Claude Projects are workspaces: each one holds its own chat history, its own instructions and a knowledge base of uploaded files for one client, product or recurring piece of work. Claude Skills are reusable instructions packaged as a SKILL.md file and uploaded as a .zip; Claude loads a skill whenever a task matches its description, whichever chat or Cowork session the person is in. The working rule for a team: put what you are working on in a Project and how you always do it in a Skill. A proposal for one client lives in that client's Project; the house method for writing proposals is a Skill that every Project can draw on.
Claude is Anthropic's AI assistant. Projects exist on every plan, including Free; the sharing, provisioning and review controls that make Projects and Skills a team system are on the Claude Team and Enterprise plans, and that is the version this post describes. For the workshop that gets a department using both, see Claude employee training: a practical workshop plan; the paid version is NetSys's Claude training for employees and business teams.
What is the difference between a Claude Project and a Claude Skill?
From Anthropic's Projects article and its skills provisioning article, as of October 2026:
| Claude Project | Claude Skill | |
|---|---|---|
| What it holds | Instructions, uploaded knowledge files and the chat history for one body of work | A SKILL.md with instructions and a description, plus any supporting files, zipped |
| Scope | One client, product, report or initiative | One repeatable procedure, used across many projects and chats |
| How it is invoked | The person opens the project and chats inside it | Claude decides to use it when the task matches the skill's description |
| Where it works | Chat and Cowork; a new version in beta turns a project into one conversation with parallel cloud threads | Chat and Cowork, and provisioned skills also load in Claude Code for signed-in users; skills require code execution |
| Who creates it | Any member (Free accounts are capped at five projects) | Any member by default; Owners can turn off user-created skills, and Enterprise can require a "Create skills" capability |
| Sharing | Team and Enterprise: "Can view" or "Can edit," by email, list or organization-wide | Owners provision organization-wide; members share peer to peer, view-only for recipients; Enterprise can assign to groups through plugins |
| Review | The project owner controls edit access | Library publishing can be set to "Requires review," "Open" or "Off" |
| Migration from a personal account | Projects, instructions and files move | "There's no migration path for skills" |
| Best for | Context: who the client is, what the data says, what has been decided | Method: the checklist, the format, the standards a deliverable must meet |
What belongs in a Project?
Everything that is true for this piece of work and nothing else. Three parts:
- Instructions. Who the audience is, the tone, the constraints, the definitions that matter for this client or product. Keep them to what differs from your company default; the default belongs in a skill.
- Knowledge. The contract, the brief, the style guide for this account, the data extracts, last quarter's report. On paid plans Anthropic says a retrieval mode expands knowledge capacity by up to ten times as the project nears its context limit, so a project can hold more than fits in one conversation.
- History. The chats inside the project, which is where decisions accumulate.
Sensible units: one project per client; one per recurring report, with each month's source files added; one per hiring round; one per software product. The wrong unit is "Marketing," which becomes a pile that nobody can find anything in. What should not go in: credentials, anything the project's members are not cleared to see, and data that must not leave the system it lives in, because a project's files are visible to whoever has access to the project.
What belongs in a Skill?
The procedures your team repeats across projects. A skill is a small package: a SKILL.md whose description tells Claude when the skill applies and whose body says how to do the job, with any templates or reference files beside it. Good candidates:
- The proposal method: sections, the questions each must answer, the pricing language that legal approved.
- A quality checklist for client deliverables, applied before anything is marked done.
- A status update format for project management: what changed, what is blocked, what is decided, what needs a decision.
- A findings-to-tickets procedure for a security or IT team, which takes a tool's output and turns each item into a ticket with owner, severity and the fix, written the way your service desk expects. Interpreting the findings themselves is a separate topic, covered in using Claude to interpret security logs and audit findings.
Two cautions from Anthropic's own article. Skills require code execution, so they work where that is enabled. And the description is what triggers them: a vague description means the skill fires at the wrong moments or not at all, so write it as a precise statement of the task it handles and test it on a known input before you publish it to the library.
People searching for a Claude security review skill usually mean one of two things. Anthropic ships Claude Security, a code-scanning product in public beta for Enterprise, plus a Claude Security plugin for Claude Code; those are Anthropic's products, not something you write. A team's own security review skill is the checklist version above: the standards you apply when reviewing a configuration or a pull request, encoded so every reviewer applies the same list.
Who owns, reviews and shares each?
This is where Team and Enterprise differ from personal accounts, and it is the part a business should set before anyone builds anything.
Projects
On Team and Enterprise a project can be shared with "Can view" (see the contents, knowledge, instructions and chats) or "Can edit" (change instructions and knowledge, manage members), to named people, an email list or the whole organization. Owners can turn project sharing off, per role on Enterprise; turning it off also turns off public projects, and existing shares remain. A reasonable policy: client projects shared with the account team only, internal reference projects shared organization-wide as view-only, and edit rights held by one named owner per project.
