When a Skill Beats a Prompt (and When It Does Not)
8 min read
A skill costs an hour to set up properly. A prompt costs a minute. That price difference means the honest question for any task is which rung of the ladder it deserves, and you can answer it with four questions.
The four-question test
Run any recurring headache through this checklist and count the yeses:
Score one point for each yes:
1. Repetition: do I do this weekly or more often?
2. Stable context: do the facts behind it (prices, services,
tone, policies) hold still for months at a time?
3. Multi-job scope: does this task share its background with
two or more related tasks?
4. Reused format: do I want the output in the same layout
every time?
3 or 4 points: install or build a skill.
2 points: save a master prompt, revisit in a month.
0 or 1 point: a one-off prompt is enough.Repetition matters because setup time has to be repaid, and a weekly task repays it in a month. Stable context matters because a skill is a written record: it stays right only as long as the facts do. Multi-job scope is the quiet one. If quoting, follow-ups, and scope questions all draw on the same rate card and the same service list, one skill briefs all three at once. And reused formats are where skills shine, because a format described once in the file comes back identical every run.
Where a prompt wins
The test cuts both ways, and three situations fail it for good reason:
- One-off tasks. A letter you will write once does not repay an hour of setup. Write one careful prompt, paste in the relevant documents, and move on.
- Fast-changing context. A skill with a stale price list is worse than no skill, because it is confidently wrong. If the facts shift every few days, keep them in the prompt, where you retype the current version each time.
- Fresh-judgment jobs. A sensitive staff conversation, a negotiation, a decision about whether to fire a client. The value there is thinking through this specific situation, and canned rules add little. Use the AI as a sounding board with a prompt, and keep your judgment in charge.
A worked example: three tasks, one plumbing company
Say you run a two-van plumbing company in Hamilton. Score three tasks from a normal week.
Quoting jobs from site-visit notes. Weekly or more? Yes, several times a week. Stable context? Yes: your rates, your standard exclusions, your deposit terms. Multi-job scope? Yes: quotes, follow-up nudges, and "is this in scope?" replies all lean on the same rate card. Reused format? Yes, you want every quote in the same layout so customers can compare and you can skim. Four points. This is exactly the job a proposal-writing skill exists for: it stores the rate card once and prices from it, or flags the gap.
Replying to Google reviews. Repetition? In a busy season, weekly, so call it a yes. Stable context? Yes: your tone, your policy for unhappy customers, your sign-off. Multi-job scope? Yes, it pairs with asking happy customers for reviews in the first place. Reused format? Mostly, a short reply with a consistent shape. Three to four points, so a review reply skill earns its setup hour.
A letter to the city about a permit dispute. Repetition? Once, with luck. Stable context, shared scope, reused format? None of it applies. One point at most. One good prompt with the permit documents pasted in, one careful read of the draft, done.
The two-point middle
Plenty of tasks land on two points, and the master prompt rung exists for them. A monthly newsletter, say: recurring but slow, with context that shifts each issue. Save the prompt, fill the placeholders, and if you notice the task growing neighbours that share its context, promote it to a skill then.
One caution before you install anything. Passing the test makes a task a good fit; it does not make the output correct. You still check quoted prices against your real rate card, and anything financial or legal still gets professional eyes before it ships.
Next lesson, we open a real skill file and walk through what each section does.