Elicit complete requirements in capped rounds

Rich Elicitation guides Antigravity through structured clarification before beginning an ambiguous task. It is intended for requests with at least two unresolved dimensions, each allowing several viable choices. Instead of silently choosing scope, audience, tone, format, or strategy, it gathers the answers that would materially change the deliverable.

The workflow uses a checklist to identify uncertainty about output format, audience, tone, depth, technical level, strategic direction, and constraints. It starts with blocking questions, then rechecks what remains unclear. Answers can reveal a new set of choices, so a follow-up round focuses on those newly relevant dimensions rather than repeating earlier questions.

Rounds are capped at three. The first and second allow up to three questions each, while the final round allows up to two and is used sparingly. The skill stops earlier when enough context exists. After the final round, briefly state remaining assumptions and begin work.

Use it for open-ended writing, planning, design, recommendations, or creative work when early answers could change the structure or direction of the result. It is particularly relevant when a first answer unlocks consequential follow-up questions.

Simple factual lookups, mathematics, clearly scoped requests, and minor uncertainties with safe defaults do not trigger the workflow. It is also unsuitable for latency-sensitive work where question overhead is unacceptable. Ambiguity that requires fetching external evidence, such as an unavailable file, is outside the reasoning-only process.

The skill trusts answers as given and does not check their internal consistency. Its round structure is guidance that requires judgment about when clarification is sufficient.

Inputs are the user's request, the unresolved task dimensions, and answers gathered through the question rounds. The output is a clarified task direction and any explicit assumptions needed before beginning the deliverable. For a presentation, clarify audience with:

Q1: Who is the audience?
  - Internal team / colleagues
  - External clients or partners
  - Investors or stakeholders (Recommended)
  - General / public audience

That presentation example moves from audience, purpose, and available material to fundraising stage and deck length after the first answers. A business-email example resolves tone and purpose in one round. These illustrate when follow-up questions add value and when further rounds are unnecessary.

Questions are grouped in one call with a short explanation of why they matter. Exactly one option per question is marked recommended. Mutually exclusive choices use single selection; combinations use multiple selection. Vague labels and mechanical round announcements are discouraged, and previously resolved categories should not be asked again.

The workflow is designed around Antigravity's ask_user_input_v0 interface. It performs reasoning and question presentation without shell commands, file reads, network requests, or state mutation. Environments without the question tool may offer a weaker question experience. A related ask-user-questions skill covers simpler single-round clarification.

It suits people requesting open-ended deliverables whose requirements span several meaningful choices. It helps clarify the first draft's direction while setting a firm limit on the number of question rounds.