Claude Cowork skills, plugins and connectors, and how open source Kortix owns them
In the open-source Kortix AI Operating System, a Claude Cowork skill becomes a markdown file and a config line in a git repo you own. Inside Cowork itself, skills and plugins live in your Claude account and follow the account across surfaces.
What changes when skills move into a repo
Cowork and OpenWork cells come from each vendor's own pages; the Kortix column quotes Kortix.
| Product | Open source | Where skills live | How you share them with a team | Versioning and rollback | Connectors and credential handling | Review before it goes live |
|---|---|---|---|---|---|---|
| Kortix | Yes. Open source. Read it, fork it, audit it. | skills/<name>/SKILL.md in the project git repo. | One git repo the team clones; grants in kortix.yaml. | Git history: diff any change, roll back any part. | Credentials are brokered server-side and never enter the machine. | Session work reaches main through a change request. |
| Claude Cowork | No. | In your Claude account as installed plugins. | An admin turns on sharing; share to people or groups. | Recipients pick up your latest version at next use. | Connectors call through Anthropic's cloud, not your local network. | Permissions can require approval before significant actions. |
| OpenWork | Yes. The desktop app is open source. | SKILL.md folders in the workspace's .opencode/skills/. | Share skills and plugins from OpenWork Cloud or a self-hosted control plane. | Commit the skill folders to a shared repository to version them. | Admins add MCP servers; members never see server URLs or tokens. | Runs in folders you authorize; edits land in your local files. |
What a skill or plugin is in Claude Cowork
A skill in Claude Cowork is a folder of instructions that Claude loads when a task needs it. A plugin bundles skills, connectors and sub-agents into one install. Anthropic's help center says you add one from the Customize menu: open the Plugins tab, then the Discover tab, then Add; in Cowork you open the Cowork tab first. Plugins need a paid plan, and the skills a plugin carries work in chat and in Cowork alike (Use plugins in Claude).
Where that configuration lives
A plugin you add in chat or in the desktop app is saved to your Claude account rather than to your computer, so it follows you into Cowork and into Claude Code when you sign in with the same account. On Team and Enterprise plans an owner can switch on skill and plugin sharing, and then you share a plugin you uploaded with specific people or with a group; each recipient gets a view-only copy and enables it themselves. An update you make later reaches them at their next use, and you can stop sharing at any time (Team and Enterprise plans).
What a team wants from that layer
When a skill encodes how a company does one job, the team that owns the job usually wants the skill in a repository, reviewed with the rest of its configuration and with a change it can undo. Cowork keeps that layer in the product and with the account. A connector in Cowork also reaches external services through Anthropic's cloud, not through your local network, as Anthropic documents on the Cowork product page.
How open-source Kortix represents the same layer
Kortix is the open-source AI Operating System: your agents, their skills, company memory, and connectors in one git repo you own. An agent is a markdown file under agents/ for its behavior plus an agents.<name> block in kortix.yaml for what it may reach, and a skill is a folder with a SKILL.md inside it. Connectors and triggers are declared in the same manifest, so the whole operating layer is text a person can read:
kortix_version: 2default_agent: supportagents: support: file: agents/support.md connectors: [helpdesk] secrets: [HELPDESK_API_KEY] skills: [kortix-cli, triage] kortix_permissions: [project.session.read, project.review.submit]Session work reaches main through a change request, and merge is default-deny for agents, so a change to a skill or a connector is a diff a person reads before the team gets it. The Kortix docs carry the full manifest reference, and Kortix on GitHub is open to read and fork. A team that wants the layer on its own network can follow the installation guide and the self-hosting guide. The comparison page sets the options side by side. If you are weighing the two products, Read the comparison.
The same layer as commands
$ kortix connectors apps gmail$ kortix connectors add gmail --provider composio --app gmail --apply$ kortix connectors connect gmail$ kortix connectors syncFrom the Kortix connectors documentation.
Questions about Cowork skills
Can you share a Claude Cowork skill with your team?
Yes, on Team and Enterprise plans. An owner turns on skill and plugin sharing, then you share a plugin you created with specific people or a group, and each recipient gets a view-only copy they enable themselves.
Where are Claude Cowork plugins stored?
Plugins you add are saved to your Claude account, not to your computer, so they are available in chat, Cowork and Claude Code when you sign in with the same account.
Can you version-control agent configuration?
In Claude Cowork that configuration lives in the product and the account. In the open-source Kortix AI Operating System, agents, skills and connectors are files in a git repo, so you diff a change and roll it back.
What are Kortix skills?
A Kortix skill is a folder with a SKILL.md file that encodes how a company does one job, and an agent's skills grant in kortix.yaml decides which skills it may load.
Run the agent layer in a repo you own.
Start with one skill and grow from there.
Open source · Any model, your keys · Self-host, VPC, or on-prem