<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Aadithchakkingal on Medium]]></title>
        <description><![CDATA[Stories by Aadithchakkingal on Medium]]></description>
        <link>https://medium.com/@aadithchakkingal10?source=rss-4311c0229b93------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*C6RbUarCX9CL4rOm</url>
            <title>Stories by Aadithchakkingal on Medium</title>
            <link>https://medium.com/@aadithchakkingal10?source=rss-4311c0229b93------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 06:36:43 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@aadithchakkingal10/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[SOC Architecture Part 2]]></title>
            <link>https://medium.com/@aadithchakkingal10/soc-architecture-part-2-15d7edf522d3?source=rss-4311c0229b93------2</link>
            <guid isPermaLink="false">https://medium.com/p/15d7edf522d3</guid>
            <category><![CDATA[incident-response]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[aws]]></category>
            <category><![CDATA[soc]]></category>
            <dc:creator><![CDATA[Aadithchakkingal]]></dc:creator>
            <pubDate>Thu, 03 Sep 2026 06:42:48 GMT</pubDate>
            <atom:updated>2026-09-03T06:42:48.154Z</atom:updated>
            <content:encoded><![CDATA[<h3>Testing an Autonomous Cloud SOC: Five Attack Scenarios, AI Triage, and Automated Containment (Part 2 of 2)</h3><p><em>Part 1 covered how I designed and built an AWS-hosted SOC pipeline — Wazuh, Shuffle, TheHive, and an AI triage layer, capped off with automated cloud containment. This part is about proving it actually works: five real attack scenarios, full evidentiary capture at every stage, and the honest gaps I found along the way.</em></p><h3>Why Evidence Discipline Matters</h3><p>Before getting into the scenarios, one editorial choice is worth calling out: the original project scope included a sixth scenario — Linux file-integrity/cron persistence, meant to trigger Wazuh’s FIM rule. I never captured attack execution, detection, or case screenshots for it in this pass. Rather than describe it based on the design documentation alone, I’ve excluded it from the results below and listed it as an open item for a future test cycle. A report should reflect what was actually evidenced, not what was planned.</p><h3>Scenario A — SSH Credential Access (Brute-Force Attempt)</h3><p><strong>Tactic/Technique:</strong> Credential Access — T1110.001 (Password Guessing), T1021.004 (SSH) <strong>Execution:</strong> Hydra dictionary attack from the Kali attacker against the Ubuntu target’s SSH service, 5 login tries across 4 parallel tasks <strong>Detection Rule:</strong> Wazuh Rule 5710 — “sshd: Attempt to login using a non-existent user” (Level 5) <strong>AI Classification:</strong> TP — severity reported as LOW in Slack, MEDIUM in TheHive</p><p><strong>Verdict: True Positive.</strong> The sampled Hydra run itself found zero valid passwords — it’s the <em>volume</em> of failed/invalid attempts, not a successful breach, that trips detection. Rule 5710 fired repeatedly for the invalid user, the firedtimes counter climbing with each attempt, and the alert correctly reached both Slack and TheHive.</p><p><strong>A worth-flagging discrepancy:</strong> the original project design described this scenario at Wazuh Level 10/High severity. The evidence actually captured shows Rule 5710 firing at Level 5, with the AI assigning LOW severity in the Slack card and MEDIUM in the TheHive case — a minor internal inconsistency between the two notification paths for the same alert. This report reflects what was captured rather than the original design figure.</p><h3>Scenario B — Windows Privilege Escalation &amp; Defense Evasion</h3><p><strong>Tactic/Technique:</strong> Privilege Escalation/Persistence — T1484 (Domain Policy Modification); Defense Evasion (log clearing) <strong>Execution:</strong> PowerShell on the Windows target — create user backdoor_user, add to local Administrators group, then clear the Windows Security event log (wevtutil.exe cl Security) <strong>Detection Rule:</strong> Wazuh Rule 60154 — “Administrators Group Changed” (Level 12), from Windows Security Event IDs 4732/4733 <strong>AI Classification:</strong> TP — Severity MEDIUM (both Slack and TheHive)</p><p><strong>Verdict: True Positive.</strong> A terminal grep of alerts.json confirmed both Event ID 4733 (member removed) and 4732 (member added) firing Rule 60154, Level 12, MITRE T1484 — and integrator.log confirmed HTTP 200 dispatch to Shuffle.</p><p><strong>Two discrepancies worth flagging here too:</strong> 1. The original design document labeled this scenario CRITICAL, but every captured triage artifact shows the AI assigning MEDIUM severity — despite the underlying Wazuh rule level (12) sitting in the platform’s own high/critical band. That’s the AI model under-scoring a genuinely high-impact privilege-escalation action. 2. The PowerShell sequence included a security-log-clearing step intended to trigger Rule 63103 (Defense Evasion), but no distinct Slack or TheHive artifact for that rule was captured in this pass — only the group-change event produced a visible downstream case. I’m treating this as a <em>possible</em> detection/notification gap rather than a confirmed one, since a follow-up verification specifically targeting Rule 63103 wasn’t performed.</p><h3>Scenario C — Linux File-Integrity/Cron Persistence (Not Evidenced)</h3><p>As noted above: this scenario appears in the original project design (a crontab-based reverse-shell injection intended to trigger Wazuh Rule 550) but no attack execution, detection, or case screenshots were captured for it during this reporting pass. It’s excluded from results and remains a recommended item for a future test cycle.</p><h3>Scenario D — Routine Administration (False-Positive Triage Validation)</h3><p><strong>Objective:</strong> validate that the AI triage layer correctly suppresses benign administrative activity rather than escalating every SIEM-correlated event. <strong>Execution:</strong> sudo systemctl status apache2 on the Ubuntu target, run by the local ubuntu user <strong>Detection Rule:</strong> Wazuh Rule 5402 — “Successful sudo to ROOT executed” (Level 3), MITRE T1548.003 <strong>AI Classification:</strong> FP — Severity LOW (both Slack and TheHive)</p><p><strong>Verdict: False Positive, correctly suppressed.