write-prdlisted

# write-prd — Research-Informed PRD Generator Write a PRD for: $ARGUMENTS ## Overview Fully autonomous research-first PRD generator. Researches the domain, competitors, best practices, and technical landscape, then makes all design decisions autonomously based on research findings + codebase context. Never asks the user questions — decides based on available evidence. **Outputs:** 1. A complete PRD (`./tasks/prd-[name].md`) 2. A JSON status tracking file (`./tasks/prd-[name]-status.json`) ## Execution Flow `INTAKE → RESEARCH → DECIDE → STRUCTURE → WRITE → FINALIZE` Print `[Phase N/6] PHASE_NAME` before each phase. Fully autonomous — no user questions at any phase. ## Phase-by-Phase Execution ### Phase 1 — INTAKE **1a. Validate input:** If `$ARGUMENTS` is empty or fewer than 5 words, fail with a clear error: "Usage: /write-prd [feature description]. Provide at least a sentence describing the feature." **1b. Parse `$ARGUMENTS`:** Extract: **Domain** (area: auth, payments, UI, data pipeline...), **Keywords** (for web research), **Implied users** (who will use this), **Implied scope** (small feature / full product / platform). **1c. Scope detection — fast mode check:** If the implied scope is very small (likely < 3 stories — a single endpoint, config change, or isolated component), switch to FAST MODE: - Phase 2: web-researcher only (skip agent-explorer and docs-researcher) - Phase 3: merge Rounds 1-3 into a single focused round, keep Edge Cases + Quality Gates + D