google_antigravity_config.json

{
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "$id": "google-antigravity-configuration-v1.14",
  "title": "Google Antigravity Configuration & Attribute Specification",
  "version": "1.14.2-preview",
  "lastUpdated": "2026-01-20",

  "identity": {
    "agentName": "Google Antigravity",
    "vendor": "Google",
    "releaseChannel": "Public Preview (v1.14.2)",
    "formFactor": "Standalone IDE (Modified VS Code Fork)",
    "licenseModel": "Proprietary (Free during preview for individuals)",
    "costArchitecture": {
      "individuals": "Free for personal Gmail accounts during preview",
      "enterprise": "Integration via Gemini Code Assist"
    },
    "installation": {
      "methods": ["System Installer (.dmg, .exe, .deb, .rpm, snap)"],
      "downloadPage": "https://antigravity.google/download",
      "commandLine": "agy (installed via Configure your Editor settings)"
    }
  },

  "configurationFiles": {
    "globalRules": {
      "file": "GEMINI.md",
      "location": "~/.gemini/GEMINI.md",
      "description": "Global rules that apply to all projects",
      "purpose": "Personal coding philosophy, language preferences, documentation standards"
    },
    "workspaceRules": {
      "location": "<workspace>/.agent/rules/",
      "format": "Markdown files",
      "description": "Project-specific rules that override global rules"
    },
    "globalWorkflows": {
      "location": "~/.gemini/antigravity/global_workflows/",
      "file": "global-workflow.md",
      "description": "Saved prompts available globally via / commands"
    },
    "workspaceWorkflows": {
      "location": "<workspace>/.agent/workflows/",
      "description": "Project-specific workflows triggered with /"
    },
    "globalSkills": {
      "location": "~/.gemini/antigravity/skills/",
      "file": "SKILL.md",
      "description": "User-specific skills available across all projects"
    },
    "workspaceSkills": {
      "location": "<workspace>/.agent/skills/",
      "file": "SKILL.md",
      "description": "Project-specific skills"
    },
    "browserAllowlist": {
      "location": "~/.gemini/antigravity/browserAllowlist.txt",
      "format": "Plain text, one domain per line",
      "description": "Trusted domains for browser agent"
    },
    "mcpConfig": {
      "location": "~/.gemini/antigravity/mcp_config.json",
      "description": "MCP server configurations"
    },
    "antigravitySettings": {
      "location": "~/.gemini/antigravity/",
      "accessMethod": "Antigravity — Settings (Cmd + ,)"
    }
  },

  "cognitiveArchitecture": {
    "defaultModel": {
      "value": "Gemini 3 Pro",
      "contextWindow": "2 Million Tokens"
    },
    "supportedModels": {
      "google": ["Gemini 3 Pro", "Gemini 3 Deep Think", "Gemini 3 Flash"],
      "anthropic": ["Claude Sonnet 4.5", "Claude Opus 4.5"],
      "openai": ["GPT-OSS-120B (Open-source variant)"]
    },
    "agentModes": {
      "planning": {
        "default": true,
        "description": "Agent plans before executing tasks. Use for deep research, complex tasks, or collaborative work.",
        "behavior": "Organizes work in task groups, produces Artifacts, thoroughly researches and plans"
      },
      "fast": {
        "description": "Agent executes tasks directly. Use for simple tasks like renaming variables or running commands.",
        "behavior": "Speed-focused, less thinking budget consumed"
      }
    },
    "planningMode": "Explicit - Generates Task Lists and Implementation Plans before coding",
    "selfCorrectionCapability": "High - Includes 'Explain and fix' and 'Send problems to agent' workflows",
    "orchestrationPattern": "Asynchronous Agent Manager handling parallel task execution"
  },

