Threat Hunting & IOC Analysis

Purpose

Enable Claude to assist threat hunters with proactive threat detection, IOC extraction and normalization, MITRE ATT&CK mapping, hunt hypothesis generation, and converting threat intelligence into actionable detection rules across all major SIEM platforms.

Activation Triggers

This skill activates when the user asks about:

• Extracting IOCs from threat reports, emails, or security advisories
• Mapping behaviors or TTPs to MITRE ATT&CK framework
• Generating hunt hypotheses for a specific threat actor or technique
• Creating Sigma rules, Splunk SPL queries, KQL, or EQL
• Converting threat intelligence into SIEM detection queries
• STIX/TAXII or MISP-compatible indicator formatting
• ATT&CK Navigator layer creation
• Threat intelligence correlation across multiple sources
• Proactive threat hunting in a SIEM or EDR

Prerequisites

pip install requests pyyaml stix2 taxii2-client

Optional platforms:

• MISP — Threat intelligence sharing platform
• OpenCTI — Threat intelligence platform
• YARA — Pattern matching (→ Skill 05)
• Sigma CLI — Rule conversion tool
• SIEM access (Splunk, Elastic, QRadar, Microsoft Sentinel)

Core Capabilities

1. IOC Extraction & Normalization

When the user provides a threat report, article, email, or log snippet:

Claude performs these extraction steps:

1. Parse all text for indicators using pattern matching:

1. Defang extracted indicators — refang before use:

• hxxp:// → http://
• [.] → .
• [at] → @
• [:] → :

1. Categorize by type: Network / File / Host / Identity / Vulnerability

1. Score by confidence: High (specific, sourced), Medium (inferred), Low (generic)

1. Output in multiple formats:

python scripts/ioc_extractor.py --input threat_report.txt --output iocs.json
python scripts/ioc_extractor.py --input report.pdf --format stix --output iocs.stix.json
python scripts/ioc_extractor.py --input email.eml --defang --output iocs.csv

STIX 2.1 output template:

{
  "type": "indicator",
  "id": "indicator--[uuid]",
  "created": "2025-05-28T00:00:00.000Z",
  "name": "Malicious IP — C2 Infrastructure",
  "pattern": "[ipv4-addr:value = '192.0.2.10']",
  "pattern_type": "stix",
  "valid_from": "2025-05-28T00:00:00Z",
  "labels": ["malicious-activity", "c2"],
  "confidence": 85
}

2. MITRE ATT&CK Mapping

When the user provides TTPs, behaviors, or a malware report:

python scripts/mitre_mapper.py --input techniques.txt --output attack_map.json
python scripts/mitre_mapper.py --technique T1059.001 --detection-query splunk

Mapping process:

1. Analyze each behavior against ATT&CK technique descriptions
2. Map to specific Tactic → Technique → Sub-technique (T1059 → T1059.001)
3. Assign confidence level based on evidence quality

ATT&CK Tactics Reference:

ATT&CK Navigator Layer format (JSON for visualization):

{
  "name": "Threat Hunt Layer — [Threat Actor/Campaign]",
  "versions": {"attack": "14", "navigator": "4.9"},
  "domain": "enterprise-attack",
  "techniques": [
    {
      "techniqueID": "T1059.001",
      "color": "#ff6666",
      "comment": "Observed PowerShell download cradle",
      "enabled": true,
      "score": 100
    }
  ]
}

3. Hunt Hypothesis Generation

When the user asks for hunt hypotheses:

Use this structured hypothesis template:

## Hunt Hypothesis — [ID]: [Short Name]

**Hypothesis Statement:**
"We believe [Threat Actor/TTPs] may be present in [Environment] based on
[Threat Intelligence / Recent Incidents / Industry Reports]."

