v1.14 · MIT · PowerShell 5.1 and 7

IDWolf

A hygiene audit for Active Directory and Microsoft Entra ID. One read-only collection with the native modules, the analysis runs anywhere else, offline, and the report says what is wrong, what it could be chained into and what could not be judged at all.

186Active Directory checks
92Entra ID checks
53MITRE techniques
43attacks seen in the wild
4compliance frameworks
0network calls in the analysis

Every number on this page is read from the check catalogs in the repository, so it cannot promise a check, a technique or a release the tool does not ship.

See it before you run it

The report below is the real output of IDWolf on a synthetic domain, shipped in the repository as examples\demo_report.html. It is one HTML file with no server and no CDN behind it: open it, read the score, follow an attack path, filter the findings, mail it as it is.

A tour of the IDWolf report, one captioned slide per section

Synthetic data, start to finish: the domain is moosebank.local and the tenant is fictional. No real directory appears on this page.

How it works

1 · Collect

Read-only, native cmdlets

Collect-ADSnapshot.ps1 uses the RSAT ActiveDirectory module and nothing else; Collect-EntraSnapshot.ps1 uses the Microsoft Graph SDK with Global Reader and GET requests only. No offensive tooling, nothing written to the directory, and secrets are masked before the snapshot is saved. A section that fails is recorded as a collection error, never as a clean result.

2 · Analyse

Anywhere, offline

The snapshot is judged on your own machine, with no AD or Graph module in sight: 186 AD checks across 9 categories - Kerberos, privileged access, account hygiene, domain configuration, permissions, AD CS (ESC1-15), DNS, host hardening read from the GPOs, operational health - plus 92 Entra checks. Password quality comes from a separate hashless export, so no password hash is ever handled.

3 · Report

For the people who have to act

One self-contained HTML file: a score and a grade, the attack-path and trust maps drawn as server-side SVG, the Tier 0 members, framework cards, every finding with its fix. Plus a CSV and JSON Lines for a SIEM, an exceptions file for what you accept on purpose, and a -CompareTo diff against the previous run.

Attack paths, not a list of flaws

Tier 0 is resolved once - nested groups, cycles, foreign principals, primary groups - and matched by SID, never by group name, so a localized directory still works. Every finding is then placed against that map, and the report draws how a weakness reaches domain dominance.

Any domain user: the condition is in the snapshotAny user Kerberoast svc_backup: Kerberoastable account in Tier 0svc_backup GenericAll Helpdesk: dangerous ACL on a Tier 0 objectHelpdesk DCSync Tier 0: domain dominanceTier 0 assumed A step the snapshot cannot confirm: shown as unverified, never counted as a pathunverified
the finding is in the snapshot Tier 0 drawn, but not verified

A path is drawn from findings that actually fired. A step the snapshot cannot confirm is labelled unverified instead of being quietly promoted, and an object that is not known to lead to Tier 0 says so on the edge.

The same discipline applies to the 43 attacks of recent years that the report names next to the checks looking for the condition they need - Kerberoasting and AS-REP roasting, noPac, the AD CS escalations, RBCD, shadow credentials, golden gMSA, SID history - and to the optional AI summary, whose chain claims are re-checked locally against the findings before you read them.

Judged against what

Frameworks and techniques

  • NIS2, ISO 27001, DORA, GDPR: every check maps to controls in all 4 frameworks, by theme rather than by hand, so the mapping cannot fall behind the catalog.
  • MITRE ATT&CK: 53 techniques, cited by the checks that look for them.
  • CIS Microsoft 365: the Entra checks follow the benchmark's own wording.
  • Attacks seen in the wild: 43 of them, named on the finding that enables them rather than as a generic threat list.

Check by check: COVERAGE.md.

And what it refuses to claim

A snapshot is a directory's configuration, not a stream of events. A collection gap must never look like an all-clear, so every check the snapshot could not answer is listed as not evaluated, with the reason: a permission missing, a section that failed, a switch not passed, a deep collection not run.

No framework panel is a statement of conformity, the score is a hygiene score and not a risk number, and the findings you accept on purpose go into an exceptions file with an expiry date instead of disappearing.

Design rules

Nothing is written to the directory

The collectors use native read-only cmdlets. Test-IDWolfSafety.ps1 parses the shipped scripts and fails on any write verb, on any external command that is not on its allow-list, on downloads, iex and Base64, and on any URL outside Graph, Gemini and MITRE. The regression suite runs it on the real code and on a deliberately tampered copy.

The same snapshot gives the same findings

Every age is computed against the collection time stored in the snapshot, never against today, and the findings come out in a stable order. The suite analyses a snapshot in two separate processes and compares the CSV, the JSON Lines and the AI prompt byte for byte, on PowerShell 7 and on Windows PowerShell 5.1.

The AI step is opt-in and tokenized

The deterministic report is on disk before anything is sent. Then every user, host, group and domain name is replaced by a token, and the outbound prompt is re-scanned for the original identifiers: if one survived, the request is aborted. The answer is de-tokenized locally, and anything it cites that did not fire is dropped.

Nothing updates itself

The checks, the attack list and the framework mappings live in the shipped catalogs, and the documents are generated from them. The launcher's only network call is one version check against the GitHub tag list - a notice, never a download - and -NoUpdateCheck turns it off.

Run it in three lines

PS> .\Collect-ADSnapshot.ps1   # on a domain-joined machine, read-only
PS> .\IDWolf.ps1 -Snapshot .\adsnapshot_contoso.json   # anywhere, offline
PS> .\Start-IDWolf.ps1   # or just the menu, if you prefer prompts

No installer, no module manifest, no build step: the scripts dot-source what they need. A scheduled-task wrapper collects, analyses, diffs against the previous run and appends JSON Lines to one file for your SIEM with no prompts - and it never registers the task itself.

Who wrote it

IDWolf is written and maintained by Gabriele Piccini. It is tested on a real throw-away Active Directory lab, not only on synthetic data: a domain populated with Vulnerable-AD and with the project's own scripts, which plant ACL chains, delegation and the rest, so the checks are judged against misconfigurations that behave like the ones in production.