Host Forensics

Remote, read-only live response through the Wazuh agent's inventory (syscollector) and alert history. You cannot image disks or run commands; you work from what the agent reports. Be explicit about that limitation in findings.

Procedure

1. Identify the agent: get_wazuh_agents (match name/IP) → agent_id, OS, version, last keepalive.
2. check_agent_health — a disconnected or stale agent means inventory may be old; record the inventory timestamp. Syscollector inventories run on an interval (1 hour by default), so a short-lived process may never appear, and a killed one may still be listed until the next scan. Compare scan.time with the incident time.
3. get_agent_processes(agent_id) — review with the heuristics below.
4. get_agent_ports(agent_id) — listening and established sockets.
5. get_agent_configuration(agent_id) — confirm FIM/rootcheck/log collection is enabled; attackers disable monitoring.
6. get_wazuh_alerts(agent_id=..., timestamp_start="now-7d") — FIM (rules 550, 553, 554; registry rules 594, 598, 750, 752), rootcheck (rule 510; rule 513 Windows malware), SCA (rule 19011 check passed → failed) and audit events for persistence evidence. Load wazuh-fim-investigation and wazuh-malware-detection for detail.
7. get_wazuh_vulnerabilities(agent_id) and get_sca_policy_checks(agent_id) — the likely way in (unpatched service, weak configuration) matters for eradication.

Suspicious process heuristics

Signal	Why
Interpreter children of web/database servers (w3wp, httpd, nginx, java, sqlservr → cmd, powershell, sh, bash)	web shell / exploitation (T1505.003, T1059)
Binaries running from /tmp, /dev/shm, /var/tmp, %TEMP%, %APPDATA%, C:\Users\Public	staging / malware
Names mimicking system binaries in wrong paths (svchost.exe outside System32, [kworker] with a real exe path)	masquerading (T1036)
Encoded / obfuscated command lines (-enc, FromBase64String, `curl	sh`)
Known tools: mimikatz, procdump targeting lsass, nc -e, socat, chisel, rclone	credential access / tunneling / exfil
Processes owned by root/SYSTEM that are normally user-level	privilege escalation

Syscollector process fields to use: name, cmd + argvs, ppid (look up the parent), euser / ruser, start_time. Port fields: local.ip, local.port, remote.ip, remote.port, state, process, pid.

Suspicious port heuristics

• Listeners on high ports bound to 0.0.0.0 not in the host's known baseline.
• Established connections to external IPs on uncommon ports, or from server processes that shouldn't initiate outbound traffic.
• Common backdoor/tunnel ports (4444, 1337, 31337, 8081 on non-web hosts) — hint only.

Persistence locations to check (via FIM/audit alerts)

Linux: /etc/crontab, /etc/cron.*, /var/spool/cron, /etc/systemd/system, ~/.ssh/authorized_keys, /etc/rc.local, /etc/ld.so.preload, shell profiles (~/.bashrc, /etc/profile.d), new users in /etc/passwd, sudoers changes.

Windows: Run/RunOnce registry keys (FIM registry alerts, rules 750, 752), services (Event ID 7045, Event ID 4697), scheduled tasks (Event ID 4698), Startup folders, WMI event subscriptions (Sysmon Event ID 19–21), new local admins (Event ID 4732 → rule 60144; Administrators group change → rule 60154).

FIM only reports paths it monitors (get_agent_configuration → syscheck). An unmonitored persistence path produces no alert, so record it as a monitoring gap rather than "clean".

Evidence handling

Follow NIST 800-86-style practice even for remote data:

• Record for each artefact: source tool, agent id, timestamp of collection, and the inventory timestamp the agent reported.
• Quote exact values (paths, PIDs, hashes); don't paraphrase.
• Note anything that could be lost (volatile process/socket state) so humans can prioritize a proper acquisition before containment actions like reboot.
• Recommend isolation over kill/restart when forensic acquisition is still needed.

Output

• add_entities for suspicious processes, files, hashes, remote IPs (validated).
• link_mitre for confirmed techniques.
• add_finding titled Host forensics: <host> with: agent status, suspicious processes/ports table, persistence evidence, monitoring gaps, evidence log, verdict (clean / suspicious / compromised) and confidence. standard_refs: NIST-800-61r3.