Execute Blueprint Skill

Purpose

Execute blueprint plans by driving each step through a strict build, review, and verify cycle using dedicated subagents. The main conversation is a coordination hub between the user and the subagents — it asks the user the questions defined in this skill, dispatches subagents to do the work, reports results, and handles git according to the user's chosen mode. The main conversation never performs build, review, or verification work itself.

Pushing every build, review, and verification into a subagent is what keeps this skill viable across long, multi-milestone plans: the bulk of the token cost — reading source files, running tools, capturing full output — is spent inside subagents whose context is reclaimed the moment they return. The coordinator's context holds only the plan and the distilled summary each subagent reports back, never the file contents or tool output, so it stays lean no matter how large the plan grows.

Iron Laws

These rules govern every execution session. They override any local reasoning that suggests cutting corners.

Every build, every review, and every verification is dispatched to a subagent via Claude Code's Agent tool. The main conversation only: