promptdojo_

When a skill is worth it, and when it isn't

A skill is not free. Someone writes it, someone reviews it, someone keeps it current. That cost is small, but it's real, and it means not every task should become a skill. Here's how to judge.

Build a skill when the task is

  • Recurring. It happens regularly, weekly, or many times a week. A skill amortizes its setup cost over many uses. A task done once never earns it back.
  • Standardized. There's a right way to do it, and you want that way followed. If the task is genuinely different every time, there's no fixed playbook to capture.
  • Done by more than one person. The payoff of a skill is consistency across people. A task only you ever do has less to gain, though a skill can still be a useful personal checklist.
  • Costly to get wrong. The more a sloppy or inconsistent result hurts, the more a skill's reliability is worth.

A task that hits most of those should be a skill.

Just prompt when the task is

  • A genuine one-off, or something you do rarely.
  • Exploratory: you're thinking, not following a procedure.
  • Still changing: you haven't settled on the right way yet, so freezing it into a skill would freeze the wrong thing.

The honest failure mode

The mistake teams make is skill sprawl: building a skill for everything, ending up with fifty skills, half of them stale, and nobody sure which is current. A skill you don't maintain is worse than no skill, because people trust it. Build skills for your real, recurring, standardized work. Leave the rest as prompts. A small set of well-kept skills beats a big pile of neglected ones.

read, then continue.