**Rationale:**
[Why this threat is relevant to this organization — industry, exposure, recent news]

**ATT&CK Techniques Covered:**
- T1059.001 — PowerShell
- T1053.005 — Scheduled Task/Job
- T1021.001 — Remote Services: Remote Desktop Protocol

**Data Sources Required:**
- Windows Event Logs (Security, System, PowerShell/4104)
- EDR process execution telemetry
- DNS query logs
- Proxy/firewall logs

**Detection Logic:**
[SIEM query or pseudocode]

**Success Criteria:**
- POSITIVE: We find evidence of the technique → escalate to IR (Skill 07)
- NEGATIVE: No evidence after thorough search → document as cleared hunt
- INCONCLUSIVE: Insufficient data → identify logging gaps

**Estimated Hunt Duration:** [X hours]
**Priority:** [High / Medium / Low]
**Analyst:** [Name]

4. SIEM Detection Query Library

When the user asks to build detection queries for specific techniques:

#### Splunk SPL Queries

// T1059.001 — PowerShell Execution with suspicious flags
index=windows (source="WinEventLog:Microsoft-Windows-PowerShell/Operational" EventCode=4104)
| search ScriptBlockText IN ("*DownloadString*", "*IEX*", "*EncodedCommand*", "*bypass*", "*WebClient*")
| stats count by ComputerName, UserName, ScriptBlockText
| where count > 0

// T1003.001 — LSASS Memory Dump
index=windows EventCode=10 TargetImage="*lsass.exe"
| where NOT (SourceImage IN ("C:\\Windows\\System32\\*", "C:\\Program Files\\*"))
| table _time, SourceImage, TargetImage, GrantedAccess, CallTrace

// T1547.001 — Registry Run Key Persistence
index=windows EventCode=13 TargetObject IN ("*\\Run\\*", "*\\RunOnce\\*")
| where NOT (Image IN ("C:\\Windows\\System32\\*", "C:\\Windows\\SysWOW64\\*"))
| table _time, ComputerName, Image, TargetObject, Details

// T1021.002 — Lateral Movement via SMB Admin Shares
index=windows EventCode=5140
| where ShareName IN ("\\\\*\\ADMIN$", "\\\\*\\C$", "\\\\*\\IPC$")
| stats count by SubjectUserName, IpAddress, ShareName, ObjectType
| where count > 3

#### Microsoft Sentinel KQL

// T1110.001 — Brute Force Login Attempt
SecurityEvent
| where EventID == 4625
| where TimeGenerated > ago(1h)
| summarize FailCount=count() by TargetAccount, IpAddress=replace(@"\.", "[.]", tostring(parse_json(EventData).IpAddress))
| where FailCount > 20
| join kind=leftouter (
    SecurityEvent | where EventID == 4624
    | summarize SuccessCount=count() by TargetAccount
) on TargetAccount
| project TargetAccount, IpAddress, FailCount, SuccessCount
| where isnotnull(SuccessCount)  // Brute force succeeded!

// T1190 — Exploit Public-Facing Application
AzureDiagnostics
| where Category == "ApplicationGatewayFirewallLog"
| where action_s == "Blocked"
| where ruleSetVersion_s startswith "3."
| summarize count() by clientIp_s, requestUri_s, ruleId_s
| where count_ > 100
| order by count_ desc

#### Elastic EQL

// T1055 — Process Injection
sequence by host.name
  [process where process.name : "notepad.exe" and event.type == "start"]
  [process where event.type == "start" and process.parent.name : "notepad.exe"
   and not process.name in ("conhost.exe")]

// T1566.001 — Spearphishing with attachment
sequence by user.name within 5m
  [file where file.extension in ("doc", "xls", "pdf") and 
   process.name : ("outlook.exe", "WINWORD.EXE")]
  [process where process.name : ("cmd.exe", "powershell.exe", "wscript.exe", "cscript.exe")]

