By AllanHarlen
Run agentic Antigravity CLI (AGY) sessions for autonomous file creation, editing, and refactoring in your codebase, with a read-only subagent for architecture mapping and security audit support.
Use this agent for coding tasks that should be executed by Antigravity CLI (AGY) through the plugin bridge. It can create, edit, delete, move, format, and search files by delegating to AGY's native tools. Prefer this agent over running the work yourself when the caller wants the harness to track the run as a subagent, when they ask for AGY/Gemini to be the implementation engine, or when they name a specific AGY model. The agent has no Write/Edit and no broad Bash: its only file-acting tool is the bridge, so file generation happens in AGY/Gemini and does not burn Claude tokens. <example> Context: User wants files created or refactored across the codebase user: "Create the React frontend screens for this module" assistant: "I will use antigravity-coder so AGY performs the file edits natively through the bridge while the harness tracks the subagent run." </example> <example> Context: User asks for a specific model to do the implementation user: "Use gemini 3.1 pro and build the backend endpoints" assistant: "I will use antigravity-coder with --model gemini-3.1-pro-high in agentic mode." </example>
Use this agent only for read-only Antigravity CLI (AGY) analysis, planning, architecture mapping, audit, and refactor-impact questions. Do not use this agent for tasks that create, edit, delete, move, or format files. For coding work, call the plugin command/skill directly so execution goes through the bridge into AGY: /cc-antigravity-plugin:antigravity --parallel --add-dir <dir> <task> Skill("cc-antigravity-plugin:antigravity", "--parallel --add-dir <dir> <task>") <example> Context: User wants files created or refactored across the codebase user: "Create the React frontend screens for this module" assistant: "I will use the cc-antigravity-plugin antigravity skill directly with --parallel, not antigravity-agent, so AGY performs the file edits natively." </example> <example> Context: User wants a high-level architecture map user: "Help me understand the architecture of this project" assistant: "I will use antigravity-agent in read-only mode for a large-context architecture pass before we decide on changes." </example> <example> Context: User asks about breakage risk user: "What would be affected if I refactor the auth module?" assistant: "I will use antigravity-agent read-only to trace callers, dependencies, and likely collateral changes." </example>
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
This plugin requires configuration values that are prompted when the plugin is enabled. Sensitive values are stored in your system keychain.
timeoutDefault timeout for AGY print mode, for example 3m, 300s, or 5m0s.
${user_config.timeout}coding_policyWhen on, a SessionStart hook injects the plugin's coding-delegation policy as session context (delegate file work to AGY via the skill/command or antigravity-coder, --parallel for big front-end, model guidance, and the AskUserQuestion image flow). Set to off/false/0/no to suppress injection.
${user_config.coding_policy}default_modelModel identifier to use when --model is not specified. Available models: gemini-3.5-flash-low, gemini-3.5-flash-medium (default), gemini-3.5-flash-high, gemini-3.1-pro-low, gemini-3.1-pro-high, claude-4.6-sonnet-thinking, claude-4.6-opus-thinking, gpt-oss-120b-medium, nano-banana (image generation via --generate-image), auto.
${user_config.default_model}
A plugin for Claude Code and Codex that integrates the Antigravity CLI (AGY) as an agentic coding assistant — creates, edits, searches files, and executes commands autonomously over your codebase.
📖 Documentation in other languages:
Fork: This plugin is a fork of gemini-cli-plugin, originally created by thepushkarp for process automation with Gemini CLI.
AGY is Google's long-context CLI terminal (2M token window). This plugin bridges AGY to Claude Code and Codex via a shared Node.js bridge, exposing AGY as a tool_use endpoint that Claude can invoke.
