Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Um comando direto ignora o prompt de revisão. Verifique a origem antes de executá-lo.
Instruções da origem · Visualização somente leitura
name
identity-provider-recon
description
Organization-grade identity-fabric mapping: tenant/federation fingerprinting and the pre-auth user-ENUMERATION oracle methodology — enumeration and fingerprint only, never credential submission. Covers domain-to-tenant resolution (Microsoft getuserrealm.srf Managed/Federated namespace check, Entra OIDC metadata tenant-GUID extraction, Autodiscover v2), keyless Microsoft tenant-federation mapping (GetFederationInformation SOAP -> sibling-domain discovery, discover-only ROE, FEDERATED_WITH provenance edge held out of attack-path pivoting), Okta org-slug derivation + OIDC fingerprint + governed custom-domain enumeration, ADFS passive/active fingerprint + version inference, Google Workspace MX-correlated detection, generic OIDC (Auth0/Keycloak/Ping Identity/OneLogin/Duo) discovery, SAML metadata (5 paths), Azure AD Seamless-SSO Negotiate-challenge detection, Microsoft Defender for Identity (MDI) sensor-API presence check, the user-enumeration oracle methodology for Microsoft GetCredentialType (IfExistsResult semantics: exists / doesn't-exist / exists-in-federated-tenant / throttled) and Okta /api/v1/authn (errorCode differential), Medium-detectability discipline with a hard 20-candidate-per-tenant cap and admin/role interest-based ranking, and name x confirmed-email-pattern login-candidate synthesis that FAILS CLOSED with zero output when no org pattern is confirmed. Grounded directly in a production ASM implementation's sso_idp.py, tenant_recon.py, and core/email_patterns.py modules. Deepens — does not duplicate — offensive-osint skill's Identity Fabric endpoint reference with the tenant-federation MAP, the oracle WORKFLOW, and the candidate-SYNTHESIS methodology that reference lacks. Use when fingerprinting an organization's identity provider, mapping its tenant/federation boundary, running an authorized pre-auth user-enumeration pass, or synthesizing login candidates from harvested names to feed that oracle — never for password spray, credential submission, or auth bypass.
version
1
sources
asm_reference_impl, public_research
triggers
["identity fabric","identity provider recon","IdP recon","IdP fingerprinting","tenant fingerprinting","tenant recon","SSO discovery","SSO fingerprinting","federation mapping","federation boundary","tenant federation map","domain to tenant resolution","GetFederationInformation","M365 tenant federation","Entra tenant recon","Azure AD tenant recon","Azure AD enum","entra enum","getuserrealm","Managed vs Federated namespace","okta enum","Okta org slug","Okta governed domains","Okta OIE vs Classic","ADFS enum","ADFS version fingerprint","ADFS mex endpoint","Google Workspace enum","SAML metadata discovery","generic OIDC discovery","Auth0 fingerprint","Keycloak fingerprint","[Truncated]"]
Companion skills: osint-methodology (§6.2 detectability tagging, §11
identity-fabric pointer — the "how to think" skill this plugs into) and
offensive-osint (§22 Identity Fabric — the concrete endpoint/payload
reference this skill builds a workflow on top of, rather than re-listing). This skill answers the
question those two don't: how do the tenant, its federation partners, its IdP, and its
user-enumeration oracle fit together as one map — and where exactly enumeration stops and
credential submission begins.
0. When to Use / When NOT
Use this skill when:
You need to resolve a domain to its identity tenant (Entra/Okta/ADFS/Google Workspace) and
determine whether auth is Managed or Federated.
You need to map an org's federation boundary — every sibling domain that shares the same
M365/Azure AD tenant trust, not just the seed.
You need to distinguish Entra vs. Okta vs. ADFS vs. Google Workspace vs. a generic OIDC IdP from
passive/low-detectability signals.
You need to detect Seamless SSO or Microsoft Defender for Identity (MDI) presence — both change
the risk calculus of anything downstream.
You have authorization for a Medium-detectability, log-generating pass and need to run a
pre-auth user-enumeration oracle (GetCredentialType / Okta /api/v1/authn) to build a valid-account
list — without ever submitting a password.
You have harvested employee names and a confirmed org email-format pattern and need to
synthesize ranked login candidates to feed that oracle.
Do NOT use this skill when:
You need to submit credentials, replay breach creds, forge/replay a token, or confirm an
auth-bypass — that is a different, higher-authorization tier. See §14.
The target's authorization for active, logged probing isn't established — the passive half of
this skill (§7–§10) needs the same soft-scope posture as every companion skill; the active half
(§11) needs it explicitly, because it generates tenant-side audit-log events (§1, §11.4).
You already have concrete endpoints and just need the reference table — go straight to
offensive-osint §22.
1. Authorization & Legal Posture
Same base posture as osint-methodology §1: intended for assets the operator owns or has to assess.
written
authorization
This skill carries a sharper posture than most of the pack because §11 is not passive. Every
domain-resolution, federation, IdP-fingerprint, Seamless-SSO, and MDI probe in §7–§10 is Low
detectability (a metadata GET/POST that any browser makes) — but the user-enumeration oracle in §11
is Medium detectability: it is a targeted, per-account POST against a live authentication
endpoint, and both Microsoft and Okta log it in the tenant's own sign-in/audit trail. Treat §7–§10
and §11 as two separate authorization tiers even within one engagement.
