A modular AI development toolkit that gives your AI assistant specialized knowledge for Vue3 + ASP.NET Core + Azure development.
This kit transforms a general AI assistant into a specialized development team with:
- 12 Expert Agents - AI personas with domain expertise
- 20 Skills - Reusable knowledge modules
- 10 Workflows - Slash commands for common tasks
- Validation Scripts - Automated quality checks
| Without Kit | With Kit |
|---|---|
| Generic AI responses | Specialized expertise for your stack |
| Inconsistent patterns | Enforced best practices |
| Manual quality checks | Automated validation |
| One-shot responses | Multi-step workflows with checkpoints |
# One-time use with npx
npx antigravity-devkit init
# OR install globally for frequent use
npm install -g antigravity-devkit
antigravity-devkit init# Copy to your project root
cp -r .agent-antigravity .agent| Command | Description |
|---|---|
antigravity-devkit init |
Install kit to .agent/ folder |
antigravity-devkit init --force |
Overwrite existing .agent/ |
antigravity-devkit update |
Update kit (backs up existing) |
antigravity-devkit help |
Show help and usage |
After installation, start using slash commands:
/plan user authentication feature
/code create login component
/test generate tests
/deploy to production
1. User Request → Everything starts with your request to the AI assistant.
2. Classification Layer (rules/GEMINI.md)
- Always Active - This runs before every response
- Analyzes your request to determine the type (question, code, design, etc.)
- Routes to the most appropriate specialist agent
- Enforces global coding standards and best practices
3. Agent Selection Layer (agents/*.md)
- The right specialist is automatically selected based on your request
- Available specialists: Frontend, Backend, Database, DevOps, Test Engineer, and more
- Each agent brings domain-specific expertise
- Agents load their required skills for the task
4. Skills Loading Layer (skills/*/SKILL.md)
- Agents load relevant knowledge modules
- Skills include: clean-code, vue3-patterns, aspnet-patterns, azure-devops, testing-patterns, architecture, and more
- Each skill provides patterns, templates, and best practices
- Skills can reference additional documentation and examples
5. Output Generation
- AI generates specialized responses using the loaded knowledge
- Output types: Code files, Documentation, Tests, Configuration, etc.
- All outputs follow the patterns and standards from the loaded skills
Optional Workflows (Right Panel)
- Slash commands like
/plan,/create,/code,/test,/deploy - These trigger pre-defined multi-step procedures
- Can be used to streamline common development tasks
Validation Scripts (Left Panel)
checklist.py- Quick validation during developmentverify_all.py- Comprehensive checks before deployment- Automatically run after certain operations or manually triggered
.agent-antigravity/
├── README.md # You are here
├── ARCHITECTURE.md # Quick reference for all components
├── rules/ # GLOBAL RULES (Always Active)
│ └── GEMINI.md
├── agents/ # SPECIALIST AGENTS (Selected per task)
│ └── *.md
├── skills/ # KNOWLEDGE MODULES (Loaded by agents)
│ └── */SKILL.md
├── workflows/ # SLASH COMMANDS (User-triggered)
│ └── *.md
└── scripts/ # VALIDATION TOOLS (Run manually or auto)
└── *.py
What: Rules that apply to EVERY interaction.
When Applied: Automatically, before any response.
Contains:
GEMINI.md- Master rules file with:- Request classification (question, code, design, etc.)
- Agent routing logic (which agent handles what)
- Universal coding standards
- Human-in-the-loop checkpoints
- Socratic Gate (ask before assuming)
Priority: Highest - these rules override everything else.
Rule Priority: GEMINI.md > Agent.md > SKILL.md
What: AI personas with specialized expertise for specific domains.
When Applied: When request matches agent's domain (auto-routed or manually invoked with @agent-name).
