How to use the Abusing DPAPI for Credential Access agent skill

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

An agent skill is a folder of instructions and optional scripts that an AI agent loads when a request matches the skill's description. This skill is built around SKILL.md, which describes the workflow, plus reference files and a Python helper, scripts/agent.py.

DPAPI is the built-in Windows encryption service that protects saved Credential Manager passwords, browser logins and cookies, Wi-Fi keys, KeePass keys and similar secrets. The skill walks the agent through three decryption routes: as the logged-in user, with a user's password or NTLM hash, or with the domain DPAPI backup key. It covers tools named in the source, including SharpDPAPI, SharpChrome, Mimikatz and Impacket's dpapi.py. The helper script finds DPAPI files in an exported user profile and runs those tools against them.

• Maps to MITRE ATT&CK T1555.004, T1555.003 and T1003.
• Covers Credential Manager, Vault, RDP, KeePass, certificate and browser secrets.
• Targets authorized engagements after a foothold, or triage of DPAPI files already pulled from a host.

How do you install the skill?

Download the skill ZIP from this page and unzip it. For a personal setup, place the unzipped folder in ~/.claude/skills/. To scope it to one 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 install: ~/.claude/skills/abusing-dpapi-for-credential-access/
• Project install: .claude/skills/abusing-dpapi-for-credential-access/ inside your project.
• Confirm the folder contains SKILL.md at its top level.

What do you need before using it?

The skill assumes you already have authorized access to a Windows host and some target-user material. The source lists these requirements. Check each one before you start, because the agent cannot supply them.

• A Windows host where the SharpDPAPI and SharpChrome tools are built from source or taken from release binaries. The source mentions msbuild or Visual Studio for building.
• Mimikatz (optional), obtained from its official releases.
• Python 3 and Impacket, installed with pipx install impacket. DonPAPI is optional for Linux triage.
• The target user's SID, plus one of: a session, password, NTLM hash, or a domain DPAPI backup key (.pvk) for Domain Admin work.
• Written authorization and a defined rules of engagement.

How do you trigger the skill with a natural request?

Describe your authorized task in plain language and name the Windows context. Mention the goal, the material you have, and the scope. The agent should match the request to the skill's description, which covers credential access after a foothold and triage of DPAPI blobs.

Good requests state the engagement and the data you hold. Vague requests about getting passwords from a computer are less likely to match and should be refused unless you can show authorization.

• Example: For our authorized red-team engagement, I have a foothold on a Windows workstation and the user's SID. Triage the host for DPAPI-protected Credential Manager and browser secrets.
• Example: I have exported C:\Users\jdoe from an approved disk image. Find the DPAPI master keys and blobs and explain which decryption route fits what I hold.
• Example: We are running a purple-team exercise. Help me check whether our detections fire on DPAPI master-key access.

What safety limits and side effects should you know?

This skill is offensive security tooling. Its legal notice states it is for authorized penetration testing, red-team engagements and education only. Extracting credentials from systems you do not own or lack written authorization to test is illegal in many places. The skill is not meant for personal or employer devices where you lack permission, because it targets saved passwords and session cookies of real users.

The skill runs scripts and uses the network. The source notes that the Domain Admin backup-key route involves MS-BKRP backup-key RPC to the domain controller, and that Protect-folder and LSASS access are detectable. Log every action under your rules of engagement.

• Runs scripts: yes (scripts/agent.py).
• Uses network: yes.
• Use only on hosts and accounts named in your written scope.
• Keep records of commands, targets and timestamps for your report.
• Treat recovered secrets as sensitive and store them according to your engagement terms.

What tips help you get reliable results?

Start with triage to see what DPAPI material exists before choosing a decryption route. Route choice depends on what you hold: a live user context, the user's password or hash, or the domain backup key. Verify each tool is built from an official upstream, since the source says to obtain tools only from official sources.

For purple-team work, pair each action with the detection it should trigger, so the exercise produces useful findings.

• Triage first, then pick a route based on your material.
• Use official upstream builds of SharpDPAPI, SharpChrome and Mimikatz.
• Ask the agent to explain each command before it runs.
• If a tool is missing, ask the agent which requirement is unmet rather than guessing a substitute.