</strong> This scenario matches the original project design exactly, including the resulting TheHive case number (#212). The AI reasoning explicitly noted no malicious file/network activity and no threat-intel hits — a clean demonstration of the triage layer functioning as intended: a SIEM-correlated event that would otherwise have generated an escalation was correctly down-classified with a clear evidentiary rationale, avoiding unnecessary analyst paging.</p><h3>Scenario E — Multi-Vector Threat Intelligence Stress Test</h3><p><strong>Objective:</strong> exercise the full threat-intelligence integration chain (AbuseIPDB + VirusTotal + URLScan) simultaneously via a single synthetic high-severity payload, using publicly routable indicators — a known Tor exit-node IP and the EICAR standard antivirus test file hash — rather than internal lab addresses. <strong>Tactic/Technique:</strong> Command and Control (T1071.001), Ingress Tool Transfer (T1105), Web Shell (T1505.003) <strong>Execution:</strong> direct REST/JSON injection into the Shuffle webhook via curl, simulating a Wazuh alert with rule level 15 <strong>Detection Rule:</strong> Synthetic Rule 100200 — “CRITICAL: Malicious C2 Communication, Exploit Payload &amp; Web Shell Dropped” (Level 15) <strong>AI Classification:</strong> TP — Severity HIGH</p><p><strong>Verdict: True Positive.</strong> The Slack notification showed AbuseIPDB 100%, VirusTotal 49/49 malicious detections, and 108 URLScan scans against the synthetic indicators. This scenario matches the original project design closely and confirms all three threat-intelligence integrations return usable enrichment data when supplied with valid public indicators — correctly driving the AI classification to TP/HIGH.</p><h3>Scenario F — AWS IAM Privilege Escalation with Automated Containment (NEW)</h3><p>This is the centerpiece of the update. <strong>Objective:</strong> validate the newly added cloud detection surface end to end — confirm a CloudTrail-sourced IAM privilege-escalation event is ingested and correlated by Wazuh, correctly classified as a critical true positive by the AI triage layer, and that the SOAR workflow’s new automated containment node revokes the over-privileged grant without human intervention, in addition to filing a TheHive case and posting a Slack alert.</p><p><strong>Tactic/Technique:</strong> Privilege Escalation — T1098 (Account Manipulation), T1078.004 (Valid Accounts: Cloud Accounts); also mapped by the AI/TheHive case to T1068 and T1484 <strong>Execution:</strong> from AWS CloudShell, the account root identity ran aws iam attach-role-policy --role-name soc-ssm-target-role --policy-arn arn:aws:iam::aws:policy/AdministratorAccess, granting the role full administrative privileges. Executed twice within the reporting window, from two different source IPs, both against the lab AWS account. <strong>Detection Rule:</strong> Custom Wazuh Rule 100102 — “AWS IAM/PAM: CRITICAL — Administrative Policy Attached” (Level 13), decoded from the CloudTrail AttachRolePolicy event via the aws-s3 module <strong>AI Classification:</strong> TP — Severity CRITICAL (TheHive case: SEVERITY:CRITICAL, TLP:AMBER, PAP:AMBER) <strong>Automated Response:</strong> Shuffle’s “AWS IAM Containment” node executed detach_role_policy against the role/policy pair extracted from the alert — status SUCCESS, HTTP 200 — removing AdministratorAccess from the role automatically.</p><p><strong>Verdict: True Positive — automated containment executed.</strong></p><p>Walking through the full chain:</p><ol><li><strong>Baseline</strong> — before the simulated attack, soc-ssm-target-role held only its two intended policies (soc-wazuh-cloudtrail-policy and AmazonSSMManagedInstanceCore).</li><li><strong>Attack execution</strong> — the attacker attached AWS-managed AdministratorAccess to the role from CloudShell, then re-listed the role’s policies to confirm the escalation succeeded — three policies now attached.</li><li><strong>Detection</strong> — the Wazuh manager’s aws-s3 module picked up the AttachRolePolicy event within its normal ~60-second fetch cadence. A Wazuh Discover query on rule.id:100102 showed 14 hits over the reporting window, reflecting repeated test executions during validation.</li><li><strong>SOAR orchestration</strong> — the alert was forwarded to Shuffle, enriched, and classified. The updated canvas routed the TP branch through the AWS IAM Containment node ahead of TheHive TP Alert.</li><li><strong>Automated containment</strong> — the containment node’s execution output confirmed: status: SUCCESS, action: detach_role_policy, role: soc-ssm-target-role, http_status: 200.</li><li><strong>Post-containment verification</strong> — a follow-up aws iam list-attached-role-policies call confirmed the AdministratorAccess grant was removed, matching the pre-attack baseline exactly — with no manual analyst action taken.</li></ol><p><strong>One real bug surfaced here too:</strong> the Slack AI-analysis card for this scenario showed the Groq model’s raw internal reasoning/scratchpad text (beginning “We must classify alert as TP or FP…”) instead of the clean, structured “AI Report” block produced for every other scenario. The underlying classification was still correct — confirmed cleanly in the TheHive case and by the containment action actually firing — so this looks like a JSON-extraction/regex issue in the Shuffle Python node for this specific alert shape, not a triage error. It’s flagged as an open item below.</p><h3>Detection Effectiveness Summary</h3><ul><li><strong>A — SSH Brute Force:</strong> Rule 5710 (Level 5) — TP/LOW–MEDIUM — no automated response — fully evidenced</li><li><strong>B — Windows PrivEsc:</strong> Rule 60154 (Level 12) — TP/MEDIUM — no automated response — fully evidenced; Rule 63103 gap noted</li><li><strong>C — Cron/FIM Persistence:</strong> Rule 550 — n/a — not evidenced, excluded</li><li><strong>D — Routine sudo (FP):</strong> Rule 5402 (Level 3) — FP/LOW — no automated response — fully evidenced</li><li><strong>E — Multi-Vector Stress Test:</strong> Rule 100200 (Level 15) — TP/HIGH — no automated response — fully evidenced</li><li><strong>F — AWS IAM PrivEsc (NEW):</strong> Rule 100102 (Level 13) — TP/CRITICAL — automated, policy detached (SUCCESS) — fully evidenced; AI output-parsing gap noted</li></ul><h3>Issues Encountered &amp; Resolution</h3><p><strong>Threat Intelligence API Exceptions on Private IOCs.