Superpowers Sage — AI Agent Plugin for Sage/Acorn/Lando

showcase
, , ,
1

Hey Roots community! :waving_hand:

I’ve been building with Sage + Acorn + Lando for a while now, and like many of you, I started integrating AI assistants into my workflow. The problem? Every session I’d end up re-explaining the same things: how ACF Composer works, that we use Poet for CPTs, that routes go through Acorn, not register_rest_route(), that assets rebuild via Lando… It got old fast.

So I built something to fix that.

Superpowers Sage

An agent plugin that gives your AI assistant deep, opinionated knowledge of the Roots ecosystem — and keeps it that way across sessions.

GitHub: https://github.com/codigodoleo/superpowers-sage

It works with Claude Code, VS Code (GitHub Copilot Chat), and Cursor. Same plugin, same skills, installed once.

What it actually does

Instead of pasting docs or re-explaining conventions every time, the plugin injects that knowledge automatically and gives you slash commands for the full development workflow:

/architecture-discovery  →  produces an approved Architecture Decision Record
/plan-generator          →  turns the ADR into an executable plan with assets
/building                →  implements from the plan, verifies each component
/reviewing               →  convention audit + design alignment check

For day-to-day work, individual skills like /modeling, /debugging, /designing, and /verifying can be called directly.

The opinionated part

The plugin enforces patterns I care about for Roots projects:

Instead of Use
register_post_type() Poet (config/poet.php)
ACF GUI ACF Composer
register_rest_route() Acorn Routes
CF7 / Gravity Forms Livewire + HTML Forms
Raw cron Action Scheduler + Queue/Job
Shortcodes Blade Components

Your AI assistant won’t suggest the “wrong” way because it simply doesn’t know one — the plugin corrects the frame before any code gets written.

The plan system

One thing I’m particularly happy with: plans survive context compression. When the AI session gets long and the context window compresses, hooks re-inject the active plan path automatically. No more “wait, what were we building again?”

Plans live in docs/plans/YYYY-MM-DD-<topic>/ and include the architecture record, content model, design assets, component sub-plans, and an auto-populated activity log.

Design tool integration

If you use Figma or Google Stitch, the /designing and /verifying skills connect directly to them. The AI can extract design specs and then screenshot your implementation to compare — pixel-by-pixel visual verification before you mark a component done.

Zero-token hooks

There are lifecycle hooks that run without consuming LLM context at all:

  • post-edit — runs lando flush for PHP files, lando theme-build for assets, automatically

  • session-start — health check, detects design tools, injects the ecosystem guide

  • pre-commit — reminds you to visually verify against design reference before committing

Installation

# Claude Code
claude plugin marketplace add codigodoleo/superpowers-sage
claude plugin install superpowers-sage@superpowers-sage

# Then in your Sage project:
/onboarding

/onboarding analyzes the project, detects your stack and packages, and suggests next steps.

VS Code and Cursor instructions are in the README.

This is still early — I’m actively building on it and would love feedback from people who live in this stack daily. If you try it and something doesn’t match how you work with Sage/Acorn, I want to know.

Happy to answer questions here. :seedling:

1 Like