# antigravity-cli

Adopt this skill: https://proskills.md/skills/other/amirkiarafiei-subagent-cli-skills-skills-antigravity-cli

Page: https://proskills.md/skills/other/amirkiarafiei-subagent-cli-skills-skills-antigravity-cli
This document is the public skill bundle for that page. It contains SKILL.md and every helper file the catalog carries. No login is required.

## File: skills/antigravity-cli/SKILL.md
size: 10699
sha: 82f724baac155bb4c6767e1e4d335736ccb66f1a
upstream: https://github.com/amirkiarafiei/subagent-cli-skills/blob/main/skills/antigravity-cli/SKILL.md

````text
---
name: antigravity-cli
description: Use Antigravity CLI as a subagent. Lets the main agent prompt Antigravity CLI from the terminal in headless mode with the `agy` command, and send the goal, the decisions and the scope with the task. Use when the user asks to delegate work to Antigravity CLI, when a task needs a second independent agent, or when a plan needs a fresh perspective.
allowed-tools:
  - Bash
  - Read
  - Write
  - Grep
  - Glob
---

# Antigravity CLI (subagent / task delegation)

<!-- =============================================================================
     GLOBAL HALF — sections 1 to 5.
     These sections are identical in every skill in this repository.
     Copy them without any change. Do not put a vendor name or a flag in them.
     A change here must be made in all skills at the same time.
     ============================================================================= -->

## What is it

This skill teaches your agent to use an agent from another vendor as a subagent.

Almost every coding agent has a headless mode. Headless mode runs the agent from the terminal with
one command and returns the answer to stdout. This skill gives the command pattern for one such
agent, and the protocol to transfer enough context with the task.

You stay the main agent. You keep the plan, the decisions and the conversation with the user. The
subagent does one bounded task and reports back.

## When to delegate

- **The user asks for it.** The user names another agent, or asks for a second opinion.
- **Large or multi-step work.** The task covers several files, phases or checkpoints.
- **Code review with a fresh mind.** Another agent reviews the work with a different lens and no
  memory of the decisions that produced it.
- **Planning that needs different ideas.** A second agent proposes options that you did not consider.
- **An internal council.** Several subagents answer the same question, and you compare the answers.
- **Deep research.** The other CLI has web search, web extraction or browser automation.
- **Background investigation.** A long task runs in the background while you continue in the
  foreground.

## When not to delegate

- **Small or single-step tasks.** One or two edits, or a short explanation, are faster in this thread.
- **Tight feedback loops.** The user wants quick back-and-forth in one conversation.
- **Secrets or policy-sensitive work.** Do not pipe credentials into a subagent. Redact first.
- **Context you already hold.** Repeating the whole plan in a prompt adds no value.
- **Low return on investment.** If writing the handoff costs more than doing the edit, do the edit.
- **Work that needs high precision and full judgement.** A subagent without your context will break a
  task that depends on a detail only this session knows.

## Delegation and context transfer protocol

An isolated subagent saves tokens, but it splits the story. The subagent does not see this
conversation. It has no memory of the goal you agreed with the user, the decisions you already made,
or the paths you agreed not to touch. A poor handoff causes misread tasks, wrong assumptions about
the stack or the style, and wasted edits.

Treat every call as **stateless delegation**: the subagent keeps nothing between calls, so each
command must carry its own state. A good handoff happens when you keep these practices in mind:

- **Shared state handoff.** The prompt carries the original goal, the decisions already made, and the
  explicit scope, to bridge the subagent's missing session history.
- **Contextual preservation.** The high-volume work stays in the isolated process, so the bulk never
  enters this conversation and no "split story" forms from missing history.
- **Result reconciliation.** What comes back is folded into this thread afterwards, so there is a
  single source of truth.

Write the prompt as a transfer of shared state, not as a title. Include all six fields:

| Include | Why |
|---|---|
| **Goal** | The same objective as the user, not only the immediate micro-task. |
| **Decisions** | Framework, patterns, naming, "use X not Y" — anything that would otherwise be guessed wrong. |
| **Scope** | The absolute paths and modules to touch, and the areas to leave alone. A relative path may not resolve where you expect. |
| **Constraints** | Performance, accessibility, compatibility, review gates, "no new dependencies". |
| **Verification** | The exact command the subagent must run and pass before it returns. |
