Your API key. Your repos. Your data.
AI.D is an overlay for Claude Code that gives your whole team — PM, Dev, Tech Lead, QA — a repeatable process from idea to production.
Created by Ilan Dahan
|
AI accelerates, you decide. You approve every phase gate. You own every decision. Claude suggests — you commit. |
Clarity before code. No code until requirements are approved. No deployment until tests pass. Phase gates enforce discipline. |
Context compounds. Every session builds on the last. Context tracking means you never restart from zero. |
- What is AI.D?
- The Three Pillars
- Phase Gate System
- Sub-Agents (4 Specialized)
- Skills System (23 Skills)
- MCP Integrations
- Commands Reference
- Quick Start
- Project Structure
- Documentation
AI.D (AI Development Methodology) is a structured overlay for Claude Code that transforms how teams build software. It's not a framework you install into your code — it's a methodology that guides Claude to work with your team through every phase of development.
| Without AI.D | With AI.D |
|---|---|
| AI generates code without context | Phase gates enforce requirements before code |
| Losing track across sessions | Context tracking remembers everything |
| Each team member works differently | One process: PM → Dev → QA all aligned |
| Scope creep and endless revisions | Approved PRDs lock scope |
| No learning from mistakes | Feedback system improves over time |
| Role | Focus |
|---|---|
| Product Manager | Requirements, user stories, DDD (Document Driven Development) |
| Tech Lead | Architecture, code review, technical direction |
| Developer | Implementation, TDD (Test Driven Development) |
| QA Engineer | Test strategy, BDD (Behavior Driven Development) |
AI.D enforces 6 mandatory phases (0-5) with quality gates between each. No phase can be skipped.
Phase 0 ──► Gate ──► Phase 1 ──► Gate ──► Phase 2 ──► Gate ──► Phase 3 ──► Gate ──► Phase 4 ──► Gate ──► Phase 5
Discovery ✓ PRD ✓ Tech Spec ✓ Impl Plan ✓ Dev ✓ QA & Ship
| Phase | Name | What You Do | Output |
|---|---|---|---|
| 0 | Discovery | Research, stakeholder mapping, Go/No-Go decision | Research report |
| 1 | PRD | Define requirements, user stories, acceptance criteria | Product Requirements |
| 2 | Tech Spec | Architecture, database design, APIs, security | Technical Specification |
| 3 | Impl Plan | Task breakdown, Jira population, sprint planning | Implementation Plan |
| 4 | Development | TDD implementation, code review, component building | Production code |
| 5 | QA & Ship | Validation, acceptance testing, deployment | Release |
| Phase | Allowed | Blocked |
|---|---|---|
| 0 Discovery | Research, stakeholders, competitive analysis | PRD, architecture, code |
| 1 PRD | + Requirements, scope, user stories | Architecture, code, Jira |
| 2 Tech Spec | + Architecture, schemas, APIs | Code, Jira issues |
| 3 Impl Plan | + Task decomposition, Jira creation | Production code |
| 4 Development | + Code, tests, components | Deployment |
| 5 QA & Ship | Everything | - |
You approve each gate. Claude cannot advance without your /phase-approve.
AI.D uses 4 specialized sub-agents — isolated AI processes that provide objective evaluation without bias from the main conversation.
| Main Agent | Sub-Agent |
|---|---|
| Full conversation context | Isolated — sees only what you pass |
| Knows its own reasoning | Evaluates output objectively |
| Might agree with itself | Catches issues the main agent missed |
| Agent | Purpose | Triggered By |
|---|---|---|
| reflection-agent | Quality evaluation for all significant outputs | Automatic (Quality Check) |
| phase-review-agent | Phase gate validation before advancement | /gate-check command |
| qa-validator-agent | Task completion validation | QA hooks (Phase 4) |
| memory-analysis-agent | Analyzes feedback and suggests improvements | /aid-improve command |
Every significant output goes through the reflection-agent:
╭─────────────────────────────────────────────────────────────╮
│ 🔍 Quality Check │
├─────────────────────────────────────────────────────────────┤
│ ✅ WHY Alignment 9/10 Addresses stated goal │
│ ✅ Phase Compliance 10/10 Appropriate for Phase 2 │
│ ✅ Correctness 8/10 Verified against specs │
│ ✅ Security 9/10 No vulnerabilities found │
│ ⚠️ Completeness 7/10 Missing error handling │
│ ══════════════════════════════════════════════════════ │
│ 📊 Overall: 8.6/10 │
│ STATUS: ✅ PASSED │
╰─────────────────────────────────────────────────────────────╯
AI.D uses 23 specialized skills organized by role and phase.
