/blueprint:init
# Bootstrap from codebase
/blueprint:new
"use PostgreSQL"
# Record
/blueprint:review
3
# Challenge it
Devil’s advocate found 3 blind spots:
• No migration strategy for 50M rows
• JSONB queries bypass indexes at scale
• Team has zero PostgreSQL ops experience
/blueprint:transition
accept 3
/blueprint:audit
# Is the code following it?
Software architecture is not a document. It is the set of decisions that constrain all future decisions. Every technology choice, every boundary drawn, every pattern adopted closes some doors and opens others. The tragedy of most software projects is not that these decisions are made poorly, but that they are made invisibly. A decision made in a Slack thread at 4pm on a Friday becomes the foundation of a system that runs for a decade.
Architecture Decision Records were invented to solve this. But in practice, most ADR implementations fail for the same reason most New Year’s resolutions fail: there is no enforcement mechanism. Writing the decision down is the easy part. Researching it before committing, challenging it before accepting, verifying that the codebase actually follows it, and revisiting it when circumstances change. That’s where ADR processes quietly die.
Blueprint doesn’t let them die. It decomposes governance into 21 specialized agents, each with one job. The cranky senior engineer persona ensures findings are stated with the clarity and conviction required to actually change behavior. And the config DSL encodes domain knowledge as data, not prose, because data outlives code.
/blueprint:new/blueprint:audit/blueprint:new --research spawns a researcher agent that evaluates alternatives with evidence before you commit. No more vibes-based architecture.
/blueprint:review challenges proposed decisions across 5 dimensions: assumptions, alternatives, consequences, codebase fit, and team capability. Hegelian dialectic, not rubber-stamping.
/blueprint:audit verifies the codebase follows accepted decisions. /blueprint:guard catches violations at the point of creation. Architecture-as-code, not architecture-as-wishful-thinking.
Researcher. Devil’s advocate. Five evaluation dimensions. Reflexion models. Strategic analysis. Each agent has one job and does it with the depth that comes from focus.
DDD bounded contexts. ATAM tradeoff analysis. Wardley Maps. Reflexion models. C4 diagrams. Not invented here. Curated from 109 sources across 20+ paradigms.
/blueprint:retro classifies root causes and catches band-aid fixes. /blueprint:debt tracks deferred decisions. /blueprint:drift detects slow erosion. Nothing forgotten.
Built on Architecture Decision Records (Nygard, 2011), the Hegelian Dialectic, Domain-Driven Design (Evans, 2003), ATAM (SEI/CMU, 1998), Reflexion Models (Murphy et al., 1995), and the principle that decisions, not code, are the load-bearing walls of software.