10 Best Claude Skills for Security Work
Compare Claude skills for secrets, dependencies, permissions, auth, RBAC, threats, headers, PII, data retention, and LLM guardrails.
A practical Claude security toolkit uses separate skills for exposure, access, design, and data lifecycle. Start with secret-scanner, dependency-audit, least-privilege-auditor, and auth-flow-reviewer; add threat modeling, RBAC, headers, PII redaction, retention, or LLM guardrails when those boundaries are in scope.
Key takeaways
- Keep security review read-only first; authorize remediation only after confirming evidence, scope, and side effects.
- Secrets, vulnerable packages, excessive permissions, and broken authorization are different failure classes that need different checks.
- Data-retention-auditor follows sensitive data beyond the primary database into logs, analytics, caches, backups, and vendors.
- LLM guardrails need layered input, tool, output, and monitoring controls rather than one prompt instruction.
- No skill certifies compliance or replaces qualified security, privacy, or legal review.
The best Claude security skills split review into specific trust questions. Did a secret escape? Can a dependency be exploited? Does a principal have more access than necessary? Can one identity act as another? Where does sensitive data persist? A single “security audit” prompt usually answers all of these too shallowly.
Use focused, evidence-first skills and keep mutation outside the initial review.
| Skill | Best for | Main output | Default posture |
|---|---|---|---|
| secret-scanner | Exposed credentials | Findings and containment steps | Read-only |
| dependency-audit | Package vulnerabilities | Prioritized remediation | Read-only |
| least-privilege-auditor | IAM and token scopes | Excess-access report | Read-only |
| auth-flow-reviewer | Identity and sessions | Trust-boundary review | Read-only |
| rbac-designer | Role and permission models | RBAC matrix | Design |
| threat-model-builder | Architecture risk | Threat model | Design |
| security-headers-hardener | Browser response policy | Header changes | May edit |
| prompt-pii-redactor | LLM-bound sensitive data | Redaction pipeline | May edit |
| data-retention-auditor | Data lifecycle and deletion | Lineage and control gaps | Read-only |
| llm-guardrails-designer | Agent and LLM controls | Layered guardrail design | Design |
1. secret-scanner: contain exposed credentials
secret-scanner inspects tracked files and relevant history for keys, tokens, passwords, and high-entropy values. A confirmed secret requires revocation or rotation before repository cleanup; deleting the string does not invalidate a credential already copied elsewhere.
2. dependency-audit: prioritize third-party risk
dependency-audit turns scanner output and manifests into a risk-ranked upgrade plan. It considers reachability, exploitability, runtime exposure, fix availability, and breaking-change cost instead of treating every advisory as equally urgent.
3. least-privilege-auditor: reduce excessive access
least-privilege-auditor traces principals, actions, resources, conditions, inheritance, and credential paths across IAM, CI workflows, service accounts, Kubernetes RBAC, and application tokens. It proposes a narrower policy only after identifying what the workload actually performs.
4. auth-flow-reviewer: follow identity end to end
auth-flow-reviewer reviews credential entry, session creation, token validation, identity propagation, authorization checks, refresh, logout, recovery, and account linking. Use it whenever authentication middleware or a privileged user journey changes.
5. rbac-designer: make authorization understandable
rbac-designer maps resources and actions to roles while avoiding role explosion and hidden privilege inheritance. It is strongest before implementation or when an existing permissions model has become impossible to explain.
6. threat-model-builder: find missing controls early
threat-model-builder identifies assets, actors, entry points, trust boundaries, data flows, threats, existing mitigations, and residual risk. It focuses effort on realistic abuse paths rather than producing a generic threat checklist.
7. security-headers-hardener: protect browser boundaries
security-headers-hardener reviews Content Security Policy, transport security, framing, MIME sniffing, referrer behavior, browser capabilities, and cross-origin isolation. Roll out restrictive policies with reporting and compatibility checks rather than breaking production integrations at once.
8. prompt-pii-redactor: minimize LLM exposure
prompt-pii-redactor identifies sensitive values before they reach model providers, logs, traces, or eval datasets. It defines detection, replacement, reversibility, false-positive handling, and safe observability around the redaction step.
