gkt 3.11.3

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 209+ packaged AI Skills organized into 22 focused groups — each powered by 18 specialist agents, 47 workflows, and a persistent brain that remembers decisions across sessions.

---

⚡ Quick Start

Get started with GravityKit in 3 easy steps:

1. Install GravityKit

Install the CLI tool globally or in your virtual environment:

pip install gkt
cd /path/to/your-project

2. Initialize Skills & Agents

Choose your AI IDE to scaffold the required agent files and skill sets.

Goal	Antigravity Command	Kiro Command
Init all skills	gkt init antigravity	gkt init kiro
Init specific group	gkt init antigravity --group general-dev	gkt init kiro --group general-dev
Where files go	.agent/ folder	.kiro/ folder

Tip: You can replace general-dev with other groups like uipath, ai-agent, or seo-marketing.

3. Enable Semantic Code Graph & MCP Servers

Wiring up the Model Context Protocol (MCP) gives your agents native ability to search code via FAISS, read office documents, and access the Brain Manager.

Goal	Command (Both IDEs)
Generate & Wire MCP	gkt mcp

(Note: gkt mcp automatically detects your initialized IDEs and configures them all at once).

Requirements: Python 3.9+ Supported IDEs: antigravity, kiro, cursor, windsurf, cline, kilocode, copilot, claude (Claude Code)

For development / contributing:

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

---

🔌 Core Features & MCP Integration ⭐ New

GravityKit ships with powerful Model Context Protocol (MCP) servers out of the box. Running gkt mcp will automatically index your codebase and wire up these servers to your IDE:

1. Code Graph (code-graph): Builds a structural graph of your codebase using Tree-sitter. AI agents can traverse function calls, imports, and classes without reading entire files.
2. Semantic Search (faiss-code-index): Generates an embedded vector index using FAISS. AI agents can perform deep semantic searches across your code.
3. Document Reader (document-reader): Allows AI agents to natively read .pdf, .docx, and .xlsx files without you having to copy-paste the contents.
4. Brain Manager (brain-manager): Grants agents direct read/write access to project architecture decisions, conventions, and checkpoints.

---

📦 Skill Groups

Install only the skills your project needs:

gkt groups   # List all available groups

Group	Skills	+Default	Description
general-dev	39	+9	Backend, Frontend, Full-stack development
n8n-dev	18	+11	n8n workflow automations
nocobase-dev	32	+12	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	29	+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	23	+11	Claude Code skills, prompt engineering, MCP
context-data-rag	29	+12	Context engineering, RAG, data pipelines
database	26	+11	SQL/NoSQL, Postgres, migrations, CQRS
observability-report	22	+13	Monitoring, Grafana, Prometheus, SLO, incidents
uipath	6	+13	UiPath RPA — XAML generation, REFramework, Orchestrator, coded workflows, AI agents
gen-doc ⭐ New	1	+14	Document Generation — PPTX pipeline

+Default: Each group automatically includes 14 base skills for memory management, lifecycle, error handling, code graph indexing, and cross-platform compatibility.

Usage:

gkt init general-dev                    # Install group (default: Antigravity IDE)
gkt init antigravity --group uipath     # UiPath skills for Claude Code
gkt init kiro --group uipath            # UiPath skills for Kiro IDE
gkt init antigravity                    # Install ALL skills

---

🛠️ 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 mcp	Wire MCP servers (Code Graph, FAISS, Document Reader) into IDEs
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

GravityKit installs into 7 AI IDEs from a single CLI command:

IDE	Command	Creates
Antigravity	gkt init antigravity	.agent/ — agents, workflows, skills, brain
Kiro	gkt init kiro	.kiro/ — skills, steering (always-loaded), hooks, specs, agents, brain
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
All	gkt init	All of the above

IDE-specific features

Kiro receives the full agent toolkit:

• .kiro/steering/brain.md — always-loaded; instructs Kiro to load brain context and check in-progress workflow sessions at every session start
• .kiro/steering/product.md, structure.md, tech.md — project context templates
• .kiro/brain/ — full brain directory including workflow_sessions/ for session continuity
• .kiro/specs/ — workflow files for Kiro spec execution
• .kiro/agents/ — 18 specialist agent definitions

GitHub Copilot receives per-role instruction files (.github/instructions/*.instructions.md) covering all 14 specialist roles. Each role file includes a Session Init block that loads brain context before starting work.

