PRD to Plan

Overview

This skill reads a PRD and the codebase it will be built in, and writes one plan file that a developer or a coding agent can follow from the first commit to the last. The plan names the choices that would be expensive to reverse, splits the work into phases that each leave the product in a working, demonstrable state, and says how each phase is verified. It differs from a ticket breakdown: the output is a single living document for sequential work, kept next to the code and updated as the build teaches something new.

Instructions

1. Establish the inputs

Ask where the PRD is (file, issue, pasted text), who will execute the plan (one developer, a team, an agent working unattended), and any fixed constraint: a date, a release freeze, a platform that must be supported first. Find where plans live in this repository: an existing plans/, docs/plans/ or docs/rfcs/ directory wins; otherwise use plans/ at the root. Name the file after the feature in kebab case, for example plans/team-invitations.md.

2. Build the requirement ledger

Give every requirement an id (R1, R2, …), reusing the PRD's numbering when it exists. List non-goals and open questions separately. Sort the open questions: one that changes the order or the shape of the phases must be answered before planning continues; the rest go into the plan with an owner and the phase that needs the answer.

3. Read the code the feature will touch

Record facts, each one verified by opening the file: