interview-me

State a one-sentence hypothesis with a confidence number, then ask one focused question at a time with your best guess attached. Continue until you can predict the user's reactions, then restate the intent covering outcome, user, success criteria, and constraints for explicit confirmation.

Use it when a request is missing who it is for, why it is needed, what success looks like, or what the binding constraint is. It also applies when the user explicitly asks to be interviewed or when you catch yourself silently filling in ambiguous requirements.

Skip it for unambiguous self-contained tasks like renaming a variable or fixing a typo, pure information requests, mechanical operations, and non-interactive contexts like CI pipelines or scheduled runs where no live user can respond.

Batched questions encourage skim-reading and surface-level answers, and later questions often depend on earlier answers. One at a time lets the user react to your attached guess, which is faster than generating answers from scratch and keeps your assumptions visible.

You are done when you can predict the user's reaction to the next three questions you would ask, roughly 95% confidence. Then write a structured restatement including an explicit out-of-scope line and wait for an unambiguous yes, not "sounds good" or "whatever you think."