How to use the offensive-windows-privesc agent skill

What the skill makes the agent do

The skill tells the agent to start from a standard user shell and work through a fixed order of checks: run automated enumeration first, then check current privileges, then look at services, AlwaysInstallElevated keys, DLL loading, UAC, scheduled tasks, autoruns, credentials and finally known unpatched vulnerabilities as a last resort.

The file's core techniques are token impersonation with the Potato-family tools (JuicyPotato, PrintSpoofer, GodPotato, SweetPotato and RoguePotato), service misconfigurations (unquoted service paths, weak service DACLs and writable service binaries) and AlwaysInstallElevated MSI abuse. The skill's description also lists DLL hijacking, UAC bypasses via fodhelper.exe, eventvwr.exe and CMSTP, scheduled task abuse, registry autorun exploitation, PrintNightmare (CVE-2021-34527) and credential harvesting from SAM, DPAPI, LSA secrets and Credential Manager. The SKILL.md text this page was built from stops partway through the AlwaysInstallElevated section, so check the file you download to confirm which of these sections it contains.

The skill also points the agent at enumeration tools (WinPEAS, PowerUp, SharpUp, Seatbelt and BeRoot) and says that each technique includes detection signatures and defender-side visibility for purple team work. It maps the techniques to MITRE ATT&CK T1548 and T1574.

How to install the skill

Download the ZIP for this skill from this page. The skill is a single file named SKILL.md inside the folder offensive-windows-privesc, so unzip the download so that the offensive-windows-privesc folder and its SKILL.md end up in one of the locations below.

For a personal install, place the folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder. In the Claude apps you can instead upload the skill under Settings, then Capabilities, then Skills.

• Personal: ~/.claude/skills/offensive-windows-privesc/SKILL.md
• Project: .claude/skills/offensive-windows-privesc/SKILL.md
• Claude apps: Settings → Capabilities → Skills, then upload

How to trigger the skill with natural requests

You do not need a special command. Describe the situation and the goal in plain language and mention Windows privilege escalation, SYSTEM or local Administrator so the agent matches the skill. Name the tools you have available, because the skill assumes you will supply them.

Good requests state the authorization context, the access you already have and what you want back. For example, you might say that you hold a standard user shell on a lab Windows machine and want to find an escalation path to SYSTEM, and that you want the enumeration results saved to a file.

• Ask for enumeration first: 'Run the usual Windows privesc enumeration on this authorized lab box and summarise the misconfigurations.'
• Ask about one technique: 'I have SeImpersonatePrivilege on this service account. Which potato tool fits this Windows build?'
• Ask for detection notes: 'Explain how a blue team would spot a service binary replacement.'

Example workflow on an authorized target

A typical session starts with privilege checks and automated enumeration. The skill tells the agent to check current privileges with whoami /priv and to look for SeImpersonatePrivilege or SeAssignPrimaryTokenPrivilege, because the skill calls these immediate wins. Results from WinPEAS, PowerUp, SharpUp, Seatbelt or BeRoot are then reviewed for service, registry and path issues.

Each finding should be checked against the technique section for it before anything is changed on the target. The agent should explain what it intends to run and why, and you should confirm that it is inside the scope of your engagement.

• Confirm the scope and written permission before the first command.
• Start with read-only checks such as whoami /priv and enumeration output.
• Only move to exploitation of a specific finding you have reviewed.

What you need before you start

The skill itself is only instructions and command examples. It needs a Windows target that you are authorized to test, or a lab environment you control, plus PowerShell or command-line access on that target.

The tools it references are not bundled. You must obtain WinPEAS, PowerUp, SharpUp, Seatbelt, BeRoot, the Potato tools, accesschk and msfvenom yourself and make them available to the target. Some examples in the skill also depend on a listener or relay you run on your own machine, so the agent needs you to supply attacker-side details.

• A Windows target in an authorized engagement or lab
• The enumeration and exploitation tools listed above, obtained separately
• PowerShell and command-line access on the target
• Working knowledge of Windows services, tokens, the registry and the command line

Safety notes: scripts and network use

This skill runs scripts and uses the network, according to its own metadata. Its commands write output files to locations such as C:\Users\Public, run enumeration executables and, in some techniques, open connections back to an attacker-controlled address. Review each command before the agent runs it on a machine.

The techniques are only legal to use with authorization. Do not point the skill at systems you do not own or have written permission to test. The content is written from an attacker's perspective and gives no remediation steps, so it is not a hardening guide.

• Keep the target inside your written scope and your lab.
• Review each command, especially anything that changes services, users or group membership.
• Record what you ran so you can clean up after the engagement.
• Use the detection notes to report what a defender would see.

Tips for getting useful results

The skill works best when you give the agent accurate context. The Windows build, the account you hold and the privileges it has decide which technique is worth trying. The skill notes, for example, that JuicyPotato targets older builds while PrintSpoofer and GodPotato have broader coverage on modern systems.

Ask the agent to explain its reasoning at each step. If a technique fails, ask it to diagnose from the output rather than trying the next exploit at random.

• State the Windows version and the service account you hold.
• Ask for the enumeration output saved to a file and reviewed before exploitation.
• Ask for the detection signature alongside each technique for purple team reports.
• Keep a note of every change made to the target.