Why add a network sensor
Wazuh catches brute force, lateral movement and process attacks once they touch a monitored host. But a plain Nmap SYN scan writes nothing to the Windows event log, so a host-only setup never sees the recon. Suricata on a span port closes that gap.
- Host onlyNmap SYN scan
- Windows Server
- Nothing loggedscan is invisible
- With SuricataNmap SYN scan
- Suricata8+ scan alerts in 60s
- Rule 100011level 10 · T1046
Detection rules
All 11 rules in local_rules.xml, ordered from recon to exploitation. Bar length is the Wazuh alert level, 0 to 15.
| Rule | Level | Detects | ATT&CK |
|---|---|---|---|
| 100010 | 0 | Suricata low-priority anomaly (base rule, no alert) | |
| 100011 | 10 | Port scan: 8+ Suricata scan alerts from one source in 60s | T1046 |
| 100012 | 10 | Port scan by signature: 8+ scan signatures in 60s | T1046 |
| 100013 | 3 | Windows Firewall dropped a connection | |
| 100014 | 10 | Port scan via firewall: 15+ distinct blocked connections | T1046 |
| 100020 | 10 | Brute force: 6+ failed logons (4625) from one source in 120s | T1110 |
| 100021 | 12 | Brute force followed by a successful logon (4624) | T1110 T1078 |
| 100030 | 13 | Suspicious process access to LSASS | T1003 |
| 100031 | 12 | Encoded or hidden PowerShell execution | T1059.001 T1027 |
| 100032 | 12 | New service installed from a temp path | T1543.003 T1021 |
| 100040 | 15 | Multi-stage attack: brute force, then exploitation, same IP | T1110 T1003 |
The kill-chain rule
Rule 100040 only fires when one IP completes the chain: a brute force that succeeds, then exploitation-stage activity within 15 minutes.
- Brute force1000206+ failed logonswithin 120 seconds
- Access100021Then a successful logonsame source
- Any of, within 900s100030LSASS access100031Encoded PowerShell100032Suspicious service
- Alert100040 · level 15Multi-stage attackfrom $(srcip)
<!-- local_rules.xml -->
<rule id="100040" level="15" timeframe="900">
<if_matched_sid>100021</if_matched_sid>
<if_sid>100030,100031,100032</if_sid>
<same_source_ip />
<description>Multi-stage attack
from $(srcip)</description>
<mitre><id>T1110</id><id>T1003</id></mitre>
</rule>
- Chained, not counted. Single alerts are noisy; requiring the sequence cuts false positives.
- Same source.
same_source_ipkeeps two unrelated events from combining. - Bounded window. A 900-second window keeps the chain tied to one intrusion.
From alerts to an incident report
The correlation engine rebuilds the same story from raw alerts, independently of rule 100040, and writes it up.
- Pull alertsWazuh Indexer REST APIwazuh_api_client.py
- Group by source IPone attacker per incidentcorrelate.py
- Window by timeevents close togethercorrelate.py
- Tag attack stagerecon, brute force, exploitationcorrelate.py
- Write the reportMarkdown or PDFgenerate_report.py
Safety rails
- Attack scripts refuse any target outside private (RFC 1918) address space and ask for confirmation before firing.
- Metasploit runs only scanner modules and lab-only exploits against Metasploitable, never production software.
- The Python tooling is read-only against the Wazuh API.