plan-to-issues — PRD, then vertical slices

Synthesize a plan from what's already been discussed, then cut it into thin end-to-end slices an agent (or human) can grab independently. (Distilled in our own way from Matt Pocock's to-prd + to-issues, MIT — with his personal issue-tracker setup stripped out.)

Do NOT interview — synthesize the current conversation + codebase understanding. If the repo has a domain glossary or ADRs, use them; otherwise carry on.

Phase 1 — the PRD

Explore the repo for current state, then write a PRD with these sections:

Problem statement — from the user's perspective.
Solution — from the user's perspective.
User stories — a long numbered list, each "As an <actor>, I want <feature>, so that <benefit>." Extensive; cover every aspect.
Implementation decisions — modules built/modified and their interfaces, technical clarifications, schema changes, API contracts. No file paths or code snippets (they go stale) — exception: a prototype-derived state machine / reducer / schema / type that encodes a decision more precisely than prose.
Testing decisions — what makes a good test (test external behaviour, not implementation details), which modules get tested, and similar tests already in the codebase.
Out of scope.
Further notes.

Seam checkpoint (the one interactive gate): before finalizing, name the seams you'll test the feature through — prefer existing seams, use the highest one possible, aim for one. Confirm with the user they match expectations.