antigravity-agent

Dispatch headless parallel coding tasks to the Antigravity CLI from an orchestration session.

Updated Mar 22, 2026
One-click install
npx skills add https://github.com/diazMelgarejo/orama-system --skill antigravity-agent-diazmelgarejo
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: antigravity-agent
Source: https://github.com/diazMelgarejo/orama-system/tree/main/bin/orama-system/skills/antigravity-agent
Command: npx skills add https://github.com/diazMelgarejo/orama-system --skill antigravity-agent-diazmelgarejo

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Running light, mechanical coding tasks in the main orchestration session blocks higher-level work; this Skill lets you fan out those tasks to the Antigravity CLI (agy) as a background worker while the main session handles architecture and review. ## Core Features & Use Cases - Headless task fan-out: Run agy -p prompts in the background with per-task stdout/stderr logs and independent exit-status collection. - Sandboxed plan-first workflow: Default to --mode plan --sandbox read-only passes, with an explicit approval gate before any --mode accept-edits run. - Model selection and runtime discovery: Discover the agy binary via command -v, smoke-test auth, and pick models from agy models (e.g., Claude Sonnet 4.6 (Thinking), Gemini 3.1 Pro (Low)). - Use Case: While the main session designs a refactor, dispatch one agy worker to add type annotations to a script and another to rename variables in a test file, then collect both logs and verify results. ## Quick Start Ask the agent to smoke-test the Antigravity CLI with a headless prompt and then fan out a sandboxed plan-only task to agy for a specific file edit.

Frequently Asked Questions about antigravity-agent

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run the Antigravity CLI headless for a single prompt?

Use `agy -p "<prompt>"` for headless single-turn execution; it prints the response to stdout. Verify auth first with a smoke test like `agy -p "reply with exactly: agy-ok"` before dispatching real work.

How do I run parallel background tasks with the agy CLI?

Launch each `agy -p` command backgrounded with its own log directory, capture the PIDs, and `wait` on each one independently so one task's failure is not masked by another's success. Preserve stdout and stderr logs before parsing results.

What models does the Antigravity CLI support?

Run `agy models` to list them; verified options include Gemini 3.5 Flash variants, Gemini 3.1 Pro (Low/High), Claude Sonnet 4.6 (Thinking), Claude Opus 4.6 (Thinking), and GPT-OSS 120B. Pass the exact string via `--model`; there is no automatic fallback chain.

Should I route Antigravity through OpenClaw or use the CLI directly?

Use the `agy` CLI directly. There is no real Antigravity provider plugin for OpenClaw, and a fabricated provider stanza cannot authenticate or run inference; the CLI works from PATH with zero extra setup once the machine is authenticated.

Is it safe to let agy edit files automatically?

Default to a sandboxed plan-only pass with `--mode plan --sandbox` and require explicit human approval before running `--mode accept-edits`. Reserve `--dangerously-skip-permissions` for trusted CI use, and scope access with `--add-dir` limited to the target repository.

Does the Antigravity CLI work on Windows?

Windows support is unverified in this Skill; only macOS discovery via `command -v agy` has been tested. Do not assume PowerShell or Windows executable discovery behaves the same way.