AI Guides › Skills & Agents
Building A Skill Around A Repeated Mistake, Not A Repeated Task
By Nigel Guy · 3 min read
Most guidance on building skills starts from the task you do often. There's a second, quieter starting point that's often more valuable and much less used: the mistake you make often. Not the task itself, which might vary each time — the specific way it keeps going wrong, which tends to be far more consistent than the task is.
The rule: if you keep making the same kind of mistake across otherwise different tasks, build the skill around catching that mistake specifically — a targeted check applied broadly beats a broad skill that happens to include a check.
The mechanism
- Notice the pattern, not the instance. You forget to check a date format. You send a draft before rereading the last paragraph. You quote a figure without checking whether it's current. These show up across many different tasks — the task changes, the mistake doesn't.
- Name the mistake precisely. "I make errors sometimes" isn't a mistake you can build a check for. "I state a number as current without checking when I last verified it" is specific enough to write a rule against.
- Build the smallest possible check, not a full replacement for the task. A skill built around a mistake doesn't need to do the whole task — it needs to sit at the point where the mistake usually happens and interrupt it. A one-step check that catches the recurring error is more useful than a large skill that happens to include the check somewhere in the middle.
- Apply it across every task where the mistake occurs, not just the one that prompted you to notice it. The value of this approach is that one small, well-aimed skill can cover ground that several separate task-specific skills would each need to duplicate.
- Track whether the mistake actually stops. The test of a mistake-focused skill isn't whether it feels thorough — it's whether the specific error you built it to catch actually shows up less often afterwards. If it doesn't, the skill is checking the wrong thing, or checking it in the wrong place.
What to skip
Skip building a mistake-check so broad it tries to catch every possible error — that's just a vague quality-control skill wearing a more specific name, and it will miss the actual pattern by trying to cover everything. And skip assuming a mistake is fixed once you've built a skill for it; keep noticing whether it still happens, because some habits resurface in a slightly different shape the check doesn't cover.
Guardrails
- This approach depends on honestly noticing your own repeated mistakes, which is harder than it sounds — a colleague or a second read of your own past work is often more reliable at spotting the pattern than your own memory of it.
- A mistake-focused skill is a narrow tool by design. Don't let it grow into a general-purpose review skill without deliberately deciding to widen its scope — that's a different, bigger thing, and it deserves its own guardrails written fresh.
- Catching a mistake mechanically doesn't address why it keeps happening. If the same error keeps recurring despite the check, that's sometimes worth treating as a process or workload problem, not only a checklist problem.
All 751 AI guides · JulieMango plans from £17/mo