences of description?
Week 0 covered the reason: Claude predicts from patterns, and an example is a pattern in its purest form. Three sentences describing your board memo ask Claude to imagine the result; one real memo shows the headings, the length, the sign-off, and a dozen things you'd never think to mention. When a Skill drifts off format, swap in a better example before you add more instructions.
When does a Skill go stale?
The moment the real process changes and the written method doesn't. A new template, a new approver, a new deadline: the Skill keeps producing last season's version, confidently, because Claude only sees what you wrote. The cure is the same curation habit you built for your Project in Week 3: give the Skill an owner, and reread it whenever the process it describes moves.