The fix for vague prompts like "make it better." Claude can't hit a target you didn't name — so name it.
Add one line of structure to any request:
GOAL — what "done" looks like, measurably
CONSTRAINT — what Claude is NOT allowed to change
SUCCESS — how you'll verify it actually worked
Goal: <what done looks like, measurable>
Constraint: <what you must NOT change / break>
Success: <how we verify it worked>
Task: <your actual request>
Code — performance
Goal: cut initial load under 1s. Constraint: don't add new dependencies. Success: Lighthouse Performance ≥ 90. Task: speed up the dashboard page.
Copy — landing headline
Goal: a headline a technical founder gets in 3 seconds. Constraint: no buzzwords ("revolutionary", "next-gen"), under 10 words. Success: states the outcome, not the feature. Task: rewrite the hero headline.
Design — component
Goal: a pricing card that reads as premium. Constraint: keep the existing color tokens and spacing scale. Success: clear visual hierarchy price → benefits → CTA. Task: redesign the $69 tier card.
Precise in, less back-and-forth out.
Want the full Claude operator setup — the CLAUDE.md + skills + memory system that makes Claude remember your whole project every session? → https://get.caffier-agency.com