9. data-retention-auditor: trace every copy
data-retention-auditor follows sensitive fields through databases, object stores, queues, logs, analytics, search indexes, caches, exports, backups, and vendors. It compares actual TTL, deletion, and restoration behavior with the supplied policy without pretending to provide legal certification.
10. llm-guardrails-designer: layer model controls
llm-guardrails-designer designs controls around input classification, retrieval, prompt injection, tool authorization, output validation, content policy, rate limits, human approval, and monitoring. No single model instruction should be the only boundary protecting a privileged action.
Recommended security stack
Install the four broad review skills first, then add specialists based on architecture:
npx agentscamp add skills/secret-scanner
npx agentscamp add skills/dependency-audit
npx agentscamp add skills/least-privilege-auditor
npx agentscamp add skills/auth-flow-reviewerRequire each report to include evidence location, impact, exploit or failure preconditions, confidence, remediation, and verification. A finding without a reproducible path should remain a hypothesis, not become an urgent production change.
Continue exploring
- 10 Best Claude Skills for Building AI Applications — Compare Claude skills for structured output, fallbacks, evals, hallucinations, routing, token cost, semantic caching, PII, guardrails, and tools.
- 9 Best Claude Skills for Code Review — Choose Claude skills for security, correctness, maintainability, and focused pull request review without relying on one vague review prompt.
Frequently asked questions
- What Claude security skills should I install first?
- Start with secret-scanner, dependency-audit, least-privilege-auditor, and auth-flow-reviewer. They cover accidental exposure, third-party risk, excessive access, and identity enforcement—the most common review surfaces.
- Can Claude automatically fix every security finding?
- It should not by default. Credential rotation, policy changes, dependency upgrades, and authentication edits can disrupt live systems. Review the evidence and approve a scoped remediation with verification and rollback.
- Does data-retention-auditor provide legal compliance advice?
- No. It maps actual technical data flows and controls against requirements you provide. Legal interpretation, jurisdiction, exemptions, and compliance certification remain with qualified professionals.
- Which skill should I use for an LLM application?
- Use llm-guardrails-designer for layered controls, prompt-pii-redactor for sensitive inputs, threat-model-builder for the full architecture, and least-privilege-auditor for agent tools and service credentials.
Related
- Data Retention AuditorAudit where personal or sensitive data is collected, copied, retained, backed up, logged, exported, and deleted, then compare actual lifecycle behavior with stated policy. Use when adding a data field, preparing a privacy review, implementing account deletion, reducing stored data, or checking whether retention controls work across primary and secondary systems.
- 9 Best Claude Skills for Code ReviewChoose Claude skills for security, correctness, maintainability, and focused pull request review without relying on one vague review prompt.
- Least Privilege AuditorAudit application, CI, cloud, database, and agent permissions against observed usage, then produce a safe reduction plan with verification and rollback. Use when credentials have broad scopes, service roles grew organically, CI tokens can write too much, an agent or MCP server has excessive tools, or before a security review or compliance audit.
- LLM Guardrails DesignerDesign input and output guardrails for an LLM app — decide what to check (injection patterns, PII, secrets, policy, schema, leakage, toxicity), place them as input vs. output rails, implement with a library like NeMo Guardrails or LLM Guard, and fail closed. Use when adding a safety/validation layer around an LLM, not relying on the prompt alone.
- Secret ScannerScan a repo or a diff for committed secrets — API keys, tokens, private keys, .env files, and high-entropy strings — then triage real leaks from fixtures. Use before pushing, in review, or when a credential may have leaked.
- Threat Model BuilderBuild a practical threat model for a feature or system using STRIDE — diagram the data flow, mark trust boundaries, enumerate concrete threats where data crosses them, and prioritize by likelihood × impact so security is reasoned about before shipping instead of bolted on after. Use when designing a feature that touches auth, money, or sensitive data, running a security design review, or hardening before a launch.
- 10 Best Claude Skills for Building AI ApplicationsCompare Claude skills for structured output, fallbacks, evals, hallucinations, routing, token cost, semantic caching, PII, guardrails, and tools.