</strong> Shuffle raised an unhandled exception whenever a threat-intelligence node was passed a private/internal indicator — AbuseIPDB and VirusTotal both return HTTP 400/422 on RFC 1918 private IPs or empty hash strings, which happens naturally with internal-only lab attack traffic. Fixed by: normalizing threat-intel handling defensively in the workflow, substituting publicly routable indicators for enrichment-chain testing (the Tor exit node and EICAR hash used in Scenario E), and scoping the Wazuh &lt;integration&gt; block to a specific rule_id whitelist to reduce noise and avoidable API failures.</p><p><strong>AI Response Parsing Gap on the New CloudTrail Alert Shape (NEW).</strong> As described above — the CloudTrail-derived alert JSON has a different field shape than the host-based Wazuh alerts the extraction regex was originally built against. Classification/severity extraction still worked correctly downstream, but the raw model output wasn’t filtered before reaching the Slack formatter. Not yet resolved — the fix is to harden the Python extraction node to strip non-JSON scratchpad content regardless of alert source.</p><h3>What This Confirms — and What’s Still Open</h3><p><strong>Findings:</strong> — The end-to-end pipeline functions correctly across all five evidenced scenarios, with alerts consistently reaching TheHive and Slack within the same workflow run. — The AI triage layer successfully distinguishes malicious from benign activity — Scenario D confirms correct FP suppression, while A, B, E, and F confirm correct TP escalation with MITRE ATT&amp;CK mapping. — Cloud control-plane detection and automated containment are now both validated end to end: CloudTrail events ingest within roughly a minute, correlate correctly, and a confirmed critical privilege-escalation TP triggers an unattended detach_role_policy action that measurably reverts the account to its pre-attack state — in the same run that produces the case and Slack alert. - Two internal AI severity-consistency gaps (Scenarios A and B) and one new AI output-formatting gap (Scenario F) don’t affect underlying TP/FP classification accuracy, but they do affect the cleanliness of analyst-facing output.</p><p><strong>Recommendations:</strong> — Standardize severity output across all SOAR notification branches to eliminate the Slack/TheHive mismatch seen in Scenario A. — Introduce a severity floor tied to the underlying Wazuh rule level for account-manipulation and persistence-technique alerts, so AI-assigned severity can’t fall below the platform’s own risk tier. — Add or verify a dedicated detection path for Rule 63103 to close the possible defense-evasion visibility gap from Scenario B. — Execute the outstanding cron/FIM persistence scenario (Rule 550) with full evidence capture. — Harden the Shuffle AI-response extraction node so raw model reasoning can never reach the Slack card, regardless of alert source. — Extend automated containment coverage cautiously — a small number of additional narrow, reversible response actions behind the same TP/CRITICAL gating logic, while keeping anything destructive behind a human-approval gate.</p><p><strong>Roadmap items now partially implemented by this update:</strong> automated SOAR response actions (IAM policy detachment is live; host isolation and security-group blocking remain future work) and broader cloud-log coverage (CloudTrail is onboarded; VPC Flow Logs and GuardDuty findings are next).</p><p><strong>Still on the roadmap:</strong> EDR integration for process-tree context and host-side isolation, a labeled evaluation set to benchmark the AI triage prompt over time, a MITRE ATT&amp;CK coverage heatmap generated from TheHive case tags, a threat-hunting dashboard for proactive querying, multi-cloud/hybrid extension, red-team automation to regression-test detection coverage after any rule or prompt change, and cost/performance tuning as alert volume and CloudTrail polling scale up.</p><p><em>This project — paired with the separate Wazuh and ELK SIEM labs I’ve documented — represents my progression from single-tool detection engineering toward a full autonomous SOC: SIEM, SOAR, AI-assisted triage, case management, and now automated cloud response, all working together end to end. Questions on the architecture or the detection logic — drop them in the comments.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=15d7edf522d3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[SOC Architecture Part 1]]></title>
            <link>https://medium.com/@aadithchakkingal10/soc-architecture-part-1-1e1c5fddbeae?source=rss-4311c0229b93------2</link>
            <guid isPermaLink="false">https://medium.com/p/1e1c5fddbeae</guid>
            <category><![CDATA[cloud-security]]></category>
            <category><![CDATA[soar]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[aws]]></category>
            <category><![CDATA[siem]]></category>
            <dc:creator><![CDATA[Aadithchakkingal]]></dc:creator>
            <pubDate>Thu, 03 Sep 2026 06:31:20 GMT</pubDate>
            <atom:updated>2026-09-03T06:31:20.879Z</atom:updated>
            <content:encoded><![CDATA[<h3>Building an Autonomous Cloud SOC: Architecture, SIEM, and SOAR Design (Part 1 of 2)</h3><p><em>How I designed and deployed an AWS-hosted Security Operations Center that detects, enriches, and triages threats with an AI analyst in the loop — and, in the latest update, automatically contains a cloud privilege-escalation attack without a human touching a keyboard.</em></p><h3>The Goal</h3><p>Most home-lab SOC projects stop at “SIEM detects an alert.” I wanted to go further and build the full loop a real SOC runs: telemetry → detection → enrichment → AI-assisted triage → case management → analyst notification — and, as of this update, <strong>automated response</strong>.</p><p>This lab integrates real-time host telemetry from Linux and Windows endpoints <em>and</em> cloud-control-plane telemetry from AWS CloudTrail, all feeding a centralized SIEM (Wazuh), a SOAR orchestration engine (Shuffle), multi-source threat intelligence enrichment (VirusTotal, AbuseIPDB, URLScan.io), an AI reasoning layer for alert triage (Groq LLaMA), a case management platform (TheHive 5), and — new in this update — an automated cloud-response capability that revokes an over-privileged IAM policy the moment a critical true positive is confirmed.</p><p><strong>What’s new in this update:</strong> 1. <strong>AWS CloudTrail ingestion</strong> — the Wazuh manager now polls a dedicated CloudTrail log bucket via the aws-s3 module, extending detection from host telemetry to the AWS cloud control plane. 