BMAD-METHOD: SPEC-DRIVEN AI CODING FRAMEWORK (2-week project)

PHASE 0: SPEC DEFINITION (Day 1, 4 hours)
→ Problem statement: 1 paragraph, max. What is the problem you're solving?
→ Success criteria: 5-8 acceptance tests (not 'code is clean', but 'API returns 200 and result.count equals X')
→ Tech stack: Explicit. Language, framework, DB, hosting. No "whatever works".
→ Scope boundary: IN-scope (3-5 features), OUT-of-scope (5+ explicitly excluded features)
→ Constraints: Performance targets, data-size limits, external API rate limits
→ Deliverable: Spec document in plaintext or Markdown, shared with AI agent before writing code

PHASE 1: ARCH DESIGN (Day 2-3, given spec)
→ Agent task: 'Given this spec, design the data model and API endpoints. Return ERD and endpoint list.'
→ Validation: You review. Does it satisfy acceptance tests? Any missing edge cases?
→ Gate: Agent only codes after you approve design

PHASE 2: IMPLEMENTATION (Day 4-9, spec + arch as input)
→ Agent task: Break into 3-day sprints (week 1: core API, week 2: UI or integrations)
→ Validation: Each sprint ends with you running tests against success criteria
→ Go/no-go: If sprint tests fail, fix before next sprint. No debt carried.

PHASE 3: INTEGRATION + TEST (Day 10-13)
→ All pieces + integration tests. Agent tests against full acceptance-criteria list.
→ You spot-check 3 high-risk paths (error handling, data boundary cases)

PHASE 4: VALIDATION (Day 14)
→ Final test run. Does it meet success criteria? Ship or cycle spec.
→ Outcome: Spec-to-shipped = 14 days, zero rewrites