Iterate: clarify, decompose, compare

The first answer is a draft. That isn't a failure — you gave an incomplete brief, and Claude filled the gaps to produce something. The skill is reading the result for what it had to guess, because every guess marks a piece of context you didn't supply.

The diagnostic move is to ask directly: "What did you assume about my situation to answer that?" The list that comes back is your missing context, itemised. Put the two or three that matter into the prompt and ask again.

Make it ask first. For anything underspecified, require questions before the answer: "Ask me up to three questions before you start." This inverts the failure mode — instead of a confident answer built on wrong assumptions, you get the assumptions surfaced while they're still cheap to fix. It's most valuable exactly when you're least sure what you want.

Guessing:
  Write a migration plan for moving our app off Heroku.

Asking:
  I want a migration plan for moving our app off Heroku.
  Before you write it, ask me up to three questions whose
  answers would most change the plan.

Break down what feels too big. An intimidating goal usually hides an unclear first step. Ask for a sequence of steps where the first one is small enough to start now — a plan you can begin in fifteen minutes is worth more than a complete one you postpone.

Compare fairly. When you test two versions of a prompt, change exactly one thing — the context, the format, or the constraint — and keep everything else identical. Change two and you learn that something helped without learning what, which is the same as learning nothing.

Loading your workspace…