Overview
AG Kit is a comprehensive agent engineering kit designed for the Google Antigravity runtime. It empowers developers to build sophisticated agent-based systems with a focus on robust rules, specialized skills, persistent memory, and intelligent orchestration. AG Kit prioritizes safety with a native hook for high-confidence destructive command prevention.
Key Features
- Antigravity-First Runtime: Seamless integration with Google Antigravity for production deployment.
- Modular Agent Architecture: Features a structured
.agents/workspace with dedicated directories for rules, skills, workflows, and memory. - Specialist Agents & Routing: Includes 20 pre-defined role definitions and intelligent routing skills for task delegation.
- Persistent Memory: Utilizes
.agents/memory/for durable storage of project conventions, decisions, and feedback, with context-compression guidance. - Advanced Orchestration: Supports
/coordinate,/orchestrate, and Antigravity's/agentsand/tasksfor complex workflow management. - Native Safety Hook: A critical
PreToolUsegate that blocks high-risk commands (e.g., root filesystem deletion, drive formatting) while allowing safe project cleanup. - Workflow Automation: A rich set of common workflows like
/plan,/create,/debug, and/deployto streamline development processes. - Merge-Aware Updates & Rollback: Safe update mechanisms that preserve user-owned and modified files, with built-in rollback capabilities.
- Reproducible Toolkits: Manages dependencies and detects drift through manifest files and a dependency graph.
- Packaging & Validation: Bundles components into local Antigravity plugin packages and provides extensive validation tools.
Typical Use Cases
- Automated Software Development: Generate, debug, enhance, and deploy applications using intelligent agents.
- Complex Task Orchestration: Break down and manage intricate workflows, delegating tasks to specialist agents.
- Knowledge Management: Persistently store and retrieve project-specific information, decisions, and conventions.
- Safe Command Execution: Ensure that agent actions are controlled and that potentially destructive commands are prevented by default.
- Rapid Prototyping: Quickly explore architecture, brainstorm ideas, and create structured features with agent assistance.
- Production Readiness: Leverage a production-ready runtime with built-in safety checks and comprehensive validation.