  "settings": {
    "terminalExecutionPolicy": {
      "key": "Terminal Command Auto Execution",
      "location": "Antigravity — Settings → Advanced Settings → Terminal",
      "options": {
        "off": {
          "description": "Never auto-execute terminal commands except those in Allow list",
          "securityLevel": "Most secure"
        },
        "auto": {
          "description": "Agent decides whether to auto-execute based on internal safety models",
          "default": true,
          "behavior": "Asks permission for risky commands"
        },
        "turbo": {
          "description": "Always auto-execute terminal commands except those in Deny list",
          "warning": "High Risk - Not recommended for enterprise use"
        }
      }
    },
    "javascriptExecutionPolicy": {
      "key": "JavaScript Execution policy",
      "options": {
        "alwaysProceed": {
          "description": "Agent will not stop to ask for permission to run JavaScript in browser",
          "warning": "Highest exposure to security exploits"
        },
        "requestReview": {
          "description": "Agent will always stop to ask for permission to run JavaScript",
          "default": true
        },
        "disabled": {
          "description": "Agent will never run JavaScript code in browser"
        }
      }
    },
    "presetModes": {
      "secureMode": {
        "description": "Enhanced security controls restricting access to external resources",
        "features": "Several security measures enforced to protect environment"
      },
      "reviewDrivenDevelopment": {
        "description": "Agent frequently asks for review",
        "recommended": true,
        "note": "Good balance - agent makes decisions and returns for approval"
      },
      "agentDrivenDevelopment": {
        "description": "Agent will never ask for review",
        "warning": "Maximum autonomy"
      },
      "customConfiguration": {
        "description": "Fully customizable policies"
      }
    },
    "allowListTerminalCommands": {
      "location": "Advanced Settings → Terminal → Allow List Terminal Commands",
      "usage": "With 'Off' policy, only listed commands auto-execute",
      "securityModel": "Positive - everything forbidden unless permitted",
      "example": ["ls -al", "git status", "npm test"]
    },
    "denyListTerminalCommands": {
      "location": "Advanced Settings → Terminal → Deny List Terminal Commands",
      "usage": "With 'Turbo' policy, listed commands require approval",
      "securityModel": "Negative - everything allowed unless forbidden",
      "recommendedBlocks": ["rm", "rmdir", "sudo", "curl", "wget"]
    },
    "browserUrlAllowlist": {
      "location": "Advanced Settings → Browser → Browser URL Allowlist → Open Allowlist File",
      "file": "~/.gemini/antigravity/browserAllowlist.txt",
      "purpose": "Prevent prompt injection from compromised sites",
      "format": "Domain names, one per line"
    },
    "telemetry": {
      "key": "Enable Telemetry",
      "location": "Settings → Account",
      "default": true
    }
  },

  "rulesAndWorkflows": {
    "rules": {
      "purpose": "System instructions - passive, persistent guidelines always 'on'",
      "behavior": "Agent considers these before generating any code or plan",
      "scope": {
        "global": "~/.gemini/GEMINI.md",
        "workspace": "<workspace>/.agent/rules/<rule-name>.md"
      },
      "useCases": [
        "Language preferences (e.g., Always use TypeScript over JavaScript)",
        "Documentation standards (e.g., Always add docstrings)",
        "Security guidelines (e.g., Never generate hardcoded API keys)",
        "Code style preferences"
      ]
    },
    "workflows": {
      "purpose": "Saved prompts triggered on demand with / commands",
      "behavior": "User-triggered, not automatic",
      "scope": {
        "global": "~/.gemini/antigravity/global_workflows/",
        "workspace": "<workspace>/.agent/workflows/"
      },
      "triggering": "/ + workflow name",
      "analogy": "Macros of the agentic world"
    },
    "precedence": "Workspace rules in .agent/rules/ override global GEMINI.md rules"
  },

  "skills": {
    "purpose": "Agent-triggered capabilities invoked via natural language semantic matching",
    "distinction": "Unlike slash commands - agent decides when to use based on task matching",
    "location": {
      "global": "~/.gemini/antigravity/skills/<skill-name>/",
      "workspace": "<workspace>/.agent/skills/<skill-name>/"
    },
    "structure": {
      "SKILL.md": "Required - Brain of the skill with metadata and instructions",
      "scripts/": "Optional - Executable scripts for complex operations",
      "templates/": "Optional - Template files for code generation",
      "resources/": "Optional - Reference materials"
    },
    "skillMdFormat": {
      "frontmatter": {
        "name": {
          "required": false,
          "description": "Unique identifier, defaults to directory name if not provided"
        },
        "description": {
          "required": true,
          "description": "Most important field - trigger phrase for LLM semantic matching",
          "bestPractice": "Be precise, not vague. Include specific use cases."
        }
      },
      "body": {
        "goal": "Clear statement of what skill achieves",
        "instructions": "Step-by-step logic",
        "examples": "Few-shot examples of inputs and outputs",
        "constraints": "'Do not' rules"
      }
    },
    "scriptIntegration": {
      "location": "scripts/ subdirectory",
      "referencing": "Relative path from SKILL.md",
      "purpose": "Delegate execution for binary operations, calculations, legacy system interaction"
    },
    "vsRulesAndWorkflows": {
      "rules": "Passive constraints, always on",
      "workflows": "Active, user-triggered sequences",
      "skills": "Agent-triggered capabilities, semantic matching"
    }
  },

  "mcpSupport": {
    "native": true,
    "feature": "MCP Store for one-click installation of connectors",
    "examples": ["BigQuery", "AlloyDB"],
    "configLocation": "~/.gemini/antigravity/mcp_config.json",
    "viewing": "Agent Manager → View raw config",
    "remoteServers": {
      "github": {
        "example": "Remote GitHub MCP Server",
        "oauth": "Supports OAuth flow for authentication"
      }
    },
    "refreshing": "Click Refresh button in MCP panel after config changes"
  },

  "artifactGeneration": {
    "taskLists": "Structured plans before coding",
    "implementationPlans": "Detailed implementation steps",
    "walkthroughs": "Post-task summaries",
    "browserRecordings": "Video of UI validation (proof of work)",
    "screenshots": "UI state capture",
    "purpose": "Build trust through behavioral review (watching videos) not just code review (reading diffs)"
  },

