blueprintlisted

# Blueprint ## Overview One objective in, a **construction plan** out: numbered tasks, each executable by an agent that has read nothing but the task itself. Blueprint's product is not prose. It is a set of task briefs plus a dependency graph, and it is finished when `plancheck` passes and every task exists as a roadmap child item. A plan nobody can dispatch cold is not a plan. ## When to use Fire when **any** of these is true of a single item: | Predicate | Why blueprint and not superplan | |---|---| | more than ~2 days of work | one plan doc stops being holdable in context | | more than one deployable surface (schema + API + UI) | each surface wants its own review and its own PR | | not verifiable until several pieces land together | the verification story has to be designed, not assumed | | superplan produced more than ~12 steps | that is a decomposition failure reported as a step list | | work will span sessions or agents | context loss between sessions is the thing briefs prevent | **Not** for: a single-PR change (`superplan`), deciding *what* to build (`roadmap`), choosing between options (`decision-matrix`), or a fork-analysis planning document — see *Two kinds of plan* below. ## Two kinds of plan Odin writes two artifacts both called "plan", and they have different shapes: | Artifact | Written by | Shape | Gated by | |---|---|---|---| | **Decision plan** (`.claude/docs/plans/*.md`) | `writing-plans` after `grilling` | forks, alternatives, recommendations, a