I Built a CLI for Reusable AI-Agent Workflows

The Problem: Agent Workflows Are Still Hard To Share

AI coding agents are getting better, but the workflows around them are still surprisingly manual. If you have a good workflow, it probably looks something like this:

• clarify the request
• write a proposal
• review the design
• create an implementation plan
• build with tests
• verify the result
• archive the learning

That process might work well for one person. The problem is making it repeatable for a team, another project, or even your future self. Most agent workflows still live in prompt history, docs, or muscle memory.

What I Built

I built GetSuperpower, a small CLI for packaging a complete AI-agent workflow as one installable skill tree. The idea is simple:

• define a workflow in workflow.json
• provide one entry skill the user calls
• include local workflow-specific skills when needed
• declare dependencies on reusable skills
• install the workflow into supported agent clients

The user does not need to manually call every step. They install the workflow, restart their agent, and invoke one entry skill.

Example

Install the OpenSpec delivery workflow:

npx getsuperpower@latest install 'https://github.com/0xroylee/getsuperpower.git#examples/workflows/openspec-superpowers'

Then call:

$openspec-delivery implement this OpenSpec change

That entry skill coordinates: proposal -> design -> plan -> TDD -> verification -> archive. The workflow can use local skills and reusable skills together, so authors can package their own process without rebuilding everything from scratch.

Why Not Just Use A README?

A README is useful for humans, but it is not enough for repeatable agent execution. An agent workflow needs:

• a clear entry point
• explicit dependencies
• install targets
• validation
• a way to inspect what will be installed

GetSuperpower keeps the README, but adds a manifest and CLI around it.

What It Is Not

GetSuperpower is not trying to be a full agent runtime. If you are building a production agent application, tools like LangGraph, CrewAI, or n8n may be a better fit. GetSuperpower is more like a package manager for the process around an existing coding agent. It packages the instructions, dependencies, and workflow shape that you want the agent to follow.

Supported Targets

GetSuperpower currently supports skill installs for:

• Claude
• Codex
• Cursor
• opencode / OpenCodex
• GitHub Copilot

Creating Your Own Workflow

You can scaffold your own workflow:

npx getsuperpower@latest init my-workflow
npx getsuperpower@latest validate my-workflow
npx getsuperpower@latest deps my-workflow

The goal is to make agent workflows shareable the way packages are shareable: installable, inspectable, versioned, and easy to try.

Repo: https://github.com/0xroylee/getsuperpower

I would especially love feedback on the manifest shape and dependency model.