How It Works:
- Request comes in: "Create a Vue component for user profile"
- GEMINI.md routes to
frontend-specialist - Agent file is loaded with its skills
- Response follows agent's patterns and rules
Available Agents:
| Agent | Domain | When to Use |
|---|---|---|
orchestrator |
Multi-agent coordination | Complex tasks spanning multiple domains |
project-planner |
Discovery, planning | New features, architecture decisions |
frontend-specialist |
Vue3, TypeScript | Components, state, UI logic |
backend-specialist |
ASP.NET, C# | APIs, services, business logic |
database-architect |
SQL Server | Schema design, queries, migrations |
devops-engineer |
Azure DevOps, AKS | Pipelines, deployments, infrastructure |
security-auditor |
Security, KeyVault | Audits, vulnerability checks |
test-engineer |
xUnit, Vitest | Unit tests, integration tests |
debugger |
Bug fixing | Root cause analysis, debugging |
performance-optimizer |
Grafana, metrics | Performance tuning, monitoring |
documentation-writer |
Docs, README | API docs, technical writing |
explorer-agent |
Codebase analysis | Understanding existing code |
Manual Invocation:
@backend-specialist create a user service
@security-auditor review this controller
What: Reusable knowledge packages that agents load for specific tasks.
When Applied: When an agent needs specialized knowledge (defined in agent's frontmatter).
Structure:
skills/
└── skill-name/
├── SKILL.md # Main skill file (required)
├── references/ # Templates, examples (optional)
└── scripts/ # Validation scripts (optional)
How Loading Works:
# In agent file (e.g., backend-specialist.md)
---
skills: clean-code, aspnet-patterns, api-patterns
---When backend-specialist is activated, these skills are loaded.
Available Skills by Category:
| Category | Skills | Purpose |
|---|---|---|
| Core | clean-code, architecture |
Universal coding standards |
| Planning | brainstorming, plan-writing |
Discovery and task breakdown |
| Frontend | vue3-patterns, vitest-testing |
Vue3 + Tailwind patterns |
| UI/UX Design | frontend-design |
Design thinking, UX psychology, anti-AI-slop, animations, Tailwind theming |
| Backend | aspnet-patterns, csharp-patterns, api-patterns |
ASP.NET best practices |
| Database | sqlserver-design |
SQL Server patterns |
| Azure | azure-devops, azure-aks, azure-keyvault |
Azure services |
| DevOps | gitops-patterns, grafana-logging |
CI/CD and monitoring |
| Testing | xunit-testing, testing-patterns |
Test strategies |
| Security | vulnerability-scanner |
Security checks |
| Industry | english-education |
Domain-specific (customizable) |
What: Pre-defined procedures for common development tasks.
When Applied: When user types a slash command (e.g., /plan, /code).
Available Commands:
| Command | Purpose | Output |
|---|---|---|
/brainstorm |
Clarify requirements | Questions → Understanding |
/plan |
Create project plan | docs/PLAN-{slug}.md file |
/create |
Build complete feature | Code + Tests + Verification |
/code |
Generate specific code | Code for one task |
/test |
Generate tests | Test files |
/debug |
Fix bugs | Root cause → Fix → Verify |
/review |
Code review | Review report |
/deploy |
Trigger deployment | Pipeline trigger |
/status |
Check project health | Status report |
/orchestrate |
Multi-agent task | Coordinated execution |
Choosing the Right Command:
| Your Situation | Use This | Why |
|---|---|---|
| "I have a vague idea" | /brainstorm |
Clarify before building |
| "I know what to build" | /plan |
Create task breakdown |
| "I have a plan, build it all" | /create |
Full feature with checkpoints |
| "Just this one thing" | /code |
Quick, focused output |
| "Something is broken" | /debug |
Systematic fix |
| "Is this code good?" | /review |
Quality check |
| "Ready to ship" | /deploy |
Trigger pipeline |
Key Differences:
/create |
/code |
|---|---|
| Full feature (many files) | Single task (few files) |
| Follows a plan file | No plan needed |
| Multiple agents | One specialist |
| Checkpoints for approval | Quick execution |
/brainstorm |
/plan |
|---|---|
| Discover requirements | Define tasks |
| Output: understanding | Output: PLAN-*.md file |
| Use when unclear | Use when clear |
What: Python scripts for automated quality checks.
When Applied:
- Manually: Run from terminal
- Auto: Agents may run after completing tasks
Available Scripts:
| Script | Purpose | When to Use |
|---|---|---|
checklist.py |
Quick validation (lint, types, tests) | During development, pre-commit |
verify_all.py |
Comprehensive check (all validations) | Before deployment, releases |
Usage:
# Quick check during development
python .agent/scripts/checklist.py .