#### Sigma Rule Template

title: Suspicious PowerShell Download Cradle
id: a3c2f1b4-8e9d-4a2c-b7f6-1234567890ab
status: stable
description: Detects PowerShell commands used to download and execute code from the internet
author: Threat Hunter
date: 2025/05/28
modified: 2025/05/28
references:
  - https://attack.mitre.org/techniques/T1059/001/
tags:
  - attack.execution
  - attack.t1059.001
  - attack.defense_evasion
  - attack.t1027
logsource:
  category: ps_script
  product: windows
  definition: Script Block Logging enabled (EventID 4104)
detection:
  selection:
    ScriptBlockText|contains|all:
      - 'DownloadString'
      - 'IEX'
  selection2:
    ScriptBlockText|contains:
      - '-EncodedCommand'
      - '-enc '
      - '-WindowStyle Hidden'
      - 'Net.WebClient'
      - 'WebProxy'
  condition: selection or selection2
falsepositives:
  - Legitimate software installations
  - Administrative scripts
level: high

5. Threat Intelligence Correlation

When the user asks to correlate IOCs or identify threat actors:

1. Cross-reference infrastructure across known campaigns:

• Same registrar + similar registration dates → likely related infrastructure
• IP hosting multiple C2 domains → infrastructure cluster
• Certificate SAN fields → reveal connected domains

1. Map to threat actor groups:

• MITRE ATT&CK Groups: https://attack.mitre.org/groups/
• VirusTotal/OpenCTI actor tracking
• Mandiant / CrowdStrike / SentinelOne threat intel reports

1. Generate Threat Assessment:

   ## Threat Assessment — [Campaign Name]
   
   **Threat Actor:** [APT Group / Criminal Group / Unknown]
   **Confidence:** [High / Medium / Low]
   **Motivation:** [Espionage / Financial / Hacktivism]
   **Targeting:** [Industries / Countries / Organization types]
   
   **Campaign IOCs:**
   - Infrastructure: [IPs, domains]
   - Malware: [Family names, hashes]
   - TTPs: [ATT&CK technique IDs]
   
   **Relevance to Organization:**
   [Why this threat is or isn't relevant]
   
   **Recommended Actions:**
   1. Block IOCs in firewall/proxy
   2. Hunt for T1XXX in SIEM
   3. Deploy YARA rules for detection

Script Reference

ioc_extractor.py

python scripts/ioc_extractor.py --input threat_report.txt --output iocs.json
python scripts/ioc_extractor.py --input report.pdf --format stix --output iocs.stix.json
python scripts/ioc_extractor.py --input email.eml --defang --output iocs.csv

mitre_mapper.py

python scripts/mitre_mapper.py --input techniques.txt --output attack_map.json
python scripts/mitre_mapper.py --technique T1059.001 --detection-query splunk
python scripts/mitre_mapper.py --actor "APT29" --output apt29_layer.json

Skill Integration

References

• MITRE ATT&CK Framework
• ATT&CK Navigator
• Sigma Rules Repository
• STIX/TAXII Standards
• Threat Hunting Playbook
• MISP Threat Intelligence
• OpenCTI Platform
• CISA Advisories

v3.0 Enhancements (2026 Update)

Threat-informed, repeatable hunting:

• ATT&CK current version — map to the latest Enterprise matrix (incl. updated cloud, identity, and containers techniques); call out sub-techniques explicitly.
• PEAK hunting framework — structure hunts as Prepare → Execute → Act with a documented hypothesis, data sources, and ABLE (Actor/Behavior/Location/Evidence) baselining.
• Identity-centric hunting — Entra ID / Okta logs: impossible travel, illicit OAuth consent grants, MFA-fatigue, token theft & replay, and risky sign-in correlation.
• Living-off-the-land — baseline LOLBin/LOLBAS usage and hunt deviations rather than static signatures.
• Detection-as-code — express hunt findings as Sigma rules under version control with test data, then promote validated hunts into Skill 12/15 detections.
• Hunt maturity — track from ad-hoc → data-driven → automated; record which hunts became scheduled detections.

Precision rule: every hunt yields a hypothesis, the query, the result (found/not-found/inconclusive), and a disposition (new detection / tuned alert / closed).