claude-cybersecurity (Claude Skill) — 설치 & 라이브 데모

왜 쓰나요

핵심 기능

8 parallel specialist agents for different threat surfaces
OWASP 2025, CWE Top 25, MITRE ATT&CK mapping
PCI-DSS and other compliance framework coverage
Business-logic flaw detection (what static tools miss)
AI-generated code patterns flagged (e.g. hallucinated auth)
11+ programming languages, zero dependencies beyond Claude Code

라이브 데모

실제 사용 모습

claude-cybersecurity-skill.replay ▶ 준비됨

0/0

설치

작성자 저장소 · Claude Skills 공식 문서

사용 사례

실전 활용법: claude-cybersecurity

Pre-release security review of a new service

👤 Engineering teams before shipping to production ⏱ ~60 min intermediate

언제 쓸까: Before a release with a new or heavily changed service.

흐름

Run the full 8-agent pass

Use claude-cybersecurity on services/payment. Full pass — vulns, authz, secrets, supply chain, IaC, business logic.✓ 복사됨

→ Report with findings per agent, severity, OWASP/CWE mapping
Triage and route

Group by severity. For each critical/high, draft a GitHub issue with reproduction steps.✓ 복사됨

→ Issue drafts ready to paste

결과: A defensible pre-release sign-off with a paper trail.

함정

Treating business-logic flags as false positives automatically — Those are the ones static tools miss — read them twice

함께 쓰기: github · repo-forensics-skill

Review AI-generated code for the mistakes LLMs make

👤 Teams accepting lots of PR code from Copilot/Claude/Cursor ⏱ ~45 min intermediate

언제 쓸까: Auditing merged or pending PRs that were AI-assisted.

흐름

Focus on the AI-code agent

Use claude-cybersecurity — ai-code patterns only. Flag: hallucinated middleware, plausible-but-wrong crypto, trust-all defaults.✓ 복사됨

→ Pattern-specific flags
Cross-check authz flags

Now run the authorization agent — AI tends to miss role checks on non-happy paths.✓ 복사됨

→ Authz matrix showing gaps

결과: Catch the class of mistakes LLMs make before they ship.

함께 쓰기: github

Sweep infrastructure-as-code for dangerous defaults

👤 Platform teams ⏱ ~30 min intermediate

언제 쓸까: Before merging Terraform / K8s / CloudFormation changes.

흐름

Run IaC agent

Use claude-cybersecurity IaC agent on terraform/ — flag public S3, open security groups, privileged pods.✓ 복사됨

→ Misconfigs with severity + fix diffs

결과: Clean IaC before it hits the cluster.

Scan for malware / C2 / backdoor patterns in dependencies

👤 Security engineers auditing third-party code ⏱ ~30 min intermediate

언제 쓸까: New dependency coming in, or suspicious dep update.

흐름

Run threat-intel agent

Run claude-cybersecurity threat-intel on node_modules — known malicious patterns.✓ 복사됨

→ Pattern matches (or clean)

결과: Early warning before a malicious package slips in.

함정

False positives on obfuscated legitimate libs (e.g. minified SDKs) — Cross-check with published source; don't auto-remove

함께 쓰기: repo-forensics-skill

조합

다른 MCP와 조합해 10배 효율

claude-cybersecurity-skill + repo-forensics-skill

Layer supply-chain + install-time audits with app-level code review

Run repo-forensics on the dep tree, then claude-cybersecurity on our app code.✓ 복사됨

claude-cybersecurity-skill + github

Auto-file issues for each high-severity finding

After the review, open a GitHub issue per high finding with CWE + remediation.✓ 복사됨

도구

이 MCP가 노출하는 것

도구	입력	언제 호출
agent_vulns	code path	Standard review
agent_authz	code path	Multi-role apps
agent_secrets	code path	Always
agent_supply_chain	manifest files	Dep updates
agent_iac	iac dir	Infra PRs
agent_threat_intel	code/deps path	Third-party audits
agent_ai_code	code path	AI-assisted PRs
agent_business_logic	code path, domain hints	High-trust workflows (payments, auth)

비용 및 제한

운영 비용

API 쿼터: None
호출당 토큰: High — 8 agents in parallel
금액: Free
팁: Scope to the diff of a PR rather than whole repo when you can

보안

권한, 시크릿, 파급범위

최소 스코프: Read on the code under review

자격 증명 저장: None

데이터 외부 송신: None beyond LLM call — findings stay local

절대 부여 금지: Write access to the audited repo during review

This is defensive. Do not use agent output as proof of exploitation — these are findings, not confirmed vulns.
Complement, don't replace, established SAST/SCA tools.

문제 해결

자주 발생하는 오류와 해결

Too many findings to process

Filter by severity=high,critical first; revisit rest iteratively

Agent mislabels framework conventions as vulns

Provide framework context: 'this is Rails, strong_params is in use'

대안

claude-cybersecurity 다른 것과 비교

대안	언제 쓰나	단점/장점
repo-forensics	You need install-time / supply-chain vetting, not app code review	Different threat model
GitHub Advanced Security (CodeQL)	You want SAST integrated into the PR workflow	Misses business logic and AI patterns

claude-cybersecurity