Soft scope check — ask once before running §11:
"I can build a valid-account list by probing Microsoft/Okta's pre-auth user-existence oracle. This
never submits a password, but it IS logged in the tenant's own audit trail and is rate-limited on
my end to stay under the radar. Confirm you want me to run this active pass?"
Always-on guardrails specific to this skill:
Never submit a real password, a guessed password, or any credential to any login endpoint. §11's
oracles work by reading the shape of a rejection (§11.2, §11.3), not by trying to succeed.
Cap user-enumeration at 20 candidates per tenant (§11.4) — this is not a suggestion, it is the
production cap in the module this skill is grounded in.
A discover_only federated sibling domain (§8) is a lead, never a target. Confirm ownership before
extending §7–§11 to it.
Nothing in this skill authorizes §14's excluded techniques, no matter how interesting the oracle
results look.
2. Confidence Levels
Same three-tier model as the companion skills, applied to identity-fabric assertions:
Level
Meaning
Identity-fabric example
TENTATIVE
Plausible, unverified.
Okta org-slug guess ({stem}-prod.okta.com) not yet confirmed by a live OIDC response; an 8-permutation passive email guess (core/email_patterns.infer_patterns_for_name) for enrichment only.
Independently corroborated or an oracle differential fired.
Entra tenant GUID extracted from OIDC issuer AND matched by a second probe (getuserrealm/autodiscover); an Okta org's OIDC issuer contains okta.com (self-corroborating — the provider's own metadata endpoint answered); a GetCredentialType/Okta-authn oracle differential fired for a specific address (§11) — direct verification, not inference.
Rule of three still applies to anything you'd report as organizational fact ("this org uses
Entra") beyond a single tenant-level metadata hit — a single OIDC response is enough to say "an Entra
tenant answers for this domain," but attributing that tenant to a specific subsidiary or brand
needs the same corroboration discipline as org-attack-surface §2.
3. Output Format
Standard Finding schema (companion skill §3), with category: SSO_EXPOSURE for every identity-fabric
finding in this skill — that is the category the reference implementation this skill is grounded
in uses uniformly, from an INFO tenant-identified breadcrumb through a LOW confirmed-user-enumeration
result:
UTC timestamps everywhere. For §11 oracle results specifically, the evidence block MUST carry
probed_count, the enumerated address list(s) split by outcome, and — critically — a
credential_submitted: false marker, so a downstream reader (or auditor) can never mistake an
enumeration result for a proven-live credential.
4. Source Hygiene & Citations
Same discipline as the companion skills: URL + UTC timestamp + tool version + run_id on every
artifact.
Record the exact request body sent to GetCredentialType / Okta authn, not just "I probed the
endpoint" — the oracle's entire evidentiary value is in the response differential, which is
worthless without the paired request.
Cache raw SOAP/JSON responses from GetFederationInformation, OIDC metadata, and getuserrealm.srf —
these are the primary evidence for the tenant-federation map and should survive a re-run months
later even if the live tenant configuration has since changed.
Never log a probed password value in any artifact, even the fixed placeholder Okta's authn oracle
requires (§11.3) — redact it in stored evidence as <oracle-placeholder>.
5. Do NOT
Do NOT submit a real, guessed, or breach-sourced password to any login endpoint under this skill.
§11's oracles are read-only-by-design: Microsoft's GetCredentialType and Okta's /api/v1/authn
both reveal account existence from the shape of a rejection, never from a successful login.
Do NOT exceed the 20-candidate-per-tenant cap on user enumeration (§11.4).
Do NOT auto-promote a discover_only federated sibling domain (§8) into active scope. Confirm
ownership first.
Do NOT treat a STRONG-looking Okta-org-slug guess as confirmed until its own OIDC metadata
answers and its issuer field actually contains okta.com (§7.4) — a guessed slug that happens to
resolve to someone else's Okta org is a namesake collision, not a hit.
Do NOT synthesize login candidates from names without a confirmed org email pattern (§12.2) —
return nothing rather than spray 8 unverified permutations per name at a live oracle.
Do NOT run §11 (the active oracle) without the sharper authorization posture in §1.
Do NOT proceed past §11 into password spray, credential replay, or auth-bypass confirmation — that
is explicitly out of scope everywhere in this skill (§14).
6. The Identity-Fabric Map — Mental Model
Most identity recon stops at "what login page does this domain redirect to." That single fact
misses everything this skill is built to surface: the tenant boundary, who else federates into it,
what defensive posture it runs, and — under explicit authorization — which specific accounts on it
are real.