---

🤖 Agent Team (18 Specialists)

GravityKit ships 18 specialist agents as native sub-agent definitions (.agent/agents/*.md), compatible with Claude Code's sub-agent format:

Agent	Role
leader	Orchestrates the full team; delegates to specialists via sub-agents
quickstart	Fully automated project build — no approvals needed
planner	Requirements analysis, PRD, task breakdown
architect	System design, database schema, API spec, diagrams
designer	UI/UX design system, wireframes, design tokens
frontend-dev	React/Vue/Tailwind — components, layouts, state
backend-dev	Node/Python/Go — APIs, DB queries, auth
mobile-dev	React Native/Expo — iOS and Android
devops	Docker, CI/CD, Kubernetes, cloud deployment
qa-engineer	Test cases, API tests, performance, automation
code-reviewer	Code quality, design patterns, security, performance
security-engineer	Threat modeling, SAST, pen-test, incident response
tech-writer	Docs, API references, ADR, RFC, tutorials
researcher	Market analysis, web search, competitive intelligence
meta-thinker	Creative advisor, brainstorming, vision expansion
knowledge-guide	Codebase onboarding, session handoffs, knowledge transfer
release-manager	Version bumps, changelog, git tagging, release notes
seo-specialist	Meta tags, schema markup, Core Web Vitals, sitemap

Agents are installed to .agent/agents/ (antigravity) or .kiro/agents/ (kiro) during gkt init.

---

🚀 How It Works

Mode 1: /wf-leader — Smart Delegation

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

Phase	Agent(s)	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 (46)

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

General Development

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-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-release-manager	Changelog generation, version bumping
/wf-tech-writer	Documentation & API refs
/wf-doc-writer	Professional technical documentation, reports, RFC, ADR
/wf-knowledge-guide	Code explainer, note taker, handoff specialist

Specialized

Workflow	Description
/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-seo-specialist	SEO audit, meta tags, schema markup, Core Web Vitals
/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-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-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
/wf-gen-doc ⭐ New	Document Generator — orchestrates the PPTX generation pipeline

UiPath RPA ⭐ New

Workflow	Description
/wf-uipath-project ⭐	All-in-one: business analysis → brainstorm → plan → XAML implementation
/wf-uipath-analyst	Business analysis, process decomposition, architecture selection
/wf-uipath-developer	XAML generation, REFramework wiring, UI inspection, validate-and-fix loop
/wf-uipath-reviewer	Code review for XAML quality, rule compliance, security, naming conventions
/wf-uipath-deploy	Package, publish to Orchestrator, schedule, monitor

---

🤖 UiPath RPA Group

The uipath skill group provides end-to-end RPA project automation using UiPath Studio. Use gkt init antigravity --group uipath or gkt init kiro --group uipath.

Skills included

Skill	Description
uipath-core	XAML generation scripts, scaffold, validate, lint, decomposition rules
uipath-rpa-workflows	REFramework patterns, Dispatcher+Performer, Simple Sequence templates
uipath-shared	CLI reference, UI Automation prerequisites, selectors, debugging guide
uipath-platform	Orchestrator queues, assets, processes, folder management
uipath-coded-workflows	C# coded workflows, unit testing, OOP patterns in UiPath Studio
uipath-agents	UiPath AI agents with LLM reasoning and dynamic decision making

/wf-uipath-project — All-in-One Workflow

The flagship UiPath workflow covers the full project lifecycle in one command:

Phase 0: Session Resume Check
    ↓ (loads brain/workflow_sessions/ if previous session exists)
Phase 1: Business Analysis & Brainstorm
    → 10 elicitation questions
    → Key Business Points: value, complexity, constraints
    → Architecture selection: REFramework / Simple Sequence / Dispatcher+Performer
    ↓ [Checkpoint saved to brain]
Phase 2: Plan & Task Breakdown
    → Workflow tree decomposition
    → Argument map (In/Out per workflow)
    → Config.xlsx keys
    → Dev Sequence A→E
    ↓ [Checkpoint saved to brain]
Phase 3: Implement XAML
    → Scaffold project (scaffold_project.py)
    → UI Inspection — MANDATORY before any generation
    → Generate workflows (generate_workflow.py) + validate loop
    → Wire Main.xaml / Process.xaml
    ↓ [Checkpoint saved to brain per sub-phase]
Phase 4: Final Validation & Handoff
    → XAML quality checklist
    → Handoff document
    ↓ [Session marked complete in brain]

---

📝 Gen-Doc Group ⭐ New

The gen-doc skill group provides an automated pipeline for generating AI-designed PPTX presentations. Use gkt init antigravity --group gen-doc or gkt init kiro --group gen-doc.