| Skill | Purpose |
|---|---|
why-driven-decision |
Foundational — understand WHY before any action |
reflection |
Quality check system with sub-agent evaluation |
phase-enforcement |
Enforce phase gates, refuse out-of-phase work |
context-tracking |
Track tasks, steps, progress across sessions |
| Skill | Phase | Purpose |
|---|---|---|
pre-prd-research |
0 | Business analysis, competitive research, problem validation |
aid-discovery |
0 | Stakeholder identification, Go/No-Go decision |
aid-prd |
1 | User stories, acceptance criteria, scope definition |
aid-tech-spec |
2 | Architecture, API contracts, security design |
aid-impl-plan |
3 | Task breakdown, Jira population, sprint planning |
aid-development |
4 | Implementation guidance, TDD practices |
aid-qa-ship |
5 | Validation, release preparation, deployment |
| Skill | Focus Areas |
|---|---|
role-product-manager |
Requirements, user stories, scope, stakeholder alignment |
role-developer |
Code quality, TDD, debugging, technical feasibility |
role-qa-engineer |
Test strategy, BDD/Gherkin, acceptance testing |
role-tech-lead |
Architecture review, code review, technical direction |
| Skill | Purpose |
|---|---|
atomic-design |
Figma-to-code component system (atoms → pages) |
atomic-page-builder |
Compose pages from existing components only |
system-architect |
Security-first architecture (ISO 27001, OWASP) |
test-driven |
TDD methodology, test patterns, coverage |
code-review |
Quality review, security audit, production readiness |
| Skill | Purpose |
|---|---|
memory-system |
Session feedback, learning, improvement cycles |
learning-mode |
Decision transparency, debate invitations |
figma-design-review |
Design system compliance scoring |
AI.D integrates with Model Context Protocol (MCP) servers for seamless tool connectivity:
| MCP | Purpose |
|---|---|
| Filesystem | File operations, directory management |
| Chrome DevTools | E2E testing, visual testing, performance |
| Jira | Epics, stories, tasks, sprint management |
| Confluence | Documentation, knowledge base |
| Figma | Design tokens, component specs, assets |
| GitHub | Repos, PRs, issues, code review |
Setup: See INSTALLATION.md for MCP configuration and environment variables.
Setup commands: See INSTALLATION.md
┌─────────────────────────────────────────────────────────────┐
│ /aid-start → Work → /aid-end → /aid-improve │
│ ↑ │ │
│ └───────────── Next Session ────────────┘ │
└─────────────────────────────────────────────────────────────┘
| Step | Command | What Happens |
|---|---|---|
| 1 | /aid-start |
Select role (PM/Dev/QA/Lead) + phase → Skills loaded |
| 2 | Work | Claude applies relevant skills, enforces phase gates |
| 3 | /aid-end |
Rate session (1-5), describe what worked/didn't |
| 4 | /aid-improve |
System learns and updates skills (optional, weekly) |
| Command | Description |
|---|---|
/good-morning |
Morning startup — check systems, load context |
/context |
Show current work context |
/aid-start |
Start session — select role & phase |
/aid-end |
End session and provide feedback |
/aid-status |
Show current state |
| Command | Description |
|---|---|
/phase |
Show current phase status |
/gate-check |
Check if ready to advance (spawns sub-agent) |
/phase-approve |
Human sign-off for current phase |
/phase-advance |
Move to next phase |
| Command | Description |
|---|---|
/discovery |
Start Phase 0 research |
/prd |
Create Product Requirements Document |
/tech-spec |
Create Technical Specification |
/jira-breakdown |
Break spec into Jira issues |
/design-system |
Build design system from Figma |
/build-page |
Compose pages from components |
/write-tests |
Write tests (TDD methodology) |
/code-review |
Review code quality |
/qa-ship |
QA validation and release |
| Command | Description |
|---|---|
/aid-improve |
Run learning cycle (spawns sub-agent) |
/reflect |
Show detailed quality check breakdown |
See INSTALLATION.md for complete setup instructions.
Once installed, your daily workflow:
/good-morning # Start your day
/aid-start # Select role + phase
[work] # Claude applies skills
/aid-end # Rate session
AI.D/
├── .claude/ # ALL Claude Code content
│ ├── agents/ # 4 sub-agents
│ ├── skills/ # 23 specialized skills
│ ├── commands/ # Slash commands
│ ├── hooks/ # QA gate hooks
│ ├── references/ # Reference documentation
│ ├── rules/ # Behavioral rules
│ └── templates/ # State file templates
├── docs/ # AI.D methodology docs
├── integrations/ # MCP setup guides
├── CLAUDE.md # Main instructions
├── INSTALLATION.md # Setup guide
└── README.md # This file
your-project/
├── .aid/ # Runtime state
│ ├── state.json # Current phase
│ └── context.json # Work context
└── .claude/ # Linked from AI.D
├── skills/
├── agents/
└── commands/
| Document | Description |
|---|---|
| INSTALLATION.md | Complete setup guide |
| CLAUDE.md | Critical instructions for Claude |
| docs/PHASE-GATES.md | Phase system details |
| docs/MORNING-STARTUP.md | Daily workflow guide |
1. /good-morning # Load context, check systems
2. /aid-start # Select PM role, Phase 0
3. /discovery # Research and Go/No-Go
4. /phase-approve # Human approval
5. /phase-advance # → Phase 1
6. /prd # Create requirements
7. /phase-approve # Human approval
8. /phase-advance # → Phase 2
9. /tech-spec # Create architecture
10. /design-system # Extract Figma tokens
11. /phase-approve # Human approval
12. /phase-advance # → Phase 3
13. /jira-breakdown # Create Jira tasks
14. /phase-advance # → Phase 4
15. /aid-start # Select Developer role
16. /write-tests # TDD - tests first
17. [implement code]
18. /code-review # Quality check
19. /phase-advance # → Phase 5
20. /qa-ship # Final validation & deploy
21. /aid-end # Collect feedback
22. /aid-improve # Learn from session
Created by Ilan Dahan
Built with Claude Code by Anthropic.
AI.D Community License v1.0 — Free to use, adapt, and share. Cannot be sold.
See LICENSE for full details.
| Allowed | Not Allowed |
|---|---|
| Use in any project | Sell the methodology |
| Customize for your needs | Remove attribution |
| Share and teach | |
| Commercial use |
Credit: Ilan Dahan / theaid.ai