┌─────────────────────────────┐
│ DOMAIN → TENANT │ §7 — getuserrealm.srf, Entra OIDC metadata,
│ (who owns the login?) │ Autodiscover v2, Okta org-slug + OIDC,
└──────────────┬───────────────┘ ADFS fingerprint, Google Workspace MX,
│ generic OIDC, SAML metadata
▼
┌─────────────────────────────┐
│ FEDERATION MAP │ §8 — GetFederationInformation SOAP:
│ (who else shares this trust?)│ every sibling domain in the SAME
└──────────────┬───────────────┘ M365/Azure AD tenant, discover-only
│
▼
┌─────────────────────────────┐
│ IdP POSTURE │ §9 distinguishing signals + §10 —
│ (Managed/Federated, Seamless │ Seamless SSO Negotiate challenge,
│ SSO, MDI monitoring) │ MDI sensor-API presence
└──────────────┬───────────────┘
│ authorization gate — §1 tightens here
▼
┌─────────────────────────────┐
│ CANDIDATE SYNTHESIS │ §12 — harvested names x CONFIRMED org
│ (who do we probe?) │ pattern → ranked login candidates,
└──────────────┬───────────────┘ fail-closed with no pattern
│
▼
┌─────────────────────────────┐
│ USER-ENUMERATION ORACLE │ §11 — GetCredentialType / Okta authn,
│ (which accounts are real?) │ valid-account list OUTPUT, capped
└──────────────┬───────────────┘ at 20/tenant, never a password sent
│
▼ ← THE HARD BOUNDARY (§14) —
[ everything below spray / credential submission / auth bypass
this line is OUT is a different authorization tier entirely
of this skill's and lives in the engagement's authenticated-
scope ] testing phase, not here.
Every stage feeds the next. A tenant you can't resolve (§7) has no federation map to walk (§8). A
federation map with no confirmed tenant SERVICE asset gives §11 nothing to enumerate against — the
production implementation this skill is grounded in gates the oracle specifically on tenants that
were resolved via oidc_metadata, getuserrealm, autodiscover, or okta_oidc (§7.9), not on a
bare domain guess.
7. Domain → Tenant Resolution
Six independent probes, all Low detectability, run concurrently against the same seed domain. Any
one confirms tenant existence; running all of them cross-corroborates and often fills in fields a
single probe leaves blank (e.g. getuserrealm confirms Federated but not the tenant GUID; OIDC
metadata fills that in separately).
7.1 Microsoft — getuserrealm.srf (Managed vs. Federated)
GET https://login.microsoftonline.com/getuserrealm.srf?login={probe-user}@{domain}&xml=1
Needs one plausible username local-part — use a harvested email if you have one, else a synthetic
admin@{domain} seed (this is a Microsoft-side probe; it never touches the client's own infra).
Cheap regex-extract the response fields rather than pulling in a full XML parser:
Response fields: NameSpaceType (Managed / Federated / Unknown — Unknown means Microsoft has
no record of the domain, stop here), FederationBrandName, AuthURL (present only when Federated —
the upstream ADFS/IdP URL), CloudInstanceName.
Why Managed vs. Federated matters for everything downstream:Managed means authentication
terminates directly at login.microsoftonline.com — no on-prem ADFS proxy sits in front of it, so
§11's GetCredentialType oracle and any later authenticated-testing phase hit Entra's own (more
permissive) IP-based throttling with no ADFS Extranet Lockout buffer. Federated routes through
whatever AuthURL points to (usually ADFS — cross-reference §7.5). Severity note: a Managed
namespace on its own is a MEDIUM SSO_EXPOSURE finding (spray/MFA-fatigue hits Entra with no ADFS
buffer), FIRM confidence — it is a posture fact, not a vulnerability, but it changes risk on
everything that follows.
A hit here is CONFIRMED confidence, INFO severity — it's the federation root every downstream SaaS
integration trusts, not a weakness by itself, but the single most valuable pivot anchor in the whole
identity fabric (a leaked credential + this tenant GUID = a pre-auth pivot into every federated
SaaS).
GET https://autodiscover-s.outlook.com/autodiscover/autodiscover.json/v1.0/{email}?Protocol=Autodiscoverv1
Needs a probe email (harvested, or synthetic admin@{domain}). A response containing a Protocol or
Url field confirms the mailbox — and therefore the M365 tenant — exists. FIRM confidence, no
finding by itself (feeds the tenant SERVICE node); this is a corroboration probe, not a standalone
signal.
Autodiscover IP correlation (companion skill offensive-osint §22.1) is a complementary passive
check when MX is masked by an inbound filter (Mimecast/Proofpoint/Barracuda) — resolve
autodiscover.{domain} and check whether it lands in Exchange Online IP space. Use it alongside this
probe, not instead of it.
Okta has no domain-keyed discovery endpoint the way Microsoft does — you derive candidate org slugs
and confirm each by fetching its OIDC config.
Candidate derivation (from the target's own domain stem + any subdomain that is literally an
*.okta.com host already in your graph):
candidates = { stem, stem-prod, stem-dev, stemcorp, stem-corp } ∪ { okta-slug matches in known subdomains }
[a-z0-9][a-z0-9-]{1,40}\.okta\.com
Confirmation probe (Low detectability — Okta's own public metadata endpoint):
GET https://{slug}.okta.com/.well-known/openid-configuration
Only treat a slug as CONFIRMED when issuer actually contains okta.com — a slug guess that
resolves to a different company's Okta org (namesake collision) is common with short/generic
stems; don't attribute it without this check (§5). Distinguish Classic vs. OIE (Okta Identity
Engine) by whether the issuer contains /oauth2/ — this affects which downstream endpoints and
policies apply.
Governed custom domains (call ONLY after the org slug is confirmed real — never on a guess):
GET https://{slug}.okta.com/api/v1/domains
Returns every custom domain the org has verified as an identity-provider surface
({"domain": "...", "validationStatus": "VERIFIED"}). This is the Okta equivalent of §8's federation
map — it reveals which branded login portals (login.acmecorp.com, sso.subsidiary.com) all route
into the same Okta tenant, closing the "co-federation via a non-Microsoft IdP" gap that §8's
Microsoft-only federation probe structurally cannot see. Each verified domain is a genuine tenant-map
node, not a guess — emit it as its own domain asset governed by this Okta org.
