Building Detection Rules With Sigma

Building Detection Rules with Sigma

When to Use

Use this skill when:

• SOC engineers need to create detection rules portable across multiple SIEM platforms
• Threat intelligence reports describe TTPs requiring new detection coverage
• Existing vendor-specific rules need standardization into a shareable format
• The team adopts Sigma as a detection-as-code standard in CI/CD pipelines

Do not use for real-time streaming detection (Sigma is for batch/scheduled searches) or when the target SIEM has native detection features that Sigma cannot express (e.g., Splunk RBA risk scoring).

Common Misconfigurations & Verification

• Logsource/field-mapping mismatch: the rule's logsource (category: process_access, product: windows) only maps to real fields when the matching pipeline runs. Converting with SplunkBackend() and no splunk_windows_pipeline() emits raw Sigma field names (TargetImage, GrantedAccess) that don't exist in your Sysmon sourcetype, so the search runs but never matches. Always pair the backend with the pipeline and diff the output fields against your actual data.
• Category vs index gap: category: process_creation assumes Sysmon EventCode 1 (or 4688 with command-line auditing enabled). If the host only ships Security 4688 without Audit Process Creation + command-line capture, CommandLine-based selections silently fail — verify the source data has the fields the rule keys on.
• Backend conversion drift: the same rule yields different operators per backend (|contains → *0x1010* wildcards in SPL but a match in EQL). A rule that tests clean in Splunk can over/under-match in Elastic. Convert and unit-test against each target, don't assume portability.
• Filter logic inversion: condition: selection and not 1 of filter_* makes broad |endswith: '\svchost.exe' filters swallow true positives (malware named svchost.exe in a non-system path). Anchor filters on full path/signature, not basename.
• Verification: run sigma check for syntax, fire a known-true event (e.g., mimikatz LSASS access) to confirm 4/4 test cases match, then run a 7-day non-alerting backtest to measure the FP rate before promoting to a correlation/analytics rule.

Prerequisites

• Python 3.8+ with pySigma and appropriate backend (pySigma-backend-splunk, pySigma-backend-elasticsearch, pySigma-backend-microsoft365defender)
• Sigma rule repository cloned: git clone https://github.com/SigmaHQ/sigma.git
• MITRE ATT&CK framework knowledge for technique mapping
• Understanding of target SIEM log source field mappings

Workflow

Step 1: Define Detection Logic from Threat Intelligence

Start with a threat report or ATT&CK technique. Example: detecting Mimikatz credential dumping (T1003.001 — LSASS Memory):

title: Mimikatz Credential Dumping via LSASS Access
id: 0d894093-71bc-43c3-8d63-bf520e73a7c5
status: stable
level: high
description: Detects process accessing lsass.exe memory, indicative of credential dumping tools like Mimikatz
references:
    - https://attack.mitre.org/techniques/T1003/001/
    - https://github.com/gentilkiwi/mimikatz
author: mahipal
date: 2024/03/15
modified: 2024/03/15
tags:
    - attack.credential_access
    - attack.t1003.001
logsource:
    category: process_access
    product: windows
detection:
    selection:
        TargetImage|endswith: '\lsass.exe'
        GrantedAccess|contains:
            - '0x1010'
            - '0x1038'
            - '0x1fffff'
            - '0x40'
    filter_main_svchost:
        SourceImage|endswith: '\svchost.exe'
    filter_main_csrss:
        SourceImage|endswith: '\csrss.exe'
    filter_main_wininit:
        SourceImage|endswith: '\wininit.exe'
    condition: selection and not 1 of filter_main_*
falsepositives:
    - Legitimate security tools accessing LSASS
    - Windows Defender scanning
    - CrowdStrike Falcon sensor

Step 2: Validate Sigma Rule Syntax

Use sigma check to validate the rule:

# Install pySigma and validators
pip install pySigma pySigma-validators-sigmaHQ

# Validate rule
sigma check rule.yml

Alternatively, validate with Python:

from sigma.rule import SigmaRule
from sigma.validators.core import SigmaValidator

rule = SigmaRule.from_yaml(open("rule.yml").read())
validator = SigmaValidator()
issues = validator.validate_rule(rule)
for issue in issues:
    print(f"{issue.severity}: {issue.message}")

Step 3: Convert to Target SIEM Query

Convert to Splunk SPL:

from sigma.rule import SigmaRule
from sigma.backends.splunk import SplunkBackend
from sigma.pipelines.splunk import splunk_windows_pipeline

pipeline = splunk_windows_pipeline()
backend = SplunkBackend(pipeline)

rule = SigmaRule.from_yaml(open("rule.yml").read())
splunk_query = backend.convert_rule(rule)
print(splunk_query[0])

Output:

TargetImage="*\\lsass.exe" (GrantedAccess="*0x1010*" OR GrantedAccess="*0x1038*"
OR GrantedAccess="*0x1fffff*" OR GrantedAccess="*0x40*")
NOT (SourceImage="*\\svchost.exe") NOT (SourceImage="*\\csrss.exe")
NOT (SourceImage="*\\wininit.exe")

