Skip to content

interview

interview scopes one feature or problem by asking — one question at a time — until you and the agent actually share the same picture. It is the greenfield intake: no existing plan, no docs assumed, just a conversation that walks the decision tree until the intent is concrete enough to build on.

What it solves

Most underspecified work isn’t underspecified on purpose — the gaps are simply invisible until someone asks. A wall of questions at once gets skimmed and half-answered; a vague “build X” turns into the agent guessing the requirements and you discovering the wrong guess in review. interview removes the guessing by making the unknowns explicit, one decision at a time, before any code or spec.

How it works

  • One question at a time — each answer narrows the next question, so the conversation walks the decision tree instead of dumping a checklist.
  • No document dependency — it doesn’t need a CONTEXT.md or ADRs to exist; it works on a blank slate, which is what makes it the greenfield tool.
  • Confirmed summary — it ends with a summary you approve, which becomes clean input for the next step: doc-prd to publish it, doc-align to stress-test it against existing docs, or implement to build it.

It sits next to two siblings: doc-align is the counterpart for an existing plan (validated against the project’s docs), and doc-research explores a whole area and produces several backlog items, where interview scopes a single one.

When to use

At the start of a feature or problem that isn’t concrete yet — before writing a spec, a PRD, or code. When the plan already exists and needs checking against the project’s decisions, reach for doc-align instead.