security-skills MCP — ユースケース, インストール & ライブデモ

なぜ使うのか

主な機能

Threat modeling template based on STRIDE
Dependency CVE cross-referencing workflow
IaC misconfiguration checks (Terraform, Docker, k8s)
Security code review patterns
Defensive focus — no exploitation tooling

ライブデモ

実際の動作

security-skill.replay ▶ 準備完了

0/0

インストール

クライアントを選択

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "security-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/eth0izzle/security-skills",
        "~/.claude/skills/security-skills"
      ],
      "_inferred": true
    }
  }
}

Claude Desktop → Settings → Developer → Edit Config を開く。保存後、アプリを再起動。

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "security-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/eth0izzle/security-skills",
        "~/.claude/skills/security-skills"
      ],
      "_inferred": true
    }
  }
}

Cursor は Claude Desktop と同じ mcpServers スキーマを使用。プロジェクト設定はグローバルより優先。

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "security-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/eth0izzle/security-skills",
        "~/.claude/skills/security-skills"
      ],
      "_inferred": true
    }
  }
}

Cline サイドバーの MCP Servers アイコンをクリックし、"Edit Configuration" を選択。

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "security-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/eth0izzle/security-skills",
        "~/.claude/skills/security-skills"
      ],
      "_inferred": true
    }
  }
}

Claude Desktop と同じ形式。Windsurf を再起動して反映。

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "security-skill",
      "command": "git",
      "args": [
        "clone",
        "https://github.com/eth0izzle/security-skills",
        "~/.claude/skills/security-skills"
      ]
    }
  ]
}

Continue はマップではなくサーバーオブジェクトの配列を使用。

~/.config/zed/settings.json

{
  "context_servers": {
    "security-skill": {
      "command": {
        "path": "git",
        "args": [
          "clone",
          "https://github.com/eth0izzle/security-skills",
          "~/.claude/skills/security-skills"
        ]
      }
    }
  }
}

context_servers に追加。保存時に Zed がホットリロード。

claude mcp add security-skill -- git clone https://github.com/eth0izzle/security-skills ~/.claude/skills/security-skills

ワンライナー。claude mcp list で確認、claude mcp remove で削除。

ユースケース

実用的な使い方： security-skills

Produce a STRIDE threat model for a new feature

👤 Security engineers, staff+ engineers owning a feature's security review ⏱ ~45 min advanced

使うタイミング： Design review for a feature that touches user data, auth, or external networks.

前提条件

Feature design doc or architecture diagram — Paste into chat or point Claude at the file
Skill installed — git clone https://github.com/eth0izzle/security-skills ~/.claude/skills/security-skills

フロー

Walk the system

Use security-skills. Here's the design for [feature]. Enumerate trust boundaries and data flows.✓ コピーしました

→ Named trust boundaries, explicit data flow arrows
Apply STRIDE

For each boundary, call out Spoofing, Tampering, Repudiation, Info Disclosure, DoS, Elevation risks.✓ コピーしました

→ Table with threats scored, not generic 'consider XSS'
Propose mitigations

Rank the threats by likelihood * impact and propose concrete mitigations for the top 5.✓ コピーしました

→ Specific controls (rate limit, signed requests, audit log)

結果： A threat model doc ready for a security review meeting.

注意点

STRIDE becomes checkbox theater — Demand concrete attacker stories — 'attacker with X wants Y, does Z'

組み合わせ： filesystem

Sweep a repo's dependencies for known CVEs

👤 Dev teams responding to a security advisory ⏱ ~30 min intermediate

使うタイミング： A new CVE drops and you need to know where you're exposed.

フロー

Point at the manifest

Use security-skills. Audit package-lock.json for dependencies with known CVEs published in the last 90 days.✓ コピーしました

→ List of packages, advisory IDs, severity
Triage by usage

For each vulnerable dep, grep the codebase — is it actually reachable?✓ コピーしました

→ Reachable vs unreachable classification
Draft an upgrade plan

Plan the minimum-risk upgrade order so CI doesn't explode.✓ コピーしました

→ Ordered upgrade sequence with version pins

結果： A concrete patch plan — not 'run npm audit fix' and pray.

注意点

Transitive dependencies missed — Use full lockfile, not just top-level manifest

組み合わせ： filesystem · github

Review Terraform and k8s manifests for common misconfigs

👤 Platform engineers, DevSecOps ⏱ ~30 min intermediate

使うタイミング： Before merging IaC changes into main, or auditing what's already there.

フロー

Point Claude at the IaC directory

Use security-skills. Scan terraform/ for S3 buckets without encryption, IAM wildcards, public security groups.✓ コピーしました

→ File+line references with severity
Get fixes

For each finding, propose the minimal Terraform diff to fix it.✓ コピーしました

→ HCL snippets showing before/after

結果： PR-sized security fixes you can merge incrementally.

注意点

False positives on intentional public assets — Tag resources that are intentionally public; the skill should skip them

組み合わせ： github · filesystem

組み合わせ

他のMCPと組み合わせて10倍の力を

security-skill + github

Post findings as PR comments automatically

Review this PR and leave line comments for each security finding.✓ コピーしました

security-skill + filesystem

Recursive scans across a monorepo

Walk the entire services/ tree and aggregate findings by service.✓ コピーしました

ツール

このMCPが提供する機能

ツール	入力	呼び出すタイミング	コスト
threat_model	system description	Design-phase review	0
cve_audit	lockfile path	After a public CVE drops	0 (uses public advisory data)
iac_review	directory path	Pre-merge IaC audits	0
secret_scan	path or diff	Pre-commit or pre-push	0

コストと制限

運用コスト

APIクォータ: None built-in
呼び出しあたりのトークン: Moderate — depends on files inspected
金額: Free
ヒント: Scope scans to specific directories; don't point at node_modules.

セキュリティ

権限、シークレット、影響範囲

認証情報の保管： None in skill itself

データ送信先： None — skill is prompts; be aware inspected code goes to Claude API

Defensive only. Do not use for exploitation, credential harvesting, or red-team operations.
Never paste production secrets into prompts — even when asking the skill to 'find them'.

トラブルシューティング

よくあるエラーと対処法

Skill suggests exploitation techniques

Re-prompt for defensive framing: 'as a defender, how do I prevent X'

Too many false positives on IaC scan

Provide context (prod vs staging; intentional public resources)

Missed a known vulnerability

Claude's knowledge cutoff may lag. Pair with a proper scanner (Trivy, Snyk) for authoritative CVE data.

代替案

security-skills 他との比較

代替案	代わりに使う場面	トレードオフ
Trivy / Grype	You need an authoritative, up-to-date vulnerability database	No LLM reasoning — just raw scan output
tfsec / checkov	Specific IaC scanning with a curated rule set	Rule-bound; misses nuanced misconfigs

その他

リソース

📖 GitHub の公式 README を読む

🐙 オープンな issue を見る

🔍 400以上のMCPサーバーとSkillsを見る