Clarify project requirements before writing any code

Brave Man flips the usual build order: clarify exhaustively first, build later. Most people describing a project only give a partial picture - humans don't think in complete specs, and even trying, they forget details that become expensive problems once a project has grown. If an agent starts from a thin description, it fills gaps with silent guesses that are costly to undo once wrong. Brave Man's only job is to run a structured interview until the project is fully understood, then write that understanding down as a single, self-contained prompt.md file a fresh agent session can execute later - it never writes code, scaffolds files, or produces an implementation plan itself.

Use it when a user describes wanting to build a website, app, software, or tool - phrases like "build me a website" or "I want an app for X" - and use it BEFORE writing any code or implementation plan. Never skip straight to building because a request "sounded simple": simple requests still go through triage, which decides how short the interview gets, not whether it happens.

A five-step process: Triage (2-3 quick questions sizing up scope, audience, and existing preferences to calibrate interview depth); Phased interview (work through eight relevant phases one at a time, batching 3-5 questions per round rather than dripping single questions) covering Purpose & Users, Core Features & Flows, Data & Content Model, Tech Stack & Environment, Integrations & Auth (explicitly skipped and stated as such if triage shows no accounts/external services needed), Non-Functional Requirements, Edge Cases & Error States, and Definition of Done; Track completion via a visible checklist using [x] confirmed, [~] defaulted-and-accepted, [ ] open, and [-] skipped markers, never moving to synthesis while any relevant phase is still open; Synthesize the final prompt.md once every relevant phase closes, structured as a Project Brief with overview, prioritized core features, and further sections addressed directly to whichever agent reads it next; and Hand off by telling the user to start a new chat, tag prompt.md, and have that agent execute it.

Best practices: batch questions per themed round rather than endless single-question ping-pong, use plain language over jargon unless the user has shown technical fluency, offer options when a question has few sensible answers, never re-ask redundant or already-inferable questions, and handle "I don't know" by proposing a sensible named default rather than blocking. Even when a user says "just use your judgment," Phase 3 (data) and Phase 5 (auth/integrations) still require at least a default-and-confirm pass rather than being silently skipped.

Users starting a new build who want their vague idea turned into a complete, unambiguous specification before any code gets written - and the downstream agent session that will execute the resulting prompt.md as its complete spec.