How to use the Auditing Entra ID with AADInternals agent skill

What is an agent skill and what does this one do?

An agent skill is a folder containing a SKILL.md file and supporting files that an AI agent loads when a request matches the skill's description. This skill's SKILL.md describes an assessment workflow built around AADInternals cmdlets.

Once installed, the agent can run unauthenticated reconnaissance of a domain's tenant and federation settings, optionally test whether usernames exist, acquire tokens for Microsoft APIs, enumerate directory objects with a valid token, and check whether federation signing certificates would allow Golden SAML forgery. It then helps document findings and hardening recommendations. The skill is built for authorized red-team and defensive validation work, not general use.

Install the auditing-entra-id-with-aadinternals skill

Download the skill ZIP from this page and unzip it. For a personal install, place the unzipped skill folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder.

Alternatively, in the Claude apps go to Settings → Capabilities → Skills and upload the ZIP there.

The skill also needs AADInternals itself. Install it from the PowerShell Gallery on a Windows host:

How do you trigger the skill with natural requests?

You do not need a special command. Describe the task in plain language and include the domain and scope, for example: ask the agent to run outsider reconnaissance on a domain you are authorized to test, or to check whether the AD FS token-signing certificate is protected against Golden SAML. Mention that the work is authorized and under written rules of engagement so the agent treats it as an assessment.

Name the stage you want. The workflow has distinct stages (recon, user enumeration, token acquisition, directory enumeration, federation inspection, backdoor testing), and asking for one stage at a time keeps the evidence clear.

• Run unauthenticated recon on target.com and summarise the federation type and SSO status.
• Find the tenant GUID for target.com and list verified domains.
• Using my cached Graph token, enumerate users and flag synced accounts.
• Check whether the AD FS signing certificate on adfs.target.com is protected against Golden SAML and list hardening steps.

What the workflow covers, step by step

The SKILL.md workflow runs in order. Step 1 is outsider reconnaissance using Invoke-AADIntReconAsOutsider, Get-AADIntLoginInformation and Get-AADIntTenantID, which need no credentials. Step 2 is optional external user enumeration with Invoke-AADIntUserEnumerationAsOutsider, which the source describes as noisy.

Step 3 acquires tokens with Get-AADIntAccessTokenForAADGraph, Get-AADIntAccessTokenForMSGraph and Get-AADIntAccessTokenForEXO, where -SaveToCache lets later cmdlets reuse the token. Step 4 uses that token for directory reads such as Get-AADIntTenantDetails and Get-AADIntUsers.

Step 5 inspects federation and AD FS configuration, and Step 6 covers the federation backdoor and Golden SAML test. Step 6 is only for controlled tenants with explicit authorization and needs a Global Administrator token.

What do you need before running it?

Requirements from the skill: a Windows host with PowerShell 5.1+ (or PowerShell 7 on the supported subset), the AADInternals module from the PowerShell Gallery, Python 3 for the bundled scripts/agent.py helper, and written authorization or rules of engagement for the target tenant.

For backdoor and federation tests you also need Global Administrator or an equivalent account in the target tenant. The skill assumes familiarity with SAML, OAuth tokens and Azure AD Connect, so it is not a starting point for beginners.

• Windows host with PowerShell 5.1+ or PowerShell 7 (supported subset)
• AADInternals module from the PowerShell Gallery
• Written authorization / rules of engagement for the target tenant
• Global Administrator or equivalent for backdoor/federation tests
• Python 3 for scripts/agent.py

Safety notes: scripts and network access

This skill runs scripts and uses the network. The bundled scripts/agent.py helper is Python 3, and the AADInternals cmdlets query Microsoft endpoints, including public ones such as getuserrealm and OpenID configuration. Read the scripts and understand what they send before you run them, especially on a machine that holds production credentials.

The toolkit can forge SAML tokens and install persistent federation backdoors that impersonate any tenant user. Use it only against tenants you own or have explicit written authorization to test. Unauthorized use violates the Computer Fraud and Abuse Act and equivalent laws. External user enumeration is noisy and will likely appear in logs, so plan for it in your rules of engagement.

Tips for getting useful output

Keep the scope explicit in every request, including the domain names and which stages are allowed. Ask the agent to save evidence and the recommendations it produces, since the skill is designed to document exposure and hardening steps.

If you are a blue team, the skill can also help you produce realistic AADInternals telemetry to build detections. Pair it with a Graph-log hunting skill if you have one, as the source suggests.

• State the authorization and scope in each request
• Run one stage at a time and review output before moving on
• Ask for findings and hardening recommendations, not only raw output
• Consult the references/api-reference.md file for cmdlet parameters