| **Output** | For example: "report only, no edits", "apply edits with a minimal diff", "list the files changed". |

Prefer sequential delegations with explicit carry-over. Parallel runs diverge unless every run gets
the same briefing.

**Run the subagent in the mode that does the work without asking for approval.** Headless mode has
nobody to answer a permission prompt. Any mode that stops to ask will stall, or return an empty
answer with a success exit code. Each CLI names this mode differently — take the flag from the vendor
card below.

**Never call a subagent in plan mode.** Many CLIs have a plan or read-only mode. That mode makes the
other agent write a plan *for itself*, which is not what you asked for: you want its work or its
answer, and you keep the planning. Plan mode also waits for someone to approve that plan, so the run
comes back with nothing. If you want no file changes, keep the auto-approving mode and write "report
only, no edits" in the **Output** field. Do not use plan mode to make a run read-only.

After the subagent returns:

- **Report to the user.** Give a short summary. Do not paste long logs unless the user asks.
- **Reconcile the context.** Record the decisions it made, the files it changed, and the open risks,
  so this session stays the single source of truth.

## CLI failure modes

Every vendor has its own terminal rules. A headless run fails quietly more often than it fails
loudly.

- **Wrap the call in an external timeout.** A headless CLI can stall before its own timeout starts.
- **Exit code 0 is not success.** If stdout is empty, treat the run as failed and read stderr. A
  permission the CLI could not ask for, and a prompt that never arrived, both look like success.
- **Never carry a flag habit from one CLI to another.** The same short flag means different things in
  different tools. In OpenCode, `-p` is `--password`: passing a prompt to it empties the message and
  the run waits on stdin forever. Confirm every flag against the CLI's own `--help`.
- **An unrecognized-flag error means this skill is stale, not that the task is impossible.** Read
  `--help`, continue with the flags that exist, and tell the user which line in this file is wrong.

<!-- =============================================================================
     VENDOR HALF — section 6.
     Everything below is specific to this CLI.
     Keep all seven sub-headings, in this order, even if the answer is "none".
     Take every flag from the installed binary (`<cli> --help`), not from memory
     and not from another skill in this repository.
     ============================================================================= -->

## Vendor card

### Binary and prompt form

- **Binary:** `agy`
- **Install:** `curl -fsSL https://antigravity.google/cli/install.sh | bash` (macOS and Linux). Windows commands are in the vendor getting-started page.
- **Prompt form:** **flag.** The prompt is the argument of `-p` (aliases `--print`, `--prompt`).
  A bare `agy "prompt"` opens the interactive interface and never returns.

### Headless and output flags

| Need | Flag |
|---|---|
| Run once and exit | `-p`, `--print`, `--prompt` |
| Plain text | `--output-format text` (the default) |
| One JSON object | `--output-format json` |
| Event stream | `--output-format stream-json` |
| Structured answer | `--json-schema <schema or path>` |
| Limit a run | `--print-timeout <duration>` — the default `0s` waits until the turn completes. Set an external timeout too. |

### Approvals and permissions

Headless mode cannot ask, so a tool that needs approval is **auto-denied**, and the run still **exits
0** with an empty answer and a notice on stderr. Grant access before the run, from the narrowest
option upward:

1. **`--add-dir <absolute path>`** — repeatable. File reads are *not* granted by the working
   directory, and a relative path such as `.` does not resolve to your current directory. Use
   absolute paths.
2. **Allow-rules** under `permissions.allow` in `~/.gemini/antigravity-cli/settings.json`, written as
   `action(target)`, for example `command(git)`. Inspect the effective rules with
   `agy -p "/permissions"`, which answers without starting a run.
3. **`--dangerously-skip-permissions`** — approves every tool, including writes and shell commands.
   This is the documented unattended path. `--mode` also accepts `accept-edits` and `auto` (and `plan`,
   which you must not use for delegation). `agy --help` does not say what `auto` approves, so this card does not rely on it.

### Models and how to list them

Run **`agy models`** for the identifiers this build accepts.

Selection is `--model <identifier>` together with `--effort low|medium|high|xhigh|max`. A model family on its
own is rejected — pass the full tiered identifier, or the family plus an effort level. An unknown
model does not fall back silently: the run exits non-zero and lists what is available.

If the user names a model or an effort level, use it. Otherwise ask the binary first. For capability
and price comparisons, check [Artificial Analysis](https://artificialanalysis.ai/) and the vendor's
own documentation.

### Command pattern

