DECEPTION CHECK / SKYKICK / SPLUNK AND CROWDSTRIKE ADAPTER NOTES / 2026-10-19

START HERE
The SPL and CQL files contain three read-only hunts each: application/IP sign-ins,
broad authority/credential changes across all apps, and a confirmed credential-key
pivot. These are mapping-required templates, not native-schema detections. No
Splunk, Falcon, or LogScale runtime or customer-tenant execution is represented.

The dc_* names are this companion's deliberately small normalization contract.
They are not Splunk CIM, CrowdStrike CPS/ECS, or built-in connector field names.
Have the SIEM owner map the source records before running the searches. A saved
search or view name alone does not create these fields. Preserve the original raw
event and record the parser version/source so an analyst can verify each mapping.

TELEMETRY REQUIRED
Ingest both Entra service-principal sign-ins and directory audits for each tenant.
An endpoint sensor alone does not supply either cloud event stream. A connector
being installed does not prove that these event types, dates, tenants, or credential
fields are collected. Inspect a known event from each stream and verify retention,
ingestion health, permissions, and first/last available times before hunting.

The exact source mapping below is for Azure Monitor table-shaped records from
AADServicePrincipalSignInLogs and AuditLogs. A raw Event Hub export, Graph feed,
Splunk add-on, or Falcon connector can use a different envelope or field spelling.
Unwrap the actual source first. For example, published Falcon Entra examples use
Vendor.properties.* fields; its Graph examples use Vendor.* for some properties.
Those examples do not establish the full service-principal sign-in mapping here.
Map a verified sample, including nested objects; never blindly substitute prefixes.
Splunk's documented azure:monitor:aad example uses lowercase tenantId/time and a
properties object, illustrating why table spellings cannot be assumed on ingest.
Microsoft 365 Management Activity records are a separate schema.

NORMALIZATION CONTRACT
Each normalized row represents one source event. All dc_* fields below are scalar
strings; *_json fields contain complete serialized JSON. Missing source values
remain null/missing, not invented success values. GUID fields must be canonical
lowercase, without braces/whitespace; keep the original source alongside them.
Keep source IPv4 and IPv6 as strings. The three reported IPv4s use exact equality;
alternative IP encodings require a verified mapping and the original address.

Common mapping (both source tables)
  dc_event_kind       constant from verified source: entra_sp_signin OR
                      entra_directory_audit; never infer from outcome or app ID
  dc_tenant_id        AADTenantId (Entra tenant, NOT workspace TenantId)
  dc_event_time       TimeGenerated, preserved UTC timestamp string
  dc_event_id         Id
  dc_correlation_id   CorrelationId
  Splunk _time        parse TimeGenerated as UTC epoch seconds at ingestion
  CQL @timestamp     parse TimeGenerated as UTC epoch milliseconds at ingestion

Service-principal sign-ins (AADServicePrincipalSignInLogs)
  dc_activity_time       CreatedDateTime
  dc_app_id              AppId
  dc_sp_id               ServicePrincipalId
  dc_sp_name             ServicePrincipalName
  dc_source_ip           IPAddress
  dc_result_type         ResultType
  dc_result_signature    ResultSignature
  dc_result_description  ResultDescription
  dc_resource_name       ResourceDisplayName
  dc_credential_type     ClientCredentialType
  dc_key_id              ServicePrincipalCredentialKeyId
  dc_thumbprint          ServicePrincipalCredentialThumbprint
  dc_user_agent          UserAgent

Directory audit (AuditLogs)
  dc_activity_time       ActivityDateTime
  dc_category            Category
  dc_operation           OperationName
  dc_result              Result
  dc_result_reason       ResultReason
  dc_initiated_by_json   entire InitiatedBy object serialized without loss
  dc_targets_json        entire TargetResources array, including every target's
                         modifiedProperties, serialized without loss

TimeGenerated is the investigation filter used by this companion; AuditLogs
documents it as record creation time. Keep the separate activity timestamp shown
above and the platform's ingestion timestamp. Do not silently replace one with
another. Validate timezone, seconds versus milliseconds, delays, and the earliest
and latest source events. Extend the window when source timing requires it.

IMPLEMENTATION AND COMPLETENESS
Splunk: supply real index/sourcetype selectors, then create and verify the dc_*
fields through your source-specific extraction/normalization. If a field contains
raw JSON, explicit spath input/path/output extraction can be used; first confirm
the exact path and that the complete payload is retained. Do not depend on an
unverified automatic extraction or on a universal sourcetype. The supplied files
are SPL, not SPL2. Their triple-backtick comments are SPL search comments.

