Projects, skills, plugins and connectors in Claude

A project holds knowledge and instructions for one area, a skill is a procedure Claude loads when needed, a plugin is a bundle of skills, connectors and commands, and a connector links one of your systems. The playbook belongs in the project's knowledge and the procedure in a skill. Here are the four concepts as Anthropic's own pages describe them, a rule for what goes where, and what stays with you if you change tools.

Updated by Nils-Erik Jansson

Four concepts

Concept What it is What goes there What comes with you if you switch
Projects A workspace with its own knowledge (files and text you upload) and its own instructions, which apply only to chats in the project. Its own memory. On Team and Enterprise, projects can be shared with permission to use or to edit [1]. For a legal department, one contract type or one area of work Your documents (templates, playbooks, examples of reviewed contracts, the AI policy) as knowledge, one project per contract type or area. Your rules on tone, language, what must never be proposed and when to suggest escalation, as instructions The documents are your files and sit in your own templates outside the tool from day one. The instructions are text that can be moved into another tool, rewritten
Skills Folders of instructions, scripts and resources that Claude loads when they are relevant. Anthropic supplies built-in ones and users can create their own. Procedural knowledge, as opposed to connectors, which give access to data [2]. A skill can contain or install third-party software, so know who wrote it [2] Your procedures, the things done the same way every time whatever the project: the order of review, the structure of a contract summary, the format of the watch list. Several projects can use the same skill Text you wrote, which can be moved into another tool with equivalent features, rewritten
Plugins Bundles that can contain skills, connectors, commands and subagents, available in chat, in Cowork and in Claude Code under the same account [3]. Anthropic's legal plugins are such bundles, supplied without any national law and without your rules Anthropic's legal structure, where it fits, with your projects and skills on top Does not come with you; the plugins are Anthropic's
Connectors Links to other systems: contract archive, document management, e-signature, email. On Team and Enterprise an Owner adds the connector for the organisation and members connect with their own accounts [4]. The permissions in the connected system and your instructions govern what the tool may do Your systems, with a documented data path for each system in the connector register and permissions that match the user's own Configuration that has to be redone

All four are available on Team and Enterprise according to Anthropic's pricing page [5].

What goes where

The third column of the table is the rule for what goes where. What never goes anywhere: the matters on the list of matters kept out of the tool; see the checklist.

Why it matters

Three reasons. Data protection: what sits in a project's knowledge is covered by the project's retention period on Enterprise, and projects are kept indefinitely by default; see the insight on retention in Claude Enterprise. Quality: a playbook in the project's knowledge is read in every review, while a playbook pasted into a chat is forgotten when the chat ends. Maintenance: when the playbook changes, it should change in one place, and everyone needs to know where that place is.

What is yours if you switch

The fourth column of the table shows what comes with you. What does not come with you is whatever exists only in chats. That is why the rule in the setup is that no governing content lives only in a chat.

The structure of projects, skills and connectors is set up in steps three and four of the setup of Claude for Legal, and the connector register with the data path for each system is part of the deliverable. How to write the playbook is covered in the guide on the contract review playbook. Which matter types should be covered by skills of your own is decided in the needs interview.

Frequently asked questions

Where should the playbook live, in a project or in a skill?
The playbook as a document lives in the project's knowledge, so that it is read in every review within the project and can be changed by whoever owns it. The procedure for the review, in what order and with what output, lives in a skill, so that it can be used in several projects and carried over to Cowork. Content in the project, procedure in the skill. That is the rule of thumb.
Do we need Anthropic's legal plugins if we have our own skills?
Not necessarily. The plugins are Anthropic's bundles of skills, connectors and commands for different areas of law, written without reference to any national law. They give you a structure to start from. A department with its own skills for its contract types can do without them, or use them only for the parts that lack a procedure of their own. That is decided in the needs interview.
Who can add connectors?
On Team and Enterprise an Owner adds the connector for the organisation, and members then connect with their own accounts to the systems they already have access to [4]. So the administrator controls which systems can be connected at all, and the permissions in the connected system decide what each user sees. That is why the connector register, with the data path for each system, is part of the data protection documentation.

Statements about Anthropic's terms and products were read on Anthropic's pages on 10 November 2026 and are checked at the start of every engagement.

Sources

  1. [1]What are projects?, Anthropic Help Center
  2. [2]What are skills?, Anthropic Help Center
  3. [3]Use plugins in Claude, Anthropic Help Center
  4. [4]Connectors directory, Anthropic
  5. [5]Pricing, Anthropic

Back to Claude for Legal

More under Claude for Legal