Claude adds Cowork plugins to turn AI into specialized teammates
Claude just rolled out plugin support for Cowork, letting teams package skills, connectors, slash commands, and sub-agents to tailor Claude for specific jobs like sales and legal. Anthropic also open-sourced 11 internal plugins to get people started. The feature is now live as a research preview for paid Claude users, with files currently stored locally.

If you have used Claude Code, you already know the magic of handing an AI a complex goal and watching it execute across local files and terminal commands. Now, Anthropic is bringing that exact workflow to regular office work.
Anthropic just rolled out plugin support for Cowork, its collaborative workspace tool designed to make Claude Code accessible to non-engineers. The new feature allows users to bundle skills, data connectors, custom slash commands, and specialized sub-agents into reusable packages.
The update is available now as a research preview for all paid Claude subscribers.
Custom agents in a single folder
The biggest friction with using AI at work is context switching. Every time you open a prompt, you have to remind the model what data to check, what tone to take, and what steps to follow.
Cowork plugins tackle this by turning Claude into a domain expert for specific roles like legal, sales, or finance. A sales plugin, for instance, can connect straight to your CRM and internal knowledge base, arming Claude with the playbook needed to research leads and draft follow-ups automatically.
Everything in these plugins is file-based. That means your skills, connectors, and sub-agents live as plain files that can be edited with simple prompts or modified directly using a built-in plugin creator.
To seed the ecosystem, Anthropic open-sourced 11 ready-to-use plugins on GitHub that its own team uses daily. The lineup covers enterprise search, finance modeling, legal doc review, marketing, product management, and even biology research.
Right now, plugins are strictly saved locally to your machine. Anthropic noted that team-wide sharing tools and private enterprise marketplaces will arrive in the coming weeks.
Why it matters
We think Anthropic is making a brilliant architectural bet by keeping everything file-based and modular.
Instead of building a rigid enterprise software suite, Anthropic is essentially giving users a system prompt on steroids that can spin up specialized sub-agents. Because the components are just files, creating an entirely new workflow requires almost no engineering overhead. You can literally ask Claude to edit the plugin for you.
For paid users, this is definitely worth trying right now if your job involves repeatable, multi-step tasks like drafting deal memos, reviewing contracts, or parsing product specs.
However, the lack of cloud syncing and admin controls right now shows this is very early. The real test will come when teams start sharing plugins org-wide. Once you plug sub-agents into sensitive CRMs and corporate data, governance and security guardrails will quickly become Anthropic's biggest challenge.
Sources (1)
- claude.com(opens in a new tab)Originalclaude.com