Modular AI Agent Capability Kit
22 skills · 15 workflows · 14 rules — plug into any AI coding assistant
Quick Start · Skills · Workflows · Profiles · Architecture
Most AI coding assistants have zero memory between sessions. They forget your stack, your conventions, your patterns. You end up re-explaining everything, every time.
Antigravity Kit solves this by giving your AI assistant a persistent, structured brain:
- Skills — Domain expertise (frontend, backend, AI, security...) loaded on-demand
- Workflows — Step-by-step process chains for common tasks (brainstorm → implement → test → commit)
- Rules — Always-on guardrails (coding standards, git conventions, research protocols)
- Manifest routing — The right skill gets loaded at the right time, automatically
Think of it as a USB drive of expertise you plug into any AI assistant.
# npm
npx @jhm1909/ag-kit init
# pnpm
pnpm dlx @jhm1909/ag-kit init
# With a profile
npx @jhm1909/ag-kit init --profile fullstack-saas
pnpm dlx @jhm1909/ag-kit init --profile fullstack-saasThat's it. Your .agent/ directory is now populated with skills, workflows, and rules.
22 production-grade skills across 8 domains. Each skill includes expert knowledge, reference materials, and executable patterns.
| Skill | Size | Key Features |
|---|---|---|
| frontend-developer | 292K | React/Vue/Angular, performance optimization, accessibility. Includes Vercel React rules (65 patterns) |
react-nextjs |
180K | Server Components, App Router, Next.js 16 |
threejs |
520K | WebGL, 3D scenes, camera systems |
| backend-developer | 36K | API design, system architecture, security (Node/Python/Go/Java) |
| mobile-developer | 236K | React Native, Expo, native features, 9 sub-skills |
| devops-engineer | 40K | CI/CD, Docker, Kubernetes, Terraform, AWS/GCP |
| Skill | Size | Key Features |
|---|---|---|
| designer | 640K | BM25 searchable database: 50+ styles, 161 palettes, 57 fonts, 99 UX rules. Anti-AI-slop aesthetic guide (Anthropic) |
| app-builder | 80K | Full-stack scaffolding orchestrator, 13 project templates |
| Skill | Size | Key Features |
|---|---|---|
| qa-tester | 72K | Unit/E2E/Security/Performance testing. Trail of Bits security audit methodology |
| code-review | 32K | Verification gates, differential review. Trail of Bits blast-radius analysis |
| debugging | 28K | Scientific debugging: root cause analysis, 4 sub-methodologies |
| modern-python | 16K | uv + ruff + ty + pytest. Trail of Bits standard toolchain |
| Skill | Size | Key Features |
|---|---|---|
| ai-engineer | 24K | GenAI, RAG, agentic systems, evaluation pipelines |
| mcp-builder | 92K | MCP server development (TS/Python). Anthropic official guide + eval framework |
| prompt-engineer | 20K | System prompts, few-shot, chain-of-thought, prompt optimization |
| self-improvement | 16K | Feedback descent, evolutionary skill discovery, capability gap analysis. EvoSkill methodology |
| context-engineering | 36K | Token optimization, progressive disclosure, multi-agent coordination |
| Skill | Size | Key Features |
|---|---|---|
| project-manager | 96K | PRDs, roadmaps, RICE/Kano prioritization. Dean Peters PM frameworks (discovery, JTBD) |
| business-analysis | 80K | Requirements analysis, specs, agile documentation |
| lead-architect | 48K | System design, ADRs, RFCs, scalability decisions |
| Skill | Size | Key Features |
|---|---|---|
| blockchain-engineer | 20K | Smart contracts, DeFi, tokenomics, EVM/Solana |
| game-development | 48K | Game dev orchestrator, 10 platform sub-skills |
| marketer | 80K | SEO, content strategy, video production (Remotion) |
| research-first | 4K | Meta-skill: always check docs via context7 MCP before coding |
15 chainable workflows that guide you from idea to deployment:
New Project /brainstorm -> /documentation -> /break-tasks -> /implement-feature -> /commit
New Feature /break-tasks -> /implement-feature -> /gen-tests -> /commit
Bug Fix /development -> /gen-tests -> /commit
Deep Debug /debug -> /commit
UI/UX /ui-ux-design -> /break-tasks -> /implement-feature -> /commit
Study Kit /absorb -> /break-tasks -> /implement-feature -> /commit
