agystack skill
An Antigravity plugin that runs poteto-mode playbooks, isolated subagents and cloud swarms on Google Cloud Run.
- 131 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Subagents and `/loop` edit files and run verify commands on their own.
- `/swarm` creates containers in your Google Cloud project and spends budget.
- `watch-pr` acts in your pull requests through `gh`.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/agystackDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Subagents and `/loop` edit files and run verify commands on their own.
- `/swarm` creates containers in your Google Cloud project and spends budget.
- `watch-pr` acts in your pull requests through `gh`.
This is third-party code. Review the repository files before installing.
What it does
agystack is a plugin for Google Antigravity built on Lauren Tan's (poteto) pstack set. `/agystack` switches on `poteto-mode`, picks one of 23 playbooks and hands large edits (over 50 lines or several files) to the isolated `poteto-agent` subagent. Results land in Antigravity artifacts: a plan, `walkthrough.md` and a report. On top of pstack it adds `/swarm` (parallel workers in Google Cloud Run containers, patches delivered to GCS), `/loop` (hypothesis and verify against a `--verify "<cmd>"` command, reverting regressions), `watch-pr` (watching CI, Graphite stack conflicts and review comments), `/interrogate` and `/arena` (adversarial review and competing implementations) and `/deslop`.
Who it is for. Developers on Google Antigravity who want a rigorous process verified against real runs and several agents working in parallel.
Good fit when
- A task touches several files and the edits should go to a subagent with a clean context.
- A test fails and `/loop` with a verify command iterates edits until it turns green.
- A pull request must reach merge, with CI watched and comments triaged.
- Two implementations are compared in parallel worktrees through `/arena`.
Not a fit when
- Work happens in Claude Code or Cursor, while the plugin targets Antigravity.
- Cloud execution is off limits, and `/swarm` uses Google Cloud Run containers.
- The edit fits one line, where the full process is overkill.
Example request
`/agystack` Find the cause of the failing test, fix it through a subagent and confirm with the test command.Limitations
Works only in Antigravity. `watch-pr` needs `bun` 1.0 or newer and `gh`, `gt` (Graphite) is recommended for stacked PRs, and Node.js is not supported. Cloud swarms run in your Google Cloud project, so containers and GCS patch storage are billed. `/loop` executes your verify command and reverts edits itself, so run it on a clean branch. The plugin grew from pstack, which has its own card. The playbooks are strict and verbose and slow down small tasks.
How to disable. Remove the plugin with Antigravity tools or the clone folder `~/.gemini/config/plugins/agystack`. Delete `/swarm` cloud resources in Google Cloud separately.
FAQ
What install options exist?
Three: `agy plugin install agystack`, asking the agent to install from the link, and cloning into `~/.gemini/config/plugins/agystack` or `.agents/plugins/agystack`.
What does `watch-pr` need?
`bun` 1.0 or newer and `gh`; `gt` (Graphite) is advisable for stacked PRs.
How does `/swarm` differ from a subagent?
A subagent works locally, while `/swarm` starts parallel workers in Cloud Run and returns isolated patches through GCS.
Where do results end up?
In Antigravity artifacts: `implementation_plan.md`, `walkthrough.md` and a topic report.
What is poteto-mode?
A strict engineering mode: the agent reads the principles, picks one of 23 playbooks and hands large edits to the `poteto-agent` subagent.
Discussion
No questions or reviews yet.
Related
SkillsA .agents workspace for Google Antigravity: 20 roles, 47 skills, 13 commands, project memory, a safety hook and a CLI with update rollback
Delegating to an intern: Google's Antigravity CLI (agy) with a fast Gemini model for Claude Code, Codex, Pi and OpenCode. Code and tasks go to Google
Google's terminal AI agent: edits code across files, runs commands and keeps session history, started with the agy command
A plugin that calls Antigravity CLI from inside Claude Code: quick prompts, delegated tasks and a second opinion on your diff.
pstack for Claude Code and Codex
pstack-claude
An unofficial port of Lauren Tan's pstack skill set (Cursor) for Claude Code, Codex, Pi, OpenCode and Gemini: strict workflows and the poteto-mode
An MCP bridge: Claude Code hands heavy tasks to the Antigravity CLI and only the finished answer returns to Claude context
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| agystack (this card) | no | 131 | |
| pstack | no | 4,065 | a mirror of the Cursor pstack set; agystack carries the approach into Antigravity and adds cloud swarms, `/loop` and `watch-pr` |
| Antigravity for Claude Code | no | 484 | a Claude Code plugin that hands bulk work to Gemini; agystack runs inside Antigravity itself |
README in short
The README covers cloud swarms, the `/loop` cycle, pull request babysitting, adversarial reviews and subagent context isolation, followed by installation and the list of native Antigravity tools.