2. <strong>Automated SOAR containment</strong> — the Shuffle workflow gained an execute_python action node that detaches an over-privileged IAM policy grant on confirmed critical true positives, without human intervention.</p><p>Five attack/activity scenarios were executed end-to-end with full evidentiary capture across this pipeline. A sixth scenario (Linux file-integrity/cron persistence) was scoped in the original design but not executed with supporting evidence in this pass — more on why that distinction matters in Part 2.</p><h3>Architecture</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oxIFdOsqS0_Ri9GbWG7Vmg.png" /></figure><p>Autonomous Cloud SOC architecture and alert data flow</p><p>The lab runs inside a single dedicated AWS VPC, isolating all lab traffic from production infrastructure. Telemetry flows one direction — from endpoints and the AWS control plane toward the SIEM. Alerting and response flow the other direction — from the SOAR engine out to threat-intel APIs, the AWS IAM API for containment, the case management system, and the analyst notification channel.</p><p><strong>High-level data flow:</strong> — Attacker (Kali) → Windows Target + Ubuntu Target (attack traffic) — Windows Target + Ubuntu Target → Wazuh SIEM Manager (agent telemetry) — AWS CloudTrail (soc-cloudtrail-logs S3 bucket) → Wazuh SIEM Manager (~60s poll cycle, aws-s3 module) - Wazuh SIEM Manager → Shuffle SOAR Engine (webhook, alert normalization) - Shuffle → AbuseIPDB, VirusTotal, URLScan.io (parallel enrichment) - Shuffle → AI Threat Analysis (Groq API, LLaMA 3.3 70B) - AI verdict → TP/FP Router - <strong>FP path</strong> → TheHive FP Case (Low severity, no paging) - <strong>TP path (other rules)</strong> → TheHive TP Alert (High/Critical) → Slack #soc-alerts - <strong>TP path (critical IAM rule)</strong> → AWS IAM Containment (detach_role_policy via boto3) → TheHive TP Alert → Slack</p><p>For host-based detections, the branch is straightforward case-management and notification. For the new cloud IAM scenario, a confirmed critical true positive also triggers an automated containment action against the AWS IAM API <em>before</em> the case is even filed — so the offending privilege grant is revoked within the same workflow run that produces the alert.</p><h3>Deployed AWS Compute &amp; Network Layout</h3><p>Four EC2 instances make up the lab, all confirmed running with passing status checks: the Wazuh manager host, the Windows target, the Kali attacker box, and the Ubuntu target.</p><p>The environment runs inside a dedicated VPC (soc-lab-vpc, CIDR 10.0.0.0/16) with two subnets in us-east-1a — a public subnet and a private subnet — three route tables, and both an Internet Gateway and a NAT Gateway. That combination enables controlled outbound access for threat-intelligence lookups while keeping target hosts off the public internet directly.</p><p>Two Wazuh agents report as Active to the manager, one per operating system family in scope: an Ubuntu 24.04.4 LTS agent and a Windows Server 2022 Datacenter agent.</p><h3>Tools and Technologies</h3><ul><li><strong>SIEM — Wazuh</strong> (Manager, Indexer, Dashboard): central log collection, correlation rule engine, file-integrity monitoring, agent management</li><li><strong>SOAR — Shuffle</strong>: workflow orchestration — webhook ingestion, enrichment fan-out, AI triage call, conditional branching, case/notification dispatch, automated response</li><li><strong>Case Management — TheHive 5</strong>: alert and case tracking, severity tagging, investigation workflow, SOC operations dashboard</li><li><strong>AI Triage — Groq API (LLaMA 3.3 70B)</strong>: natural-language reasoning over alert + threat-intel context to classify TP/FP and assign severity</li><li><strong>Threat Intelligence — AbuseIPDB API v2, VirusTotal API v3, URLScan.io API</strong>: IP reputation, file hash reputation, URL/domain scanning</li><li><strong>Notification — Slack</strong>: real-time analyst-facing alert cards in #soc-alerts</li><li><strong>Host Telemetry — Sysmon + Windows Event Forwarding</strong> (Windows), <strong>Apache2 access logs, sshd/auth logs, syscheck FIM</strong> (Linux)</li><li><strong>Cloud Telemetry (NEW) — AWS CloudTrail</strong> via wazuh-modulesd:aws-s3: cloud control-plane visibility, IAM/API-call/management-event telemetry</li><li><strong>Automated Response (NEW) — Shuffle </strong><strong>execute_python (boto3)</strong>: “AWS IAM Containment” node, automatically detaches an over-privileged IAM policy on confirmed critical true positives</li><li><strong>Attack Simulation — Hydra, curl, native PowerShell/bash, AWS CLI/CloudShell</strong>: credential brute-forcing, direct webhook/API payload injection, native OS commands, IAM policy attachment for cloud privilege escalation</li><li><strong>Cloud Platform — AWS</strong> (EC2, VPC, Security Groups, IAM, CloudTrail): isolated compute and network hosting, plus IAM/CloudTrail for the cloud detection &amp; response scenario</li></ul><h3>Network Security Configuration</h3><p>Five AWS Security Groups enforce least-privilege network segmentation within the VPC. Management-plane access (RDP, HTTPS, SSH, the TheHive/Wazuh dashboards) is restricted to a single administrator source IP, while inter-host traffic between the attacker, targets, and SIEM is scoped to the internal 10.0.0.0/16 range only.</p><ul><li><strong>soc-siem-sg</strong> — allows agents to send telemetry to Wazuh: TCP 1514–1515 (agent traffic), 55000 (API), 8080, 9000, all from 10.0.0.0/16</li><li><strong>soc-attacker-sg</strong> — used by the Kali Linux machine: SSH (22) from the admin IP only</li><li><strong>soc-targets-sg</strong> — applied to target machines: all traffic from 10.0.0.0/16 and from soc-attacker-sg; SSH and RDP from admin IP; TCP 8080 from 10.0.0.0/16</li><li><strong>soc-admin-access-sg</strong> — restricts management access to the admin IP: RDP, HTTPS/443 (Wazuh Dashboard), TheHive UI ports, SSH — all scoped to a single /32</li></ul><h3>Configuration Steps</h3><p><strong>Wazuh Manager → SOAR Webhook Integration</strong></p><p>Configured inside ossec.conf on the manager host. This forwards any alert matching the listed rule IDs (severity level 3+) to the Shuffle webhook as JSON:</p><pre>&lt;integration&gt;<br>  &lt;name&gt;custom-shuffle&lt;/name&gt;<br>  &lt;hook_url&gt;https://shuffler.io/api/v1/hooks/webhook_ced5eb82-2aa3-4913-baf0-7bd162ab3ffb&lt;/hook_url&gt;<br>  &lt;rule_id&gt;550, 554, 5402, 5710, 5712, 5716, 31101, 31103, 31106, 31508, 31533, 60106, 60109, 60154, 100102&lt;/rule_id&gt;<br>  &lt;level&gt;3&lt;/level&gt;<br>  &lt;alert_format&gt;json&lt;/alert_format&gt;<br>&lt;/integration&gt;</pre><p>Rule ID 100102 (“AWS IAM/PAM: CRITICAL — Administrative Policy Attached”, Level 13) was added to this whitelist so CloudTrail-sourced IAM privilege-escalation events get forwarded to the SOAR pipeline alongside the original host-based rule set.