convez.ai
SkillSegurançaAvançado
CavenderProjects

Multi-Modal Vulnerability Scanner

CavenderProjects/Multi-Modal-Scanner

Scanner de vulnerabilidades multi-modal com IA para ambientes regulados

Ver no GitHub
InicialPluginScore · de 0 a 100

Skill de segurança para Claude Code que realiza avaliações de vulnerabilidades em ambientes regulados usando inteligência artificial. Cobre websites, APIs, código-fonte, agentes IA, STIG e análise de sistemas interconectados, com mapeamento de 12+ frameworks de compliance (NIST, CIS, OWASP, HIPAA, PCI-DSS).

Instalação

$npx convez-ai CavenderProjects/Multi-Modal-Scanner

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
2

Casos de uso

  • Avaliação de segurança de aplicações web e APIs em ambientes regulados
  • Análise de risco de agentes IA e skills Claude em workflows enterprise
  • Revisão estática de código-fonte com enfoque em riscos de segurança
  • Verificação de compliance STIG e mapeamento contra frameworks regulatórios
  • Detecção de cadeias de ataque em sistemas interconectados

Veredito

Ferramenta enterprise-grade de segurança com profundidade técnica legítima. Desenvolvida por profissional sênior de GRC com 20 anos de experiência em serviços financeiros e healthcare. Preenche lacuna real em avaliação de risco de agentes IA em ambientes regulados — caso de uso pouco coberto no mercado.

README

A tradução para o português deste README ainda está na fila.

Multi-Modal Vulnerability Scanner

A regulated-environment security assessment platform built and maintained by a senior information security professional with 20 years of GRC and security management experience across financial services, healthcare, and real estate.


Two Versions

VersionRepoWhat It Is
Claude Code SkillMulti-Modal-ScannerAI-augmented assessment workflows inside Claude Code — install once, runs wherever Claude Code runs
Standalone AppMulti-Modal-Scanner_StandaloneFull PyQt6 desktop application — runs independently, no Claude Code required, persistent scan history, offline capable

Both versions share the core controls libraries (website/agent, API, code review, interconnected), compliance framework mappings, and report format. The standalone app is the production-ready version of the scanner with a full GUI, database-backed scan history, and an interactive report review interface.


What This Does

Across both versions the platform provides seven assessment workflows:

  • Website Vulnerability Assessment — Evaluates web applications against 67 controls across 13 families
  • AI Agent Assessment — Evaluates Claude skills, OpenAI GPTs, MCP servers, LangChain/LangGraph apps, Bedrock agents, and other AI agents against the same controls library, identifying risks specific to AI-augmented workflows (both versions)
  • API Vulnerability Assessment — Tests APIs against OWASP API Security Top 10 and 53 controls across 17 families, covering authentication, authorization, rate limiting, data exposure, and SSRF
  • Source Code Review — Static analysis of codebases against 51 controls across 12 families covering security flaws, complexity risks, and development practice gaps
  • STIG Compliance Assessment — Imports DISA STIG XCCDF files, parses rules into a structured controls library, and produces a compliance checklist report (both versions)
  • OS & Software Assessment — Scans Windows/Linux hosts for patch compliance, EOL software, insecure services, and CVE exposure (Standalone only)
  • Connected Systems Assessment — Correlates findings from two or more completed assessments to detect multi-step attack chains spanning connected systems, with CVSS re-scoring and reachability promotion analysis

Vulnerability assessment workflows (Website, AI Agent, API, Source Code, Connected Systems) produce interactive HTML reports with:

  • CVSS v3.1 scoring and vector strings
  • Severity pill filters (multi-select: Critical / High / Medium / Low)
  • Status filters (Compliant / Needs Review / Suppressed)
  • Expandable finding cards with evidence, remediation, and review procedures
  • Per-control triage actions: Confirm / Mark Compliant / Suppress as False Positive
  • Expandable Review Procedure for every Needs Review finding — specific numbered steps tailored to the control
  • Report save with original filename + timestamp preserved

The STIG Compliance Assessment produces a separate checklist report in CAT I/II/III severity format. The Standalone App supports prior report import to carry forward false positive decisions and notes across reassessments.


Compliance Framework Coverage

Every finding is cross-referenced against 12+ compliance and regulatory frameworks:

FrameworkCoverage
OWASP Top 10 (2025)All workflows
NIST SP 800-53 Rev 5All workflows
ISO/IEC 27001:2022All workflows
PCI-DSS v4.0.1All workflows
SOC 2 Type IIAll workflows
HIPAA Security RuleAll workflows
CMMC v2.0 Level 2All workflows
DoD Cloud SRGAll workflows
FedRAMP ModerateAll workflows
SEC/FINRAAll workflows
EU DORAAll workflows
EU AI ActAll workflows

Why It Exists

Security assessments generate noise. Regulated environments generate liability. After 11 years managing GRC programs across financial services, healthcare, and real estate, I built this to structure what I was doing manually: running findings against compliance frameworks, flagging ambiguous results for explicit review, and producing reports that hold up when they surface in an audit or board-level discussion.


