agentleFS
Liberty91LTDLiberty91LTD/cti-skills/skills/lookup-sentinel/SKILL.md

lookup-sentinel

Skill35 starsChanged 47 days agoMITView on GitHub

Use to hunt in the organisation's own Microsoft Sentinel workspace: sweep logs for IOC hits (IP, domain, hash, URL, account), hunt ATT&CK techniques, run KQL on live data, or check which tables are ingested. Discovers tables first and only writes KQL for ones that exist. Chained by the /*-investigation skills to answer \"were we exposed?\". Read-only. Reads $SENTINEL_TENANT_ID, $SENTINEL_CLIENT_ID, $SENTINEL_CLIENT_SECRET and $SENTINEL_WORKSPACE_ID.

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 1 is distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read-only. The CLI can only run queries and list schema — it cannot modify the workspace, close incidents, or create analytics rules.

    line 16
  2. Very specific:

    Run tables (schema-known) and/or ingestion (actually receiving data) once, and reuse the result for the rest of the session.

    line 24
  3. Very specific:

    All commands accept --dry-run.

    line 61
  4. Very specific:

    Prior art check: Before hunting, a quick SecurityAlert | where ...

    line 153

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

lookup-sentinel

Query bridge to the organisation's own Microsoft Sentinel (Azure Log Analytics) workspace. Every other lookup in this pack asks the world about an indicator; this one asks your own telemetry: has this indicator been seen in our environment? Is this ATT&CK technique's behaviour present in our logs? It is scoping/exposure-checking, not enrichment — chain it after external lookups have told you what an indicator is, to establish whether it touched you.

Read-only. The CLI can only run queries and list schema — it cannot modify the workspace, close incidents, or create analytics rules.

The cardinal rule: discover before you query

No two Sentinel workspaces have the same tables. Table availability follows deployed connectors (Defender for Endpoint brings Device*, Entra ID brings SigninLogs, AMA/CEF brings CommonSecurityLog, …). A query referencing an absent table fails outright; worse, a hunt written for tables the environment doesn't ingest silently proves nothing.

So, in every session, before generating any hunt KQL:

  1. Run tables (schema-known) and/or ingestion (actually receiving data) once, and reuse the result for the rest of the session.
  2. probe the specific candidate tables for the hunt — schema-known tables can be empty.
  3. Generate KQL only against tables the probe confirmed populated. When a preferred table is missing, walk the fallback ladder (below). When no suitable table exists, report the telemetry gap as a finding — do not fabricate a query, and do not quietly substitute a table that can't answer the question.

When to invoke

  • An investigation skill (/ip-investigation etc.) produced a malicious/suspicious verdict and you need the exposure answer: was it seen here?
  • The user asks to hunt for an ATT&CK technique, actor TTP set, or behaviour ("hunt for T1059 PowerShell abuse", "look for Kerberoasting")
  • The user has IOCs (from a report, a feed, /ioc-enrichment-workflow) and wants them swept across the environment
  • The user asks what telemetry/tables the Sentinel workspace has, or whether a planned detection would even have data
  • The user asks to run or adapt a KQL query against live logs

Do NOT invoke for:

  • Writing portable detection rules for later deployment — that's /kql-writing (author) and /sigma-writing (vendor-neutral). This skill runs queries; compose with /kql-writing for authoring conventions.
  • External reputation questions ("is this IP bad?") — that's the other lookup-* skills.
  • Anything write-shaped (closing incidents, adding analytics rules, updating watchlists) — out of scope by design.

How to invoke

Single Python CLI (stdlib only — no install). Credentials setup: tools/integrations/sentinel.md.

