Writing · 2026-08-31 · BUILDING IN PUBLIC

Why I started turning AI workflows into skills

Turning one-off prompts into a way of working that can be repeated, explained and steadily improved.

I want to collect fewer and fewer prompts. What is really valuable is writing down a method you use again and again: when to call it, what context it needs, what judgements to make along the way, and what kind of result counts as good enough.

01 / A prompt is not a method

A prompt usually describes only “what to do this time”. It may work well in the current chat, but give it a different task or a different set of material and the results start to drift. We saved the sentence but not the judgement.

A workflow is closer to real work: first check whether the input is sufficient, then process the information in order, stop to check at the key points, and finally judge against clear criteria whether the output can be used.

What I want to keep is not a clever sentence, but a way of working that can be run again.

02 / The skeleton of a skill

I now write down four things first: what problem it solves, when it should be triggered, what steps it takes, and what evidence decides its quality. Leave out any one of the four and a skill easily degrades into just another long prompt.

  • PURPOSE: the specific problem to remove
  • TRIGGER: the situations it suits
  • METHOD: steps that are stable and checkable
  • PROOF: the criteria for a good result

03 / Treat a skill as a small product

If other people are going to use it, you cannot just throw the file over the wall. They need to know what it is good for and what it is not, what material to provide, and what they will get in the end.

So on my personal site, AI skills are presented as a type of work of their own: a clear one-line statement of value, the flow from input to output, real examples, and the version and current status. Not a decorative “I use AI” label, but a piece of work that can be understood and verified.

04 / Next: test it on real tasks

The next step is not more explanation but running it on a few real tasks and recording where a person has to fill in, where it fails and whether the results are stable. Only after these tests does it go from “a skill I designed” to “a skill others can use too”.

If you enjoyed this, you can buy me a coffee.Leave a tip

Comments

No comments yet.