WRG Sigma Rules — Claude Code Plugin

Status

Production — actively maintained rule corpus. Not yet listed in the community plugin marketplace; install directly from this repo (see Installation).

Listing is submitted through clau.de/plugin-directory-submission, which feeds Anthropic's review pipeline. anthropics/claude-plugins-community is a read-only mirror of the approved list and closes pull requests opened against it automatically, so a PR there is not the route.

Production-grade sigma detection rule writing, validation, and conversion for SOC analysts, threat-intel teams, and detection engineers using Claude Code.

Related MCP server: Security Detections MCP

TL;DR

• 3 MCP tools: draft_rule (NL → sigma YAML) + validate_rule (pySigma + best-practice linter) + convert_rule (sigma → Splunk/Elastic/OpenSearch/Wazuh/Kibana query)

• 3 Claude Code skills: sigma-rule-writer + sigma-rule-reviewer + threat-coverage-gap-analyzer

• 80 published sigma rule corpus: 13 ATT&CK tactic categories (templates + observed campaign rules). Every rule carries a sigma status: — none of them stable; see rule status for the breakdown and what it means before you deploy one

• Multi-backend conversion: Splunk SPL, Elastic/Kibana Lucene, OpenSearch Lucene + PPL, Wazuh verified (pySigma 1.x + 3 backend packages). Measured against the full corpus on 2026-07-29: Splunk and OpenSearch-PPL convert every rule; the Lucene-family targets (Elastic, Kibana, Wazuh, OpenSearch) fail on the 10 correlation rules, because that backend cannot express them — convert_rule reports this as a capability gap and names the targets that can

• Logsource-aware output: config={"pipeline": "sysmon"} maps Sigma's abstract logsource to the product's real event selection — without it a process_creation rule converts to a query that matches events of every type

• WRG ecosystem anchor: 6+ months threat-intel discipline + 100+ actor TTP corpus + observed_* rules (Mini Shai-Hulud npm worm, Nx campaign 4-vector cluster, SOCKS5 silent-fix, ClawHavoc Claude Skills, Lazarus, LockBit, LAPSUS, AI-fingerprint)

• Live demo: see DEMO.md for end-to-end tool invocation on Mini Shai-Hulud rule (pySigma 1.x + Splunk + Elastic real output)

Why this plugin exists

The sigma-rule niche in the Anthropic Claude Code plugin marketplace is empty: re-counted against anthropics/claude-plugins-community on 2026-07-29, 0 of 2283 community plugins mention sigma. That niche being empty is not the same as the area being empty — 315 of those plugins are security-themed — so the claim here is specific: nobody is doing sigma rule authoring, validation and multi-backend conversion, not "nobody is doing security".

(For scale, the same count on 2026-05-23 found 200+ plugins and one security plugin. The marketplace grew roughly tenfold in two months; the sigma count stayed at zero.)

WRG (WinstonRedGuard) has accumulated 6+ months of threat-intel infrastructure: 80 canonical sigma rules + actor catalog + pySigma integration + Pattern-driven detection-engineering discipline. This plugin packages that capability for the broader Anthropic ecosystem.

What's included

MCP tools (3)

• mcp__plugin_wrg-sigma-rules_wrg-sigma-rules__draft_rule — NL description → sigma YAML scaffold

• mcp__plugin_wrg-sigma-rules_wrg-sigma-rules__validate_rule — YAML schema + pySigma compat + best-practice linter

• mcp__plugin_wrg-sigma-rules_wrg-sigma-rules__convert_rule — sigma → Splunk/Elastic/OpenSearch/Wazuh/Kibana query

Claude Code skills (3)

• sigma-rule-writer — guided rule writing workflow

• sigma-rule-reviewer — paste rule for quality review + improvement suggestions

• threat-coverage-gap-analyzer — MITRE ATT&CK coverage analysis vs your existing corpus

Sigma rule corpus (80 rules across 13 ATT&CK tactic categories)

See resources/examples/INDEX.json for full enumeration.

Rule status

Sigma's status: field says how far a rule has been proven, and this corpus uses it literally rather than aspirationally:

stable in the sigma specification means a rule is running in production and well tested. Nothing here has earned that: the template layer is pattern material to adapt, not detections that have been validated against a real environment's telemetry. Marking them stable would make the corpus look more finished than it is, which is the same failure as writing "Unknown" in falsepositives:.

Treat any rule here as a starting point to bind to your own logsource and tune — the falsepositives: block on each one names the benign activity to expect first.

Resources

• wrg-sigma://patterns/canonical-5 — canonical detection-pattern definitions

• wrg-sigma://patterns/canonical-5/{pattern_id} — individual pattern by ID (01–05)

• wrg-sigma://coverage/mitre-attack-matrix — corpus ATT&CK coverage state (technique-by-tactic rollup, observed/template split, rules contributing no coverage), computed from the corpus at read time

Installation

Direct from this repo

git clone https://github.com/WRG-11/wrg-sigma-rules.git
# Follow Claude Code plugin install path per https://code.claude.com/docs/en/plugins

Quick example

Validate + convert a corpus rule end-to-end, from the repo root (commands from DEMO.md, captured against pySigma 1.x + the Splunk and Elasticsearch backends):

pip install pysigma pysigma-backend-splunk pysigma-backend-elasticsearch
python -c "
import sys, json
sys.path.insert(0, '.')
from tools.validate_rule.validate_rule import validate_rule_body
from tools.convert_rule.convert_rule import convert_rule_body

rule = open('resources/examples/command_and_control/observed_mini_shai_hulud_npm_supply_chain_c2_t1071.yml', encoding='utf-8').read()

print(json.dumps(validate_rule_body(rule), indent=2))
print(json.dumps(convert_rule_body(rule, target='splunk'), indent=2))
print(json.dumps(convert_rule_body(rule, target='elasticsearch'), indent=2))
"

Full captured outputs (validate JSON + Splunk SPL + Elasticsearch Lucene) are in DEMO.md.

Quality discipline

• 4-Layer self-audit per WRG audit methodology (trust-but-verify self-audit)

• 14 Python test modules covering rule validation + tool integration smoke

• pySigma 1.x compat + multi-backend conversion verified (pysigma-backend-splunk + pysigma-backend-elasticsearch + pysigma-backend-opensearch)

• LLM-safe output discipline: ASCII-only output + error-path structure preserve

• claude plugin validate PASS (re-verified 2026-07-29)

• Live demo evidence: DEMO.md — 3 real tool invocations on Mini Shai-Hulud rule

Tested environments

• Windows 11 + Claude Code

• WSL2 Ubuntu 24.04

Contributing

Sigma rule contributions welcome. Submit YAML to resources/examples/<tactic>/ with:

• ATT&CK TTP mapping in tags: field (e.g., attack.t1071)

• observed_* prefix for incident-specific rules

• template_* prefix for canonical pattern templates

• pySigma validation passing via mcp__plugin_wrg-sigma-rules_wrg-sigma-rules__validate_rule

Read CONTRIBUTING.md first if the rule claims to detect something observed in the wild. It sets out the sourcing bar — attribution, platform, and manifestation all matched against the cited source — and the three upstream rejections that produced it.

References

• Anthropic Claude Code plugin marketplace

License

MIT — see LICENSE file.

Part of the WRG-11 ecosystem

• mcp-objauthz-lab — vulnerable-by-design MCP server for learning BOLA/IDOR

• osint-trust-envelope — honest trust envelopes for OSINT results

Full index → github.com/WRG-11