gkt 3.8.0

GravityKit — The AI-Native Software House in a Box

🌌 Anti-Gravity Kit

The AI-Native Software House in a Box. Build enterprise-grade software with a team of 893+ AI Skills organized into 21 focused groups, each enhanced with 13 default base skills for memory, lifecycle, and cross-platform compatibility.

---

⚡ Quick Start

# Install from PyPI
pip install gkt

# Init in your project (all skills)
cd /path/to/your-project
gkt init antigravity

# Or install only the skills you need
gkt init general-dev     # 30 core dev skills
gkt init n8n-dev         # 18 n8n automation skills
gkt init ai-agent        # 28 AI/LLM/RAG skills

Requirements: Python 3.9+

For development / contributing:

git clone https://github.com/OrgGem/VibeGravityKit.git
cd VibeGravityKit
pip install -e .

---

📦 Skill Groups

Instead of loading all 887 skills, install only the group you need:

gkt groups   # List all available groups

Group	Skills	+Default	Description
general-dev	31	+9	Backend, Frontend, Full-stack development
n8n-dev	18	+10	n8n workflow automations
nocobase-dev	26	+11	NocoBase plugin development
general-doc	20	+11	Technical docs, RFC, ADR, wiki
research	18	+12	Deep research, market analysis
cloud-deploy	26	+12	Cloud, DevOps, CI/CD, Kubernetes
security-audit	14	+13	Threat modeling, SAST, vulnerability analysis
security-pentest	14	+13	Offensive security, red team, exploitation
seo-marketing	28	+13	SEO, CRO, content marketing
ai-agent	28	+13	LLM apps, RAG, multi-agent
saas-crm	8	+13	HubSpot, Salesforce, Stripe connectors
saas-comms	6	+13	Slack, Discord, WhatsApp, Gmail connectors
saas-project	8	+13	Jira, Asana, Trello, Notion connectors
saas-marketing	6	+13	Twitter, LinkedIn, Google Suite connectors
startup-biz	26	+13	Market analysis, financial modeling
api-graphql	24	+12	API design, GraphQL, REST, OpenAPI, FastAPI
claude-code	22	+11	Claude Code skills, prompt engineering, MCP
context-data-rag	28	+12	Context engineering, RAG, data pipelines
database	26	+11	SQL/NoSQL, Postgres, migrations, CQRS
observability-report	22	+13	Monitoring, Grafana, Prometheus, SLO, incidents

+Default: Each group automatically includes 13 base skills for memory management, lifecycle, error handling, and cross-platform compatibility. The +N value shows how many unique defaults are added (some groups already include overlapping skills).

Usage:

gkt init general-dev                    # Install group (default: Antigravity IDE)
gkt init antigravity --group n8n-dev    # Specify IDE + group
gkt init antigravity                    # Install ALL skills (legacy behavior)

---

🛠️ CLI Commands

Command	Description
gkt init [target]	Install skills for an IDE or a skill group
gkt init [target] --group <name>	Install a specific skill group for an IDE
gkt groups	List all available skill groups
gkt list	List all available AI agents and their roles
gkt doctor	Check environment health (Python, Node, Git, npm)
gkt update	Update GravityKit to the latest version
gkt version	Show current version
gkt brain	Manage project brain — context, decisions, conventions
gkt journal	Knowledge journal — capture lessons, bugs, insights
gkt skills list [--all]	List active skills (or all including disabled)
gkt skills search <query>	Search skills by keyword
gkt skills enable <name>	Enable a disabled skill
gkt skills disable <name>	Disable a skill
gkt skills count	Show total skill count
gkt validate [--strict]	Validate all SKILL.md files
gkt generate-index	Regenerate skills_index.json

Alias: gravitykit works the same as gkt.

---

🌐 Multi-IDE Support

IDE	Command	Creates
Antigravity	gkt init antigravity	.agent/ (workflows + skills)
Cursor	gkt init cursor	.cursor/rules/*.mdc
Windsurf	gkt init windsurf	.windsurf/rules/*.md
Cline	gkt init cline	.clinerules/*.md
Kilo Code	gkt init kilocode	.kilocode/rules/*.md
GitHub Copilot	gkt init copilot	.github/instructions/*.instructions.md
Kiro	gkt init kiro	.kiro/ (skills, hooks, steering, specs)
All	gkt init	All of the above

---

🚀 How It Works

Mode 1: @[/wf-leader] — Smart Delegation

You → Leader → Agents → Report per phase → You approve → Next phase

Phase	Agent	Mode
📋 Planning	@[/wf-planner]	Sequential
🏗️ Architecture + 🎨 Design	@[/wf-architect] + @[/wf-designer]	⚡ PARALLEL
💻 Development	@[/wf-frontend-dev] + @[/wf-backend-dev]	⚡ PARALLEL
🧪 QA & Fix	@[/wf-qa-engineer]	Sequential
🚀 Launch	@[/wf-devops] + @[/wf-security-engineer] + @[/wf-seo-specialist] + @[/wf-tech-writer]	⚡ PARALLEL

Mode 2: @[/wf-quickstart] — Full Autopilot

One command, complete project. No approvals needed. Best for MVPs and prototypes.

💡 Tip: After gkt init <group>, a workflow guide is displayed showing all available workflows with descriptions and sample prompts.

---

🔄 Workflows (41)

Naming convention: All workflows use the wf- prefix. Type /wf- in your AI chat to filter only workflows.

