## Core Concept A Claude Code skill pattern where the AI conducts a structured interview (via the AskUserQuestion tool) before any implementation. Instead of front-loading all context into one prompt, the skill actively pulls requirements through targeted, sequential questions, then compiles the answers into a specification file that serves as the contract between user intent and AI execution. **Three-phase flow**: 1. **Interview**: structured questions on scope, constraints, preferences, edge cases 2. **Specification**: answers synthesized into a detailed spec document 3. **Implementation**: coding proceeds against the spec, not a vague prompt ## Why It Works It directly attacks the scope-misinterpretation problem (14 misunderstood requests across 8,832 sessions). The root cause is ambiguity in the initial prompt; the interview eliminates it by making the AI responsible for asking clarifying questions rather than guessing. The spec file also becomes an auditable artifact for course correction. Design considerations: sequence questions broad→specific; keep the spec human-editable; support skip/default answers to avoid blocking. ## Related Concepts - [[Plan-First Workflow Gates]] — interview-first is a specific implementation - [[Scope Misinterpretation as Trust Boundary]] — the problem it solves - [[Skill Chaining for Workflow Automation]] --- *Source: [[This is single-handedly the best Claude Skill someone ever shared]]* (Danny Postma, [x.com/dannypostma/status/2017864243895538012](https://x.com/dannypostma/status/2017864243895538012))