When to use instead of native Claude Code:
--parallel)agy directly vs via pluginClaude can call agy directly via Bash (agy --print "task" --dangerously-skip-permissions --add-dir .) without any intermediary. However, the plugin delivers capabilities that raw agy cannot:
| Capability | agy direct | Via plugin (bridge) |
|---|---|---|
| Headless model selection | No — lacks --model flag | Yes, via settings.json patch |
| Guaranteed coding-agent behavior | No — AGY tends to respond in text | Yes — <constraints> block instructs use of write_to_file, grep_search, etc. |
| Structured quota/auth signals | No — free-form text, no exit codes | Exit codes 10/11 + JSON-line parseable |
| Automatic file ingestion | Manual | --dirs, --files with binary detection and truncation |
| Parallelism via Gemini subagents | Manual | --parallel + --subagent-model |
| Fallback for 28k char limit (Windows) | Silent breakage | Auto-drop of inline files |
| Auditable logging | No | JSONL in %LOCALAPPDATA%\agy\cc-plugin-logs\ |
| Process overhead | None | Node.js + ConPTY + version check |
| Visibility of AGY actions | Full — direct output | Black box — Claude doesn't validate before exec |
| Quota dependency | Claude only | Claude + AGY/Gemini |
Summary: for automated workflows, skills, and coding tasks where consistent agentic behavior is required, the bridge is the right choice. For simple ad-hoc invocations, raw agy suffices.
# macOS / Linux
curl -fsSL https://antigravity.google/cli/install.sh | bash
# Windows PowerShell
irm https://antigravity.google/cli/install.ps1 | iex
After installing, run agy once to log in and verify it works:
agy --print "what is 2+2"
The
SessionStarthook automatically checks if AGY is installed and accessible on each Claude Code session start.
Via CLI (terminal):
# Add the GitHub repository as a marketplace source
claude plugin marketplace add AllanHarlen/cc-antigravity-plugin
# Install the plugin
claude plugin install cc-antigravity-plugin@AllanHarlen/cc-antigravity-plugin
Via slash command (inside Claude Code):
/plugin marketplace add AllanHarlen/cc-antigravity-plugin
/plugin install cc-antigravity-plugin@AllanHarlen/cc-antigravity-plugin
To test a local copy of the repository:
cc --plugin-dir /path/to/cc-antigravity-plugin
git clone https://github.com/AllanHarlen/cc-antigravity-plugin.git \
~/.agents/skills/cc-antigravity-plugin
Restart Codex after cloning.
For any coding request (creating, editing, deleting, moving, or formatting files), always use the direct command/skill path:
/cc-antigravity-plugin:antigravity --parallel --add-dir ./frontend "implement the requested components"
Do not use antigravity-agent for coding. That agent is read-only and exists
only for analysis, planning, audits, and refactor-impact work.
If you specifically want the coding run tracked by the harness as a subagent,
use the antigravity-coder agent instead of the read-only one. It is the
sanctioned subagent path for implementation: it has no Write/Edit and no
broad Bash — its only file-acting tool is the bridge, so AGY/Gemini performs
the file generation and it does not spend Claude tokens writing file contents.
The plain command/skill path remains the simplest option when you do not need a
separate subagent layer.
Agentic coding assistant powered by the direct Kiro CLI bridge. Use the command/skill path or kiro-coder for implementation; kiro-agent is read-only analysis.
npx claudepluginhub allanharlen/cc-antigravity-plugin --plugin cc-antigravity-pluginRun the Antigravity CLI (Gemini) as a collaborating AI inside Claude Code, with intelligent model routing across the SDLC. Claude conducts — requirements, architecture, the hard 20%, verification, review — and routes deterministic, high-volume work (scaffolding, tests, first-pass review, migrations, web/Vertex AI Search) to Antigravity. Hybrid agentic engineering; lower token cost as a financial lever.
One plugin to bridge and delegate across Claude Code, Codex CLI, and Antigravity CLI (agy) — single-source AGENTS.md, shared skills, mirrored hooks and MCP servers, and bidirectional delegation.
Personal Claude Code + Codex dev stack: security hooks, AI-first code conventions, /security-review, /repo-map, /stack-check, portable statusline. Designed to complement other skills-based plugins, not replace them.
Multi-agent orchestration framework for Claude Code. Routes tasks to specialized Haiku/Sonnet subagents while Opus orchestrates — inspired by speculative decoding. Includes 10 specialized heads, environment preflight checks, and ~50% API cost reduction.
HelloAGENTS — The orchestration kernel that makes any AI CLI smarter. Adds intelligent routing, unified QA gates, safety guards, and notifications.
Multi-agent development workflow (/crew) — planning, implementation, review, documentation, evals, and audit gates. Commands, agents, and skills are auto-discovered from the repo.