Use the same agent skill in Claude Code, Cursor, and Codex

You wrote a skill that works. Then you open Cursor instead of Claude Code and it is not there, so you copy the file across. Later you sharpen one line in one copy and forget the other two. Now the same skill behaves differently depending on which tool you happened to open, and adding a second laptop and one teammate turns three copies into six.

Copying more carefully does not solve this. Keeping one hosted copy that every tool reads does. This guide walks through that: publish once, point each tool at it, and pull.

Publish the skill once

Goodeye is the hosted service the skill lives in. You reach it with the goodeye CLI, an MCP server, or a REST API. The CLI needs an account and a one-time sign-in, both covered in getting started.

Start from the local skill file you already have and publish it:

goodeye skills publish ./SKILL.md

That creates the hosted skill and prints its id, its version number, and a version token. Keep the token: every later publish to the same skill passes it back with --expected-version-token, which is what stops two people from silently overwriting each other. Roll back a bad agent skill update shows how to read it back when you need it.

A note on naming, because the two things share a word. The local skill file is the SKILL.md on your disk. The hosted skill is the versioned object in Goodeye. From here on, every tool reads a copy of the hosted skill, and you stop editing the local files directly.

Point each tool at its skills directory

Every agent loads skills from a directory. You tell Goodeye which directories to mirror into, once per machine:

goodeye skills sync target add --preset claude    # ~/.claude/skills
goodeye skills sync target add --preset codex     # ~/.agents/skills
goodeye skills sync target add --preset cursor    # ~/.cursor/skills

codex and agents are the same preset under two names, both pointing at ~/.agents/skills, which is the shared location Codex reads. If your tool reads somewhere else, pass a path instead of a preset:

goodeye skills sync target add ~/.config/my-agent/skills

This step is local configuration only. It does not move any files yet. Check what a machine is configured to mirror with:

goodeye skills sync target list

A target mirrors the skills you own unless you say otherwise. Choose the scope when you create the target, because it cannot be changed in place:

goodeye skills sync target add --preset cursor --scope selected --only "release-notes-*"

--scope takes owned (the default), all, or selected, and --only takes a skill slug or a glob and can be repeated. If you already created the target and want a different scope, remove it and add it again: goodeye skills sync target remove ~/.cursor/skills, then re-add with the scope you want.

Pull, and every tool sees the same skill

goodeye skills sync pull

That materializes the current hosted version of every in-scope skill into every configured target. Run it on each machine. Running goodeye skills sync with no subcommand does the pull and then prints where each skill stands, which is usually the one command you want.

Edit the hosted skill, publish, and the next pull on any machine updates Claude Code, Codex, and Cursor together. Nobody has to be told to update.

Check where a machine actually stands

Before you trust that a machine is current, ask it:

goodeye skills sync status

Nothing is written and nothing on disk changes. It compares what Goodeye reports against what was last mirrored and what is on disk, and classifies each skill as clean, edited locally, behind the registry, conflicted, deleted upstream, or a local directory Goodeye does not track yet.

"Edited locally" is the interesting one. It means someone changed the file on disk instead of publishing a new version, which is the exact drift you are trying to eliminate. You have two options: publish that local change so it becomes the hosted version everyone gets, or throw it away.

goodeye skills sync pull --force

--force overwrites local edits with the hosted copy. Use it when you want the machine to match the hosted version and you do not want to keep what is on disk.

The verifiers travel with the skill

A skill is a runbook, and a runbook alone still lets an agent decide it did a good job. If the skill references verifiers, those are checks you wrote that its output has to pass, and they are hosted and versioned the same way the skill is.

That matters here because the binding travels. When you grant a skill to a teammate, the verifiers it references cascade with the grant at the same version, so their agent runs your runbook and your checks rather than an approximation of them. If you have not written a check yet, run a rubric check on AI output from the CLI covers authoring one, binding it to a skill, and pinning it to a version.

Automatic sync, and what it actually does

Adding a sync target switches automatic sync on by itself, at a default interval of one hour. You do not have to turn it on.

It is worth knowing how it runs, because the name suggests a background daemon and it is not one. The automatic pull rides along after a goodeye command finishes, at most once per interval. A machine that never runs a goodeye command never pulls automatically, so on a laptop you have not touched in a week, run goodeye skills sync pull yourself rather than assuming it kept up.

Change the interval, or turn it back on after switching it off:

goodeye skills sync auto on --interval 900
goodeye skills sync auto off
goodeye skills sync auto

The bare command prints the current setting.

Where to start

Publish the one skill you already copy between tools, add the preset for each tool you use, and run goodeye skills sync pull on every machine you work from. Then make your next edit to the hosted skill instead of the file on disk, and confirm it landed with goodeye skills sync status. When you want to hand it to a person or a team rather than just your own machines, share an AI agent skill privately and revoke access picks up from here.