</p><p><strong>Ubuntu Agent → Log Collection &amp; FIM</strong></p><p>Apache access logs are ingested for web-layer visibility, and syscheck monitors the web root in real time plus two high-value system files periodically:</p><pre>&lt;!-- Apache Log Monitoring --&gt;<br>&lt;localfile&gt;<br>  &lt;log_format&gt;apache&lt;/log_format&gt;<br>  &lt;location&gt;/var/log/apache2/access.log&lt;/location&gt;<br>&lt;/localfile&gt;</pre><pre>&lt;!-- Real-Time Webroot File Integrity Monitoring --&gt;<br>&lt;syscheck&gt;<br>  &lt;directories realtime=&quot;yes&quot; check_all=&quot;yes&quot;&gt;/var/www/html&lt;/directories&gt;<br>  &lt;directories check_all=&quot;yes&quot;&gt;/etc/crontab,/etc/sudoers&lt;/directories&gt;<br>&lt;/syscheck&gt;</pre><p><strong>Wazuh Manager → AWS CloudTrail Ingestion (NEW)</strong></p><p>Configured on the manager’s aws-s3 wodle, this polls the CloudTrail bucket roughly once per minute, fetching new CloudTrail objects and decoding them into wazuh-alerts-* under the data.aws.* field namespace. The manager’s ossec.log shows a continuous cycle: “Starting fetching of logs” → “Executing Bucket Analysis” → “Fetching logs finished,” repeating at ~60-second intervals.</p><p><strong>Shuffle → AI Triage Prompt (Groq LLaMA Node)</strong></p><p>The AI Threat Analysis node calls the Groq API with a system prompt engineered to reduce false-positive bias. Two details matter here: the prompt explicitly states that a private/internal source IP must never <em>by itself</em> determine a TP/FP classification, and that authorized administrative or scanning activity should default to FP even when Wazuh correctly detected the event. Severity criteria (Low/Medium/High/Critical) are tied to evidence strength rather than raw Wazuh rule level alone:</p><pre>POST https://api.groq.com/openai/v1/chat/completions<br>{<br>  &quot;model&quot;: &quot;llama-3.3-70b-versatile&quot;,<br>  &quot;messages&quot;: [<br>    {&quot;role&quot;: &quot;system&quot;, &quot;content&quot;: &quot;You are a SOC AI Analyst. Analyze all available<br>    security alert telemetry and threat intelligence. ... Classify the alert as TP or FP<br>    based on whether the activity is malicious, unauthorized, suspicious, security-relevant,<br>    or demonstrably benign. ... Private RFC1918 addresses ... must be evaluated using the<br>    same behavioral security criteria as public addresses. Do not assume malicious intent<br>    without supporting evidence. ...&quot;}<br>  ]<br>}</pre><p><strong>Shuffle → Dynamic TP/FP Branching Logic</strong></p><p>A Python execution node parses the AI’s JSON response via regex to extract classification and severity, feeding a TP_FP_Router condition node:</p><ul><li><strong>True Positive path</strong> — condition: ai_threat_analysis.body.classification EQUALS &quot;TP&quot;. Outcome: case created in TheHive TP Alert with High/Critical severity and a Slack broadcast.</li><li><strong>False Positive path</strong> — condition: ai_threat_analysis.body.classification EQUALS &quot;FP&quot;. Outcome: low-severity tracking case in TheHive FP Case, tagged FP, no analyst paging.</li></ul><p><strong>Shuffle → AWS IAM Containment Node (NEW)</strong></p><p>For the new AWS IAM/PAM critical rule family, the TP branch now routes through an additional execute_python node before reaching TheHive TP Alert. This node uses boto3 with a dedicated, scoped SOAR IAM identity to call detach_role_policy against the role and policy ARN extracted from the CloudTrail event — revoking the over-privileged grant automatically.</p><h3>Log Integrations</h3><ul><li><strong>Ubuntu sshd/auth.log</strong> → Wazuh localfile + built-in sshd decoder → detects invalid-user and brute-force authentication attempts (Rules 5710/5712)</li><li><strong>Ubuntu sudo/journald</strong> → built-in sudo decoder → tracks privileged command execution (Rule 5402)</li><li><strong>Ubuntu Apache2 access.log</strong> → web server request visibility for the hosted backdoor scenario</li><li><strong>Ubuntu </strong><strong>/etc/crontab, </strong><strong>/etc/sudoers, </strong><strong>/var/www/html</strong> → syscheck (real-time + periodic) → file-integrity monitoring for persistence and web-shell tampering (Rule 550)</li><li><strong>Windows Security Event Log</strong> → Sysmon + Windows Event Forwarder → account/group management events (Event IDs 4732/4733) and log-clearing activity (Event ID 1102 / Rule 63103)</li><li><strong>Wazuh Manager </strong><strong>integrator.log</strong> → confirms successful HTTP 200 dispatch of each qualifying alert to the Shuffle webhook</li><li><strong>AWS CloudTrail (NEW)</strong> → wazuh-modulesd:aws-s3 wodle → cloud control-plane visibility, detects IAM privilege-escalation actions such as AttachRolePolicy (Rule 100102)</li></ul><h3>SOAR Workflow Alignment</h3><p>The Shuffle workflow (Wazuh_To_TheHive) implements the full detect-enrich-triage-respond chain as a single canvas: a webhook trigger receives normalized Wazuh alerts (including CloudTrail-sourced alerts), fans them out in parallel to three threat intelligence nodes, merges the enrichment results into the AI Threat Analysis node, and branches on the AI’s verdict. For the new AWS IAM/PAM critical rule family, the TP branch now passes through the AWS IAM Containment node before reaching one of the two TheHive endpoints and notifying Slack.</p><p>Over the last three months, the TheHive SOC Operations Dashboard shows 182 total cases (159 High-severity FP tracking cases, 23 Low) and 24 True Positive alerts (18 Medium, 3 Low, 3 High) — a snapshot taken just before this update’s Scenario F was added.</p><p>The #soc-alerts Slack channel serves as the analyst-facing feed, receiving both FP triage summaries and CRITICAL true-positive alert cards in the same channel, differentiated by classification and severity fields.</p><p><strong>In Part 2</strong>, I walk through all five attack scenarios executed against this pipeline end-to-end — SSH brute force, Windows privilege escalation, a deliberate false-positive test, a multi-vector threat-intelligence stress test, and the new AWS IAM privilege-escalation scenario with automated containment — plus the real issues I hit along the way and what I’d fix next.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1e1c5fddbeae" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[ELK SIEM Lab]]></title>