Skills
Organization-wide skill management is on Team and Enterprise. Owners provision skills at Organization settings > Plugins & skills, where they can upload or create one; provisioned skills are on by default and members can toggle them off; only owners add or remove organization-wide skills. Members can also publish to the organization library, and the setting that governs that is the one to decide deliberately: "Requires review," "Open" or "Off." Team organizations defaulted to "Open" and Enterprise to "Off," and Anthropic's article says organizations that made no choice moved to "Requires review" on October 2, 2026. Peer sharing is on by default on Team; shared skills are view-only for recipients, update automatically when the author changes them, and disappear when the recipient leaves. On Enterprise, skills can be assigned to groups through plugins, and provisioned skills also load in Claude Code for users signed in with their Claude account unless the organization's Claude Code settings turn that sync off.
Set the library to "Requires review" for anything that touches client data or produces output that leaves the company. Name a reviewer per department. Keep a one-line changelog in the SKILL.md so the next reviewer knows what changed.
How should a business combine Projects and Skills?
Illustrative patterns, not client stories:
- A consulting firm. One project per engagement with the statement of work, the client's documents and the working notes. Three skills shared organization-wide: the proposal method, the deliverable quality checklist and the weekly status format. A new consultant opens the client project and the skills apply themselves.
- A property management company. One project per building with leases, vendor contracts and inspection reports. A lease summary skill that extracts the same twelve fields every time, and a tenant letter skill with the approved tone and the legal footer.
- An IT and security team. One project per tenant review holding the exported tool output and the notes. A findings-to-tickets skill, provisioned by the owner and set to "Requires review," so every reviewer produces tickets in the same shape.
- Project management. One project per initiative with the plan, the decisions log and the meeting notes; a status update skill and a risk register skill shared with the whole PMO.
What are the limits and gotchas?
- Anthropic's pages state no numeric limit on skill count or size, and no file-count or size figure for project knowledge; the practical limit is whether Claude can find the right thing, so prune both.
- Free accounts are capped at five projects; the Projects article gives no cap for paid plans.
- Skills do not migrate when an employee's personal account is merged into the organization; projects do. Have people re-upload skills they wrote, or better, provision the good ones centrally.
- Skills require code execution, and in Anthropic's HIPAA configuration for local Claude Code and Cowork they are turned off; skills are not covered under the BAA.
- Shared skills are view-only for recipients, so improvements have to go back to the author or to the owner who provisioned the skill.
- A project's chat history is visible to everyone with view access.
How does NetSys teach this?
NetSys's Claude training service runs role-based workshops in which a department builds its first two projects and two skills on its own work during the session, agrees the sharing and review policy with the workspace owner, and leaves with a one-page guide for the next person. The workshop sits inside our AI services, alongside workspace setup and ongoing administration, and we train on Claude, ChatGPT business plans and Microsoft Copilot depending on what the company runs.
Frequently asked questions
Are Claude Skills the same as Projects?
No. A project is a workspace for one body of work, with its own files, instructions and chats. A skill is a reusable procedure that Claude applies when a task matches its description. Projects hold context; skills hold method.
Can a skill be used inside a project?
Skills are not attached to a particular project. Anthropic describes them as loading when a task matches the skill's description, and provisioned skills are organization-wide, so a person working inside a client project has the organization's skills available alongside that project's knowledge.
Who can create skills on the Team plan?
By default any member can create their own. Owners can turn off user-created skills on the Policy tab, and only owners can provision or remove organization-wide skills. Publishing to the library can be set to require review.
Do skills move when an employee joins the Team plan from a personal account?
No. Anthropic's migration article says there is no migration path for skills. Projects, with their instructions and files, do move. Re-upload the skills or provision them centrally.
What is a Claude security review skill?
Either Anthropic's Claude Security product (code scanning, in public beta for Enterprise, plus a plugin for Claude Code) or a team's own skill that encodes its review checklist. The second is what most businesses mean, and it is a good first skill for an IT or security team.
Can project knowledge be shared with the whole company?
On Team and Enterprise, yes: a project can be shared organization-wide with view or edit rights, and owners can turn organization-wide sharing off if the company prefers named sharing only.
Does Claude Code use the same skills?
Provisioned organization skills load in Claude Code for users signed in with their Claude account, unless the organization's managed Claude Code settings turn that sync off.
NetSys can plan a workshop in which your team builds its first Claude Projects and Skills on real work and leaves with a sharing policy the owner has agreed. Request a team workshop plan.
Discuss ai services for your business.
Tell us about your current systems, the result you need and your timeline. We will discuss the work, responsibilities and pricing before you decide on an engagement.