7.5 ADFS — passive fingerprint + version + mex (active)
Passive fingerprint:
GET https://adfs.{domain}/adfs/ls/idpinitiatedsignon.aspx
A response body is version-fingerprintable from resource-reference strings: adfs/portal/css/style.css2012 → ADFS 2012 R2 (3.0); adfs/portal/illustration/illustration.png → ADFS 2016/2019 (4.x/5.x);
bare federation service string with neither marker → ADFS confirmed, version unknown. Severity:
2012 R2/3.0 → MEDIUM (weak-crypto defaults, historical golden-SAML entry point); newer/unknown →
INFO.
Mex endpoint (active — this one logs on the target side, unlike the fingerprint GET above):
GET https://adfs.{domain}/adfs/services/trust/mex
A response containing wsdl:definitions confirms the mex endpoint is live and leaks federation
metadata (endpoint URLs, signing certs, supported claim types). Gate this probe on the passive
fingerprint already having confirmed ADFS — don't fire it speculatively.
7.6 Google Workspace — MX-correlated detection
Google Workspace has no per-tenant OIDC endpoint — accounts.google.com's
/.well-known/openid-configuration is provider-wide, not tenant-specific, so it cannot confirm this
specific domain is a Workspace customer on its own. Detection therefore hinges on MX:
dig +short MX target.example
Resolve-DnsName -Name target.example -Type MX
Any MX host containing aspmx.l.google.com or googlemail.com confirms Google Workspace. Emit the
tenant node with tenant_id = domain (there is no separate tenant GUID to extract) and
issuer = https://accounts.google.com for consistency with the other tenant nodes, but be explicit
in the evidence block that the corroborating signal is MX, not a tenant-scoped metadata response.
Probe /.well-known/openid-configuration on the target apex and on subdomains matching the common
IdP-prefix set (auth, login, sso, idp, iam, identity, accounts, oauth) — both
discovered subdomains that match and the fixed prefix list applied directly to the root domain, capped
around 40 candidate hosts. Skip any hit already attributed to Microsoft/Okta/Google (issuer contains
microsoftonline, okta.com, or accounts.google.com) — those are covered by the dedicated probes
above.
Product
Signal in issuer
Auth0
auth0.com
Keycloak
keycloak or /realms/<realm> present anywhere in the issuer URL
Ping Identity
pingidentity or pingone
OneLogin
onelogin.com
Duo SSO
duosecurity.com
(unclassified)
falls back to "Generic OIDC IdP" — still a valid tenant node, just unbranded
7.8 SAML metadata discovery
Same host candidate set as §7.7 (apex + IdP-prefix subdomains), probed against 5 conventional metadata
paths:
Structural confirmation (don't bother parsing full XML): body contains entitydescriptor or
entitiesdescriptor (case-insensitive). Extract entityID via a simple attribute regex:
entityid=["']([^"']+)["']
Severity note: public SAML metadata is INFO, CONFIRMED — and the remediation is explicitly not
"block this." Publishing SP↔IdP trust metadata is conventional; the risk is entirely in signing-key
compromise, which this passive probe cannot assess. Don't over-call a routine metadata endpoint as a
finding worth escalating.
7.9 Assembling the tenant SERVICE node
Every probe above upserts into the same tenant node when they resolve to the same product —
merge, don't fragment, on (placeholder-IP, port 443, product) so "Microsoft Entra (Azure AD)"
discovered via OIDC metadata AND via getuserrealm.srf AND via MX inference lands as one node with
accumulating attrs and sources, not three disconnected fragments. Attach a domain OWNED_BY tenant edge. Track which discovery_method values are considered strong enough to gate §11's oracle
downstream: oidc_metadata, getuserrealm, autodiscover, okta_oidc — a tenant node that only
ever got an mx_inference hit (Google Workspace's usual path, or a weak Microsoft breadcrumb) should
not, by itself, authorize spending oracle budget against it without at least one of the four
stronger methods also confirming.
8. Federation Mapping — the Tenant Boundary
Domain resolution (§7) tells you a tenant exists. It says nothing about who else shares that same
trust. This is the highest-value, most commonly skipped step in identity-fabric recon: an attacker
who compromises credentials on any domain inside the federation boundary can pivot across all of
them, because they share one authentication trust — but inward DNS/CT enumeration of the seed alone
never surfaces sibling roots with no DNS trail back to it.
8.1 GetFederationInformation — the one keyless, Microsoft-only federation-map endpoint
A single unauthenticated SOAP POST to Microsoft's own infrastructure returns every domain federated
into the same Azure AD / M365 tenant as the seed:
POST https://autodiscover-s.outlook.com/autodiscover/autodiscover.svc
SOAPAction: "http://schemas.microsoft.com/exchange/2010/Autodiscover/Autodiscover/GetFederationInformation"
Content-Type: text/xml; charset=utf-8
Lowercase, de-duplicate, strip trailing dots. Two derived facts matter:
Canonical tenant name — the one domain ending .onmicrosoft.com (but NOT
.mail.onmicrosoft.com, a routing variant) gives you <prefix>.onmicrosoft.com; strip the suffix
for the bare tenant prefix. You need this prefix for §10.2's MDI presence check.
