Antigravity Agentic Coding Plugin 🚀

The Antigravity Agentic Coding Plugin is a powerful, autonomous coding assistant framework designed specifically for OpenCode. It transforms your development environment into an agentic powerhouse with structured task management, semantic memory, AST-level safety rules, and session-based undo capabilities.

🌟 Key Features

• 🧠 Autonomous Planning: Decomposes complex requests into a structured task graph with dependencies.
• 💾 Semantic Memory: Utilizes ChromaDB to remember past sessions, findings, and decisions.
• 🛡 AST-Level Safety: A rules engine that analyzes Python Abstract Syntax Trees to prevent dangerous operations.
• 🔍 Hybrid Search: Combines blazingly fast grep keyword search with semantic memory retrieval.
• ⏪ Advanced Session Revert: Full-session rollback capability. Revert multiple file changes and even deletions in one go.
• 📂 Artifact Management: Organized session-prefixed storage for generated files, logs, and research.
• ⚙️ OpenCode Native: Deep integration with OpenCode's tool and rule system.

🚀 Installation

Install the plugin into your project with a single command:

npx @lens.kamdem/agentic-coding-plugin

What happens during installation?

1. Core Logic: All complex backend logic is placed in a hidden .agentic-coding/ directory.
2. Configuration: An .opencode/ folder is created at your project root, pre-configured to hook into the Antigravity framework.
3. Clean Root: Your project root remains clean; only the configuration files and GEMINI.md are added.

🛠 Setup

1. Environment Variables: Create a .env file (inspired by .env.example) and add your LLM API keys.
2. Python Environment:

   python3 -m venv .agentic-coding/venv
   ./.agentic-coding/venv/bin/pip install -r .agentic-coding/python/requirements.txt

3. Run OpenCode:

   opencode

📖 How to Use

Antigravity Features in OpenCode

The plugin exposes several custom tools and commands that you can use directly within your OpenCode sessions:

1. Task Planning (/plan)

Ask the agent to plan a complex task. It will return a task graph.

• Command: /plan <request>
• Example: /plan "Refactor the authentication module to use JWT and add unit tests"

2. Status Tracking (/status)

Check the progress of your current tasks.

• Command: /status

3. Unified Search (/search)

Perform a hybrid search across code and session memory.

• Command: /search <query>
• Example: /search "How did we handle CORS in the last session?"

4. Safety Audit (/checkRules)

Manually trigger a rule check for a specific action.

• Command: /checkRules <action> --content <optional_code>

5. Change Management & Undo

Review your work and rollback if something goes wrong.

• /changes show: View all diffs recorded in the current session.
• /undo preview: See a summary of what will be reverted.
• /undo confirm: Execute the rollback (restores files and deletes newly created ones).

🧩 Advanced: Customizing Rules

Rules are defined as Markdown files with YAML frontmatter in .opencode/rules/.

Example rule (.opencode/rules/no-eval.md):

---
name: forbid-eval
description: Block usage of eval()
condition: "AST contains 'Call(eval)'"
action: block
---
The use of `eval()` is strictly forbidden for security reasons.

📁 Project Structure

your-project/
├── .opencode/                  # OpenCode integration hooks
│   ├── tools/                  # Compiled JS tools
│   ├── rules/                  # Safety rules
│   └── opencode.json           # Main plugin config
└── .agentic-coding/            # Antigravity Core (Hidden)
    ├── venv/                   # Python Virtual Environment
    ├── python/                 # Backend logic (planning, memory, rules)
    ├── chroma/                 # Semantic memory database (ChromaDB)
    ├── file_tracker.db         # SQLite change tracking
    └── artifacts/              # Session-prefixed artifacts

📜 License

MIT © Lenskam