LearnAI home

Getting Good ยท Lesson 6

Your Skill Library

Two or three skills that earn their place, and how to keep them working.

Small, and looked after

A library of two or three skills that you trust beats a shelf of a dozen you half remember. That is the whole argument of this lesson, and it runs against instinct, because building the first one is enjoyable and the obvious next move is to build five more.

Resist that. The constraint is not how many you can write. It is how many you can keep true. A skill is a claim about how you work, and claims go stale. Every one you own is a small standing commitment to check, occasionally, whether it is still right.

Why stale is worse than absent

A skill nobody has read in a year is not doing nothing. It is applying last year's rules to this year's work, silently, without appearing in the conversation to be argued with.

That is the specific danger of automation. A saved prompt you paste is visible: you see the outdated line as it goes in, and you notice. A skill applies itself. When the report format changed in March and your skill still says otherwise, the output is confidently wrong in a way that looks exactly like the output being confidently right.

The moment to review a skill is the moment the underlying process changes. If your team revises a format, a template or an approval step, ask yourself immediately whether a skill of yours still describes the old one.

Your capstone, built where it belongs

This site has no accounts and stores nothing you do. That is deliberate, and it means the artifact for this course lives on your own machine or your own drive, not here. Nothing you build below is saved by LearnAI, and nothing here needs to be.

Here is the build, and it takes a fortnight rather than an afternoon.

Run the tally from lesson 2 for two weeks. Every time you retype a preference you have typed before, mark it. At the end, you will have a short and slightly surprising list.

Take the single item with the most marks. Write one skill for it, using the template from lesson 4. Instructions first, description last, in the words you genuinely use.

Use it for a week without adding anything else. Watch when it fires and when it does not. Fix the description before you touch the instructions, and change one thing at a time.

Only then consider a second, and only if the tally justifies it. Most people find that two skills cover most of the repetition, and the third is where enthusiasm starts outrunning need.

Your library index, kept alongside the folders

SKILL LIBRARY

meeting-notes Job: writing up or tidying notes from any internal meeting Added: [date] Last reviewed: [date] Depends on: our three-section notes format Retire if: we stop using that format

board-papers Job: drafting or revising a paper for the monthly board Added: [date] Last reviewed: [date] Depends on: the one-page limit agreed by the board Retire if: the limit changes or the meeting stops

The "depends on" line is the useful part. It records what the skill is quietly assuming, so when that thing changes you know which file to open. Most stale skills are stale because nobody ever wrote the assumption down.

In plain English

Library:
The small set of skills you actually own, ideally two or three.
Stale skill:
One still applying rules that a changed process has made wrong.
Library index:
A one-page note recording each skill's job, its assumptions and when you last checked it.
Retire:
Removing a skill because the thing it encoded no longer applies.

The review, twice a year

It takes twenty minutes. Open each skill, read it as if somebody else wrote it, and ask three things: is this still what we do, has it earned its place since the last review, and is anything in it now wrong.

Retiring is the part people avoid, because deleting something you wrote feels like waste. It is not. The skill did its job for as long as the process it described was real. Keeping it past that point does not preserve the work, it just gives you an unreliable colleague you have stopped supervising.

Prompt you can copy: the library review

Here are the skills I currently have, pasted in full below.

For each one, tell me:

  1. Its job, in one sentence, in your words not mine
  2. Anything it assumes about how we work that it does not state
  3. Anywhere two of these skills give conflicting instructions
  4. Any rule inside it that is vague enough to be unusable

Do not suggest new skills. Do not suggest improvements yet. Just tell me what is there.

MY SKILLS: [paste each SKILL.md]

Prompt you can copy: retire or keep

This is a skill I wrote [how long ago]. Since then, [what has changed about the process, or nothing has].

Ask me the questions you would need answered to tell whether this skill is still correct. Ask them one at a time, and wait for my answer before the next one. At the end, give me a plain recommendation: keep as is, revise, or retire.

THE SKILL: [paste it]

That is the course. Finish the quiz and the Skill Smith badge is yours. If you leave with one working skill and the habit of retiring the ones that stop being true, you have taken everything this course had to give.

Checkpoint

Keep two or three skills you trust rather than a dozen you half remember. Record each one's job and what it assumes, review twice a year, and retire anything the process has outgrown. Build the library on your own machine, because this site stores nothing.

๐Ÿ“ Quiz

Question 1 of 4

Why is a small library of two or three skills better than a dozen?

Found this useful? Pass it on.