Inter-office memo
- To
- Everyone who uses Claude Code beside another agent
- From
- Mo, Chief Executive
- Date
- November 19, 2026, 10:41 PM CST
- Re
- Claude Code skills in Codex
How to use Claude Code skills in Codex and other agents in 2026: 6 ways compared
Codex and Antigravity read skills from .agents/skills, and their docs don’t list Claude Code’s .claude/skills. Keep each skill once in .agents/skills and link it into .claude/skills, and all eight agents we checked find it. Use /import to move once, skillshare to sync many agents, or Enjoy to hand any skill to any of its six agents. Cursor, Copilot, OpenCode, Muse Code and Grok Build already read .claude/skills.
Disclosure: Enjoy is our product, and it can hand one agent’s skill to another. We checked every way below against its maker’s own pages on November 19, 2026, and we say where another way is the better pick, including the free ones that need nothing but a folder.
Where each agent looks for skills
A skill is the same thing in every agent now: a folder with a SKILL.md file of instructions, which the agent loads when a task calls for it. Claude Code’s docs say its skills “follow the Agent Skills open standard, which works across multiple AI tools,” and Codex’s say its skills “build on the open agent skills standard.” The standard asks for two fields: a name of up to 64 lowercase letters, numbers and hyphens that matches its folder, and a description of up to 1,024 characters.
What differs is the folder. Claude Code reads .claude/skills. Codex and Antigravity read .agents/skills, and their skills pages don’t mention Claude’s folder. That’s why a skill you made in Claude Code doesn’t show up in Codex. The other five agents here read both. So the short answer is to keep each skill once in .agents/skills and link it into .claude/skills.
| Agent | Reads .claude/skills, checked November 19, 2026 |
Reads .agents/skills |
Its own folders | A limit in its docs |
|---|---|---|---|---|
| Claude Code | Yes, its own | Not on its skills page | .claude/skills, ~/.claude/skills |
A skill’s description is “truncated at 1,536 characters in the skill listing” |
| Codex | Not on its skills page | Yes, up to the repository root | .agents/skills, ~/.agents/skills |
Two skills with one name: “Codex doesn’t merge them” |
| Antigravity | Not on its skills page | Yes, up to the repository root | .agents/skills, ~/.gemini/config/skills |
Its CLI reads two folders “which the Antigravity App doesn’t read” |
| Cursor | Yes, for compatibility | Yes | .cursor/skills, ~/.cursor/skills |
“Only ~/.cursor/skills/ syncs” to its cloud agents |
| GitHub Copilot | Yes | Yes | .github/skills, ~/.copilot/skills |
“Supported locations depend on the surface” |
| OpenCode | Yes, for compatibility | Yes | .opencode/skills, ~/.config/opencode/skills |
Skills without a description “are not advertised” |
| Muse Code | Yes | Yes, its own | .agents/skills, $XDG_CONFIG_HOME/muse/skills |
Project skills “load only after you trust” the workspace |
| Grok Build | Yes | Yes, “at each tier” | .grok/skills, ~/.grok/skills |
It “inlines at most the first 25,000 tokens of a skill body” |
The rows come from Claude Code’s, Codex’s, Antigravity’s, Cursor’s, GitHub Copilot’s, OpenCode’s, Muse Code’s and Grok Build’s own pages. And these are the six ways to keep one set of skills for all of them:
| Way | Best for | Price, checked November 19, 2026 | Reaches |
|---|---|---|---|
1. One copy in .agents/skills, linked into .claude/skills |
Projects you use with more than one agent | Free | All eight agents above |
| 2. A one-time import | Moving from one agent to another | Free | Claude Code to Codex, or back |
| 3. skillshare | Your personal skills, in every agent | Free and open source | 79 targets, Muse Code not among them |
| 4. Vercel’s skills CLI | Installing a published skill into several agents | Free and open source | 79 agents, Muse Code not among them |
| 5. Rulesync | Teams that keep rules, MCP servers and skills in one source | Free and open source | Dozens of tools, Muse Code included |
| 6. Enjoy | Several labs’ agents in one app | A monthly subscription, free for 7 days (pricing) | The six agents Enjoy runs |
Free means nothing beyond the agents’ own plans, which their makers bill for separately.