  "interfaceComponents": {
    "agentManager": {
      "description": "Mission Control for managing autonomous agents",
      "capabilities": [
        "Dispatch multiple agents to work in parallel",
        "View status of each agent",
        "See produced Artifacts",
        "Handle pending approval requests"
      ]
    },
    "editor": {
      "description": "VS Code-based editor for human verification and code editing",
      "foundation": "Modified VS Code fork"
    },
    "browser": {
      "description": "Integrated browser for opening URLs, DOM interaction, JavaScript execution"
    },
    "keyboardShortcuts": {
      "openSettings": "Cmd + ,"
    }
  },

  "hostCompatibility": {
    "windows": "Windows 10/11 (64-bit)",
    "macos": "macOS Monterey 12+",
    "linux": "glibc 2.28+",
    "wsl2": "Required for Windows sandboxing"
  },

  "sandboxing": {
    "localSubprocess": {
      "default": true,
      "description": "Commands run as local subprocesses"
    },
    "wsl2": {
      "required": "Windows",
      "description": "Windows Subsystem for Linux 2 required"
    },
    "docker": {
      "optIn": true,
      "configuration": "Via template config"
    }
  },

  "securityAndGovernance": {
    "terminalPolicy": {
      "default": "Auto (Agent decides when to ask)",
      "options": ["Off (Never auto-execute)", "Auto", "Turbo (Always auto-execute - High Risk)"]
    },
    "browserPolicy": {
      "default": "Request Review",
      "options": ["Always Proceed (High Risk)", "Request Review", "Disabled"]
    },
    "networkAccessControl": {
      "requirement": "Outbound to Google services required",
      "enterprise": "Configurable firewall whitelisting"
    },
    "secretDetection": {
      "method": "Integrated IAM using Google Cloud credentials",
      "storage": "Secrets stored securely"
    },
    "promptInjectionProtection": {
      "browserAllowlist": "Restrict agent to trusted domains only",
      "purpose": "Prevent attacks from compromised documentation sites"
    }
  },

  "gettingStarted": {
    "setupFlow": {
      "steps": [
        "Choose setup flow (import from VS Code/Cursor or fresh start)",
        "Choose editor theme",
        "Configure agent autonomy settings",
        "Configure editor (keybindings, extensions, command line)",
        "Sign in to Google (personal Gmail account)",
        "Accept Terms of Use"
      ]
    },
    "openWorkspace": {
      "method": "Agent Manager → Open folder",
      "note": "Can skip and open workspace later"
    }
  },

  "undoChanges": {
    "method": "↩️ Undo changes up to this point in chat",
    "availability": "After each step"
  },

  "customizations": {
    "accessMethod": "Agent Manager → ... → Customizations",
    "options": ["Rules", "Workflows"]
  },

  "modelSelection": {
    "location": "Model Selection dropdown in Agent Manager",
    "available": ["Gemini 3 Pro", "Gemini 3 Deep Think", "Gemini 3 Flash", "Claude Sonnet 4.5", "Claude Opus 4.5", "GPT-OSS-120B"],
    "quotaNote": "Gemini 3 Pro has limited free quotas during preview"
  },

  "workspaceAwareness": {
    "level": "Deep - 'Mission Control' over entire project structure",
    "indexingStrategy": "Full-codebase indexing with Mission Control oversight"
  },

  "extensionEcosystem": {
    "skills": "YAML/Markdown format in .agent/skills/",
    "vscodeExtensions": "Supports standard VS Code extensions (with some limitations)"
  },

  "bestPractices": {
    "security": [
      "Use 'Review-driven development' preset (recommended)",
      "Configure Allow List when using 'Off' policy",
      "Configure Deny List to block rm, rmdir, sudo, curl, wget",
      "Set up Browser URL Allowlist for trusted domains only",
      "Restart Antigravity after changing allow/deny lists"
    ],
    "rules": [
      "Keep global GEMINI.md for personal preferences",
      "Use workspace .agent/rules/ for team standards",
      "Be specific with rule instructions"
    ],
    "skills": [
      "Write precise descriptions for semantic matching",
      "Include Goal, Instructions, Examples, Constraints in SKILL.md body",
      "Use scripts/ for complex operations",
      "Place global skills in ~/.gemini/antigravity/skills/"
    ],
    "workflows": [
      "Create reusable prompts as workflows",
      "Share workspace workflows with team via version control"
    ],
    "agentUsage": [
      "Use Planning mode for complex tasks",
      "Use Fast mode for simple, localized tasks",
      "Review Artifacts (task lists, recordings) before accepting",
      "Use Undo to revert unwanted changes"
    ]
  },

  "comparisonToOtherAgents": {
    "paradigm": "Mission Control GUI - maximalist, agent-first approach",
    "philosophy": "AI is primary actor, human is supervisor/architect/manager",
    "differentiator": "Explicit Planning Mode with Task Lists and Implementation Plans before coding",
    "artifactTrust": "Proof of work via browser recordings and screenshots",
    "contextStandard": "GEMINI.md (proprietary) + .agent/ structure"
  }
}