MVP Sprint /brainstorm -> /documentation -> [/break-tasks -> /implement -> /commit] x N
| Workflow | Purpose |
|---|---|
/guide |
Interactive discovery — finds the right workflow for your task |
/brainstorm |
Idea analysis, high-level roadmap + PRD |
/documentation |
Architecture docs, API specs, technical specs |
/break-tasks |
Requirements to actionable task breakdown |
/implement-feature |
Spec-to-code implementation |
/development |
Quick fixes and minor changes |
/debug |
Scientific debugging with hypothesis testing |
/ui-ux-design |
Requirements to comprehensive UI/UX deliverables |
/qa |
Test plans and test case documents |
/gen-tests |
Auto-generate unit, E2E, security, and perf tests |
/absorb |
Study reference projects and extract best practices |
/commit |
3-mode git workflow (Quick / Safe / Custom) |
/bootstrap |
Project scaffolding from architectural specs |
/install-skill |
Install skills from GitHub repos or registries |
/custom-behavior |
Modify agent rules and workflows |
Install only what you need:
npx @jhm1909/ag-kit init --profile <name>
# or
pnpm dlx @jhm1909/ag-kit init --profile <name>| Profile | Included Skills | Optional |
|---|---|---|
| web-frontend | frontend-developer, designer | react-nextjs, threejs |
| fullstack-saas | frontend, backend, devops, PM, QA | react-nextjs, ai-engineer, lead-architect |
| mobile-app | mobile-developer, designer, QA | backend, devops |
| ai-powered-app | ai-engineer, frontend, backend | lead-architect, prompt-engineer |
| video-content | marketer, frontend, remotion | threejs, designer |
| blockchain-dapp | blockchain, frontend, backend | devops, lead-architect |
USER -> /guide (Discovery)
-> Workflow Chain (Orchestration)
-> Invoke [skill] (Expertise)
-> skills-manifest.json (Routing)
-> Rules (Guardrails)
.agent/
├── skills-manifest.json <- Single source of truth
├── ARCHITECTURE.md <- Detailed architecture docs
├── skills/ <- 22 expert skills
│ ├── frontend-developer/
│ │ ├── SKILL.md
│ │ ├── react-nextjs/ <- Sub-skill
│ │ ├── threejs/ <- Sub-skill
│ │ └── references/ <- Vercel React performance rules
│ ├── mcp-builder/
│ │ ├── SKILL.md
│ │ └── references/ <- Anthropic MCP guides (TS, Python, eval)
│ ├── designer/
│ │ ├── SKILL.md
│ │ ├── scripts/ <- BM25 search engine
│ │ └── references/ <- Anti-AI-slop aesthetics guide
│ └── ...
├── workflows/ <- 15 chainable workflows
├── rules/ <- 14 always-on rules
└── known-registries.json <- External skill sources
This kit integrates production-grade methodologies from industry leaders:
| Source | License | Skills Enhanced |
|---|---|---|
| Trail of Bits | — | qa-tester v3 (security audit), code-review v2 (differential review), modern-python (uv/ruff/ty) |
| Vercel | — | frontend-developer v4 (65 React performance patterns, composition rules) |
| Dean Peters | — | project-manager v5 (discovery process, PRD framework, Jobs-to-be-Done) |
| Anthropic | Apache 2.0 | mcp-builder (MCP server guide + eval), designer v4 (anti-AI-slop aesthetics) |
| Superpowers | MIT | debugging (rationalization prevention, defense-in-depth), code-review (anti-performative review), qa-tester (TDD discipline), /brainstorm (HARD-GATE) |
| EvoSkill | Apache 2.0 | self-improvement (feedback descent, evolutionary loop), prompt-engineer (abstraction ladder), ai-engineer (skill evolution), code-review (root cause analysis) |
Works with any AI coding assistant that reads .agent/ directories:
| Platform | Status |
|---|---|
| Claude Code (Anthropic) | Supported |
| Gemini CLI (Google) | Supported |
| Cursor (Anysphere) | Supported |
| GitHub Copilot (Microsoft) | Supported |
Skills follow the agentskills.io specification.
ag-kit init # Install full kit
ag-kit init --profile <name> # Install specific profile
ag-kit init --force # Overwrite existing installation
ag-kit update # Smart update — add new files only
ag-kit update --force # Update all files including changed
ag-kit list # Show available profiles and skills
ag-kit status # Check installed version
ag-kit verify # Run integrity checks
ag-kit help # Show all commands22 skills · 15 workflows · 14 rules · 6 profiles
6 absorbed sources · 2,800K+ knowledge base
MIT — jhm1909