How to use the Active Directory offensive testing agent skill

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

An agent skill is a folder containing a SKILL.md file that gives an AI agent specialised instructions for a task. When the skill is installed and relevant, the agent reads these instructions and follows the workflow they describe. This skill is a single file, SKILL.md, with no bundled scripts.

The skill is an Active Directory attack methodology. Its source covers reconnaissance with tools such as BloodHound, PowerView and ADExplorer; credential abuse such as Kerberoasting, ASREProasting, NTLM relay and LLMNR/NBT-NS poisoning; privilege escalation through ACL abuse, GPO abuse and Kerberos delegation; lateral movement; persistence; forest trust attacks; ADCS certificate abuse; and evasion of Microsoft Defender for Identity. It names the commands and tools the agent should use for each stage.

• Recon: map AD structure, preferably offline from a snapshot, to limit live queries
• Credentials: harvest hashes or tickets where the engagement allows it
• Escalation: map paths toward Domain Admin, Enterprise Admin or Tier 0
• Persistence and documentation: record every action with timestamps

How do you install the skill?

Download the ZIP file from the skill page on this site. Unzip it so that the skill folder sits inside your personal skills directory, ~/.claude/skills/, or inside a project's .claude/skills/ folder if you want it available only to that project. Confirm that SKILL.md is inside the skill folder after unzipping.

If you use the Claude apps instead of a local setup, you can upload the skill under Settings, then Capabilities, then Skills. Either way, the skill is loaded when the agent decides it is relevant to your request.

• Personal install: ~/.claude/skills/ (available across your projects)
• Project install: .claude/skills/ inside the project folder (available only there)
• Claude apps: Settings, then Capabilities, then Skills, then upload

What tools must you have before you start?

The skill does not include the tools it refers to. The source names SharpHound, BloodHound, PowerView, Rubeus, Impacket, Responder, NetExec, Certipy, hashcat, nanodump and pypykatz as tools the instructions reference. You need to obtain and run those yourself, on Windows and Linux systems suited to security testing.

You also need written authorization and an in-scope target network before any of this applies. The skill assumes deep Windows and Active Directory knowledge, so the agent's output is only as useful as your ability to check it.

• Windows and Linux security tooling, installed separately
• Written authorization for the engagement
• A defined, in-scope target network
• Lab environment if you are learning rather than testing a live client

How do you trigger the skill with natural requests?

You do not need special syntax. Describe the authorized task in plain language and name the Active Directory area you care about. The skill's description says it applies when assessing on-prem AD, hybrid AD with Entra ID, or ADCS deployments, so mention those environments explicitly.

Good requests state the scope, the environment and the stage of work. Vague requests such as 'hack this network' do not match the skill's purpose and should be avoided.

• Ask for a plan: 'We have written authorization for an internal assessment of corp.local. Outline an AD attack path review starting from BloodHound data.'
• Ask about a stage: 'Explain how to assess ADCS templates for misconfigurations in our lab forest.'
• Ask for detection help: 'What attacker techniques against AD should our blue team write detections for?'

What safety limits should you respect?

The source is explicit that some actions can disrupt or compromise production systems. It names LSASS dumping, DCSync and GPO changes as examples. Run these only in environments where you have authorization and where you can safely test. Prefer a lab or a clearly agreed test window.

The skill favours low-noise collection, such as offline analysis of an ADExplorer snapshot and throttled or DC-only BloodHound collection. Use those options where they fit the engagement. The skill's own workflow tells the operator to validate at each hop and document every action with timestamps.

The skill's metadata states that it does not run scripts and does not use the network. Any network activity or script execution comes from the tools you run, not from the skill file itself.

• Do not use it against networks you are not authorized to test
• Do not run credential dumping, DCSync or GPO changes on production without agreed rules
• Keep a timestamped log of every action
• Stop and report if you hit something outside scope

What does a good session look like?

A typical session starts with a question about scope and reconnaissance, then moves through the stages in the skill's workflow. The agent proposes a path, explains the risk of each step, and asks you to confirm before anything intrusive. Because the skill lists concrete commands, check every command against the tool's documentation and your engagement rules before running it.

Treat the agent as a reference and planning aid. You remain responsible for judging whether a step is safe and in scope.

• Confirm scope in the first message
• Ask the agent to explain each step before running it
• Verify commands against the tool's own documentation
• Record results and timestamps as you go