Use the same AI skill in Claude, ChatGPT, and Cursor (and Claude Code, Codex, OpenCode)

You write proposals in Claude, a teammate drafts in ChatGPT, and the developer works in Cursor. You'd like all three to follow the same proposal checklist, not three versions of it. The good news: every one of these tools reads the same skill file. The catch: each keeps it in a different place.

The vendors' docs quoted here were read on November 4, 2026. OpenAI's help center refused automated reads that day, so its lines come from the pages as search engines show them.

One file format everywhere

A skill is a folder with a SKILL.md in it: a short block of front matter, then the instructions.

---
name: proposal-check
description: Checks a proposal before it goes to the client. Use when someone asks to review, check, or finish a proposal or quote.
---
# Proposal check
1. Find the client's problem in their own words. If it's missing, say so.
2. Check scope, price, and the next step against the brief.
3. List what to fix, most important first.

The Agent Skills specification sets the rules: name is "Max 64 characters. Lowercase letters, numbers, and hyphens only", and description is "Max 1024 characters" and "Describes what the skill does and when to use it." Cursor's docs: "Skills work across any agent that supports the Agent Skills standard." Anthropic's say "Claude Code skills follow the Agent Skills open standard."

Where each tool looks

So in a repository, one copy in .agents/skills/ reaches Codex, Cursor, and OpenCode. Claude Code wants its own .claude/skills/. The two chat apps want an upload or a paste. For the coding tools side by side, see Codex vs Claude Code skills.

Put one skill in every tool

1. Write the skill once as above. Keep the folder in a shared drive or a repository.
2. Coding tools: commit it to .agents/skills/proposal-check/ and to .claude/skills/proposal-check/ (or link one folder to the other).
3. Claude: zip the folder and upload it under Customize → Skills. Skills need code execution turned on.
4. ChatGPT: on Business or Enterprise, add it as a skill and Share it. On Plus or Pro, each person adds it in the ChatGPT desktop app, or pastes the instructions into a Project.
5. On every change, repeat steps 2 to 4 for each person and each tool.

Step 5 is where it breaks. A fix lands in the repository, but the copy someone uploaded to Claude last month stays as it was.

Or connect every tool to one copy

All six tools take a remote MCP server, on the plans in the table: a link they connect to and read from.

Fluxenic's AI Skill Pod serves a team's skills from one such link: the same link goes into Claude (any plan with a custom connector, Free's one included), ChatGPT (Pro, Business, or Enterprise), Claude Code, Codex, Cursor, and OpenCode, and an edit reaches everyone's next chat. Its guide also lists each tool's folders for the copy-the-file way. Fluxenic is the company that runs Convika.

One limit to know: AI Skill Pod's own help says Claude chat and Desktop "don't load skills served over MCP into the Skills list". Claude reaches them through the connector instead.

Frequently asked questions

Can the same SKILL.md work in Claude and ChatGPT?

Yes, the file format is the same. Claude takes it as a ZIP upload under Customize → Skills. ChatGPT runs it as a standalone skill in the ChatGPT desktop app, and inside a plugin on the web; the web Skills tab, with Share, is on Business, Enterprise, Healthcare, and Edu. Or paste its instructions into a Project. Or connect both to a remote MCP server that holds it.

Does Cursor read Claude Code's skills folder?

Cursor's docs list .agents/skills/ and .cursor/skills/ in a project, and ~/.agents/skills/ and ~/.cursor/skills/ for you. Put the skill in .agents/skills/ to reach Cursor, Codex, and OpenCode with one copy, and keep a copy or a link in .claude/skills/ for Claude Code.

Where does Codex look for skills?

In .agents/skills in the folder you start in, its parents up to the repository root, and ~/.agents/skills for you, per OpenAI's Codex docs.

How do I keep the copies in sync?

Keep one folder in a repository and link the tools' folders to it, for the coding tools. For Claude and ChatGPT, which take uploads and pastes, the reliable way is a remote MCP server every tool connects to, so there is one copy.

Sources

• Agent Skills specification (read November 4, 2026)
• Claude Code docs: Extend Claude with skills (read November 4, 2026)
• Claude Help Center: Using skills in Claude and custom connectors using remote MCP (read November 4, 2026)
• OpenAI's docs: Build skills (where skills run; read November 4, 2026)
• OpenAI's docs: Pricing (Skills, Plugins, and MCP by plan; read November 4, 2026)
• OpenAI for developers: Connect and test your plugin (the Add custom MCP server steps; read November 4, 2026)
• Cursor docs: Agent Skills and MCP (read November 4, 2026)
• OpenCode docs: Skills (read November 4, 2026)