Superpowers port for Antigravity 2.0, IDE & CLI

Google Antigravity
1

Hey guys,

Wanted to share this here as superpowers helps me tremendously.

I’ve ported superpowers framework ( GitHub - obra/superpowers: An agentic skills framework & software development methodology that works. · GitHub ) to work natively with Antigravity 2.0, Antigravity IDE and Antigravity CLI.

I kept all the skills and prompt templates exactly the same, meaning this fork is fully backward-compatible with Claude Code and other systems. The changes in this port are focused entirely on adapting the tool mappings and subagent orchestrations to work natively with the new Antigravity 2.0 engine.

Here is a summary of what was changed to make it work:

* Added a plugin.json manifest so the Antigravity 2.0, Antigravity IDE and Antigravity CLI automatically discover and loads the plugin on startup.

* Created an antigravity-tools.md mapping file to translate upstream skill commands into Antigravity 2.0 tools (like mapping Read to view_file, Glob to find_by_name, and Bash to run_command).

* Integrated workspace isolation. The worktree setup now natively uses the Workspace: “branch” parameter inside invoke_subagent to isolate git branches automatically.

* Documented token-saving optimizations like using define_subagent for multi-agent plans.

* Wrote a new test suite under tests/antigravity/ that verifies plugin loading, skill triggering, and subagent dispatching.

For full changes from the original superpowers check the RELEASE_NOTES.md file: superpowers/RELEASE-NOTES.md at main · roundpilot/superpowers · GitHub

Instructions on how to set this up are on the GitHub repo below.

Link to the repo: GitHub - roundpilot/superpowers: An agentic skills framework & software development methodology that works. · GitHub

2

v6.0.0 — Full native rebuild for Antigravity 2.0

The original port (linked in the post) used a tool mapping layer — skills were still written in Claude Code tool names and translated at runtime. It worked, but it cost tokens every session and added complexity.

I’ve since done a complete rebuild in a new repo: superpowers-antigravity

What’s different:

  • 86 files removed — all legacy platform code (Claude Code, Codex, Cursor, Copilot CLI, etc.) is gone

  • No more tool mapping — skills use view_file, run_command, invoke_subagent directly. Zero translation overhead

  • New capabilities — native generate_image mockups in brainstorming, browser-testing skill for UI verification, async subagent coordination with manage_task and schedule, rich plan formatting with Mermaid diagrams and diff blocks

  • Lighter subagents — reviewers use read-only TypeName: "research" instead of full agent clones

  • Manual install option — no git required, just download the ZIP

If you’re on Antigravity 2.0, switch to the new repo. The old one won’t receive updates.

Install: git clone https://github.com/roundpilot/superpowers-antigravity ~/.gemini/config/plugins/superpowers

3

For each prompt—whether coding, asking for an explanation, or requesting a specific task—do I need to execute the /superpowers: using-superpowers command every time, or is running it once per session enough to continue asking without using the command again?"

4

You should only need to use it once per session, the agent will know which skills to select when needed, but if you still have a chat open and are looking to make a big change/implementation then it doesn’t hurt to use the slash command again.

5

Thank you very much for the reply, @Kahun . I have another question: when using the Antigravity CLI and executing any superpowers command in the terminal or PowerShell, the /artefact command appears. However, I am not completely sure if this command is part of superpowers or how these two tools are meant to interact.

6

The /artifact command is a built-in feature of Antigravity CLI, nothing to do with superpowers, but they do interact together.

Superpowers workflows and skills generate agent artifacts such as implementation_plan.md, task.md and walkthrough.md. The CLI provides the /artifact command to open the text user interface to inspect these files.

7

I’ve released version 6.2.0 version today. The using-superpowers skills will always load automatically so using the slash command is no longer required. This matches the behaviour of the upstream version of superpowers.

8

Hey thank you !

Would like to add that, per my experience, it’s better to still specify the /skill if you want the model to really adhere to it (esp Gemini models)

9

With the changes amongst Antigravity and AI clients in general, to be more uniform and follow specifications.. and additions,.. Yep. Tested and confirmed[again]: there is literally no use anymore for this repo except for ‘saving space’[1] . or minor (visual) tweaks.

It is even worse on Antigravity CLI, Antigravity & Antigravity IDE compared to the official repo. It even breaks Antigravity, kills antigravity to the point where it wastes time and resources.. It isn’t even functional to the point the entire ‘superpowers’ core functionality is broken and destroyed.. its not just useless, its destructive and time and resource wasting. Amazing. And why ?[2].

With the official repo the performance, from testing over 1 week is much better and consistent. The official repo supports all the Antigravity products just fine. Even ‘interchangeable commands’, proper styling --whatever. I didn’t miss a thing. And surely not Antigravity dying. Or superpowers not even working, to the point where it worked against everything. (0.5 week x, 0.5 week y, same “long running, same repetitive tasks” from consistently the same projects…).

Can’t say the same about the 0.5 week on this repo. It has been terrible and even simple tasks were failing. Didn’t even invoke whilst slash commands were used explicitly, and global overrides. Didn’t use skills correctly. Crashing antigravity when invoking subagents and all that. Antigravity getting “stuck” and had to undo entire prompts. What a waste of my time.. Literally the “core functionality” of superpowers was broken, non consistent, non functional, even causing crashes and usage/time waste.. Whoa; that’s almost admirable..

& If some really tiny things bother u, in styling or that won’t 1:1 carry over now, why not just write the 5 lines of skills to change the 'x` is different here visually FOR YOURSELF and all it a day ?

You just seem stuck on a project – and for what ? Those months have passed where it made sense per AI client rewrites. – so much so its to the point its not only non functional, but worse than the original, and even killing Antigravity ?

:warning: I SINCERELY HOPE new Antigravity users/people unbeknownst aren’t drawn to your project, with the popularity of the original repo. That experience would get anyone to unsubscribe if they were unaware of the havoc it causes.

Wow. Just wow.

Just another one who takes a good project and throws something at it and releases it for ‘internet kudos’ while inherently not just not improving it but making it worse or destroying its core purpose. I am almost impressed by this, in the same way people cannot look away from a fire.

  1. The only thing I can think of why you’d continue this or anyone would attempt: If you are so worried about some extra files ON A ORIGINAL REPO THAT IS 2.5MB; your priorities should be elsewhere to begin ↩︎

  2. Some months+ ago, yes, this made sense, ‘per AI client, specialized, and..’, but right now its not only counter intuitive but literally causing problems, wasting time, breaks the entire purpose of the original superpowers and its functionalities. ↩︎

10

Thanks for the critical feedback, I’ll skip past the “internet kudos” jab. I made this for my own daily workflow and decided to share it with others. I’m more than happy to fix any issues you experienced, but I need some more details:

  • What version of superpowers-antigravity were you running?
  • What’s your environment (OS, IDE or CLI, is Git Bash installed)?
  • Which AI model?
  • What version of Antigravity?
  • Any steps to reproduce?

Last week’s v6.2.0 rewrote a lot of the internals. Before that some skills were teaching outdated tool calls, including a bad subagent dispatch shape and the review loop had no cap on long running tasks. You might find the latest version much more solid in your environment.

Your report also already prompted me to look into v6.2.0 and found one bug that it shipped with; an edit-time hook that could block file writes on machines without Git Bash. As of v6.2.1 (today) it’s opt-in. If your testing was after last week and you saw writes failing, that was probably it.

Either way update it, rerun the same tasks, and let me know if the problems persist. A GitHub issue with the details above would be ideal.