Sibling roots worth surfacing — every returned domain that is NOT the seed apex, NOT a
subdomain of the seed, and NOT a *.onmicrosoft.com Microsoft-owned routing domain. These are the
external co-federated brands/subsidiaries.
8.3 ROE — discover-only, never auto-scope
Emit each sibling root as a discover-onlyrelated_domain asset, exactly like org-attack-surface
treats every candidate beyond the seed:
Attach a BRAND_MATCH owner-signal, weight 0.40, confirming: false — co-federation is a
strong lead, not proof of common ownership, and should never be scored as if it were.
Tie it to the seed with a FEDERATED_WITH provenance edge — but hold that edge type out of any
attack-path pivot-weight table. It exists to close the graph for reporting ("here's who shares
this trust"), never to let an automated pivot walker treat co-federation as a live reachability
path into the sibling.
Promotion to active scope is a distinct, explicit operator action. A co-federation hit — however
many siblings it surfaces — never auto-extends §7–§11 onto them.
Attacker framing worth stating explicitly in the finding description: every federated domain
shares one authentication trust, so a credential or MFA-bypass foothold gained against any one of
them pivots across all of them. That is the entire reason this map is worth building even when every
individual sibling looks unremarkable on its own. Severity: INFO, FIRM confidence (an
unauthenticated Microsoft-side response, not yet independently corroborated per-sibling) — the value
is informational/structural, not itself a vulnerability.
8.4 Honest scope: this is Microsoft-only
There is no equivalent keyless, unauthenticated federation-info surface for Okta, Google
Workspace, or a generic SAML IdP. Say so explicitly rather than silently only covering Microsoft
tenants — §7.4's Okta governed-domains endpoint (/api/v1/domains) is the closest analog for Okta
(it maps custom-domain trust, not sibling-tenant federation), and no analog exists at all for generic
OIDC/SAML. Report a federation-map pass as "Microsoft tenant federation mapped; no equivalent surface
available for {other confirmed IdP}" rather than implying completeness that isn't there.
9. IdP Fingerprint — Distinguishing Signals at a Glance
Consolidated cross-provider comparison (endpoint detail lives in §7 above and in offensive-osint
§22 — this table is the "which one am I looking at" quick-reference):
/.well-known/openid-configuration on any alive host
Auth0 tenant subdomain
issuer contains auth0.com
Keycloak
same
Realm name
issuer contains keycloak or /realms/<realm>
Ping Identity
same
Ping environment
issuer contains pingidentity/pingone
OneLogin
same
OneLogin subdomain
issuer contains onelogin.com
Duo SSO
same
Duo tenant
issuer contains duosecurity.com
Generic SAML
one of 5 metadata paths (§7.8)
entityID
EntityDescriptor/EntitiesDescriptor present
Priority order when multiple signals fire (an org can legitimately run more than one — e.g. Okta
as the primary IdP federated into Entra for M365 specifically): trust the most tenant-specific signal
(a GUID or org slug beats an MX correlation), and emit every confirmed IdP as its own SERVICE
node rather than collapsing to "the" IdP — a hybrid identity fabric is common and each node is a
distinct pivot surface downstream.
10. Seamless-SSO & MDI Presence Detection
Both probes below touch only Microsoft-owned infrastructure — zero packets to the client's own
domain — so, unlike §7.5's ADFS mex probe or a client-hosted OIDC fingerprint, they carry no
passive-mode gate. They are Low detectability by construction: the client never sees the request.
10.1 Azure AD Seamless SSO — the Negotiate challenge
GET https://autologon.microsoftazuread-sso.com/{domain}/winauth/trust/2005/usernamemixed
Detection rule: Seamless SSO is enabled when the response is 401and its
WWW-Authenticate header contains Negotiate (case-insensitive substring match — that's the whole
oracle, no body parsing needed).
What it means: Seamless SSO relies on the on-prem AZUREADSSOACC$ computer account. If its
Kerberos key never rotates, it's a known silver-ticket-forging vector against Azure AD, and the
endpoint itself is a documented password-spray target — this leg of auth has no on-prem lockout and
no MFA. Severity: LOW, FIRM confidence (single-probe direct observation) — a posture fact that
sharpens risk on any later spray/credential-testing phase, not a vulnerability in isolation.
10.2 Microsoft Defender for Identity (MDI) presence
Requires the canonical tenant prefix from §8.2 (the .onmicrosoft.com name, suffix stripped) — this
probe only runs once §8 has resolved it.
<tenant-prefix>sensorapi.atp.azure.com — resolve for an A record
PREFIX="acme"
dig +short A "${PREFIX}sensorapi.atp.azure.com"
$PREFIX = "acme"
Resolve-DnsName -Name "${PREFIX}sensorapi.atp.azure.com" -Type A -ErrorAction SilentlyContinue
If it resolves, MDI is deployed. This is a defensive-posture signal, not a weakness — say so
explicitly in the finding, INFO severity, FIRM confidence. What it tells an attacker: on-prem
identity activity (DCSync attempts, lateral-movement reconnaissance, Kerberoasting) is
sensor-monitored, so noisy on-prem AD attacks against this tenant are more likely to be detected.
This is genuinely useful recon output — it should inform how loud any later authorized on-prem phase
is willing to be — but report it for what it is, not as a finding "against" the client.