```bash
agy -p "GOAL: [goal] | DECISIONS: [decisions] | SCOPE: [paths] | CONSTRAINTS: [constraints] | VERIFICATION: [test_command] | OUTPUT: [format]" \
  --output-format text --model <identifier> --effort medium \
  --add-dir /absolute/path/to/repo --print-timeout 15m 2>&1
```

### Prompt examples

- **Implement:** `agy -p "GOAL: [goal] | DECISIONS: [decisions] | SCOPE: [paths] | VERIFICATION: [test_command] | OUTPUT: files changed" --add-dir /abs/repo --print-timeout 15m`
- **Review:** `agy -p "GOAL: Review [scope] for [concerns] | CONSTRAINTS: do not edit any file | OUTPUT: findings with severity" --add-dir /abs/repo`
- **Investigate:** `agy -p "GOAL: Map how [feature] works | SCOPE: [paths] | CONSTRAINTS: do not edit any file | OUTPUT: file:line map" --add-dir /abs/repo`

### Docs and reference

- Flags, models, permissions and paths: [reference.md](reference.md)
- Vendor documentation: <https://antigravity.google/docs/cli/headless>
- **Checked against `agy` 1.5.1 `--help` on 2026-11-05.**
````

## File: skills/antigravity-cli/reference.md
size: 6860
sha: e3516f5389a712341f43638913ab242817eee42e
upstream: https://github.com/amirkiarafiei/subagent-cli-skills/blob/main/skills/antigravity-cli/reference.md

````text
# Antigravity CLI — reference

> **Checked against `agy` 1.5.1 `--help` on 2026-11-05.** Flags change between builds,
> and this vendor's changelog has announced flags the shipped binary rejects. `agy --help` and
> `agy models` are the only authorities. On `flags provided but not defined`, re-read `--help`, use
> what exists, and report that this file needs updating.

## Authentication

Silent keyring sign-in (local), Google Sign-In via an SSH URL (remote), or `/logout`. Headless mode
reuses cached credentials — authenticate once interactively first, or an unauthenticated run exits
with an authentication error.

## Models

Run **`agy models`** for the identifiers this build accepts. No fixed list is reproduced here: names
and tiers change, and a list written into this file goes stale.

Identifiers carry a reasoning tier. Pass either a full tiered identifier, or a family name together
with `--effort low|medium|high|xhigh|max` — a family name **alone is rejected** with
`requires --effort`, which lists the available levels. Headless mode does **not** fall back on an unknown
model: it exits non-zero with `status: "ERROR"` and lists what is available in the `error` field.

Note that `agy models` does **not** accept `--output-format` on this build, despite the changelog
announcing it.

## Essential Flags

| Flag | Purpose |
|------|---------|
| `-p`, `--print`, `--prompt` | Run a single prompt non-interactively and print the response. **Required for delegation.** |
| `--output-format` | `text` (default), `json`, `stream-json`. |
| `--model` | Model identifier for this run (see `agy models`). |
| `--effort` | Reasoning effort: `low`, `medium`, `high`, `xhigh`, `max`. |
| `--mode` | `accept-edits`, `auto` or `plan`. Do not use `plan` for delegation. `--help` does not say what `auto` approves. |
| `--add-dir` | Add a directory to the workspace (repeatable). **Absolute paths only.** |
| `--print-timeout` | Maximum wait for a response; default `0s`, which waits until the turn completes. |
| `--json-schema` | Schema string or file path to enforce structured output. |
| `--dangerously-skip-permissions` | Auto-approve every tool. Blunt — prefer `--add-dir` and allow-rules. |
| `--agent` | Select an agent for the run (see `agy agents`). |
| `--sandbox` | Run with terminal sandbox restrictions enabled. |
| `--disable-slash-commands` | Disable slash command and skill expansion in print mode. |
| `-c`, `--continue` | Continue the most recent conversation. |
| `--conversation <id>` | Resume a conversation by ID. |
| `--log-file` | Override the CLI log file path. |
| `--project`, `--new-project` | Workspace and project scoping. |

**Not flags on this build** (all rejected, though older docs or sibling skills name them): `--bare`,
`-o`, `--yolo`/`-y`, `-m`, `--worktree`/`-w`, `--resume`/`-r`. Use `--print`, `--output-format`,
