SDLC 3: Implementation
Using the spec and architecture from Phases 1–2, you write tests from acceptance criteria before any implementation — all tests fail initially. Then you build an implementation plan, execute it step by step, and run tests after each step. When tests pass, a self-review checks against the spec and architecture. Issues get fixed and re-tested in a loop until both tests and review are clean.
Audience: Developers, tech leads, and architects | Guided time: 35 minutes
- Master the prompt → review → iterate cycle
- Use Claude Code for TDD workflows
- Perform AI-assisted code review before PRs
- Handle complex refactoring with AI assistance
- Maintain code quality standards with AI
- Package workflow prompts as reusable skills
Using the spec and architecture from Phases 1–2, you write tests from acceptance criteria before any implementation — all tests fail initially. Then you build an implementation plan, execute it step by step, and run tests after each step. When tests pass, a self-review checks against the spec and architecture. Issues get fixed and re-tested in a loop until both tests and review are clean.
| File | Description |
|---|---|
docs/specs/task-comments.md | Feature spec — what to build |
docs/architecture/task-comments.md | Architecture plan — where and how it fits, error format, response shapes |
CLAUDE.md | Project conventions and constraints |
requirements/stories.md | Acceptance criteria — become executable tests before implementation |
| File | Description |
|---|---|
tests/ | Tests generated from acceptance criteria — written BEFORE implementation |
docs/plans/task-comments.md | Plan-Review-Execute plan file — reviewed and approved before coding |
src/routes/comments.js (+ related) | Core implementation — comments routes and data layer |
CLAUDE.md (updated) | Implementation notes — conventions learned during coding |
requirements/backlog.md (updated) | ACTIVE unit marked COMPLETE, next unit ready |
.claude/skills/self-review/SKILL.md | Reusable skill — workflow prompt packaged for the team |
The optional Ralph Loop reuses these artifacts and quality gates as a supervised pipeline. The backlog update signals what comes next.