11. The User-Enumeration Oracle Methodology
This is the one active, Medium-detectability technique in this skill. Read §1 again before running
it. The output of this section is a list of email addresses confirmed to exist as accounts. It is
never a list of working passwords, and no password is ever submitted to produce it.
11.1 Oracle mechanics — the shared pattern
Both Microsoft's and Okta's endpoints are, structurally, the same trick: a pre-auth endpoint that
must answer something different for "this account doesn't exist" vs. "this account exists but you
got the credential wrong" — because a real login flow needs to tell a real user "no such account" vs.
"wrong password" before MFA. That necessary UX distinction is the oracle. Feed it a candidate address
(never a real password — Okta's endpoint requires some password value in the request shape, so a
fixed non-functional placeholder is used, §11.3), read which of the two response shapes came back,
and you've confirmed account existence without ever attempting to authenticate.
This is exactly why it's Medium, not Low, detectability: every probe is a real per-account
authentication attempt as far as the tenant's audit log is concerned, even though it never succeeds
and never even tries a real password. Microsoft and Okta both log it (§11.4).
11.2 Microsoft — GetCredentialType
POST https://login.microsoftonline.com/common/GetCredentialType
Content-Type: application/json; charset=UTF-8
Body: {"Username": "<email>", "isOtherIdpSupported": true}
Response field IfExistsResult — this is the entire oracle:
IfExistsResult
Meaning
0
Account exists in this tenant.
6
Account exists, but in a different (federated/other) Entra tenant.
5
Request throttled — back off (§11.4); not a valid/invalid signal either way.
(anything else / no field)
Treat as unresolved for this address — do not count it as either outcome.
A 0 or 6 result is CONFIRMED confidence — this is direct verification, not inference, even
though the finding severity stays LOW (§11.5 explains why severity and confidence diverge here).
11.3 Okta — /api/v1/authn
POST https://{org}.okta.com/api/v1/authn
Content-Type: application/json
Body: {"username": "<email>", "password": "<oracle-placeholder>"}
Accept HTTP status 200, 401, or 403 (anything else, skip that address — treat as unresolved,
don't retry aggressively). Parse errorCode from the JSON body:
errorCode
Meaning
E0000004
Authentication failed with a known account (wrong-password shape) → account exists.
anything else, or no errorCode (e.g. E0000095)
Not confirmed as existing from this probe — do not add to the known list.
A confirmed E0000004 hit is CONFIRMED confidence, same severity/confidence divergence rationale as
§11.2.
11.4 Detectability & rate discipline
Rule
Value
Why
Cap per tenant
20 candidates
Production cap — beyond this you're trading marginal enumeration value for audit-log noise and lockout risk. Not a soft guideline; treat it as a hard stop.
Ranking before the cap
Interest-scored, highest first
When you have more candidates than the cap, rank and keep the top 20 rather than an arbitrary slice — see §11.6's scoring table. Log how many were dropped and why.
Tenant gate
Only tenants resolved via oidc_metadata, getuserrealm, autodiscover, or okta_oidc (§7.9)
Don't spend oracle budget against a tenant only weakly inferred (e.g. bare MX correlation).
On a 5/throttled Microsoft response
Back off — halve rate, add jitter, per osint-methodology §6.4's back-off ladder
A throttle response is the tenant's own defense reacting; continuing to hammer it is both noisy and unproductive.
Detectability tag
Medium (per osint-methodology §6.2)
Both endpoints log per-attempt in the tenant's own sign-in/audit trail — this is the standing detectability classification for this exact technique across the whole skill pack; don't re-derive a different tag here.
11.5 Why severity stays LOW even at CONFIRMED confidence
This is a deliberate, non-obvious design choice worth internalizing: a confirmed user-enumeration hit
is CONFIRMED confidence (you directly verified the differential — no inference involved) but only
LOW severity. The two axes measure different things. Confidence answers "how sure am I this is
true"; severity answers "how bad is this fact on its own." Account-existence disclosure is a real,
useful reconnaissance primitive — it's the input to a spray list — but it is not itself a
compromise, a credential, or an authorization bypass. Don't inflate severity to make the finding
"sound worse than it is"; the honest LOW/CONFIRMED combination is what lets a client triage this
correctly against findings that actually prove exploitation.
11.6 Interest ranking for the 20-candidate cap
When candidates exceed the cap, rank by likely operational value before truncating — admin/role
accounts first, then real person-shaped accounts, generic role-inbox addresses last:
Sort descending, take the top 20, log the count and reason for anything trimmed.
12. Name x Pattern Login-Candidate Synthesis
Feeds §11 with candidates beyond what direct email harvesting already found — turning a discovered
employee name into a login candidate the oracle can check.
12.1 Two distinct synthesis modes — do not conflate them
Mode
Input
Output
Confidence
Use
Passive enrichment
A name + domain, no confirmed pattern
Up to 8 unverified permutations ({first}.{last}, {first}{last}, {first}, {f}{last}, {first}.{l}, {last}, {first}_{last}, {first}-{last})
TENTATIVE
Breach pre-hit search, phishing-list curation, general enrichment — NOT for feeding §11's live oracle.
Oracle-feed synthesis
A name + domain + a confirmed org email pattern
Exactly ONE canonical, format-consistent address per name
FIRM candidate (becomes CONFIRMED per-address only if §11 fires on it)
The only mode that should ever reach §11.
