---
title: What is agent skills governance?
url: https://opencontroller.lyzr.com/blog/what-is-agent-skills-governance
published: 2026-11-11
updated: 2026-11-11
author: Lyzr
---

# What is agent skills governance?

**Agent skills governance is the set of controls over the packaged instructions, scripts and resources that AI agents load to perform specialised tasks. It covers six lifecycle stages: authoring, review, versioning, distribution, revocation and usage telemetry. Because skills can include executable code and change what an agent does without a model or prompt change, they need the same review as any other software an agent runs.**

Skills have become a common way to extend agents. Instead of writing a long system prompt, teams package know-how, such as "how we format board reports" or "how to file an expense", into a reusable folder an agent loads when it needs it.

## What is an agent skill?

In Anthropic's [Agent Skills](https://docs.claude.com/en/docs/agents-and-tools/agent-skills/overview) format, a skill is a folder containing a `SKILL.md` file. The file starts with YAML frontmatter that includes a `name` and a `description`, followed by instructions. The folder can also contain scripts, templates and reference files.

Agents load skills progressively. At first, only the name and description are visible to the agent. The full instructions and files are read only when the agent decides the skill is relevant. That keeps context small, but it also means a skill's description decides when it activates.

Skills can be installed in several places: shared by an organisation's admins, uploaded by individual users, or sitting on a developer's laptop for a coding agent.

## Why do skills need governance?

- **They can run code.** A skill can include scripts the agent executes. That is a software supply chain.
- **They change behaviour silently.** Editing a skill changes how every agent that uses it behaves, with no model or prompt change to review.
- **They spread by copy.** A useful skill gets copied between teams and laptops, and versions diverge.
- **Descriptions can mislead.** A badly written or malicious description can make an agent load a skill in the wrong situation.
- **They can embed data.** Reference files may contain internal data that should not travel with the skill.

These map to risks in the [OWASP Top 10 for Agentic Applications](https://genai.owasp.org/), including supply chain vulnerabilities and tool misuse.

## What does good skills governance look like?

| Lifecycle stage | Requirement | What good looks like |
|---|---|---|
| Authoring | Every skill has an owner and a stated purpose | Owner resolved in your identity provider; purpose in the description |
| Review | Scripts and instructions reviewed before shared use | Code review for scripts; check for secrets, external calls and data files |
| Versioning | Skills are versioned and changes are tracked | Skills stored in a repository; version recorded wherever installed |
| Distribution | Approved skills come from one catalogue | Organisation-provisioned skills from the catalogue; user uploads limited or reviewed |
| Revocation | A skill can be withdrawn everywhere quickly | One action removes or disables it across workspaces and devices |
| Usage telemetry | You know which agents and users invoke which skills | Usage counts per skill, per version, per team |

## How do you start governing skills?

1. **Inventory what exists.** List skills in your enterprise AI workspaces and on developer machines. Expect duplicates.
2. **Assign owners.** Any skill without an owner after two weeks is disabled.
3. **Review scripts first.** Prioritise skills that include executable code or make network calls.
4. **Publish a catalogue.** One approved source, with versions.
5. **Set an upload policy.** Decide whether users can upload their own skills, and if so, which ones need review.
6. **Measure usage.** Retire skills nobody uses; invest in the ones many teams rely on.

Skills sit next to MCP servers as the two main ways agents gain new abilities. Govern them together: see the pillar [how to govern MCP servers](/blog/how-to-govern-mcp-servers) and [how to review MCP servers agents already use](/blog/how-to-review-mcp-servers-agents-already-use). For a platform walk-through, read [how to govern skills in Claude Team and Enterprise](/blog/how-to-govern-skills-in-claude-team-and-enterprise).

## How Opencontroller does this

Opencontroller, by Lyzr, discovers and manages skills in Anthropic Enterprise workspaces and keeps them in a governed registry alongside agents, versions, tools and MCP servers. Opencontroller Edge, installed by IT through MDM, extends control to developer devices, coding agents, skills and local MCP servers, which is where many unreviewed skills live. Policies written as code can apply across the estate. See the [capability reference](/capability-reference) or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Is an agent skill the same as an MCP server?**
No. A skill packages instructions, scripts and files the agent reads and runs. An MCP server is a separate service that exposes tools and data over the Model Context Protocol.

**Should users be allowed to upload their own skills?**
For personal, instruction-only skills, often yes with monitoring. Skills that include scripts or are shared with others should go through review first.

**How do you revoke a skill that has been copied to laptops?**
You need device-level visibility, such as an endpoint agent managed through MDM, plus a catalogue that records which version is approved.

## Key facts

Agent skills governance is the set of controls over packaged instructions, scripts and resources that AI agents load for specialised tasks, across six lifecycle stages: authoring, review, versioning, distribution, revocation and usage telemetry. In Anthropic Agent Skills, a skill is a folder with a SKILL.md file whose YAML frontmatter includes name and description; agents load only the name and description first and read full content when relevant. Skills can include executable scripts, so they are a software supply chain risk. Opencontroller by Lyzr discovers and manages skills in Anthropic Enterprise and keeps them in a governed registry; Opencontroller Edge, installed by IT via MDM, extends control to skills and local MCP servers on developer devices. Pricing starts from $250K per deployment per year.

## Related

- [How do you govern skills in Claude Team and Enterprise?](https://opencontroller.lyzr.com/blog/how-to-govern-skills-in-claude-team-and-enterprise.md)
- [How do you review MCP servers your agents already use?](https://opencontroller.lyzr.com/blog/how-to-review-mcp-servers-agents-already-use.md)
- [What should an AI agent registration questionnaire ask?](https://opencontroller.lyzr.com/blog/ai-agent-registration-questionnaire.md)
- [More on Tools and MCP](https://opencontroller.lyzr.com/blog/cluster/tools-and-mcp)