Skills included

Skill	Description
gen-doc-ppt-master	Core document presentation skill: SVG templating and PPTX export

/wf-gen-doc — Document Generator Workflow

The document generator workflow takes you from requirements to a fully editable artifact:

Phase 0: Environment Validation
    ↓ Auto-installs required environment dependencies
Phase 1: Source Material Assimilation
    → Import and convert PDF/DOCX/URL/TEXT into Markdown
Phase 2: Template Selection & Strategy
    → Determine layouts, styling, and compile a design blueprint
Phase 3: Visual & Content Execution
    → Generate AI imagery and bind content into vector pages
Phase 4: Finalize & Export
    → Post-process SVG layers and export directly to native PPTX

---

📂 Project Structure

.agent/
├── agents/              # 18 specialist agent definitions (Claude Code native format)
│   ├── leader.md
│   ├── architect.md
│   └── ... (16 more)
├── workflows/           # 47 workflows (wf-* prefix)
│   ├── wf-leader.md
│   ├── wf-uipath-project.md
│   └── ...
├── skills/              # 209+ packaged skills across 22 groups + 14 default
│   ├── uipath-core/
│   ├── brain-manager/
│   └── ...
└── brain/               # Persistent project memory
    ├── default_skills.md          # Default skill reference & integration guide
    ├── skills_manifest.json       # Lightweight index of all skills (lazy-loading)
    ├── lifecycle.md               # Session phases: 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 registry
    └── workflow_sessions/         # Cross-session workflow continuity ⭐ New
        ├── SESSIONS.md            # Active session index + checkpoint format spec
        ├── wf-uipath-project-latest.md   # Latest session state (always current)
        └── wf-{name}-{date}.md           # Per-run dated artifacts

---

🧠 Brain System

Every installation includes the Brain — persistent memory that agents read at session start and write to at each checkpoint.

Core brain files

Brain File	Purpose
default_skills.md	14 default skills, usage guide, workflow wiring
skills_manifest.json	Lightweight index of all packaged 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

Workflow Session Continuity ⭐ New

brain/workflow_sessions/ stores structured session artifacts — similar to how Claude Code compacts conversations. When a workflow saves a checkpoint, a new session can resume exactly where the previous one left off.

How it works:

1. First run: workflow creates {wf-name}-{date}.md + {wf-name}-latest.md
2. Each phase end: agent appends phase summary (decisions, outputs, open questions)
3. New session: lifecycle.md init phase checks workflow_sessions/ automatically:

   📋 Found session: InvoiceAutomation — paused at Phase 2
   Resume from that point, or restart from Phase 0? (resume/restart)
   

4. Resume: skips completed phases (✅), continues from in-progress phase (🔄)

Session artifact format:

---
workflow: wf-uipath-project
project: InvoiceAutomation
session_date: 2024-01-15
last_phase: "Phase 2 — Plan"
status: in_progress
---

## Phase 1 — Business Analysis ✅
**Architecture selected:** REFramework Performer — 200+ invoices/day, needs per-item retry
**Apps:** Outlook (email), SAP (ERP entry)

## Phase 2 — Plan ✅
**Workflow tree:** 8 workflows across Outlook/ and SAP/ folders
**Next steps:** Start Phase A scaffold

IDE support:

IDE	Session file location	How it's loaded
Antigravity	.agent/brain/workflow_sessions/	lifecycle.md Init Phase instruction
Kiro	.kiro/brain/workflow_sessions/	steering/brain.md (always-loaded steering)
GitHub Copilot	.agent/brain/workflow_sessions/	leader.instructions.md Session Init block

14 Default Skills (Always Installed)

Category	Skills
Memory & Context	brain-manager, journal-manager, context-manager, codebase-navigator, code-graph-index
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 v3.8.0
git push origin v3.8.0

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