Superpowers for Google Antigravity — a complete development workflow powered by composable skills, workflows, and rules.
This repo ports the Superpowers skills library to work natively with Google Antigravity agent, with Antigravity-compatible path conventions, an auto-update workflow, and curated workflow files translated from the upstream skills.
Inside any project directory:
npx agy-superpowers@latest initThis scaffolds .agent/ with all Superpowers skills, workflows, and rules.
Open the project with Google Antigravity — Superpowers activates automatically.
To overwrite an existing .agent/:
npx agy-superpowers@latest init --forcenpm install -g agy-superpowers
agy-superpowers init.agent/
├── rules/
│ └── superpowers.md # Always-on rule: agent checks skills before acting
├── skills/ # 23 active skills (14 upstream + 9 expansion skills)
│ ├── brainstorming/
│ ├── writing-plans/
│ ├── executing-plans/
│ ├── frontend-developer/
│ ├── mobile-developer/
│ ├── subscription-billing/
│ ├── i18n-localization/
│ └── ... # + 15 more skills
├── workflows/
│ ├── brainstorm.md # /brainstorm
│ ├── write-plan.md # /write-plan
│ ├── execute-plan.md # /execute-plan
│ ├── code-review.md # /code-review
│ ├── debug.md # /debug
│ ├── publish.md # /publish
│ └── update-superpowers.md # /update-superpowers
├── patches/
│ └── skills-patches.md # AI-driven skill patches (see below)
└── .shared/
└── update-superpowers.sh # Shell script for version management
Antigravity automatically reads skills before every response. You don't need to invoke them manually — the superpowers.md rule (set to alwaysApply: true) instructs the agent to check for a relevant skill before acting.
💡 Idea → /brainstorm → /write-plan → /execute-plan → /code-review → merge
| Step | Workflow | What happens |
|---|---|---|
| 1 | /brainstorm |
Agent asks clarifying questions, refines your idea into a spec |
| 2 | /write-plan |
Spec → bite-sized tasks (2–5 min each) with exact file paths |
| 3 | /execute-plan |
Tasks run in batches with human checkpoints |
| 4 | /code-review |
Pre-review checklist, severity-based issue reporting |
| 5 | (auto) | finishing-a-development-branch — merge / PR / discard |
| Workflow | What happens |
|---|---|
/debug |
Systematic 4-phase debugging process |
/publish |
Publish to npm with version bump, git tag, and push |
/update-superpowers |
Pull latest upstream release, re-sync skills, AI-rewrite rules |
/ui-ux-pro-max |
Plan and implement UI with curated design intelligence |
/mobile-uiux-promax |
Plan and implement mobile app UI across all major platforms |
How it works: Superpowers gives your agent 23 specialized skills. You don't need to remember or invoke them — the agent automatically reads and selects the right skill based on what you ask. Just describe what you need in natural language.
Let's say you want to add dark mode to your React app. Here's what the full cycle looks like:
| Step | You type | What happens |
|---|---|---|
| 1 | /brainstorm add dark mode to my React app |
Agent asks clarifying questions, explores approaches, and produces a design spec |
| 2 | /write-plan |
Spec gets converted into a task list with exact file paths and acceptance criteria |
| 3 | /execute-plan |
Agent implements tasks one by one, pausing for your review at checkpoints |
| 4 | /code-review |
Agent reviews the completed code against the plan, reports issues by severity |
| 5 | (automatic) | Agent offers to merge, create a PR, or discard the branch |
That's it. Five steps from idea to merge-ready code.
| Command | What it does | Example |
|---|---|---|
/brainstorm |
Turn a rough idea into a validated design spec | /brainstorm add user onboarding flow |
/write-plan |
Break an approved spec into bite-sized tasks | /write-plan (after brainstorm) |
/execute-plan |
Execute tasks with human review checkpoints | /execute-plan (after write-plan) |
/code-review |
Review completed work before merging | /code-review |
/debug |
Systematic 4-phase debugging for any issue | /debug users can't log in on Safari |
The agent doesn't just write code — it has specialized knowledge in marketing, pricing, security, legal, and more. When you mention a topic, the relevant skill activates automatically.
