Writing PRDs with Claude Code and Context Assembly

shipped two complete PRDs yesterday in 4 hours normally that's a 2-week grind here's the actual workflow: - I keep a markdown folder with all my discovery docs, user research transcripts, and strategy notes - Claude Code reads the whole directory so it has context before I ask for anything - I describe the problem and target outcome in plain english - it generates a first draft with context/goals, user stories, analytics events with schema, dependencies, and draft project plan - I review the user stories hard (this is where AI makes the most mistakes) - iterate 2-3 times on edge cases and acceptance criteria - final pass: I rewrite the "why" sections in my own voice the part that used to take forever was assembling context hunting through confluence, slack threads, old docs now the LLM does that in seconds still review everything manually: scope decisions, prioritization, anything customer-facing but the assembly and formatting work that ate 80% of my time is gone if you want the exact prompts and template structure I use, I put them all in one place: