ShellGPT
Turns a plain-English request into the terminal command you need, and can run it for you
by Farkhod Sadykov
Plain-Language Step Guide
by Pavlo KharmanskyiSkillWrite code
Turns a request from your AI agent into numbered steps you can follow without coding
The problem it solves
Your agent asks you to "set the env var on the server", and you have no idea which window, command or button it means.
For People building with AI agentsReaders who do not code
- Before
- You copy a half-explained command, guess where it goes and come back with three follow-up questions.
- After
- You get what it needs and why, one action per step and a way to check each one, and the agent then confirms your step worked.
- Stars
- ★ 1.3k
- License
- MIT
- Cost
- Free
- Level
- Beginner
- Contents
- Instructions only
Overview
This skill from the Open Steps pack changes how a coding agent asks you to do something, such as run a command, paste a secret or click a button. Before asking, the agent must try the task itself and shrink the request to the part only you can do. You then get what it needs and why, numbered steps with one action each, how to tell it worked and what it does next, and afterwards it checks your step. It suits people who build with an AI agent but do not read code.
The agent has to earn the request first, and secrets go through a one-line command that hides your typing instead of through the chat.
Try these first
Start with the first one: it is a small task you can check easily.
- Walk me through it, step by step: [the setup step you asked me to do].First try
- I don't understand what to do. What exactly do you need from me?
- I need to add [an API key] to [the test server]. Give me one command that asks for it without showing it.handles a secret
- Before asking me, try it yourself and tell me only the part you can't do.
Example prompts written by Agent Flows Hub.
What you get
From the creator's documentation.
| Output | What it is | Reads |
|---|---|---|
| Earn the ask | The agent tries first, looks for another route and shrinks the request to what only you can do | This session |
| Numbered steps | One action per step, each command labelled with the machine it touches | This session |
| Secret entry | One-line commands that prompt for a value, hide your typing and confirm by size, not content | Your computer and server |
| Check afterwards | Runs the quickest check that your step worked; if not, gives only the corrected step | Your computer |
| Choices | A decision gets a plain question with two to four options, the recommended one first | This session |
Good to know
When the agent needs you to run, paste, click or approve something, get one plain-language instruction per step. Commands are single lines that prompt for any value and confirm in plain words what happened.
Limits
- The secret-entry command templates are written for bash and zsh and have not been run on Windows; there the skill asks you to use a separate Git Bash window.
- A decision, rather than an action, gets a short question with two to four options instead of steps.
Notes by Agent Flows Hub (original project name: os-step-by-step), written from the creator's documentation, not from running it ourselves. os-step-by-step SKILL.md · Open Steps README
Stuck setting it up?
Tell us what happened. We take a look first; if it needs paid work, you get an estimate before anything starts.
More to write code
Other skills
- Planning with Files Makes your AI agent keep its plan, notes and progress in files so long tasks stay on track
- PM Resume Review Gives point-by-point feedback on a product manager resume, with suggested rewrites
- PPT Master Turns a document or a topic into a PowerPoint deck you can keep editing
- PRD Writer Writes a short product requirements document from your idea, users and success measure
- Product Deep-Dive Report Researches a product, company or idea online and writes a long PDF report in Chinese
- Product Listing Writer Writes titles, bullet points, descriptions and keywords for your online shop listings