# Full verification before deploy
python .agent/scripts/verify_all.py . --url http://localhost:3000What They Check:
- Security (secrets, vulnerabilities)
- Code quality (lint, types)
- Tests (pass/fail, coverage)
- Build (compiles successfully)
- Dependencies (outdated, vulnerable)
The antigravity-dev-kit supports your entire development lifecycle with specialized workflows for each phase:
Workflows: /brainstorm, /plan
Start here when you have a new feature idea or requirement. The kit helps you:
- Clarify vague requirements through strategic questions
- Create detailed task breakdowns
- Generate a
PLAN-*.mdfile for structured implementation
When to use:
- New feature requests
- Complex requirements that need clarification
- Architecture decisions
Workflows: /create (full feature), /code (single task)
Build your features with specialized agents:
- Frontend Specialist - Vue3 components and UI
- Backend Specialist - ASP.NET APIs and services
- Database Architect - SQL schema and queries
Key features:
- Checkpoints for human approval
- Follows your plan file
- Multi-agent coordination for complex features
Workflows: /test
Scripts: checklist.py
Generate comprehensive tests:
- Unit tests (xUnit for backend, Vitest for frontend)
- Integration tests
- Test coverage reports
Automated checks:
- Linting and type checking
- Code quality validation
- Test execution
Workflows: /review
Agent: Security Auditor
Get automated code reviews covering:
- Security vulnerabilities
- Code quality issues
- Best practice violations
- Performance concerns
Output: Detailed review report with actionable recommendations
Workflows: /deploy
Scripts: verify_all.py
Deploy with confidence:
- Comprehensive pre-deployment validation
- Azure DevOps pipeline trigger
- Deployment verification
Checks before deploy:
- All tests passing
- No security vulnerabilities
- Build successful
- Dependencies up to date
Workflows: /debug, /status
Agents: Debugger, Performance Optimizer
Keep your application healthy:
- Root cause analysis for bugs
- Performance optimization
- System health monitoring
- Quick fixes and patches
Continuous improvement:
- Monitor application metrics
- Optimize slow queries
- Fix production issues
- Iterate back to Phase 1 for new features
| Scenario | Workflow Path | Duration |
|---|---|---|
| New Feature | /brainstorm → /plan → /create → /test → /review → /deploy |
Full cycle |
| Bug Fix | /debug → /test → /review → /deploy |
Fast track |
| Quick Task | /code → Done |
Immediate |
1. /brainstorm user authentication
→ AI asks clarifying questions
→ You answer and confirm understanding
2. /plan user authentication
→ Creates docs/PLAN-user-auth.md
→ Review and approve the plan
3. /create user authentication
→ AI follows the plan
→ Builds backend → [Checkpoint] → Builds frontend → [Checkpoint]
→ Generates tests → [Checkpoint]
4. /review
→ Security and quality check
→ Approve for deployment
5. /deploy
→ Triggers Azure DevOps pipeline
/code create a Vue component for user avatar
→ Routes to frontend-specialist
→ Generates component with TypeScript
→ Done
/debug login fails with 401 error
→ AI investigates root cause
→ Proposes fix
→ You approve
→ Fix applied and tested
- Create folder:
skills/your-skill/ - Add
SKILL.mdwith frontmatter:
---
name: your-skill
description: What this skill does
---
# Your Skill Name
[Your skill content - patterns, rules, templates]- Reference it in an agent's skills list
Edit the agent's .md file to:
- Change which skills it loads
- Modify its rules and patterns
- Update its domain expertise
Use skills/english-education/ as a template:
skills/your-industry/
├── SKILL.md # Main skill file
├── references/ # Industry templates
│ ├── template-1.md
│ └── template-2.md
└── scripts/ # Industry validators
└── validator.py
| Layer | Technology |
|---|---|
| Frontend | Vue3, TypeScript, Pinia, Vitest |
| Backend | ASP.NET Core, C#, xUnit |
| Database | SQL Server |
| Cloud | Azure DevOps, AKS, KeyVault |
| Observability | Grafana |
| GitOps | ArgoCD/Flux pattern |
| File | Purpose |
|---|---|
ARCHITECTURE.md |
Quick reference table of all components |
rules/GEMINI.md |
Master rules (always active) |
agents/*.md |
Specialist personas |
skills/*/SKILL.md |
Knowledge modules |
workflows/*.md |
Slash command definitions |
scripts/*.py |
Validation scripts |
MIT