The distinction matters because of precision: spraying 8 unverified guesses per name at a live,
logged oracle multiplies your Medium-detectability footprint eightfold for no proportional gain in
enumeration value, and most of those 8 will be wrong for any given org's actual convention.
12.2 The fail-closed rule
If no org email-format pattern is confirmed, oracle-feed synthesis produces nothing — not a
best-effort guess, not the 8-permutation fallback, zero candidates. This is the load-bearing
precision decision in this whole section: a confirmed pattern (e.g. from a domain-search enrichment
source returning {first}.{last} as the org's dominant format) lets you generate one
format-consistent address per name; without it, every permutation is an unverified guess, and
guessing at a live, logged, rate-capped oracle is the wrong trade. Return an empty list and say so
explicitly — do not silently fall back to passive-mode permutations for oracle-feed use.
Recognized pattern tokens: {first}, {last}, {f} (first initial), {l} (last initial). A name
that can't be cleanly decomposed into first+last (single-token names, honorifics-only) or a pattern
that still contains an unrecognized token after substitution both fail closed for that specific name
— skip it, don't guess a partial fill.
12.3 Merging with harvested emails
Combine directly harvested emails (already-confirmed real addresses, highest priority) with
synthesized candidates (name × confirmed pattern) into one order-preserving, de-duplicated list,
harvested-first. Read the input names directly from discovered PERSON-type identity assets and the
org's persisted email-format-pattern attribute — not from an intermediate inferred-email layer — so
this synthesis step doesn't silently produce zero output just because some other harvesting pass
hasn't run yet in a given execution order. The production ceiling of 40 caps the synthesized
candidates inside the synthesis step (before the merge); the merged harvested+synthesized list is
then bounded downstream by §11.6's interest ranking and the 20/tenant oracle cap.
13. End-to-End Workflow
A concrete run order tying §7–§12 together:
Resolve the tenant (§7): fan out all six probes concurrently against the seed domain. Any hit
confirms an IdP exists; merge all hits into one tenant SERVICE node per product (§7.9).
Map the federation boundary (§8), if Microsoft was confirmed in step 1 — GetFederationInformation,
extract sibling roots, emit as discover-only leads, note the tenant prefix for step 4.
Cross-reference the distinguishing-signal table (§9) if more than one IdP fired — a hybrid
fabric (e.g. Okta primary + Entra for M365) is common; keep every confirmed IdP as its own node.
Check Seamless-SSO and MDI posture (§10) — both are Microsoft-infrastructure-only, so they run
regardless of what else fired, gated only on having a tenant prefix (MDI) or the seed domain
itself (Seamless SSO).
Stop and re-confirm authorization (§1) before proceeding — everything above this line is Low
detectability; everything below generates tenant-side audit events.
Gather harvested identity — pull already-harvested email addresses and discovered PERSON names
from the graph.
Synthesize oracle candidates (§12) — apply the confirmed org pattern (if any) to the harvested
names; merge with the directly-harvested emails, order-preserving, deduped, capped at ~40.
Rank and cap (§11.6) — score by admin/role/person-shaped interest, keep the top 20 per tenant
the oracle will actually check.
Run the oracle (§11.2/§11.3) against every tenant that cleared §7.9's gate, respecting the
back-off ladder on any throttle signal.
Report the valid-account list — LOW severity, CONFIRMED confidence per positively-differentiated
address, credential_submitted: false explicit in every evidence block.
Hand off — the valid-account list is the input to a later, separately-authorized
authenticated-testing phase (§14). This skill's job ends at step 10.
14. The Hard Boundary — Enumeration Ends Here
Everything in this skill is discovery, fingerprint, or account-existence enumeration. None of it
submits a credential, forges a token, or confirms a bypass. The line is explicit and non-negotiable:
IN scope (everything above): domain→tenant resolution, federation mapping, IdP fingerprinting,
Seamless-SSO/MDI presence detection, the GetCredentialType/Okta-authn existence oracle, name×pattern
candidate synthesis.
OUT of scope — a different, higher authorization tier, not covered by this skill at all:
Password spray against the valid-account list this skill produces. Confirming account
existence (§11) is not authorization to attempt authentication against those accounts.
Credential submission of any kind — a real password, a breach-sourced password, or any
guess — to any login endpoint.
Token forging or replay — JWT forgery, SAML golden-ticket construction, Kerberos silver-ticket
use against a Seamless-SSO computer account this skill merely detected the presence of (§10.1).
Anything corresponding to a mature ASM platform's own stage-6 --validate intrusive tier
(validate_creds, validate_sso, jwt_forge_confirm) or any equivalent authenticated-confirmation
step in another toolchain.
If the engagement calls for that next step, it belongs to the authorization tier your engagement
scope defines for authenticated/intrusive testing — point the reader there rather than executing it
under this skill. A valid-account list handed off with an explicit credential_submitted: false
marker (§3, §11) is the clean, auditable handoff artifact between this skill's enumeration phase and
that separately-authorized phase.
15. Anti-Patterns & Common Failure Modes
Treating a Managed namespace finding as "no ADFS = safer." It's the opposite — no on-prem
buffer means spray/MFA-fatigue hits Entra's own (more permissive) throttling directly (§7.1).
Attributing an Okta org-slug guess without confirming its own OIDC issuer. A generic stem
guess resolving to a stranger's Okta org is a namesake collision, not a hit (§7.4, §5).