Why a skill doesn’t carry over on its own
Three things catch people out:
- The folder. Codex “scans
.agents/skillsin every directory from your current working directory up to the repository root,” per OpenAI’s docs, and Antigravity “reads.agents/skills/in each parent folder up to the repository root.” Neither page mentions.claude/skills. Claude Code’s page, for its part, lists only.claude/skillsfolders. - Claude-only fields. Claude Code “extends the standard with additional features like invocation control, subagent execution, and dynamic context injection.” In the compatibility table of Vercel’s skills CLI,
context: fork, which runs a skill in a subagent, works only in Claude Code. Even Anthropic’s other products take only the standard’s fields: “Outside Claude Code, you can use only the fields in the Agent Skills spec.” - Your personal skills live in different homes.
~/.claude/skillsfor Claude Code,~/.agents/skillsfor Codex,~/.gemini/config/skillsfor Antigravity. More on those below.
1. One copy in .agents/skills, linked into .claude/skills
Best for: projects you use with more than one agent, and teams whose members use different ones.
Move each skill into .agents/skills, the folder seven of the eight agents read, and put a link to it in .claude/skills for Claude Code:
mkdir -p .agents/skills
git mv .claude/skills/review .agents/skills/review
ln -s ../../.agents/skills/review .claude/skills/review
Claude Code’s docs say “a <skill-name> entry in the enterprise, personal, or project location can be a symlink to a directory elsewhere on disk,” and that it “loads the skill once even if several locations point at the same target.” The commands are ours, with review standing in for your skill’s folder. We checked that the link resolves, not how each agent then lists it.
Write the skill to the standard: a name that matches its folder, and a description that says what it does and when to use it. Codex picks skills by it, “when your task matches the skill description,” and most of the others do the same.
Price: free.
Watch for:
- Windows. A committed link needs Git’s
core.symlinkson. Anthropic warns of a linkedCLAUDE.mdthat “Git checks a committed symlink out as a plain text file unlesscore.symlinksis enabled,” and the same goes for a linked skill. - Agents that read both folders see it twice. Cursor, Copilot, OpenCode, Muse Code and Grok Build find the skill in
.agents/skillsand again through the link. Grok Build “deduplicates skills by name,” and OpenCode registers its folders in a fixed order of precedence; Cursor’s, Copilot’s and Muse Code’s pages don’t say what they do with two skills of one name. - Linking the other way is less certain. Codex “supports symlinked skill folders and follows the symlink target,” so you could keep the files in
.claude/skillsand link them into.agents/skillsinstead. Antigravity’s skills page doesn’t say whether it follows links.
2. Import once, when you’re moving
Best for: leaving one agent for another, not running both.
Both labs will copy the other’s skills. OpenAI’s import guide says “The desktop app can import from Claude Code, Claude Cowork, or Cursor. Codex CLI can import from Claude Code or Cursor.” Its table brings Claude Code’s skills over as skills, and its slash commands as skills too. In the CLI, “Type /import in an idle Codex CLI session.” And “Importing doesn’t change or delete your existing agent setup.”
Going the other way, Claude Code’s /import takes codex, gemini or cursor, and brings “configuration from OpenAI Codex, Google Gemini CLI, or Cursor on your machine into Claude Code, including instruction files, MCP servers, commands, subagents, and skills.” “Add --dry-run to preview without writing anything.”
Price: free.
Watch for: a copy is a second skill, out of date from your next edit. Only the ChatGPT desktop app keeps it current: “open Settings > Import and turn on automatic updates to keep imported work in sync with the original agent.” OpenAI’s page describes no such switch for the CLI. Claude Code’s /import is “Not available on Amazon Bedrock, Google Cloud’s Agent Platform, Microsoft Foundry, or Claude Platform on AWS, or through a Claude apps gateway.”
3. skillshare
Best for: one person with a folder of personal skills who wants them in every agent.
skillshare says it will “Sync skills, agents, MCP, plugins to all AI CLI tools with one command and simplify team sharing.” skillshare init creates its source folder and finds your agents, and skillshare sync puts your skills in each one: “Source → every target, as per-skill symlinks.” Its targets page says it “provides 79 built-in targets,” Claude Code, Codex, Antigravity, Cursor, Copilot, OpenCode and Grok among them, each with its own personal and project folder.
Price: free. “skillshare is free and open source,” under the MIT license.