`--dangerously-skip-permissions`, `--model`, and `--continue`/`--conversation` instead.

## Approvals and permissions

Headless mode cannot prompt, so a tool requiring approval is **soft-denied**: the run continues,
**exits 0**, `status` is `SUCCESS`, `response` is **empty**, and a notice naming the tool goes to
**stderr**. Under `--output-format text` this looks like silent success with no output — always check
for empty output rather than trusting the exit code.

Grant the narrowest access that works:

1. **`--add-dir /abs/path`** — file reads are *not* granted by the shell's working directory, and a
   relative path such as `.` does not resolve to your cwd. Absolute paths only.
2. **Allow-rules** under `permissions.allow` in `~/.gemini/antigravity-cli/settings.json`, matching
   `action(target)`:
   ```json
   { "permissions": { "allow": ["command(git)", "command(npm run (build|lint|test))", "write_file(src/)"] } }
   ```
   `trustedWorkspaces` is a *trust* setting and does **not** grant tool permissions.
3. **`--dangerously-skip-permissions`** — last resort; approves writes and command execution too.

Inspect the rules actually in effect with `agy -p "/permissions"` — read-only slash commands answer in
print mode without starting an agent turn, spending quota, or leaving a conversation behind.

## Subcommands

| Subcommand | Purpose |
|---|---|
| `agy models` | List available model identifiers. **The authority on `--model` values.** |
| `agy agents` | List available agents. **Returns empty when none are configured** — do not assume named specialists exist. |
| `agy changelog` | Release notes; useful for explaining a behaviour that changed under you. |
| `agy plugin` / `plugins` | Install, uninstall, list, enable, disable plugins. |
| `agy install` | Configure environment paths and shell settings. |
| `agy update` | Update the CLI. |

The supported way to pick a specialist is **`--agent <name>`**, discovered with `agy agents`. An
`@name` prefix inside the prompt is **not** a dispatch mechanism — if no such agent exists the text is
simply read as part of the prompt.

## Agent Skills

| Scope | Path |
|---|---|
| **Global (user), recommended** | `~/.gemini/config/skills/<skill-folder>/` — read by the App, CLI and IDE. Install here. |
| CLI-only user paths (still supported) | `~/.gemini/antigravity-cli/skills/` and `~/.gemini/skills/` |
| Plugin skills (CLI) | `~/.gemini/config/plugins/<plugin-folder>/skills/` |
| **Workspace** | `<workspace-root>/.agents/skills/<skill-folder>/`. The App and CLI also read `.agents/skills/` in each parent folder up to the repository root. |
| Legacy | `~/.gemini/antigravity/skills/` (IDE only); `.agent/skills` is still read for backward compatibility. |

Source: <https://antigravity.google/docs/skills>, and strings compiled into `agy` 1.5.1.

## Configuration and paths

| Component | Path |
|---|---|
| Configuration and permissions | `~/.gemini/antigravity-cli/settings.json` |
| CLI logs | `~/.gemini/antigravity-cli/log/`, `~/.gemini/antigravity-cli/cli.log` |
| Conversations | `~/.gemini/antigravity-cli/conversations/` |
| MCP configuration | `~/.gemini/antigravity-cli/mcp_config.json` |
| Global rules | `~/.gemini/GEMINI.md` |
| Workspace rules | `<workspace-root>/.agents/rules/` |

## Documentation

- Vendor documentation: <https://antigravity.google/docs/cli/headless>, <https://antigravity.google/docs/cli/reference>
- **Checked against `agy` 1.5.1 `--help` on 2026-11-05.** Install command from the vendor getting-started page.

## Delegation Checklist

1. **Full Goal**: clearly state the final objective.
2. **Prior Decisions**: stack, style, and API choices already made.
3. **Scope**: exact modules or directories to touch, and what not to touch.
4. **Constraints**: performance, security, or style requirements.
5. **Verification**: the explicit command the subagent must run and pass before returning.
6. **Desired Shape**: "apply changes," "report only," "return JSON," etc.
````

