Antigravity Skills, MCP, and Memory

! MCP coming soon

A reusable Git template repo for high-performance agentic coding with Antigravity. Supports legacy (PHP/WordPress/LAMP) and modern stacks (Python/FastAPI, Next.js/React Native) with modular skills, planning docs, persistent memory, MCP for external references, and optimized context loading.

Quick Start

1. Clone this repo as a template (via "Use this template" on GitHub) or fork it for a new project.

2. Open in Antigravity as a workspace.

3. Initialize Project:

   • Auto: Prompt the agent "Initialize this new project".
   • Manual: Run python .agent/tools/init_project.py "My Project Name".
   • This creates folder structures, memory.json, and initial docs.

4. Customize:

   • Required: Edit /docs/planning/prd.md with your project vision.
   • Optional: Fill out scope.md, technical-specs.md, and user-stories.md for better agent guidance.
   • Review .agent/skills/ and skills-manifest.md to see available capabilities.
   • Run local install commands for dependencies (e.g., composer install or npm install).

5. Start prompting for features:
   "Add user authentication with OAuth"
   The agent will follow the workflow in docs/planning/agent-workflow.md.

Folder Structure Overview

gravity-boots/ ├── .agent/ # Agent config │ ├── skills/ # Modular skills (flat) │ │ └── project-init/ # Example: the init skill │ │ └── SKILL.md │ ├── rules/ # Agent rules (coding standards) │ │ ├── php-standards.md │ │ └── ... │ └── tools/ # Helper scripts │ └── memory_manager.py # Memory management ├── docs/ # Documentation │ ├── planning/ # Project planning │ │ ├── prd.md │ │ ├── scope.md │ │ ├── technical-specs.md │ │ ├── user-stories.md │ │ ├── definition-of-done.md │ │ ├── current-plan.md # Active task plan │ │ └── agent-workflow.md # Workflow diagram + steps │ ├── context/ # Runtime state │ │ └── memory.json # Persistent memory │ └── mcp/ # Mounted Context Protocol (external docs) │ └── laravel-11-guide.md # Example: Laravel docs ├── sql/ # Generated SQL scripts ├── examples/ # Generated examples │ └── json/ # JSON fixtures ├── src/ # Code │ ├── backend/ # PHP / Python │ │ ├── python/ # FastAPI/Django/Flask stubs │ │ └── php/ # Laravel/Symfony stubs │ └── frontend/ # Next.js / React Native ├── .env.example # Env vars template ├── .gitignore # Standard ignores ├── AGENTS.md # Agent persona/rules ├── README.md # This file └── skills-manifest.md # Skills list/reference

Agent Workflow (Mermaid Diagram)

flowchart TB
    A["User Prompt"] --> B{"Agent Clarifies Understanding + Recommends Prompt Updates"}

    B --> C{"Is this a new project or existing / ongoing?"}
    
    C -->|New / Init / Bootstrap / Start| NewPath["Run project-init skill\nCreate structure, memory.json, stubs, first commit"]
    C -->|Existing / Add feature / Fix / Refactor| ExistingPath["Read memory.json first\nLoad relevant skills based on prompt + open files"]

    NewPath --> D["Load Memory + Planning Skills + PRD Summary"]
    ExistingPath --> D

    D --> E{"User Approves Prompt / Intent?"}
    E -->|Yes| F["Create Granular Plan\n• Suggested skills per task\n• Memory.json updates if needed"]
    E -->|No / Revise| G["Ask Clarifying Questions or Revise Prompt"]
    G --> E

    F --> P1{"User Approves Plan?"}
    P1 -->|Yes| Exec["Begin Executing Tasks"]
    P1 -->|No / Revise| P2["Ask Clarifying Questions or Revise Plan"]
    P2 --> P1

    Exec --> T1["Test Task Output"]
    T1 --> S{"Test Success?"}
    S -->|Yes| Lint{"Lint & Validation Clean?"}
    S -->|No| Fix["Fix Issues"]
    Lint -->|Yes| Mark["Mark Task Complete in current-plan.md"]
    Lint -->|No| Fix
    Fix --> T1
    Fix --> P3{"User Approval Needed?"}
    P3 -->|Yes| P4["Ask Clarifying Questions or Revise"]
    P3 -->|No| T1
    P4 --> P3

    Mark --> Git["Git Operations\n- git pull\n- Commit with Conventional Commit\n- git push (with safety checks)"]
    Git --> Done["Update Changelog / Memory / Docs\nAlert User / Loop to Next Item"]

    Done --> End[Task Complete\nLoop to Next Item or End]

    %% Styling
    classDef decision fill:#bbf,stroke:#333,stroke-width:2px
    classDef action fill:#dfd,stroke:#333
    class C,E,P1,S,Lint,P3,P4 decision
    class B,D,F,Exec,T1,Fix,P2,P4,Done,NewPath,ExistingPath,Mark,Git action

Loading

Key Features

• Modular skills with progressive loading (1–4 active at a time)
• Persistent memory.json for state across sessions
• MCP for external docs (/docs/mcp/)
• Yes/no approval loop + silent execution mode
• Artifact organization (/sql/, /examples/json/)
• Rule-based coding standards (.agent/rules/) for auto-compliance
• Lint/validation before commits
• Supports mixed stacks: PHP, Python, Next.js, etc.

Deployment & Production

Important: The .agent/ folder, docs/, and AGENTS.md are your "Source Code for the Agent".

1. Repository: Keep these files in Git. The agent needs them to understand and build your project.
2. Production Server: Exclude them. They are not needed for the runtime application.
   • Use .dockerignore (provided) to exclude them from container builds.
   • Use rsync --exclude-from if deploying via SSH.

Contributing

Pull requests welcome! Please use Conventional Commits.

License

MIT License

Last Updated: January 23, 2026