            <link>https://medium.com/@aadithchakkingal10/elk-siem-lab-006de3e5bf23?source=rss-4311c0229b93------2</link>
            <guid isPermaLink="false">https://medium.com/p/006de3e5bf23</guid>
            <category><![CDATA[elasticsearch]]></category>
            <category><![CDATA[kibana]]></category>
            <category><![CDATA[siem]]></category>
            <category><![CDATA[logstash]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <dc:creator><![CDATA[Aadithchakkingal]]></dc:creator>
            <pubDate>Fri, 31 Jul 2026 10:03:30 GMT</pubDate>
            <atom:updated>2026-07-31T10:03:30.674Z</atom:updated>
            <content:encoded><![CDATA[<h3>Building an ELK Stack SIEM Pipeline: From Apache Logs to Live Detection Rules</h3><p><em>How I built a full ingestion-to-alerting pipeline with Elasticsearch, Logstash, Kibana, and Filebeat — and the two real configuration bugs I hit along the way.</em></p><h3>The Goal</h3><p>A SIEM is only as good as the pipeline feeding it. For this lab, I wanted to go beyond “install ELK and look at a dashboard” and actually build the full data path: a real log-generating service, structured parsing, indexed storage, visualization, and — the part that actually matters for detection engineering — working alert rules validated against live traffic.</p><p><strong>What this lab covers:</strong> — Apache HTTP Server as the log source — Filebeat → Logstash → Elasticsearch → Kibana ingestion pipeline — A custom grok pipeline parsing Apache Combined Log Format into structured fields — A Kibana dashboard with three panels — Three Elasticsearch-query detection rules, all validated with live alerts</p><h3>Lab Environment</h3><ul><li><strong>Stack:</strong> Elastic Stack 8.15.0 (Elasticsearch, Logstash, Kibana, Filebeat)</li><li><strong>Deployment:</strong> Custom docker-compose.yml, single-node, security disabled for lab use</li><li><strong>Log source:</strong> Apache 2.4.58, Combined Log Format at /var/log/apache2/access.log</li><li><strong>Index pattern:</strong> apache-logs-YYYY.MM.dd</li></ul><h3>Architecture</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zWztBeGHI4NpfZ4_AMo4sQ.png" /><figcaption>elk-architecture-diagram</figcaption></figure><p>ELK stack SIEM ingestion and alerting pipeline</p><p><em>Apache access log entries are tailed by Filebeat and shipped to Logstash, which grok-parses each request into structured fields before indexing into Elasticsearch. Kibana queries that same index to power the dashboard and run three alerting rules.</em></p><h3>1. The Ingestion Path</h3><p>Apache was installed from the Ubuntu repos as the log-generating service. Filebeat then tailed the access log using the filestream input and forwarded events to Logstash on port 5044 — deliberately <em>not</em> writing directly to Elasticsearch, so all parsing logic stays centralized in one place:</p><pre>filebeat.inputs:<br>  - type: filestream<br>    id: apache-access<br>    enabled: true<br>    paths:<br>      - /var/log/apache2/access.log</pre><pre>output.logstash:<br>  hosts: [&quot;logstash:5044&quot;]</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Bsq9nC24yR1etwrxZo331w.png" /></figure><p>Logstash’s job was to grok-parse each raw Apache line into structured fields (HTTP method, status code, URL, user agent, source address) and normalize the timestamp before indexing:</p><pre>input {<br>  file {<br>    path =&gt; &quot;/var/log/apache2/access.log&quot;<br>    start_position =&gt; &quot;beginning&quot;<br>    sincedb_path =&gt; &quot;/dev/null&quot;<br>  }<br>}</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AbEb2hVt0jtRJe1Ogg8qrw.png" /><figcaption><em>Logstash container logs: a parsed Apache request showing extracted fields.</em></figcaption></figure><pre>filter {<br>  grok {<br>    match =&gt; { &quot;message&quot; =&gt; &quot;%{COMBINEDAPACHELOG}&quot; }<br>  }<br>  date {<br>    match =&gt; [ &quot;timestamp&quot;, &quot;dd/MMM/yyyy:H:m:s Z&quot; ]<br>  }<br>}</pre><pre>output {<br>  elasticsearch {<br>    hosts =&gt; [&quot;http://elasticsearch:9200&quot;]<br>    index =&gt; &quot;apache-logs-%{+YYYY.MM.dd}&quot;<br>  }<br>  stdout { codec =&gt; rubydebug }<br>}</pre><p>After bringing up the four-container stack (elasticsearch, logstash, kibana, filebeat), a direct query against Elasticsearch’s _cat/indices API confirmed the daily index existed and was populated with parsed documents — end-to-end ingestion verified before touching Kibana at all.</p><h3>2. Kibana: Data View, Discover, Dashboard</h3><p>A Data View named “Apache logs” (index pattern apache-logs-*, timestamp field @timestamp) exposed the parsed data to Kibana. Discover confirmed every expected field was present and correctly typed — http.request.method, http.response.status_code, user_agent.original, url.original, source.address.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cjdy_s7hqIzt0zHkRGRdAA.png" /><figcaption>Kibana Discover: 24 Apache log documents with all expected parsed fields</figcaption></figure><p>From there I built an “Apache Security Dashboard” incrementally, one panel at a time: — Top HTTP response status codes — Top requested URLs — Top client IPs</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_0etvw3gLU5j0xSR18nVKA.png" /><figcaption>Completed “Apache Security Dashboard”: status codes, top URLs, top client IPs.</figcaption></figure><h3>3. Three Detection Rules — Not Just One</h3><p>The assignment called for a two-rule minimum; I built four, spanning reconnaissance, volumetric, and tool-signature detection categories:</p><ul><li>****Sensitive URL Access Detection<strong>:</strong> Alerts on requests to /admin, /login, /phpmyadmin — Reconnaissance / unauthorized access</li><li>****High Request Volume Detection<strong>:</strong> Alerts when request count from a source exceeds a threshold in a short window — Scanning / brute-force / DoS-like burst</li><li>****Suspicious User-Agent Detection<strong>:</strong> Query-string wildcard match against user_agent.original for tool signatures like *curl*, *sqlmap*, *nikto* — Automated recon / pentest tool usage</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8h9Lc628oEEE4JKj3VviCg.png" /><figcaption>Three Detection Rules</figcaption></figure><p>All three ran on a 1-minute interval and reached a 100% execution success ratio with zero failures once correctly configured.</p><h3>Two Real Bugs (Not Staged)</h3><p><strong>1. Kibana Alerting needs an encryption key before it works at all.