# Connectivity + auth check (run once per session)
python3 tools/clis/sentinel.py check

# Discovery
python3 tools/clis/sentinel.py tables                       # all schema-known tables
python3 tools/clis/sentinel.py tables --filter Device       # filter by substring
python3 tools/clis/sentinel.py ingestion --days 30          # which tables actually received data (with volume)
python3 tools/clis/sentinel.py probe DeviceNetworkEvents,SigninLogs,CommonSecurityLog --days 7
python3 tools/clis/sentinel.py schema DeviceNetworkEvents   # column names + types

# Hunting — arbitrary KQL, always bounded by a server-side timespan
python3 tools/clis/sentinel.py query 'DeviceNetworkEvents | where RemoteIP == "203.0.113.42" | summarize hits=count(), first=min(TimeGenerated), last=max(TimeGenerated) by DeviceName' --timespan P30D
python3 tools/clis/sentinel.py query --file data/detection-rules/kql/t1059-encoded-powershell.kql --timespan P7D --max-rows 50

All commands accept --dry-run. The CLI exits 2 if any of the four env vars is unset (when not in dry-run) — report missing credentials and offer /cti-setup; do not fabricate results.

Workflow

intent (IOC sweep | TTP hunt | freeform query | table inventory)
  ↓
check  →  tables / ingestion (once per session)  →  probe candidates
  ↓
generate KQL per /kql-writing conventions, restricted to confirmed tables
  ↓
run: summarize-first pass (counts per device/user) → drill into hits only
  ↓
interpret → hits feed /*-investigation + rigor pipeline; misses reported with window + tables searched; gaps reported as findings

Query discipline: start with summarize/count per entity rather than raw rows; keep timespans as tight as the question allows (P7D default, widen deliberately); take/project when you do pull rows. Results are capped (500k rows / 64 MB server-side; --max-rows client-side, reported as truncated) — treat a truncated result as a signal to aggregate, not paginate.

IOC sweep — table candidates by indicator type

Probe in listed order; use every confirmed table, via union where sensible. Use let-bound dynamic lists for multi-IOC sweeps (see /kql-writing).

IOC typePreferred (EDR-rich)FallbacksMatch columns
IPDeviceNetworkEventsCommonSecurityLog, SigninLogs + AADNonInteractiveUserSignInLogs, AzureActivity, W3CIISLog, SyslogRemoteIP/LocalIP; SourceIP/DestinationIP; IPAddress; CallerIpAddress; cIP
DomainDeviceNetworkEventsDnsEvents/ASimDnsActivityLogs, CommonSecurityLog, EmailUrlInfoRemoteUrl has; Name/DnsQuery; RequestURL/DestinationHostName; UrlDomain
File hashDeviceFileEvents, DeviceProcessEventsDeviceImageLoadEvents, EmailAttachmentInfo, DeviceEventsSHA256/SHA1/MD5, InitiatingProcessSHA256
URLDeviceNetworkEventsEmailUrlInfo, UrlClickEvents, CommonSecurityLogRemoteUrl; Url; RequestURL
Account/UPNSigninLogsIdentityLogonEvents, DeviceLogonEvents, SecurityEvent (4624/4625), OfficeActivity, AuditLogsUserPrincipalName; AccountUpn; AccountName; TargetUserName; UserId
Sender/emailEmailEventsEmailPostDeliveryEvents, OfficeActivitySenderFromAddress, SenderMailFromAddress; RecipientEmailAddress

Also probe ThreatIntelligenceIndicator (or newer ThreatIntelIndicators) — if a TI connector feeds the workspace, the IOC may already be matched by built-in analytics; check SecurityAlert for prior alerts on the same indicator before declaring a novel finding.

Report every sweep with: tables searched, tables unavailable, window, and per-table hit counts. A miss means "not observed in collected telemetry over the window" — never "not compromised". Retention differs per table; note the window explicitly.

TTP hunt — mapping techniques to tables

For technique details and detection guidance, first resolve the technique via /mitre-attack (local dataset — data sources per technique). Then map data sources to this workspace's confirmed tables:

