Get Schema

Overview

get-schema turns an approved PRD into a documented database schema. It reads the PRD, derives the entities the product actually needs, defines their columns and relationships, and writes the result to docs/schema/<name>.md as tables a developer can read without opening a migration file.

The reason this exists as its own step: a PRD describes behaviour, not tables. It says "a user can record a signal against an experiment". It does not say whether a signal is a row, a column, or an event stream, and it should not — that is a data modeling decision. When nobody makes that decision explicitly, it gets made implicitly by whoever writes the first migration, and every later feature inherits it.

The schema document sits between the PRD and the task breakdown:

get-prd    -> docs/prd/<name>.md       (what and why)
get-schema -> docs/schema/<name>.md    (what the data looks like)   <- this skill
get-tasks  -> tasks/<name>/*.md        (in what order to build it)

It exists before get-tasks because tasks need to know the shape of the data they read and write. A task that says "store the signal" is not useful until "signal" has columns.

The output is a Markdown document with entity tables, a relationship table, an index table, and a constraints section. It is deliberately not SQL. DDL goes stale the moment an ORM is chosen or a migration is refactored; a table describing intent survives.