</strong> The first attempt to create a rule was blocked with <em>“Additional setup required: you must configure an encryption key to use Alerting.”</em> This is resolved by setting xpack.encryptedSavedObjects.encryptionKey in the Kibana config — a one-time step that’s easy to miss and worth including in any ELK SIEM deployment runbook.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-z4Wgryg_9-3rqq_8IZzEQ.png" /><figcaption><em>Kibana Alerting encryption-key blocker</em></figcaption></figure><p><strong>2. </strong><strong>match_phrase doesn’t support wildcards.</strong> The Suspicious User-Agent rule was first built using a match_phrase query, which silently failed to match substrings inside User-Agent strings. Switching to a query_string query with explicit wildcards fixed it:</p><pre>{<br>  &quot;query_string&quot;: {<br>    &quot;fields&quot;: [&quot;user_agent.original&quot;],<br>    &quot;query&quot;: &quot;*curl*&quot;<br>  }<br>}</pre><p>Validated by firing requests with a range of suspicious User-Agents:</p><pre>curl -A &quot;curl-test&quot; http://localhost/<br>curl -A &quot;sqlmap&quot; http://localhost/<br>curl -A &quot;nikto&quot; http://localhost/<br>curl -A &quot;python-requests&quot; http://localhost/</pre><p>The rule correctly transitioned from 0 to 1 active alert. The High Request Volume rule went further, demonstrating a full <strong>Active → Recovered</strong> lifecycle as traffic volume rose and fell — proof that Kibana’s alerting tracks state over time rather than firing once and staying stuck open.</p><h3>Findings</h3><ul><li>The pipeline (Apache → Filebeat → Logstash → Elasticsearch → Kibana) worked cleanly once Filebeat was introduced as a dedicated shipper, rather than having Logstash read the file directly.</li><li>Free-text/keyword searches and rule queries <strong>do not</strong> automatically support wildcards — an important distinction when writing detection logic on text fields.</li><li>No alerts were routed anywhere outside the Kibana Alerts UI — a real gap for production use.</li></ul><h3>Recommendations for Anyone Building This Next</h3><ul><li>Wire up an external connector (email, Slack, webhook) so alerts don’t require someone watching the dashboard.</li><li>Enable xpack.security.enabled=true with TLS before running this outside an isolated lab.</li><li>Add an aggregation-based rule on source.address to catch distributed bursts more precisely than a flat document-count threshold.</li><li>Expand grok parsing to more log sources (mod_security, Windows Sysmon, firewall logs) to broaden coverage beyond web-layer events.</li><li>Add an ILM policy on apache-logs-* before running this long-term, to control index growth.</li><li>Baseline normal traffic before trusting the 404/volume thresholds outside a lab context — as built, they’d generate false positives on any real production site.</li></ul><p><em>This lab, paired with a separate Wazuh SIEM deployment I documented, rounds out a home-lab detection engineering baseline covering host-based and web-layer monitoring. Questions or suggestions on the detection logic — drop them in the comments.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=006de3e5bf23" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Wazuh SIEM Lab]]></title>
            <link>https://medium.com/@aadithchakkingal10/wazuh-siem-lab-e2df1421a4a9?source=rss-4311c0229b93------2</link>
            <guid isPermaLink="false">https://medium.com/p/e2df1421a4a9</guid>
            <category><![CDATA[wazuh]]></category>
            <category><![CDATA[siem]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[docker]]></category>
            <category><![CDATA[linux]]></category>
            <dc:creator><![CDATA[Aadithchakkingal]]></dc:creator>
            <pubDate>Fri, 31 Jul 2026 08:57:16 GMT</pubDate>
            <atom:updated>2026-07-31T08:57:16.380Z</atom:updated>
            <content:encoded><![CDATA[<h3>Deploying Wazuh SIEM From Scratch: Agent Onboarding, Log Monitoring, and Detection Validation</h3><p><em>A hands-on home-lab writeup covering Docker deployment, endpoint enrollment, and four core SIEM detection capabilities — from the perspective of a SOC Analyst candidate building practical, evidence-backed skills.</em></p><h3>Why I Built This</h3><p>Reading about SIEM concepts only gets you so far. To actually understand what a SOC analyst does day to day — deploying detection tooling, generating test alerts, and confirming they fire correctly — you have to stand up the platform yourself and break things on purpose. This lab documents exactly that: a full Wazuh deployment, one enrolled Linux endpoint, and validation of four distinct monitoring capabilities.</p><p><strong>What this lab covers:</strong> — Installing Wazuh 4.14.6 (manager, indexer, dashboard) as a single-node Docker deployment — Enrolling an Ubuntu 24.04 endpoint as a monitored agent — Validating SSH/PAM authentication log monitoring — Validating File Integrity Monitoring (FIM) — Confirming Security Configuration Assessment (SCA) against the CIS Ubuntu 24.04 benchmark — Confirming software inventory tracking via Syscollector</p><h3>Lab Environment</h3><p>Component Detail Host OS Ubuntu (Docker 29.6.2, Compose v5.3.1) Wazuh version 4.14.6, single-node profile Deployment source Official wazuh-docker repo, tag v4.14.6 Agent OS Ubuntu 24.04.4 LTS</p><h3>Architecture</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AlK7Id4YYYrj9lR-DPf0Og.png" /><figcaption>wazuh-architecture-diagram</figcaption></figure><p>Wazuh SIEM deployment architecture</p><p><em>The Linux endpoint’s Wazuh agent forwards authentication logs, file integrity events, SCA scan results, and package inventory to the manager over port 1514. The manager writes to the indexer, and the dashboard queries the indexer for visualization and alerting — all internal traffic secured over TLS.</em></p><h3>1. Installation: Why Certificate Generation Comes First</h3><p>Wazuh’s indexer, Filebeat, and dashboard components all communicate over TLS, so before anything else I cloned the official wazuh-docker repo and pinned it to the stable v4.14.6 tag — keeping manager, indexer, dashboard, and agent versions all matched to avoid the classic version-mismatch headache.