Malwar¶

Static analysis engine purpose-built for detecting malware in agentic AI skill files.

What is Malwar?¶

Malwar is a 4-layer malware detection engine designed to catch threats embedded in natural language markdown instructions -- the kind that direct AI agents to perform malicious actions. Unlike traditional scanners that analyze binaries or source code, Malwar detects prompt injection, social engineering, obfuscated commands, and data exfiltration hidden inside SKILL.md files used by agentic AI platforms.

20% of ClawHub's 10,700+ skills are malicious. The ClawHavoc campaign alone trojanized 824+ skills to deliver the AMOS infostealer. VirusTotal sees nothing. Code scanners see nothing. Malwar was built to catch them.

How It Works¶

SKILL.md --> Rule Engine --> URL Crawler --> LLM Analyzer --> Threat Intel --> Verdict
              <50ms          1-5s           2-10s            <100ms

Layer	What It Catches
Rule Engine	Obfuscated commands, prompt injection, credential exposure, exfiltration patterns (26 rules)
URL Crawler	Malicious URLs, domain reputation, redirect chains to C2 infrastructure
LLM Analyzer	Social engineering, hidden intent, context-dependent attacks invisible to regex
Threat Intel	Known IOCs, campaign attribution, threat actor fingerprints

Full pipeline details: Architecture

Key Features¶

26 detection rules covering 7 threat categories
4-layer pipeline combining speed with depth
Sub-50ms scans for rule-based detection
SARIF 2.1.0 output for CI/CD integration
REST API with 30+ endpoints, authentication, and RBAC
Web dashboard with analytics, trend charts, and scan history
Plugin system for extending detection with third-party plugins
YAML DSL for writing custom rules without Python
ML-based risk scoring for anomaly-aware threat assessment
PostgreSQL backend support alongside SQLite
Redis caching for scan results and rule compilations
Scheduled scanning with configurable background jobs
Multi-channel notifications via Slack, email, and webhooks
Audit logging with immutable, append-only trail
Rich TUI for interactive terminal usage
Campaign tracking with IOC correlation and attribution
STIX/TAXII export for SIEM integration
Python SDK for embedding scans in your tools
LangChain integration with scan-before-execute guards
CI/CD templates for GitHub Actions, GitLab CI, and Azure DevOps
ClawHub registry integration — browse, search, and scan skills directly from the registry
Remote URL scanning — scan any SKILL.md hosted on the web
Docker and Kubernetes deployment with Helm chart

Quick Install¶

git clone https://github.com/Ap6pack/malwar.git && cd malwar
pip install -e ".[dev]"
malwar db init

Or with Docker:

docker compose up -d    # API + Dashboard at http://localhost:8000

See the full Installation Guide for all options.

Quick Scan¶

malwar scan SKILL.md                    # scan a file
malwar scan skills/                     # scan a directory
malwar scan SKILL.md --format sarif     # CI/CD output
malwar scan SKILL.md --no-llm          # skip LLM (fast + free)
malwar crawl scan beszel-check          # scan a ClawHub skill by slug
malwar crawl search "crypto"            # search ClawHub registry

$ malwar scan suspicious-skill.md

  MALICIOUS  Risk: 95/100  Findings: 4

  MALWAR-OBF-001   Base64-encoded command execution        critical   L14
  MALWAR-CMD-001   Remote script piped to shell            critical   L22
  MALWAR-EXFIL-001 Agent memory/identity file access       critical   L31
  MALWAR-MAL-001   ClawHavoc campaign indicator            critical   L14

  Scan completed in 42ms (rule_engine, threat_intel)

See the full CLI Reference and Quick Start Guide.

Documentation¶

Section	Description
Getting Started	Installation, quick start, first scan
User Guide	CLI, API, detection rules, threat campaigns
Integrations	Python SDK, LangChain, GitHub Action, STIX/TAXII
Deployment	Docker, Kubernetes, production configuration
Development	Architecture, contributing, benchmarks

Detection Accuracy¶

Tested against 44 fixtures (31 labeled, 13 real-world ClawHub samples):

Metric	Value
Accuracy	96.8%
Precision	95.8%
Recall	100.0%
F1 Score	97.9%

Every malicious sample is detected. See the full Accuracy Report.