Quickstart¶
Step 1: Bootstrap¶
Blueprint scans your codebase for existing architectural decisions (dependency manifests, convention files, planning artifacts, git history) and creates the ADR directory with inferred decisions.
Output:
Scanning codebase...
Found: package.json (React 18, TypeScript 5, Tailwind CSS)
Found: CLAUDE.md (12 conventions detected)
Found: .planning/ROADMAP.md (3 architectural choices)
Generated:
docs/adr/0001-use-react-with-typescript.md [Proposed]
docs/adr/0002-tailwind-for-styling.md [Proposed]
docs/adr/0003-monorepo-structure.md [Proposed]
docs/ARCHITECTURE.md
3 ADRs drafted. Review with /blueprint:list
Step 2: Record a New Decision¶
Blueprint interviews you about context, constraints, and alternatives, then generates a structured ADR:
The --research flag spawns a researcher agent that evaluates alternatives with evidence before generating the ADR. The research brief covers: options analysis, benchmark data, ecosystem maturity, team experience fit, and a recommendation.
Step 3: Challenge It¶
The devil's advocate challenges ADR-0004 across 5 dimensions:
Devil's Advocate Challenge: ADR-0004
1. HIDDEN ASSUMPTIONS
- Assumes PostgreSQL can handle projected write volume
- Assumes team will learn PostgreSQL operations
2. UNCONSIDERED ALTERNATIVES
- CockroachDB for horizontal scaling without sharding
- DynamoDB if query patterns are primarily key-value
3. MISSING CONSEQUENCES
- No migration strategy for existing 50M rows
- JSONB queries bypass indexes at scale
4. CODEBASE FIT
- Current ORM abstractions are database-agnostic: good
- Connection pooling not configured: needs work
5. TEAM CAPABILITY
- Zero PostgreSQL ops experience on the team
- Recommendation: pair with a DBA for first 3 months
Step 4: Accept (or Revise)¶
If the decision survives the challenge:
If it needs revision, update the ADR with the devil's advocate findings, then review again.
Step 5: Verify Compliance¶
After the code is written:
The compliance auditor checks whether the codebase actually follows the accepted decisions. Evidence-based verdicts: COMPLIANT, PARTIAL, VIOLATING, or INSUFFICIENT EVIDENCE.
What's Next?¶
/blueprint:evaluate: Run the 5-agent architecture evaluation team/blueprint:fitness: Generate CI-runnable architecture tests/blueprint:hooks install all: Enable automatic governance triggers/blueprint:status: View the governance dashboard
Tip
For a faster start with pre-built decisions, try /blueprint:quickstart. It auto-detects your stack and generates 5-8 foundational ADRs from templates covering common choices (framework, styling, state management, API patterns, testing strategy).