These are triggered via slash commands or activate automatically during the dev loop:
| Skill | When it activates |
|---|---|
brainstorming |
Adding a feature, building a component, or modifying behavior |
writing-plans |
Design is approved — breaking work into tasks |
executing-plans |
Running a plan step-by-step with checkpoints |
test-driven-development |
Implementing any feature or bugfix |
systematic-debugging |
Debugging any issue |
verification-before-completion |
Before declaring a fix or task done |
requesting-code-review |
Completing tasks or before merging |
receiving-code-review |
Receiving code review feedback |
finishing-a-development-branch |
Implementation is complete — merge / PR / discard |
using-git-worktrees |
Starting work on an isolated branch |
dispatching-parallel-agents |
2+ independent tasks that can run in parallel |
subagent-driven-development |
Executing plan tasks in the current session |
Activate when you work on specific technical domains:
| Skill | When it activates |
|---|---|
frontend-developer |
Building web UI, component architecture |
mobile-developer |
Mobile app features, React Native / Flutter / iOS / Android |
frontend-design |
Web components, pages, apps — high design quality |
mobile-uiux-promax |
Mobile app UI for iOS, Android, React Native, Flutter |
Example prompt: "Build a responsive dashboard with dark mode" → activates
frontend-developer+frontend-design
| Skill | When it activates |
|---|---|
product-manager |
Defining requirements, prioritizing features, planning roadmaps |
ux-designer |
Designing UI, wireframes, user research, information architecture |
copywriter |
Landing page copy, app store descriptions, email sequences |
i18n-localization |
Internationalization, localizing for new markets |
Example prompt: "Help me prioritize features for the next sprint based on user feedback" → activates
product-manager
| Skill | When it activates |
|---|---|
subscription-billing |
Stripe integration, webhooks, trial logic, dunning flows |
Example prompt: "Set up subscription billing with Stripe and handle trial-to-paid conversion" → activates
subscription-billing
| Situation | What to do |
|---|---|
| Multi-file feature, new component, architectural change | /brainstorm → full design cycle |
| Quick fix, rename, typo, single-line change | Just ask directly |
| Knowledge question ("What's the best auth library?") | Just ask directly |
| Not sure about scope | Start with /brainstorm — it's safe to exit early |
Rule of thumb: if it touches more than one file or involves design tradeoffs → /brainstorm.
- Provide context — mention your tech stack, project name, and constraints: "In my Next.js app with Supabase auth..."
- State goals, not solutions — "Users drop off during signup" gives the agent room to find the best fix, vs "Add a progress bar to the signup form"
- Mention specific domains to activate the right skills — "Design the UX for my onboarding flow" triggers
ux-designer, "Localize my app for Japanese" triggersi18n-localization
Full product launch cycle:
/brainstorm → /write-plan → /execute-plan → /code-review
↓ then in separate conversations:
"Write marketing copy for..." → copywriter skill
"Set up billing with Stripe..." → subscription-billing skill
"Localize for Japanese market..." → i18n-localization skill
Debugging effectively:
/debug → systematic investigation
↓ if architectural issue found:
/brainstorm → redesign the component
Per-project settings (like auto_commit) live in .agent/config.yml — see Configuration for details.
| ❌ Don't | ✅ Do instead |
|---|---|
Skip /brainstorm for large features |
Always brainstorm features that touch multiple files |
| Cram multiple goals into one prompt | One conversation = one clear objective |
Jump to /execute-plan without reviewing the spec |
Review and approve the design spec before execution |
Skip /code-review before merging |
Always review before merge — catch bugs early |
| Use long conversations for multiple tasks | Start a new conversation for each distinct task (saves tokens) |
Skills are sourced from upstream Superpowers but automatically patched to be Antigravity-native. The patching system uses a human-readable spec at .agent/patches/skills-patches.md which the AI interprets semantically — making patches resilient to upstream changes.