</p><pre>git clone https://github.com/wazuh/wazuh-docker.git<br>cd wazuh-docker<br>git checkout v4.14.6<br>cd single-node</pre><p>The certificates (root CA, admin, indexer, Filebeat, dashboard) were generated using the bundled cert-generator container <strong>before</strong> the first docker compose up. This ordering matters — skip it, and the stack comes up with broken internal TLS trust:</p><pre>docker compose -f generate-indexer-certs.yml run --rm generator</pre><p>With certs in place, pulling each image individually (to confirm clean downloads) and bringing the stack up in detached mode completed the setup:</p><pre>docker compose pull wazuh.indexer<br>docker compose pull wazuh.manager<br>docker compose pull wazuh.dashboard<br>docker compose up -d</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7wEaWcWeig2c8j__UGHdgg.png" /><figcaption><em>docker compose up -d: all three containers (manager, indexer, dashboard) started successfully.</em></figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ITJoFaVn-68j6rPaykhudw.png" /><figcaption>Wazuh Overview dashboard immediately after deployment, prior to agent enrollment</figcaption></figure><p>All three containers — manager, indexer, dashboard — started successfully, and the dashboard was reachable at https://localhost, confirming zero agents connected at this stage.</p><h3>2. Enrolling a Linux Agent</h3><p>Wazuh’s dashboard has a guided “Deploy new agent” wizard that generates the correct install command for your OS and manager address. For this lab, that meant a .deb package installed with the manager IP passed as an environment variable:</p><pre>wget https://packages.wazuh.com/4.x/apt/pool/main/w/wazuh-agent/wazuh-agent_4.14.6-1_amd64.deb \<br>  &amp;&amp; sudo WAZUH_MANAGER=&#39;&lt;MANAGER_IP&gt;&#39; dpkg -i ./wazuh-agent_4.14.6-1_amd64.deb</pre><pre>sudo systemctl daemon-reload<br>sudo systemctl enable wazuh-agent<br>sudo systemctl start wazuh-agent</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GfqtKhrOL7sPzRKdzFdfiQ.png" /><figcaption>Deploy New Agent wizard: Linux DEB amd64 package selected.</figcaption></figure><p>On restart, the agent log confirmed the SCA module and Syscollector both started, and the agent requested and received its enrollment key automatically over port 1514/tcp — no manual key exchange needed. The Endpoints view in the dashboard confirmed the agent as <strong>Active</strong>, running version 4.14.6.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*V3ZHGC1b_3JlnqDHNwSs_A.png" /><figcaption>Wazuh Endpoints page: agent shown as Active</figcaption></figure><h3>3. Detection Validation</h3><h3>SSH / Authentication Monitoring</h3><p>To trigger PAM and sshd failure events, I ran repeated SSH login attempts against the agent using an invalid username and deliberately wrong passwords. The Threat Hunting dashboard picked up <strong>6 total alerts</strong>, all authentication-related — individual sshd non-existent-user attempts, a PAM login failure, and a syslog correlation alert for repeated password failures (rule 2502, severity level 10).</p><p>That correlation alert is worth calling out: rather than surfacing five or six separate low-severity sshd events, Wazuh rolled them into a single higher-severity syslog alert — which is far more actionable for an analyst triaging a queue than a wall of individual level-5 hits.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OvoNXnW-CVJVHgC0UqagyA.png" /><figcaption><em>Events: 6 hits — rule IDs 5710, 5503, 2502</em>)</figcaption></figure><h3>File Integrity Monitoring (FIM)</h3><p>I created, appended to, and deleted a test file on a monitored path:</p><pre>sudo touch /etc/wazuh_test2<br>echo &quot;Hello again&quot; | sudo tee -a /etc/wazuh_test2<br>sudo rm /etc/wazuh_test2</pre><p>All three lifecycle events (added, modified, deleted) showed up correctly in the FIM Events view — 3 hits, exactly as expected.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4FrpxuRuLD58rgT6l3Skrg.png" /><figcaption>FIM Events: 3 hits recorded for /etc/wazuh_test2</figcaption></figure><h3>Security Configuration Assessment (SCA)</h3><p>No extra configuration needed here — the agent log confirmed SCA automatically loaded the CIS Ubuntu 24.04 benchmark policy and ran a full evaluation on agent start, giving immediate baseline hardening visibility.</p><h3>Software Inventory (Syscollector)</h3><p>Installing a new package (cowsay) surfaced an interesting gotcha: searching the Events view for the literal word “package” returned nothing. That’s because the field isn’t free text — it’s structured data (data.package, data.dpkg_status, decoder.name). Once I searched the structured fields directly, the install showed up cleanly: decoder.name: dpkg-decoder, data.package: cowsay, data.dpkg_status: install.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*sFd7ySIwjq18ymnV0ctdmA.png" /><figcaption><em>Document detail view confirming the cowsay installation was indexed via the dpkg-decoder.</em></figcaption></figure><p><strong>Lesson learned:</strong> when building detections or dashboards around software inventory, remember Syscollector data lives in structured fields, not a generic keyword index.</p><h3>Summary of Detections</h3><p>Test Detection Source Result SSH brute-force / invalid user sshd, PAM, syslog 6 alerts generated FIM create/modify/delete syscheck 3 hits Package install (cowsay) Syscollector Indexed correctly under agent.id SCA (CIS Ubuntu 24.04) sca module Ran automatically on start</p><h3>Key Takeaways</h3><ul><li><strong>Certificate generation is a hard prerequisite</strong> — always run it before the first docker compose up.</li><li><strong>Agent enrollment via the dashboard wizard is fast and reliable</strong> — automatic key exchange, no manual steps.</li><li><strong>Correlation matters more than raw alert count.</strong> A single well-tuned correlation rule (like the syslog repeated-password alert) is more useful to an analyst than a pile of individual low-severity hits.</li><li><strong>Structured fields aren’t always keyword-searchable.</strong> Syscollector data needs field-specific queries, not free text.</li></ul><p>This lab mirrors the core rhythm of SOC work: stand up detection tooling, generate and triage test alerts, and document the evidence. It’s a reusable baseline I plan to build on with custom detection rules, MITRE ATT&amp;CK mapping, and additional log sources like Windows Sysmon or cloud audit logs.</p><p><em>If you’re building your own home SOC lab, I’d recommend starting exactly here — certs first, one agent, and validate each detection type before scaling out. Happy to answer questions in the comments.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e2df1421a4a9" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>