Treating co-federation (§8) as ownership proof.BRAND_MATCH at weight 0.40 is a lead. Don't
auto-promote a federated sibling into active scope no matter how confident the federation response
looks.
Reporting MDI presence as a weakness. It is a defensive-posture signal — get the framing
backwards in a client report and you've told them their own monitoring investment is a finding
against them (§10.2).
Inflating oracle-hit severity. CONFIRMED confidence does not mean HIGH severity here — see the
explicit rationale in §11.5. Account existence is not compromise.
Falling back to 8-permutation guessing when no pattern is confirmed. The single most important
discipline in §12 — fail closed, don't guess at a live logged oracle (§12.2).
Exceeding the 20/tenant oracle cap "just this once." The cap exists because of the detectability
tier this technique carries (§1, §11.4), not as an arbitrary throttle.
Skipping the re-scope-check between §10 and §11. The passive/active authorization boundary in
this skill sits exactly there (§1) — don't carry a passive-recon go-ahead into the active oracle
without re-confirming.
Treating a throttled (IfExistsResult=5) Microsoft response as a negative result. It's neither
confirmed-exists nor confirmed-doesn't-exist — it's "back off and reconsider," full stop (§11.2,
§11.4).
16. Skill Self-Test
Drop these into a fresh session to verify the skill loads and routes correctly.
"Resolve acme.com to its M365/Entra tenant without sending anything to acme.com's own
infrastructure." → §7.1–§7.3 (all three Microsoft-side probes).
"getuserrealm.srf says NameSpaceType=Managed. Why does that matter for spray risk?" → §7.1
(Managed = MEDIUM, no ADFS buffer).
"Given a confirmed M365 tenant, find every domain federated into it." → §8.
"Derive Okta org-slug candidates for target.com and confirm which one is real." → §7.4 (slug
derivation + the issuer-confirmation guard, not a bare guess).
"Confirm whether Azure AD Seamless SSO is enabled for target.com." → §10.1.
"MDI sensor-API host resolves for the tenant. Is that good or bad news for the client?" → §10.2
(defensive-posture signal, not a weakness — get the framing right).
"Build a valid-account list for 50 harvested emails using GetCredentialType." → §11.2, §11.4
(cap at 20/tenant, rank first, Medium detectability re-scope per §1).
"I have employee names but NO confirmed org email pattern. Synthesize logins to feed the
enum oracle." → §12.2 — NO. Fail closed, return zero candidates, do not fall back to
8-permutation guessing.
"I have a confirmed {first}.{last} pattern and 30 harvested names. Build the ranked oracle
candidate list." → §12.1, §12.3, §11.6.
"GetCredentialType found 8 accounts confirmed to exist. Should I password-spray them now?" →
§14 — NO, out of scope for this skill. Existence ≠ authorization to attempt authentication;
that belongs to the engagement's separately-authorized testing phase.
"IfExistsResult came back as 6 for a few addresses. What does that mean?" → §11.2 (exists, but
in a different/federated tenant — not a negative result).
"Full identity-fabric map for target.com from zero — what's the run order?" → §13.
"Microsoft's GetCredentialType just returned IfExistsResult=5 on several probes in a row." →
§11.2, §11.4 (throttled — back off per the ladder, not a valid/invalid signal).
"ADFS mex endpoint at target.com returned a WSDL body. Does hitting that count as passive?" →
§7.5 (no — active, target-side logged, gated on the passive fingerprint already confirming ADFS
first).
17. Changelog
v1.0 (2026-08-06) — initial release. Domain→tenant resolution across six probes
(getuserrealm.srf Managed/Federated, Entra OIDC metadata + tenant-GUID extraction, Autodiscover v2,
Okta org-slug derivation + OIDC fingerprint + governed-domains, ADFS passive/active fingerprint +
version inference, Google Workspace MX correlation, generic OIDC, SAML metadata) (§7); keyless
Microsoft-only tenant-federation mapping via GetFederationInformation SOAP with discover-only ROE
and the FEDERATED_WITH pivot-exclusion discipline (§8); cross-provider distinguishing-signal
reference table (§9); Seamless-SSO Negotiate-challenge and MDI sensor-API presence detection, both
Microsoft-infrastructure-only and therefore ungated by passive mode (§10); the user-enumeration
oracle methodology for Microsoft GetCredentialType (IfExistsResult semantics) and Okta
/api/v1/authn (errorCode differential), with the 20-candidate-per-tenant cap, Medium-detectability
discipline, and the confidence/severity-divergence rationale for why a CONFIRMED oracle hit stays
LOW severity (§11); name×confirmed-pattern login-candidate synthesis with the fail-closed rule as
the load-bearing precision decision (§12); end-to-end workflow (§13); the hard boundary separating
enumeration from password spray / credential submission / auth bypass (§14); anti-pattern catalog
(§15); 16-prompt self-test including two explicit negatives (§16). Grounded directly in the
reference production implementation (modules/sso_idp.py, modules/tenant_recon.py,
core/email_patterns.py) — every endpoint, response-field semantic, cap, and severity/confidence
choice in this skill matches a shipped, tested value, not an invented placeholder. Deepens
offensive-osint §22's endpoint reference rather than duplicating it; companion to
osint-methodology §6.2 (detectability tagging) and §11 (identity-fabric pointer).
