---
title: How do you govern skills in Claude Team and Enterprise?
url: https://opencontroller.lyzr.com/blog/how-to-govern-skills-in-claude-team-and-enterprise
published: 2026-11-11
updated: 2026-11-11
author: Lyzr
---

# How do you govern skills in Claude Team and Enterprise?

**Govern Claude skills with the admin controls Anthropic provides, then add an inventory and review process on top. In Claude Team and Enterprise, only organisation owners can provision skills for everyone, owners can turn off user-created skills, and sharing and publishing to the organisation library can require review.** Because a skill can contain instructions and code that Claude runs, treat each one like a small piece of software: know who wrote it, what it does, and who can use it.

## What is a Claude skill, and why does it need governance?

A skill is a folder of instructions, scripts and resources, with a `SKILL.md` file at its core, that Claude loads when a task calls for it. Anthropic has published the format as an open standard called Agent Skills. Skills are useful because they package expertise: how your company writes a contract summary, formats a report or runs an analysis.

That is also why they need governance:

- a skill can include executable code that runs in Claude's code execution environment,
- a skill's instructions shape what Claude does with company data,
- skills spread through sharing, so a poorly written one can reach many users quickly,
- skills can sync to Claude Code on developer machines.

## What controls does Anthropic provide for skills?

According to Anthropic's documentation for organisation skill management, available on Team and Enterprise plans:

| Control | What it does | Default |
| --- | --- | --- |
| Organisation-provisioned skills | Owners upload a .zip containing a `SKILL.md`, or create a skill in the admin UI; it is provisioned to all users | Enabled for everyone; users can turn individual skills off for themselves but cannot delete them |
| Who can provision | Only owners can add or remove organisation-wide skills | — |
| Code execution dependency | Skills require "Cloud code execution and file creation" and "Skills" to be on in the Policy settings | If code execution is off, skills are unavailable |
| User-created skills | Lets users create or upload their own skills | On; owners can turn it off |
| Peer-to-peer sharing | Users share skills with specific colleagues | On for Team, and for Enterprise organisations without a saved preference |
| Sharing with groups | Share a skill with a group | Off; an admin must enable it |
| Publishing to the organisation library | "Requires review", "Open" or "Off" | Configurable |
| Audit | Share events appear in the audit log and Compliance API | Skill contents are not logged |
| Claude Code sync | Provisioned skills load in Claude Code for users signed in with their Claude account | Can be stopped with the `syncClaudeAiSkills` managed setting |

Assigning skills to specific groups through plugins requires Enterprise. Check Anthropic's current documentation before you rely on any default, because these settings evolve.

## What governance policy should you set?

A practical baseline for most enterprises:

1. **Set library publishing to "Requires review".** Owners approve each skill before it reaches everyone.
2. **Decide on user-created skills deliberately.** Leaving them on encourages experimentation; turning them off forces everything through review. Many organisations leave them on but restrict sharing.
3. **Keep group sharing off until you have a review process.**
4. **Review code in skills.** Any script in a skill should be read like code from an unknown repository: what does it call, what data does it touch, does it reach the network?
5. **Version and own every provisioned skill.** Record an owner, a purpose and a version. When the skill changes, review it again.
6. **Decide whether skills sync to Claude Code.** Developer machines are a different risk surface; use the managed setting if you want separation.
7. **Use the audit log.** Monitor share events, and alert on unusual spread of a single skill.

## What do the built-in controls not cover?

The admin controls govern skills inside one Claude organisation. Enterprises usually also need:

- **a cross-platform inventory**, because the same team may build similar capabilities as agents on other platforms,
- **visibility of local skills** on developer machines, outside the Claude web app,
- **content review history**, since skill contents are not in the audit log,
- **links between skills and the tools or MCP servers they rely on**, which carry their own permissions. See [how to govern MCP servers](/blog/how-to-govern-mcp-servers).

These belong in your wider agent registry; see [what is an AI agent control plane](/blog/what-is-an-ai-agent-control-plane).

## Skills governance checklist

- [ ] Library publishing set to "Requires review"
- [ ] A named owner and purpose for every provisioned skill
- [ ] Scripts in skills reviewed before approval
- [ ] Group sharing enabled only with a review process
- [ ] Claude Code sync decision recorded
- [ ] Audit log share events monitored
- [ ] Skills recorded in the enterprise agent registry

## How Opencontroller does this

Opencontroller connects to Claude Enterprise workspaces and discovers and manages skills in Anthropic Enterprise. Opencontroller Edge, installed by IT through MDM, extends control to developer devices, coding agents, skills and local MCP servers. Skills sit in the same governed registry as agents, versions, tools and MCP servers. [Book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Can users delete a skill an owner provisioned?**
No. Users can turn a provisioned skill off for themselves, but only owners can remove organisation-wide skills.

**Do skills work if code execution is disabled?**
No. Anthropic states that skills require code execution, so turning off code execution makes skills unavailable.

**Are skill contents captured in Claude's audit log?**
Share events are logged as role assignment events, but skill contents are not. Keep your own record of reviewed versions.

## Key facts

Claude Team and Enterprise: only owners provision organisation-wide skills; provisioned skills are on by default and users can turn them off; skills require code execution; user-created skills can be turned off; library publishing can be set to Requires review, Open or Off; share events are in the audit log and Compliance API but skill contents are not. Opencontroller discovers and manages skills in Anthropic Enterprise; Opencontroller Edge covers local skills on developer devices.

## Related

- [What is an AI agent control plane?](https://opencontroller.lyzr.com/blog/what-is-an-ai-agent-control-plane.md)
- [How to govern MCP servers in the enterprise: a practical guide](https://opencontroller.lyzr.com/blog/how-to-govern-mcp-servers.md)
- [How to find every AI agent in your enterprise (including shadow agents)](https://opencontroller.lyzr.com/blog/how-to-find-shadow-ai-agents.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