## File: README.md
upstream: https://github.com/amirkiarafiei/subagent-cli-skills/blob/main/README.md

````text
# Subagent CLI Skills 🧔🏻👶🏻

A collection of Skills that connects agents from any vendor to each other.

Let Claude Code prompt Codex, Antigravity or Pi as a subagent. All agents can call each other as subagent. Supports 24
agents.

## How It Works

Almost every coding agent has a headless mode. This mode runs the agent from the terminal with one
command. Each skill in this repository teaches your main agent how to call one other agent in this
way.

<p align="center">
  <img src="assets/banner.png" width="720" alt="Delegation Flow Diagram">
</p>

Each skill also applies a delegation and context transfer protocol. The subagent starts with an
empty context. The protocol sends the goal, the decisions and the scope with every command. The
subagent then has the state that it needs to do the task correctly.

## 🛠️ Installation

### Option 1: Interactive Installer

Run the installer with curl:

```bash
curl -sSL https://raw.githubusercontent.com/amirkiarafiei/subagent-cli-skills/main/install.sh | bash
```

Or run it locally, if you cloned the repository:

```bash
chmod +x install.sh
./install.sh
```

### Option 2: Give this prompt to your agent

```
Read https://raw.githubusercontent.com/amirkiarafiei/subagent-cli-skills/main/AGENTIC_INSTALLATION.md and follow the instructions to install the Subagent CLI Skills.
```

Your agent finds which tools you have, asks what it needs to know, and installs the skills. See
[AGENTIC_INSTALLATION.md](AGENTIC_INSTALLATION.md).

### Option 3: Download the skill folder

From `./skills/<vendor>/` — it contains `SKILL.md` and `reference.md`

## Example Usage

Name the tool in your request. Your agent then activates the correct skill. You can also type the
slash command yourself. You can name a model for each tool.

**1. Review code with two different agents and models**

```
Ask Codex with GPT-Astra and Hermes with GLM-5.3 to review the code using their subagent cli skills
```

**2. Get the latest documentation**

```
/antigravity-cli Read the latest docs for the Stripe API. List the breaking changes.
```

**3. Build two features at the same time with two different agents**

```
Build the login form with Codex. Build the signup form with OpenCode.
```

**4. Get fresh ideas on a plan from three different agents**