When you run /update-superpowers, skills are:
- Copied fresh from
superpowers/skills/ - Patched via AI to remove non-Antigravity references
- Ready to use immediately
| Skill | Triggers automatically when... |
|---|---|
brainstorming |
Adding a feature, building a component, or modifying behavior |
copywriter |
Writing landing page copy, app store descriptions, email sequences, or user-facing text |
dispatching-parallel-agents |
Facing 2+ independent tasks that can run without shared state |
executing-plans |
Running a plan step-by-step with checkpoints |
finishing-a-development-branch |
Implementation is complete and you need to merge / PR / discard |
frontend-design |
Building web components, pages, or apps with high design quality |
frontend-developer |
Building web UI, component architecture, or reviewing frontend code |
i18n-localization |
Planning i18n architecture, localizing for new markets, or managing translations |
mobile-developer |
Working on mobile app features, reviewing mobile code, or mobile architecture |
mobile-uiux-promax |
Designing or building mobile app UI for iOS, Android, React Native, Flutter, SwiftUI, or Compose |
product-manager |
Defining requirements, prioritizing features, planning roadmaps, or validating problems |
receiving-code-review |
Receiving code review feedback, before implementing suggestions |
requesting-code-review |
Completing tasks or before merging to verify work meets requirements |
subagent-driven-development |
Executing implementation plans with independent tasks in the current session |
subscription-billing |
Integrating subscription billing, Stripe webhooks, trial logic, or dunning flows |
systematic-debugging |
Debugging any issue |
test-driven-development |
Implementing any feature or bugfix |
using-git-worktrees |
Starting work on an isolated branch |
using-superpowers |
Starting any conversation — checks for relevant skills |
ux-designer |
Designing UI, wireframes, user research, or information architecture |
verification-before-completion |
Before declaring a fix or task done |
writing-plans |
Design is approved — breaking work into tasks |
writing-skills |
Creating or editing a skill |
Upstream Superpowers releases are tracked automatically. To update:
/update-superpowers
This workflow will:
- Pull the latest Superpowers release from GitHub
- Copy fresh skills from upstream
- AI-rewrite any changed workflows and rules to stay Antigravity-compatible
- Apply skill patches via
.agent/patches/skills-patches.md - Commit all changes with a versioned commit message
Per-project settings live in .agent/config.yml. Create or edit this file in your project's .agent/ folder:
# .agent/config.yml
# auto_commit: true | false
# When true (default), Superpowers skills automatically commit after completing
# tasks and writing design docs.
# When false, all git commits and staging are skipped — files are left as
# modified for you to commit manually.
auto_commit: true| Setting | Default | Description |
|---|---|---|
auto_commit |
true |
AI auto-commits after tasks and design docs. Set to false to skip all commits. |
This file is preserved across /update-superpowers runs.
The recommended way is npx (see Quick Start above). Alternatively, clone and copy:
git clone https://github.com/bonnguyenitc/antigravity-superpowers
cp -r antigravity-superpowers/.agent /your/project/This setup enforces four core principles across every task:
- Test-Driven Development — Write the failing test first, always
- YAGNI — Don't build what isn't needed yet
- Systematic over ad-hoc — Follow the skill process, don't guess
- Evidence over claims — Verify before declaring success
- Token efficiency — 23 focused skills instead of 58, optimized for the workflows you actually use
This project stands on the shoulders of many great open-source projects. A huge thank you to all of them! 🙏
Content from these repos was ported, adapted, and integrated into the skills library:
| Repository | Author | Skills |
|---|---|---|
| obra/superpowers | Jesse Vincent & Prime Radiant | Core foundation — upstream skills library this repo is built on |
| vercel-labs/agent-skills | Vercel Labs | 66 React/Next.js rules + 36 React Native rules (in frontend-developer + mobile-developer) |
| antfu/skills | Anthony Fu | 44 Vue/Nuxt rules (in frontend-developer) |
| vuejs-ai/skills | vuejs-ai | Vue ecosystem agent skill rules (upstream of antfu/skills) |
| kevmoo/dash_skills | Kevin Moore | 8 Dart & Flutter best-practice rules (in mobile-developer) |
| new-silvermoon/awesome-android-agent-skills | new-silvermoon | 17 Android/Compose/Kotlin rules (in mobile-developer) |
| AvdLee/SwiftUI-Agent-Skill | Antoine van der Lee | 19 SwiftUI reference files (in mobile-developer) |
| msitarzewski/agency-agents | msitarzewski | Deliverable-focused, workflow-oriented agent patterns (inspired indie hacker skills) |
| nextlevelbuilder/ui-ux-pro-max-skill | nextlevelbuilder | UI/UX Pro Max search databases — curated style, typography, color, UX, and stack knowledge (in .shared/ui-ux-pro-max) |
Antigravity integration, workflow adaptation & 44 expansion skills (UI/UX intelligence databases, mobile design system) by @bonnguyenitc.
MIT — see LICENSE for details.
