---
name: SharePoint AI Skills Governance Audit
description: Inventories AI Skill files across SharePoint sites, audits authoring permissions on the Agent Assets library, checks sensitivity label compliance on skill Markdown files, and maps governance controls across tenant, site, and user tiers.
---

# SharePoint AI Skills Governance Audit

> **TL;DR:** This skill inventories all AI Skill `.md` files stored in SharePoint Agent Assets libraries across the tenant, audits who can author and use skills at each governance tier, and confirms that sensitivity label controls apply to skill content.

## What does the SharePoint AI Skills governance audit cover?

SharePoint AI Skills reached public preview in April 2026. A Skill is a Markdown (`.md`) file stored in a SharePoint site's **Agent Assets** library that teaches the SharePoint AI what business logic to apply — content understanding rules, list-update actions, summarisation instructions. Because skills are authored as Markdown files within SharePoint, they are governed by existing SharePoint permissions rather than a separate Microsoft 365 admin centre toggle. This creates a governance gap: any user with Edit permissions on the Agent Assets library can author a skill, and skills can instruct AI actions on content that includes OFFICIAL:Sensitive or Highly Confidential files. This skill inventories skills across the tenant, maps the three governance tiers (tenant admin / site admin / site user), audits authoring permissions, and checks whether sensitivity labels apply to skill files.

## When should you run this skill?

- "Inventory all SharePoint AI Skills across our tenant"
- "Audit who can create or modify AI Skills on SharePoint sites"
- "Check whether skills are applied to sites containing OFFICIAL:Sensitive content"
- "Review SharePoint AI Skills for our IRAP evidence package"
- "Identify sites where AI Skills are deployed without site owner awareness"
- "Assess our SharePoint AI governance posture before enabling Skills tenantwide"

## How this skill works, step by step

1. Confirm scope: all SharePoint sites in the tenant, or a specific site collection
2. Open SharePoint admin centre → Active sites → filter for sites with AI enabled
3. For tenant-level governance, verify whether SharePoint AI is enabled or restricted via the SharePoint admin centre AI settings panel
4. Run a SharePoint search query to locate Agent Assets libraries across the tenant: use the Graph search endpoint with `contentType:SkillFile` or iterate sites via `Get-SPOSite -Limit All`
5. For each site that has an Agent Assets library: enumerate `.md` files in the library; record filename, last modified date, and last modifier
6. Query the Agent Assets library permissions for each site: identify all principals (users, groups, Entra groups) with Edit or Full Control — these are potential skill authors
7. Flag sites where Any authenticated user or Everyone Except External Users has Edit access to Agent Assets
8. Check whether sensitivity labels are applied to skill `.md` files (labels applied to files in the Agent Assets library govern who can read and edit skill content)
9. Identify sites where the Agent Assets library inherits default permissions from the site (no custom governance applied)
10. Cross-reference each site against the organisation's sensitivity classification: flag sites with Confidential, Highly Confidential, or OFFICIAL:Sensitive content where skill authoring is not restricted to site owners
11. Assess the three governance tiers and produce the output below

## Three governance tiers for SharePoint AI Skills

| Tier | Control owner | What they control | Default state |
| --- | --- | --- | --- |
| Tenant admin | SharePoint administrator | Enable or restrict AI features tenant-wide; block AI Skills in highly regulated site collections | AI enabled by default; no separate Skills toggle |
| Site admin / site owner | SharePoint site owner | Break permission inheritance on Agent Assets library; restrict Edit to named authors; set sensitivity labels on skill files | Inherits site permissions — any site editor can author |
| Site user | Individual contributor | Create, edit, and test skills within their Edit permission boundary | Can author if they have site Edit permissions |

## Output format

### Agent Assets library inventory

| Site URL | Site sensitivity tier | Skill files | Last author | Agent Assets Edit principals | Governance finding |
| --- | --- | --- | --- | --- | --- |
| /sites/ExecComms | Highly Confidential | 3 | Jane Smith | Site Members (47 users) | High — broad Edit on sensitive site |
| /sites/HR-Policies | OFFICIAL:Sensitive | 1 | IT Admin | Site Owners only | Pass |

### Governance summary

- Sites with Agent Assets library: N
- Sites with unrestricted skill authoring (Edit open to Members): N
- Sites with Confidential or above content + unrestricted skill authoring: N (Critical)
- Skill `.md` files without sensitivity label applied: N
- Sites with tenant-admin AI restrictions in place: N
- Recommended actions: restrict Agent Assets Edit to Site Owners on all Confidential and above sites; apply sensitivity labels to skill `.md` files; establish a skill review process for regulated site collections

## Scope and safety

Read-only — this skill does NOT:

- Create, modify, or delete AI Skill files
- Change permissions on Agent Assets libraries
- Enable or disable AI features on any site
- Alter sensitivity label assignments

## Licensing and permissions

### Licences and add-ons

| Capability used | Minimum licence |
| --- | --- |
| SharePoint AI Skills (Agent Assets library) | Microsoft 365 Copilot licence per site that uses skills, or SharePoint Premium |
| Sensitivity labels on SharePoint files | Microsoft 365 E3 + Microsoft Purview Information Protection |
| Graph API site enumeration | Microsoft 365 E3 or SharePoint Online Plan 2 |

### Least-privilege roles

- SharePoint Administrator (enumerate all sites, view Agent Assets library structure)
- Compliance Administrator (read sensitivity label assignments on files)

### Microsoft Graph permissions (read-only)

- `Sites.Read.All` — enumerate SharePoint sites and their document libraries including Agent Assets
- `Files.Read.All` — read skill `.md` file metadata and sensitivity label assignments
- `InformationProtectionPolicy.Read.All` — resolve sensitivity label names from label GUIDs

## Sources and compliance

- [Extend AI in SharePoint with skills (Microsoft Learn)](https://learn.microsoft.com/en-us/sharepoint/ai-in-sharepoint-skills)
- [AI Skills Are Now in Public Preview (Microsoft Tech Community)](https://techcommunity.microsoft.com/blog/spblog/ai-skills-are-now-in-public-preview-teaching-ai-in-sharepoint-what-to-know-and-h/4512532)
- [Introducing new agentic building in SharePoint (Microsoft Tech Community)](https://techcommunity.microsoft.com/blog/spblog/introducing-new-agentic-building-in-sharepoint-and-more-updates/4497987)
- Aligns with PSPF Requirement 4 — access to and processing of OFFICIAL:Sensitive content must be controlled
- Supports Essential Eight ML2 Control 7 (user application hardening — AI feature governance)
- Pair with SharePoint Advanced Management Site Governance for comprehensive SharePoint AI posture
- Output in Australian English