Workflow	Description
/wf-leader	Orchestrates the entire team from concept to production
/wf-quickstart	Fully automated project build from idea to production
/wf-planner	Analyzes requirements, writes PRD, breaks down tasks
/wf-meta-thinker	Creative advisor, brainstorming, vision development
/wf-architect	Systems design, database, API
/wf-solution-architect	Strategic technical planning, trade-off analysis
/wf-designer	UI/UX design system and assets
/wf-frontend-dev	Component, layout, state management (React/Vue/Tailwind)
/wf-backend-dev	API implementation, DB queries (Node/Python/Go)
/wf-fullstack-coder	Architecture, backend, frontend, testing in one workflow
/wf-mobile-dev	iOS/Android (React Native/Expo)
/wf-devops	Docker, CI/CD, cloud deployment
/wf-cloud-deployer	AWS deployment, CI/CD, Docker, Kubernetes, serverless
/wf-n8n-automator	n8n workflow builder with 70+ SaaS connectors
/wf-nocobase-plugin-expert	NocoBase plugin development (Server, Client, DB, API)
/wf-nocobase-plugin-build	Build NocoBase plugins
/wf-qa-engineer	Test case, API, SQL, automation, performance
/wf-quality-guardian	Code review, testing, security audit in one pass
/wf-code-reviewer	Automated code quality review
/wf-security-engineer	Security workflow (Audit/Pen-Test/Incident)
/wf-security-auditor	Penetration testing, vulnerability assessment
/wf-seo-specialist	Search engine optimization
/wf-seo-marketer	SEO optimization, content strategy, CRO
/wf-ai-agent-builder	Build LLM apps, RAG systems, multi-agent architectures
/wf-saas-connector	Automate 20+ SaaS platforms via API integrations
/wf-startup-advisor	Market analysis, financial modeling, GTM planning
/wf-tech-writer	Documentation & API refs
/wf-doc-writer	Professional technical documentation, reports, RFC, ADR
/wf-knowledge-guide	Code explainer, note taker, handoff specialist
/wf-researcher	Market analysis, web search, trend discovery
/wf-research-analyst	Deep research, analysis, file I/O, translation
/wf-deep-researcher	Comprehensive research and professional reports
/wf-release-manager	Changelog generation, version bumping
/wf-prompt-engineer	Create optimized prompts for any AI model
/wf-image-creator	AI image generation, design assets, diagrams
/wf-translator	Multi-language translation and i18n management
/wf-api-graphql-dev	API & GraphQL development workflow
/wf-claude-code-dev	Claude Code skills, prompt engineering, MCP
/wf-context-data-eng	Context engineering, RAG, data pipelines
/wf-database-eng	Database design, optimization, migrations, CQRS
/wf-observability-eng	Monitoring, tracing, Grafana, Prometheus, SLO

---

📂 Project Structure

.agent/
├── workflows/       # Instructions for each agent role (41 workflows, wf-* prefix)
├── skills/          # 893+ skills across 19 groups + 13 default
└── brain/           # Project context & memory
    ├── default_skills.md     # Default skill reference & integration guide
    ├── skills_manifest.json  # Lightweight index of all skills (lazy-loading)
    ├── lifecycle.md          # Session lifecycle (init → plan → work → checkpoint → handoff)
    ├── platform_notes.md     # Cross-platform compatibility (Windows/Linux/macOS)
    ├── project_context.json  # Project metadata, architecture, decisions, conventions
    └── agent_index.json      # Agent role definitions and handoff templates

---

🧠 Brain System

Every installation includes the Brain — a structured knowledge base that agents read at session start:

Brain File	Purpose
default_skills.md	Documents 13 default skills, usage guide, workflow wiring
skills_manifest.json	Lightweight index of ALL 893+ skills for lazy-loading
lifecycle.md	Session lifecycle: init, plan, work, checkpoint, handoff
platform_notes.md	Cross-platform guide: paths, shells, encoding, permissions
project_context.json	Project metadata template (tech stack, conventions, sprint)
agent_index.json	Agent role definitions and handoff protocol

The brain files reference the 13 default skills, allowing agents to automatically:

• Load context from previous sessions (brain-manager)
• Break tasks into atomic checklists (concise-planning)
• Apply platform-specific commands (powershell-windows / bash-linux)
• Debug proactively (debugger)
• Handle errors consistently (error-handling-patterns)
• Maintain clean code and semantic commits (clean-code, git-manager, commit)
• Lazy-load additional skills via skills_manifest.json (saves 40-60% tokens)

13 Default Skills (Always Installed)

Category	Skills
Memory & Context	brain-manager, journal-manager, context-manager, codebase-navigator
Planning & Quality	concise-planning, writing-plans, clean-code, debugger, error-handling-patterns
Version Control	git-manager, commit
Cross-Platform	powershell-windows, bash-linux

---

🧰 Token Optimization

Tool	What it does	Savings
Skills Manifest	Lightweight index — load full SKILL.md on demand	~50% tokens
Context Manager	Minifies code before AI sees it	~50% tokens
Context Router	Queries only relevant data from 34+ sources	~70% tokens
Diff Applier	Applies surgical patches instead of rewriting	~90% tokens

---

🚢 Publishing to PyPI

This package is published as gkt on PyPI. Publishing is automated via GitHub Actions.

Auto-Publish on Tag Push

1. Update the version in GravityKit/VERSION
2. Commit and push changes
3. Create and push a version tag:

git tag v3.2.1
git push origin v3.2.1

The GitHub Actions workflow (.github/workflows/publish.yml) will:

• Build the package (python -m build)
• Check with Twine (twine check dist/*)
• Publish to PyPI via OIDC trusted publishing

Note: OIDC trusted publishing is configured. No API token needed if the PyPI project is linked to this GitHub repo.

---

📄 License

MIT