Watch for: Muse Code isn’t one of its targets, though Muse Code reads ~/.agents/skills and ~/.claude/skills anyway. On Windows it uses “NTFS Junctions for folders (no admin required); file symlinks need Developer Mode, otherwise copied.” And it’s one more tool to run after you add a skill.
4. Vercel’s skills CLI
Best for: installing someone else’s published skill into all your agents in one go.
Vercel’s skills is “The CLI for the open agent skills ecosystem.” npx skills add vercel-labs/agent-skills installs from a repository, and “The CLI detects installed agents and lets you choose where skills are installed.” “Skills are installed to the current project by default; add -g to make them available across projects.” It “Supports OpenCode, Claude Code, Codex, Cursor, and 75 more,” Copilot, Antigravity and Grok Build among them, and writes each agent’s own folder.
Price: free, under the MIT license. It “Requires Node.js 22.20 or later.”
Watch for: “The CLI collects anonymous usage data. Set DISABLE_TELEMETRY=1 or DO_NOT_TRACK=1 to disable telemetry and security-audit requests.” Muse Code isn’t in its list of agents. Its sync command is an experiment: “Commands prefixed with experimental_ are previews.” It’s built for installing skills, not for keeping your own in step.
5. Rulesync
Best for: teams that already keep rules, MCP servers and hooks in one source and want skills there too.
Rulesync is “A Node.js CLI tool that automatically generates configuration files for various AI development tools from unified AI rule files.” rulesync generate --targets "*" --features "*" writes every tool’s files, and its --features list takes “rules, commands, subagents, skills, mcp, hooks, permissions, checks.” Its tools page lists Claude Code, Codex CLI, Antigravity, Cursor, GitHub Copilot, OpenCode, Grok CLI and Meta Muse Code. For tools without skills, it can “Generate simulated skills for tools that do not support them natively.”
Price: free, under the MIT license.
Watch for: it writes copies. “A symbolic link is never Rulesync’s, since the writer only ever creates real files,” so you edit the source and run the tool, or leave it running with --watch, which will “regenerate whenever rulesync source files change.” Forget the run, and the agents read old skills. For skills alone, the link in way 1 does the same with nothing to run.
6. Enjoy
Best for: people who run several labs’ agents in one app and want any of them to use a skill without moving it, Antigravity included.
Enjoy is a desktop app for Mac, Windows and Linux that runs Claude Code, Codex, Grok Build, Antigravity, Muse Code, and local models through OpenCode, on the plans you already have. Type / in a message, and its picker lists the skills of Claude Code, Codex and Muse Code, each under its agent’s name, from the folders those agents read.
- An agent’s own skills run as usual. “Choosing one of the conversation’s own agent’s skills runs it the way typing its name in the agent’s own app would.”
- Another agent’s skill goes with your message. “Choosing another agent’s skill sends its instructions along with your message, the way a workflow’s are, so any agent can follow it, Antigravity, Grok, and OpenCode included.” It arrives “with the folder they’re in so it can find files they mention.”
- Nothing to copy or link. “Enjoy reads the skill’s file when you send, so the agent gets its instructions as they are then.” The skills stay where their agents keep them, so they still work in a terminal.
- Edit and reuse them. On the desktop, select Open to edit a skill’s file, and a workflow can run skills too.
Price: Enjoy is one monthly subscription, and the first seven days are free, with a card to start. It includes your phone and any browser and one computer; each other computer costs extra each month, and each teammate you invite needs a subscription of their own. See pricing.
Watch for: a lent skill travels with the one message you choose it for. “A message runs one skill,” and choosing another replaces the first. For an agent to pick a skill by itself, it still has to be in a folder that agent reads, so way 1 is still worth doing. Enjoy lists only Claude Code’s, Codex’s and Muse Code’s skills, and of Muse Code’s, “only its skills in the project’s .agents/skills folder can go to another agent.” The skills built into Claude Code aren’t listed. Enjoy isn’t open source, it’s a monthly subscription on top of your AI plans once the free week ends, and scheduled workflows run only while your computer is awake with Enjoy open.
