Bound the Skill's Scope
Tell the skill what NOT to do — name the look-alike tasks it should refuse or escalate.
The most expensive mistake a skill makes is firing when it shouldn't — running a 12-step PRD workflow on a one-line bug fix, or applying a refactor playbook to a config tweak. Spell out the negative space: what tasks look similar but are NOT a match, what should escalate to the human, and what should fall back to a plain reply. A skill that says 'do nothing for X' is more useful than one that does the wrong thing for X.
Where you'll practice this
One Promptles scenario teaches this principle directly.
Bound the Scope
Your team's `improve-architecture` skill is well-loved when it fires correctly — it walks the agent through reading domain docs, identifying coupling smells, proposing a refactor, and opening a tracking issue. But it's also been firing on tasks it has no business touching:…