GitHub - ssteelfactor-oss/Kestrel

BloodHound shows you the path. Kestrel shows you what's already inside.

Everyone maps the same thing: the attack path - who can reach Domain Admin. It's the right question, and the tools that answer it are excellent. But a graph of "who can reach whom" is blind to an entire class of problem, because some of the worst things in an Active Directory aren't a path at all.

They're a backdoor welded into an attribute. A certificate that keeps authenticating long after the password is reset. A permission stamped into the schema itself, so every object created from now on inherits it silently. A deleted account someone quietly kept the right to bring back. An object hidden from enumeration by a single deny-read ACE.

None of that is an edge in a graph. All of it is sitting in your directory right now. Kestrel reads it - with native Windows calls, ordinary domain-user rights, no server, no Python, no agent, and a footprint on the domain controller so restrained it's documented event-by-event.

One .exe. One command. Everything below.

You've been looking the wrong way

The offensive AD ecosystem - SharpHound, impacket, Certipy, and the graph behind them - is superb, and it all looks at the same surface: reachability. Kestrel looks at the three dimensions that surface can't express.

• Time. A graph is a snapshot. Kestrel reads replication metadata - when a Tier-0 attribute changed and which DC originated it. A DCSync right granted last night is not the same finding as one that's been there for years, and only one of them means you're being attacked right now.
• Persistence in an attribute, not an edge. Shadow credentials, SID history, reanimate-tombstones, schema defaultSecurityDescriptor backdoors, hidden objects, OWNER RIGHTS deny-ACEs, AD FS DKM keys. The graph tools don't collect these because there's no edge to draw.
• Defensive posture, inverted. Not "who can attack," but "what isn't protected" - the Tier-0 account outside its silo, the sync account with a random RID that nobody tagged, the lockout policy that never throttles a spray.

It's not that Kestrel is faster than a graph. It answers questions the graph can't ask.

Every AD sin, in one native exe

Twenty-plus checks that are usually spread across a dozen scripts and two languages. Here they're one binary - and they end in a single, severity-sorted verdict.

One line

Kestrel.exe --all --report audit.html

Run everything. Get the full HTML report. And, at the very end of the console, the part that matters:

═══ Kestrel - Prioritized Findings ═══

  CRITICAL  DCSync         CORP\svc_sql - non-default principal can replicate directory changes (DCSync)
                          → fix: remove GetChanges/GetChangesAll from the principal on the domain head (dsacls) unless it is a DC
  CRITICAL  AD FS          CN=... - read access to the DKM key (Golden SAML precondition)
                          → fix: remove non-default read ACEs on the DKM object (dsacls) and rotate the token-signing certificate
  HIGH      Persistence    CN=... - object hidden from enumeration via a broad deny-read ACE
                          → fix: inspect the hidden object and remove the broad deny-read ACE (dsacls)

  [=] 2 critical · 1 high · 0 medium · 0 low

Not a wall of output. The findings that matter, ranked, each with the command to close it.

One exe. Red, blue, purple.

• Red / grey - enumerate persistence and delegation the graph misses, export straight to BloodHound CE OpenGraph (--opengraph) to fuse with your existing collection, or pivot from any principal (--from).
• Blue - a prioritized, remediated findings list; a --diff against last week's snapshot to catch what changed; and a tool whose exact detection footprint is published so you can whitelist it and tune your own alerting around it.
• Purple - one artifact both sides read the same way. Red finds it, blue fixes it, everyone points at the same line number.

Detectable by design

Kestrel is read-only. It never writes to the directory, never replicates, never needs SeSecurityPrivilege, never touches the cloud. Reading who can DCSync looks nothing, on the wire, like doing DCSync - no DRSUAPI, no replication-signature 4662. Every query it runs and every event it can (and cannot) generate is catalogued in FOOTPRINT.md. An auditor that tells the defender exactly how to catch it isn't a contradiction - it's the whole point.

Quick start

Grab a build. Every green CI run publishes a Kestrel.exe artifact - download it from the Actions tab, no toolchain required.

Or build it yourself. Visual Studio (v143+) or MSBuild:

git clone https://github.com/ssteelfactor-oss/Kestrel.git
cd Kestrel
msbuild Kestrel.vcxproj /p:Configuration=Release /p:Platform=x64

Pure C, zero third-party dependencies. Build /MT and it's a single self-contained executable - drop it on a domain-joined host, run as any ordinary user.

Run it.

Kestrel.exe --all --report audit.html      # everything, HTML report + ranked findings
Kestrel.exe --acl --adcs --adfs            # pick specific checks
Kestrel.exe --paths --from CORP\jdoe       # attack paths from one principal
Kestrel.exe --all --diff last-week.json    # what changed since the last snapshot
Kestrel.exe --all --opengraph graph.json   # export to BloodHound CE OpenGraph

Modules

Run --all, or select any subset. Full help: Kestrel.exe --help.

Output & interop: --report <file> (.html / .json / .yaml) · --opengraph <file> (BloodHound CE) · --diff <snapshot.json> · --verbose.

Design invariants

These are promises, not preferences - the reasons it's safe to run in production. Breaking any of them is a security bug (see SECURITY.md).

• Read-only - never creates, modifies, or deletes a directory object. Only read/access events, never modification events.
• Ordinary domain user - no privilege escalation, no SeSecurityPrivilege, DACL-only security masks.
• No replication - detects who can DCSync; never calls DRSUAPI itself.
• Directory & SYSVOL only - no RPC/SMB against member hosts.
• No evasion - no query fragmentation, no timing games, no log tampering. See FOOTPRINT.md.
• On-prem only - no Entra / Graph / cloud calls; it audits the footprint hybrid leaves in AD.
• Native & self-contained - pure C, zero third-party dependencies, single .exe.

Roadmap

Documentation

• FOOTPRINT.md - the exact LDAP / SYSVOL trace each scan leaves, and the DC events it does (and can't) generate.
• SECURITY.md - reporting, scope, and the design invariants as verifiable guarantees.
• LICENSE - Apache 2.0.

Related

Parent project: NetEnum - AD enumeration via ADSI / COM / LDAP.

Author

@ssteelfactor-oss - security research and COM / Windows internals.

License

Licensed under the Apache License 2.0 - see LICENSE. Provided "as is", without warranty; run it only against directories you're authorised to audit.