Skip to content
Back to skills

Antigravity Rescue

ASecurity

Use when stuck, wanting a second implementation or diagnosis pass, or handing a substantial task to Google's Antigravity CLI (agy) through a real subprocess -- a genuinely different agent harness, not just another model. Not relevant when already running under Antigravity CLI itself.

  • 3 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added September 2, 2026
ai-agentsrustgoshellbashsecurity

Works with

  • claude code
  • cli

Security analysis

A100/100

Pro scans all 2 files and shows the line behind each finding

Scanned September 2, 2026

npx -y skills add patrickserrano/lacquer --skill antigravity-rescue --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Antigravity Rescue?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Antigravity Rescue
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/patrickserrano-antigravity-rescue/badge)](https://www.skillsdirectory.com/skills/patrickserrano-antigravity-rescue)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: antigravity-rescue
description: >
  Use when stuck, wanting a second implementation or diagnosis pass, or
  handing a substantial task to Google's Antigravity CLI (agy) through a real
  subprocess -- a genuinely different agent harness, not just another model.
  Not relevant when already running under Antigravity CLI itself.
---

# Antigravity Rescue

A thin forwarding wrapper around Google's Antigravity CLI (`agy`). Your only
job is to forward the task and return its output. Do not inspect the repo,
read files, draft your own solution, or do any independent analysis — that
defeats the purpose of getting a second, independent opinion.

## Security: `agy -p` has no enforced sandboxing

Verified live (three separate tests, all on `agy` 1.0.6): in headless print
mode, `agy` executed a shell command, deleted a file inside the declared
`--add-dir` workspace, and deleted a file **outside** it — all instantly,
with zero confirmation, whether or not `--dangerously-skip-permissions` was
set. That flag appears to only affect the interactive TUI's confirmation
dialogs; headless mode has no technical permission boundary at all. `--add-dir`
is a workspace *hint* (it decides where ambiguous file operations default to
when unspecified) — it is **not** an access restriction, and does not confine
`agy` to that directory.

Practical consequence: `--read-only` below is **prompt-compliance only** —
the model choosing to honor an instruction, not anything CLI-enforced. There
is no equivalent of Claude Code's `--permission-mode plan` here. Forward a
task expecting the same trust you'd extend to giving `agy` unrestricted
filesystem access under your own user account, regardless of which flag you
pass — because that's what you're actually doing.

## When to use

- Proactively, when stuck: a recurring error, an approach that isn't
  converging, or a task that would benefit from a different agent harness's
  take. `agy` is a multi-vendor router (`agy models` lists Gemini, Claude, and
  GPT-OSS variants) — the value isn't necessarily a different underlying
  model, it's Google's own agent orchestration and tool-execution layer.
- Don't grab a task you can finish quickly yourself — this is a rescue
  mechanism, not a first resort.

## How to forward

```bash
scripts/antigravity-companion.sh task "<the task, verbatim>" [--model <name>] [--resume|--fresh] [--write|--read-only] [--timeout <duration>]
```

- Preserve the task text as-is; strip only the routing flags below before
  forwarding.
- **`--model`**: leave unset by default (agy picks its own default). Only
  pass a value on explicit request — check `agy models` for the current
  list; names span multiple vendors, not just Gemini.
- **`--write` / `--read-only`**: `--write` (default) runs with
  `--dangerously-skip-permissions`. `--read-only` prepends an explicit
  "diagnosis only, do not modify files" instruction to the task text —
  see the Security section above: this is the *only* protection either
  mode has, CLI-enforced or not. Prefer `--read-only` whenever the ask
  doesn't clearly need `agy` to make changes.
- **`--resume` / `--fresh`**: `--fresh` (default) starts a new Antigravity
  conversation. Use `--resume` (`agy -c`) only when continuing a prior
  rescue in the same repo.
- **`--timeout`**: defaults to `10m` (agy's own default is `5m`, tight for a
  substantial rescue task). Raise it further for a large ask.
- Return the script's output exactly as printed. No added commentary before
  or after it. If the call fails, surface the raw failure — don't paper over
  it or retry silently.

## Why `--add-dir` is still non-negotiable

The companion script always passes `--add-dir "$(pwd)"`. Without it, `agy`
silently operates in its own internal scratch sandbox
(`~/.gemini/antigravity-cli/scratch`) instead of the real project —
confirmed live, not a hypothetical edge case. This makes ambiguous file
operations land in the right place; it does **not** restrict `agy` to that
directory (see Security above) — never invoke `agy -p` directly without it
regardless.

Files in this skill

  • SKILL.md4 KB
  • scripts/antigravity-companion.sh2.7 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.