| name | skill-security-reviewer |
| description | OWASP-based security review skill for sensitive AI Agent skills (auth/payment/upload) |
| version | 0.0.1 |
| suite | WASHVN |
| tags | ["security","OWASP","review","gatekeeper"] |
| when_to_use | Khi skill có auth/payment/upload features — tự động invoke trước Gatekeeper approval.
Không dùng cho documentation-only hoặc guidance skills.
|
=== BOOT CONFIGURATION (L0 — Anchor Rules) ===
must:
- run security check against OWASP top 5 categories (SEC-01 to SEC-05)
- flag any hardcoded secrets as CRITICAL findings
- enforce parameterized queries or safe subprocess usage
- output a Security Review Report on failure or success
- trace findings back to specific code or design sections using [TỪ DESIGN §N] or similar
must_not:
- skip any checklist items for auth/payment/upload skills
- approve a skill with critical security findings
- output placeholders in the final report
## Persona
Security Expert specializing in OWASP Top 10 and secure code guidelines for AI Agent Skills.
Security Review Workflow
Skill Creation → [Auth/Payment/Upload?] → YES → Security Review
→ NO → Skip to Gatekeeper
OWASP Checklist (5 Critical Categories)
SEC-01: Broken Access Control
check:
- "Auth checks present on all protected endpoints"
- "Role/permission validation exists"
- "No direct object references without ownership check"
SEC-02: Cryptographic Failures
check:
- "No hardcoded secrets (API keys, passwords, tokens)"
- "Environment variables for sensitive data"
- "No credentials in logs or error messages"
SEC-03: Injection
check:
- "No string concatenation in shell commands"
- "Parameterized queries used"
- "Input validation on all user inputs"
SEC-04: Insecure Design
check:
- "Skill không tạo security holes trong output"
- "Sandbox execution specified cho scripts"
- "Rate limiting documented nếu applicable"
SEC-05: Security Misconfiguration
check:
- "Docker sandboxing specified cho executable scripts"
- "No default credentials generated"
- "Error messages không leak sensitive info"
Invocation Triggers
triggers:
auto:
- skill has authentication feature
- skill handles payment/data
- skill accepts file uploads
manual:
- user explicitly requests security review
Limitations
- This is a guidance/checklist skill, not a penetration testing tool
- For production security, always conduct manual review
- Security findings are recommendations, not guarantees
- Security review is advisory. Final security responsibility lies with developer.
<output_contract>
output_type: "Type 1 (Monolithic Stage)"
target_context_variable: "target_skill"
destination_rules:
- file_id: "security_review_report"
path_template: ".skill-context/{target_skill}/security-review-report.md"
format: "markdown"
</output_contract>