Your own skills, for every project
Everything above is about a project’s skills. Your personal ones, for every project, have a home in each agent, and ~/.agents/skills is the closest to a shared one:
- Claude Code:
~/.claude/skills - Codex:
~/.agents/skills - Antigravity:
~/.gemini/config/skills; its CLI also reads~/.gemini/antigravity-cli/skills/and~/.gemini/skills/ - Cursor:
~/.agents/skillsor~/.cursor/skills, and~/.claude/skillsand~/.codex/skills“For compatibility” - GitHub Copilot: “
~/.copilot/skillsor~/.agents/skills” - OpenCode:
~/.config/opencode/skills, and~/.claude/skillsand~/.agents/skillsfor compatibility - Muse Code:
~/.agents/skills, and it “also discovers~/.claude/skillsand$CODEX_HOME/skillsby default” - Grok Build:
~/.grok/skillsand~/.agents/skills, and~/.claude/skillsunless you turn off its Claude compatibility
So keep them in ~/.agents/skills, and link each one into ~/.claude/skills for Claude Code and ~/.gemini/config/skills for Antigravity, or let skillshare do the linking.
When to keep .claude/skills as it is
If Claude Code is your only agent, nothing needs to change. .claude/skills is where it looks, and moving your skills gains you nothing until a second agent arrives.
Your other agents may not need anything either. Cursor, GitHub Copilot, OpenCode, Muse Code and Grok Build all read .claude/skills already. Only Codex and Antigravity need the skill in .agents/skills, by a link, a copy or an import.
Keep a skill in .claude/skills alone, too, when it depends on what only Claude Code does: running in a subagent with context: fork, hooks of its own, which Vercel’s table marks as working in Claude Code but not in Codex, or “Claude Code-only body features, such as dynamic context injection,” which Anthropic says “don’t function in claude.ai chat or through the API.” Another agent would read the instructions without the parts that make them work.
How we compared
We read the Agent Skills specification; Anthropic’s Claude Code pages on skills, commands and memory, and its changelog; OpenAI’s Codex pages on building skills and importing; Google’s Antigravity skills page; Cursor’s, GitHub’s and OpenCode’s skills pages; Meta’s Muse Code pages on extending and configuration; xAI’s Grok Build skills page and the user guide in its repository; and the READMEs and docs of skillshare, Vercel’s skills CLI and Rulesync, all on November 19, 2026, and every one is linked below. The commands in way 1 are ours. Where a skills page doesn’t mention a folder, we say that, not that the agent ignores it. We quote each maker’s own words for its limits. These agents ship weekly, so we recheck this guide against its sources at least once a month, and the date at the foot of the page says when we last did.
12.Questions.
Can Codex use Claude Code skills?
- Yes, the format is the same, but not from the .claude/skills folder. OpenAI’s docs say Codex “scans .agents/skills in every directory from your current working directory up to the repository root,” and its skills page doesn’t mention Claude’s folder. Move the skill to .agents/skills and link it back, import it with /import, or use a sync tool.
Does Claude Code read .agents/skills?
- Not by its docs. Claude Code’s skills page lists only .claude/skills folders, plus plugins and the skills on your claude.ai account. It does say a skill’s entry there “can be a symlink to a directory elsewhere on disk,” so you can link each skill to where it lives.
Which agents read Claude Code’s .claude/skills folder?
- Cursor, GitHub Copilot, OpenCode, Muse Code and Grok Build, besides Claude Code itself. Codex’s and Antigravity’s skills pages don’t mention it. Both read .agents/skills, which all of these except Claude Code read too.
Are Claude Code skills and Codex skills the same format?
- Mostly. Both build on the open Agent Skills standard, a folder with a SKILL.md file that has a name and a description. Claude Code “extends the standard” with fields such as context, which runs a skill in a subagent, and Codex adds an optional agents/openai.yaml file. Claude Code “ignores a field it doesn’t recognize,” and in Vercel’s compatibility table, context: fork works only in Claude Code.
Does Antigravity support skills?
- Yes. Google’s docs call skills “an open standard for extending agent capabilities.” Antigravity reads them from .agents/skills in your project and ~/.gemini/config/skills for every project, uses them on its own when they fit, and runs one when you type a slash and its name.
How do I use one skill in every agent at once?
- Keep it in .agents/skills and link it into .claude/skills, which reaches all eight agents here. For your personal skills across many agents, skillshare links one folder into each agent’s, and Vercel’s skills CLI installs a skill for the agents you pick. In Enjoy, you can also choose one agent’s skill for a message to any of its six agents.