IntentConfirm: Clarify Before You Build

AI coding assistants can produce working software that misses the user's intended outcome. Short requests often omit choices that change the data model, workflow or scope. Users may not know how to write a detailed specification, and repeated free-text questioning adds friction. IntentConfirm is a portable Skill for clarifying requirements inside the user's existing AI conversation. It preserves explicit constraints, separates facts from assumptions, and asks one to three consequential questions per round with concrete options and free-text alternatives. It treats delegated choices as visible defaults. Once enough information is available, the host assistant continues the authorized implementation. Clear requests can proceed without extra questioning. For longer tasks, optional Python tools preserve evidence, validate structured requirement records and invalidate affected decisions after a change. Unrelated constraints remain intact. Structural checks do not prove semantic correctness or replace user judgment. The initial target users are people building software with AI coding assistants. In a real IBM Bob session, the user asked for a personal study tracker and answered BAA and C to select reading progress, browser storage, a single HTML file and delegated completion behavior. Bob summarized the brief and reported implementing the example. The user opened it and reported that it met their expectations. The reviewed Skill passed 38 Python tests, with 47 separate tests for the retained web prototype. This is one real use case, not a controlled productivity benchmark. The intended benefit is less rework caused by hidden assumptions. Broader host compatibility and average time savings remain to be evaluated. IntentConfirm uses the host model without requiring a separate API key or platform. The source uses the MIT license.