Uttam Gelot · updated 2026-10-17 · 2 min read

Payo

Payo is an open-source command-line tool by Uttam Gelot. It asks short questions about a software project. Then it writes the rule and skill files that AI coding assistants read, such as AGENTS.md and Agent Skills. It supports Claude Code, Cursor, GitHub Copilot, Codex, Windsurf, Zed and Antigravity. "Payo" is the Gujarati word for "foundation".

First release: v0.1.0, 2026-06-07
Current version: 2.4.9
Releases: 45
GitHub stars: 30
npm downloads: 322 from 2026-09-17 to 2026-10-16
License: MIT
Built with: TypeScript, Bun, @clack/prompts, Zod
Links: Website · GitHub · npm

Numbers from the GitHub and npm APIs on 2026-10-19.

Install

macOS and Linux:

curl -fsSL https://payo.uttamgelot.com/install.sh | sh

Windows (PowerShell):

irm https://payo.uttamgelot.com/install.ps1 | iex

Or run it with npm:

npx @uge/payo

The problem

Every AI coding session starts with no context. The assistant does not know your ORM, your folder layout, your test style or your commit rules. So it guesses. On a real codebase, these guesses add up to code that does not match the rest of the project. Files like AGENTS.md and Agent Skills fix this, but most people do not write them, or they write them once and never update them.

How Payo works

Payo reads your project. It finds the stack from the manifest files. It reads your local git history to find your branch names and commit message style. This git data never goes into an AI prompt.
Payo asks only the questions that apply to your stack. It has more than 200 questions across more than 100 technology modules.
Payo shows the list of skills before it writes them. You can remove any skill.
Payo runs the AI CLI that you already have (claude, codex, cursor-agent, copilot, agy) to write each skill for your project. If no AI CLI is available, Payo uses templates.
Payo writes one SKILL.md for each topic under .agents/skills/, plus AGENTS.md. Claude Code and Windsurf get links to the same files.
Optional guardrails: Payo writes git hooks and pre-tool checks, so the agent cannot skip a rule.

What makes it different

It writes skills for your project. It does not copy a general skills pack.
It makes only the skills that your project needs. No database means no data-layer skill.
It uses your real commands, for example pnpm exec eslint ., not <your lint command>.
If it cannot find a fact in your code, it skips the topic. It does not invent a default.

What is next

From the Payo roadmap: per-package skills for monorepos, more languages and frameworks, and a faster question flow.