Behaviour family (example techniques)PreferredFallbacks when EDR absent
Process execution/command-line (T1059, T1047, T1204)DeviceProcessEventsSecurityEvent EventID 4688 (needs command-line auditing enabled), Syslog auditd execve
Persistence via registry (T1547, T1112)DeviceRegistryEventsSecurityEvent 4657 (needs SACL auditing) — often absent; report the gap
Lateral movement (T1021, T1570)DeviceLogonEvents + DeviceProcessEventsSecurityEvent 4624 LogonType 3/10, IdentityLogonEvents
Credential access (T1110, T1558)IdentityLogonEventsSigninLogs (ResultType != 0 bursts), SecurityEvent 4625/4768/4769
C2 / exfil network activity (T1071, T1041, T1567)DeviceNetworkEventsCommonSecurityLog (proxy/firewall), DnsEvents, AzureNetworkAnalytics
Cloud/identity abuse (T1078.004, T1098, T1136)AuditLogs + SigninLogsAzureActivity, CloudAppEvents, OfficeActivity
Email initial access (T1566)EmailEvents + EmailUrlInfo + EmailAttachmentInfoUrlClickEvents, OfficeActivity
Defense evasion / log tampering (T1070, T1562)DeviceEvents, SecurityEvent 1102the gap itself is the signal — sudden per-table ingestion drops in Usage

/kql-writing holds the concrete query patterns for the common techniques — reuse them, then adapt the table references to what this workspace confirmed. When neither preferred nor fallback tables exist, the deliverable is the sentence "this environment cannot currently observe technique X because tables Y/Z are not ingested" — a telemetry-gap finding worth more than a query that runs on nothing.

Response format

source: sentinel
operation: check | tables | ingestion | probe | schema | query
workspace_id: <GUID>
query_time: <ISO8601>
# for hunts, the skill wraps CLI output as:
hunt:
  intent: ioc-sweep | ttp-hunt | freeform
  window: <timespan>
  tables_searched: [<confirmed tables>]
  tables_unavailable: [<candidates that were missing/empty>]
  kql: <the query as run>
  hits: <row_count, per-table counts, distilled rows>
  truncated: <bool>
telemetry_gaps: [<behaviours this workspace cannot currently observe, and why>]

Source reliability (Admiralty default) and evidence grade

Two instruments apply to a hit, and neither is converted into the other.

Admiralty default: A1 for a hit. This is the frontmatter default, used when /source-assessment rates the lookup result as a single item.

Evidence grade: "direct, established" for a hit. A positive hit is a first-party observation: the organisation saw the indicator in its own environment. It enters the claim table as an evidence item with access level direct and claim support established, claim_type: observation, access: telemetry, source "user environment", provenance_basis: first-party (grading rubric rule R12 in /quality-of-information-check). It can be load-bearing and it can raise an assessment's confidence ceiling.

Two limits:

  • A miss is not evidence. Absence of a hit is bounded by connector coverage, retention and logging depth. It is reported as "not observed in collected telemetry over this window" and never becomes a graded item.
  • Logs are an adversary target (T1070, T1562). Where tampering is suspected, grade the hit "direct, firm" and set deception_indicators_present.

Full reasoning: tools/integrations/sentinel.md.

Operational notes

  • Session-cache the discovery. tables/ingestion once per session; probe per hunt. Don't re-discover before every query.
  • Rate limit is generous (200 requests/30s) but basic-tier tables bill per GB scanned — tight timespans are a cost control, not just hygiene.
  • Custom tables (*_CL) are first-class. Organisations route firewall, proxy, or SaaS logs into custom tables; if discovery shows a well-populated *_CL table whose name suggests relevant telemetry, schema it and consider it in sweeps.
  • Results may contain personal data (UPNs, IPs, hostnames). Products built on them default to TLP:AMBER — apply /apply-tlp before sharing.
  • Prior art check. Before hunting, a quick SecurityAlert | where ... for the same indicator/technique avoids re-discovering what an analytics rule already alerted on.

Related skills

  • /kql-writing — query authoring conventions and per-technique patterns; this skill runs what that skill writes
  • /mitre-attack — resolve technique → data sources before the table mapping
  • /ip-investigation, /domain-investigation, /hash-investigation, /url-investigation — establish what an indicator is; this skill establishes whether it touched you; hits feed back into them for pivoting
  • /ioc-enrichment-workflow — bulk IOC lists; sweep the high-confidence subset here
  • /sigma-writing — a hunt that proves out here can graduate into a portable rule
  • /score-source, /apply-tlp, /confidence-language — rigor pipeline on hunt products

See also

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.