Falcon Next-Gen SIEM / LogScale: choose the repository/view with the verified
Entra feeds and expose the contract through your parser or tested query adapter.
No #repo, #event_simpleName, or Vendor.* value is asserted by these templates.
Set the UI/API time range explicitly; a query's lower bound cannot recover events
excluded by the picker. Use the recorded exclusive end for every comparison.

Copy one numbered block. Replace every REPLACE value. Default start is October 13,
2026, 00:00 UTC. Extend to October 1 where history permits. SPL provides the start
and latest=now; record execution time or insert a fixed exclusive epoch end.
CQL also requires the selected UI/API UTC end and a picker spanning the start.

The queries preserve individual rows and do not prove export completeness. CQL
select() can display only the last 200 events in a standard query; its export path
is intended for matching event rows. Verify the actual export and service limits.
Splunk table can truncate results under limits.conf search settings as well.
For both platforms, compare pre-projection match counts with exported rows, inspect
warnings, and split large windows into adjacent non-overlapping intervals if needed.
Do not treat the visible table as the complete timeline. Keep tenant plus source
event ID when reconciling duplicates; retain separate source streams.

HOW TO READ THE THREE HUNTS
01: A listed app OR a reported source address matches. Known apps on new addresses,
IPv6 app matches, failures, unknown outcomes, and IP matches for other apps remain
in scope. A listed vendor application can have legitimate activity. Confirm the
outcome, resource, tenant, credential, and surrounding actions before concluding
that access succeeded or was unauthorized. IDs/IPs are unchanged from KQL 02.

02: The category fallback plus operation phrases creates a broad review queue.
It includes other applications, all actor types, and failed/unknown outcomes;
there is no incident-app filter or success-only filter. Authorized changes match.
Generic Update application events can enter through ApplicationManagement even
when only their nested properties describe credential changes. Inspect every
target and old/new credential metadata. Events without target entries remain
one parent row. App display names and an app initiator do not establish app-only
authentication; verify initiator IDs and actor context. Unexpected operation
names/categories outside this predicate can be missed.

03: After confirming tenant, application/client ID, and key ID from the changed
object, pivot on all three. Never substitute a service-principal object ID for an
application/client ID, or paste secret values. A match can be a failed attempt;
read the preserved result. Missing key IDs and reuse of an already-issued token
can produce no new matching sign-in. A zero result is not clearance.

PRIMARY REFERENCES (opened 2026-10-19)
Microsoft source schema: service-principal sign-ins
https://learn.microsoft.com/en-us/azure/azure-monitor/reference/tables/aadserviceprincipalsigninlogs
Microsoft source schema: directory audit
https://learn.microsoft.com/en-us/azure/azure-monitor/reference/tables/auditlogs
Splunk SPL conditional predicates: in, if, coalesce, and like
https://help.splunk.com/en/splunk-enterprise/search/spl-search-reference/10.0/evaluation-functions/comparison-and-conditional-functions
Splunk lower() and text functions
https://help.splunk.com/en/splunk-enterprise/search/spl-search-reference/10.0/evaluation-functions/text-functions
Splunk table projection and truncation behavior
https://help.splunk.com/en/splunk-enterprise/search/spl-search-reference/10.0/search-commands/table
Splunk epoch/time modifiers
https://help.splunk.com/en/splunk-enterprise/search/spl-search-reference/10.0/time-format-variables-and-modifiers/time-modifiers
Splunk explicit JSON extraction
https://help.splunk.com/en/splunk-enterprise/search/spl-search-reference/9.0/search-commands/spath
Splunk Data Inputs source-type examples and Entra event envelope
https://help.splunk.com/en/splunk-cloud-platform/ingest-data-from-cloud-services/data-inputs-user-manual/1.17/getting-data-in-gdi/overview-of-source-types-for-data-inputs
Splunk search comment syntax
https://help.splunk.com/en/splunk-enterprise/search/search-manual/9.1/using-the-search-app/add-comments-to-searches
CrowdStrike CQL in(), including case-based OR and case-insensitive matching
https://library.humio.com/crowdstrike-query-language/functions-in.html
CrowdStrike CQL operators and regular-expression filtering
https://library.humio.com/crowdstrike-query-language/syntax-operators.html
CrowdStrike CQL select() and standard-query display/export distinction
https://library.humio.com/crowdstrike-query-language/functions-select.html
CrowdStrike Entra data connector availability, not a full field-map guarantee
https://marketplace.crowdstrike.com/listings/data-connector-built-for-microsoft-entra-id/
CrowdStrike examples distinguishing Entra and Graph parser envelopes
https://www.crowdstrike.com/en-us/blog/defending-against-scattered-spider-with-falcon-next-gen-siem/