Regulated Environment Considerations

1. False-Positive Risk

In regulated environments, a false positive isn't just wasted time — it can trigger unnecessary remediation spend, create misleading audit artifacts, or generate erroneous risk exceptions that become permanent record. Every finding with ambiguous scanner output is flagged for explicit manual review before being documented as confirmed.

2. Assessment Documentation

Any finding that may surface in an audit response, regulatory submission, or board-level risk report needs a clear record of: who assessed it, what context was applied, what compensating controls were considered, and what the final risk position is. Reports are structured for this and carry decisions forward across reassessment cycles.

3. Business-Constraint Remediation

Many findings in production regulated environments cannot be remediated in isolation. A vulnerability in a critical-care device, a legacy system under a multi-year vendor contract, or an integration a business unit depends on for revenue falls into this category. The tool supports suppression and false-positive documentation for findings where remediation isn't viable.


Repository Structure

Multi-Modal-Scanner/
├── README.md
├── pen-tester/
│   ├── SKILL.md                                    # Claude Code skill definition
│   ├── assets/
│   │   ├── report-template.html                    # Website & AI Agent report template
│   │   ├── api-report-template.html                # API vulnerability report template
│   │   ├── code-review-report-template.html        # Source code review report template
│   │   ├── interconnected-report-template.html     # Connected systems report template
│   │   └── stig-report-template.html               # STIG checklist report template
│   ├── references/
│   │   ├── controls-library.md                     # 67 controls, 13 families (Website/Agent)
│   │   ├── api-controls-library.md                 # 53 controls, 17 families (API)
│   │   ├── code-review-controls.md                 # 51 controls, 12 families (Code Review)
│   │   ├── interconnected-controls.md              # 27 controls, 9 families (Connected)
│   │   └── os-software-controls.md                 # OS & software security controls
│   └── test-reports/                               # Sample generated reports
└── Multi-Modal-Scanner_Standalone/                 # → See standalone repo
    # Full PyQt6 desktop app — lives at:
    # https://github.com/CavenderProjects/Multi-Modal-Scanner_Standalone

Standalone App

The standalone desktop application is maintained in a separate repository: github.com/CavenderProjects/Multi-Modal-Scanner_Standalone

It is the full, independently deployable version of this scanner — no Claude Code, no API dependency. Designed for use in air-gapped or restricted environments where a Claude API connection is not available or not permitted.

Key differences from the Claude Code skill:

FeatureClaude Code SkillStandalone App
RuntimeClaude CodePython + PyQt6 (desktop)
Claude API requiredYesNo
Scan historyPer sessionSQLite database, persistent
Report triageIn report (browser)In-app triage interface
AI Agent assessmentYesYes
STIG importYes (XCCDF)Yes (XCCDF)
OS & Software assessmentNoYes
Prior report importNoYes (FP + notes carryover)

Limitations and Caveats

This is a workflow augmentation tool, not an autonomous security assessment engine.

  • It does not perform port scanning, network discovery, or automated exploitation. Website and API assessments make live HTTP requests to the target; OS, code review, STIG, and agent assessments do not make external network requests
  • Output requires review by a qualified security professional before use in any regulatory or audit context
  • False-positive evaluation is only as good as the context provided
  • It does not replace legal review for risk acceptance decisions with significant regulatory exposure

Part of a Broader AI Governance Practice Portfolio

ArtifactStatusDescription
Multi-Modal Vulnerability Scanner (this repo)LiveRegulated-environment security assessment platform — Claude Code skill + standalone app
AI Risk Assessment TemplateIn progressMaps NIST AI RMF + ISO 42001 controls to GRC language enterprises already use
AI Vendor Risk QuestionnaireIn progress25-question due diligence framework for evaluating third-party AI vendors

Background

Christopher Cavender, CISSP, CCSP | IAPP AIGP (in progress)

20 years in information security and GRC. Former Business Information Security Officer at Anywhere Real Estate (Fortune 500); 11 years managing security programs across financial services, healthcare, and real estate. Currently Information Systems Security Manager at Tripoint Solutions. NJ/NYC.

Connect: LinkedIn


Contributing

Contributions welcome, especially from practitioners working in regulated environments with specific HIPAA, NYDFS, PCI, EU AI Act, or other framework-specific context to add. Open an issue or submit a PR.


License

MIT License. Use freely. Attribution appreciated but not required.


Built 2026 · Part of an active AI governance practice portfolio

Ver o original no GitHub

Bom para

Relacionados