1. Write the description as a trigger, not a summary

Start with: “Use when…” The description is what Claude scans first. If it’s vague, the skill never fires.

2. Be specific about inputs

URL? Product name? Dataset? Notes? Say it clearly. AI matches patterns. Clear inputs = reliable triggering.

3. Define scope

What it does. What it doesn’t do. Boundaries prevent accidental activation and messy outputs.

4. Mention outcome, not activity

“Generate a campaign optimized for lead generation.” Not just “Create ads.” Skills should aim at results, not tasks.

5. Embed the goal

Sales? Leads? Engagement? Clarity? Speed? AI needs a success definition. Otherwise it optimizes randomly.

6. Include KPIs if relevant

Target CPA, conversion intent, audience sophistication. Great skills encode performance expectations.

7. Hardcode frameworks

PAS. JTBD. AIDA. Your internal checklist. Frameworks turn generic output into structured thinking.

8. Add constraints

No jargon. No fluff. No em dashes. No generic adjectives. Constraints are what separate average skills from opinionated ones.

9. Define output format

Markdown? Structured report? Headlines + descriptions? Without format rules, outputs drift.

10. Include examples of good output

AI learns from patterns faster than rules. Examples sharpen quality instantly.

11. Encode your taste

Tone rules. Positioning logic. Strategic bias. This is how skills embed expertise.

12. Add decision logic

What to prioritize. What to ignore. Good skills don’t just produce — they decide.

13. Keep it focused

One job per skill. Don’t overload. Skills are modular. Clarity beats complexity.

14. Test the trigger

If it doesn’t fire automatically, your description is weak. The front matter is the gateway.

15. Refine after real use

Skills improve through iteration, not theory. Use → adjust → repackage.

Claude Skills