```
Ask Pi and Grok and Devin to analyze the plan to get fresh and diverse ideas.
```

## 🔌 Support

<p>
<img src=".github/readme-assets/antigravity.svg" alt="" width="18" height="18" align="texttop" /> Antigravity ·
<img src=".github/readme-assets/claude-code.svg" alt="" width="18" height="18" align="texttop" /> Claude Code ·
<img src=".github/readme-assets/cline.svg" alt="" width="18" height="18" align="texttop" /> Cline ·
<img src=".github/readme-assets/codex.svg" alt="" width="18" height="18" align="texttop" /> Codex ·
<img src=".github/readme-assets/copilot.svg" alt="" width="18" height="18" align="texttop" /> Copilot ·
<img src=".github/readme-assets/cursor.svg" alt="" width="18" height="18" align="texttop" /> Cursor ·
<img src=".github/readme-assets/deepagents-code.svg" alt="" width="18" height="18" align="texttop" /> Deep Agents Code ·
<img src=".github/readme-assets/devin.svg" alt="" width="18" height="18" align="texttop" /> Devin ·
<img src=".github/readme-assets/gemini.svg" alt="" width="18" height="18" align="texttop" /> Gemini ·
<img src=".github/readme-assets/goose.svg" alt="" width="18" height="18" align="texttop" /> Goose ·
<img src=".github/readme-assets/grok.svg" alt="" width="18" height="18" align="texttop" /> Grok ·
<img src=".github/readme-assets/hermes-agent.svg" alt="" width="18" height="18" align="texttop" /> Hermes Agent ·
<img src=".github/readme-assets/junie.svg" alt="" width="18" height="18" align="texttop" /> Junie ·
<img src=".github/readme-assets/kilocode.svg" alt="" width="18" height="18" align="texttop" /> Kilo Code ·
<img src=".github/readme-assets/kimi-code.svg" alt="" width="18" height="18" align="texttop" /> Kimi Code ·
<img src=".github/readme-assets/kiro.svg" alt="" width="18" height="18" align="texttop" /> Kiro ·
<img src=".github/readme-assets/mistral-vibe.svg" alt="" width="18" height="18" align="texttop" /> Mistral Vibe ·
<img src=".github/readme-assets/oh-my-pi.svg" alt="" width="18" height="18" align="texttop" /> Oh My Pi ·
<img src=".github/readme-assets/openclaw.svg" alt="" width="18" height="18" align="texttop" /> OpenClaw ·
<img src=".github/readme-assets/opencode.svg" alt="" width="18" height="18" align="texttop" /> OpenCode ·
<img src=".github/readme-assets/openhands.svg" alt="" width="18" height="18" align="texttop" /> OpenHands ·
<img src=".github/readme-assets/pi.svg" alt="" width="18" height="18" align="texttop" /> Pi ·
<img src=".github/readme-assets/qoder.svg" alt="" width="18" height="18" align="texttop" /> Qoder ·
<img src=".github/readme-assets/qwen-code.svg" alt="" width="18" height="18" align="texttop" /> Qwen Code
</p>

## Delegation and Context Transfer Protocol

The delegation technique and philosophy used in this project are inspired by the principles outlined
in **[Don't Build Multi-Agents](https://cognition.ai/blog/dont-build-multi-agents#applying-the-principles)**
by Cognition AI:

We implement a stateless delegation pattern that offloads tactical work to CLI subagents by injecting
a comprehensive "shared state" into every atomic command to prevent context fragmentation.

- **Shared State Handoff**: Inject the original goal, prior architectural decisions, and explicit
codebase scope into the prompt to bridge the subagent's lack of session history.
- **Contextual Preservation**: Execute high-volume tasks in isolated headless environments to save
orchestrator tokens and avoid "split story" failures caused by missing history.
- **Result Reconciliation**: Pull new technical assumptions and filesystem changes back into the main
thread after every delegation to maintain a single source of truth.

This methodology is further refined by lessons from **[Superpowers](https://github.com/obra/superpowers)**
and **[Claude Code](https://github.com/anthropics/claude-code)**, incorporating rigorous verification
loops and context hygiene to ensure subagents remain aligned with the primary orchestrator.

## License

This project is licensed under the MIT License - see the [LICENSE](LICENSE) file for details.
````