Convert to Elastic Query (Lucene):

from sigma.backends.elasticsearch import LuceneBackend
from sigma.pipelines.elasticsearch import ecs_windows_pipeline

pipeline = ecs_windows_pipeline()
backend = LuceneBackend(pipeline)
elastic_query = backend.convert_rule(rule)
print(elastic_query[0])

Convert to Microsoft Sentinel KQL:

from sigma.backends.microsoft365defender import Microsoft365DefenderBackend

backend = Microsoft365DefenderBackend()
kql_query = backend.convert_rule(rule)
print(kql_query[0])

Step 4: Map to MITRE ATT&CK and Add Coverage Metadata

Tag every rule with ATT&CK technique IDs in the tags field:

tags:
    - attack.credential_access        # Tactic
    - attack.t1003.001                # Sub-technique
    - attack.t1003                    # Parent technique

Track detection coverage using the ATT&CK Navigator:

import json

# Generate ATT&CK Navigator layer from Sigma rules
layer = {
    "name": "SOC Detection Coverage",
    "versions": {"attack": "14", "navigator": "4.9", "layer": "4.5"},
    "domain": "enterprise-attack",
    "techniques": []
}

# Parse Sigma rules directory for technique tags
import os
from sigma.rule import SigmaRule

for root, dirs, files in os.walk("sigma/rules/windows/"):
    for f in files:
        if f.endswith(".yml"):
            rule = SigmaRule.from_yaml(open(os.path.join(root, f)).read())
            for tag in rule.tags:
                if str(tag).startswith("attack.t"):
                    technique_id = str(tag).replace("attack.", "").upper()
                    layer["techniques"].append({
                        "techniqueID": technique_id,
                        "color": "#31a354",
                        "score": 1
                    })

with open("coverage_layer.json", "w") as f:
    json.dump(layer, f, indent=2)

Step 5: Test Rule Against Sample Data

Create test data and validate the rule catches the expected events:

# Use sigma test framework
sigma test rule.yml --target splunk --pipeline splunk_windows

# Or manually test in Splunk with sample data
# Upload Sysmon process_access log with known Mimikatz signature

Validate false positive rate by running against 7 days of production data in a non-alerting saved search.

Step 6: Deploy to Production SIEM

Deploy the converted query as a scheduled search or correlation rule:

Splunk ES Correlation Search:

| tstats summariesonly=true count from datamodel=Endpoint.Processes
  where Processes.process_name="*\\lsass.exe"
  by Processes.src, Processes.user, Processes.process_name, Processes.parent_process_name
| `drop_dm_object_name(Processes)`
| where count > 0

Elastic Security Rule (TOML format):

[rule]
name = "LSASS Memory Access - Credential Dumping"
description = "Detects suspicious access to LSASS process memory"
risk_score = 73
severity = "high"
type = "eql"
query = '''
process where event.action == "access" and
  process.name == "lsass.exe" and
  not process.executable : ("*\\svchost.exe", "*\\csrss.exe")
'''

[rule.threat]
framework = "MITRE ATT&CK"
[[rule.threat.technique]]
id = "T1003"
name = "OS Credential Dumping"

Step 7: Version Control and CI/CD Integration

Store rules in Git with automated testing:

# .github/workflows/sigma-ci.yml
name: Sigma Rule CI
on: [push, pull_request]
jobs:
  validate:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-python@v5
        with:
          python-version: '3.11'
      - run: pip install pySigma pySigma-validators-sigmaHQ
      - run: sigma check rules/
      - run: sigma convert -t splunk -p splunk_windows rules/ > /dev/null

Key Concepts

Tools & Systems

• SigmaHQ: Official Sigma rule repository with 3,000+ community-maintained detection rules on GitHub
• pySigma: Python-based Sigma rule processing framework with modular backends and pipelines
• ATT&CK Navigator: MITRE tool for visualizing detection coverage mapped to ATT&CK techniques
• Uncoder.IO: Web-based Sigma rule converter supporting 30+ SIEM platforms for quick translation

Common Scenarios

• New CVE Detection: Write Sigma rule for exploitation indicators (e.g., Log4Shell JNDI lookup patterns in web logs)
• Hunting Rule Promotion: Convert ad-hoc Splunk hunting query into Sigma rule for ongoing automated detection
• Multi-SIEM Migration: Converting 500+ Splunk correlation searches to Sigma for migration to Elastic Security
• Purple Team Output: Convert red team findings into Sigma rules for immediate defensive coverage
• Threat Intel Operationalization: Transform IOC-based threat reports into behavioral Sigma rules

Output Format

SIGMA RULE DEPLOYMENT REPORT
━━━━━━━━━━━━━━━━━━━━━━━━━━━
Rule ID:      0d894093-71bc-43c3-8d63-bf520e73a7c5
Title:        Mimikatz Credential Dumping via LSASS Access
ATT&CK:       T1003.001 - LSASS Memory
Severity:     High
Status:       Deployed to Production

Conversions:
  Splunk SPL:    PASS — Saved search "sigma_lsass_access" created
  Elastic EQL:   PASS — Detection rule ID elastic-0d894093 enabled
  Sentinel KQL:  PASS — Analytics rule deployed via ARM template

Testing:
  True Positives:    4/4 test cases matched
  False Positives:   2 in 7-day backtest (svchost edge case — filter added)
  Performance:       Avg execution 3.2s on 50M events/day