القائمة
Securing AI/ML infrastructure including model storage, API endpoints, and compute resources
يُحدَّد Prompt الذي يراجع المصدر أولًا بشكل افتراضي. يمكنك التبديل إلى أمر مباشر أو تنزيل نسخة محلية.
اقرأ SKILL.md وأي ملفات مرافقة يعرضها SkillsMP قبل أن تقرر التثبيت.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
يتجاوز الأمر المباشر Prompt المخصّص للمراجعة. افحص المصدر قبل تشغيله.
npx skills add https://github.com/pluginagentmarketplace/custom-plugin-ai-red-teaming --skill infrastructure-securityيبقى الأمر في سطر واحد. مرّر أفقيًا لمراجعته كاملًا قبل النسخ.
تفضّل نسخة محلية؟ نزّل الملفات المتاحة حاليًا لدى SkillsMP.
CI/CD integration and automation frameworks for continuous AI security testing
Standard datasets and benchmarks for evaluating AI security, robustness, and safety
Professional certifications, CTF competitions, and training resources for AI security practitioners
استنادا إلى تصنيف SOC المهني
عرض SKILL.md
| name | infrastructure-security |
| version | 2.0.0 |
| description | Securing AI/ML infrastructure including model storage, API endpoints, and compute resources |
| sasmp_version | 1.3.0 |
| bonded_agent | 06-api-security-tester |
| bond_type | PRIMARY_BOND |
| input_schema | {"type":"object","required":["assessment_scope"],"properties":{"assessment_scope":{"type":"string","enum":["api","storage","compute","network","full"]},"environment":{"type":"string","enum":["cloud","on_prem","hybrid"]}}} |
| output_schema | {"type":"object","properties":{"vulnerabilities":{"type":"array"},"compliance_status":{"type":"object"},"recommendations":{"type":"array"}}} |
| owasp_llm_2025 | ["LLM03","LLM10"] |
| nist_ai_rmf | ["Govern","Manage"] |
Protect AI/ML infrastructure from attacks targeting model storage, APIs, and compute resources.
Skill: infrastructure-security
Agent: 06-api-security-tester
OWASP: LLM03 (Supply Chain), LLM10 (Unbounded Consumption)
NIST: Govern, Manage
Use Case: Secure AI deployment infrastructure
[External Threats]
↓
[API Gateway] → [Load Balancer] → [Inference Servers]
↓ ↓ ↓
[Rate Limit] [DDoS Protection] [Model Storage]
↓ ↓ ↓
[Auth/AuthZ] [TLS Termination] [Secrets Manager]
Authentication:
methods:
- API keys (rotation: 90 days)
- OAuth 2.0 / OIDC
- mTLS for service-to-service
requirements:
- Strong key generation
- Secure transmission
# API Security Configuration
class APISecurityConfig:
def __init__(self):
self.auth_config = {
'type': 'oauth2',
'token_expiry': 3600,
'refresh_enabled': True,
}
self.rate_limits = {
'default': {'requests': 100, 'window': 60},
'premium': {'requests': 1000, 'window': 60},
'burst_multiplier': 2,
}
self.input_validation = {
'max_tokens': 4096,
'blocked_patterns': self._load_blocked_patterns(),
'sanitization': True,
}
Storage Security:
encryption: AES-256-GCM
access_control: RBAC
audit_logging: enabled
backup: encrypted, offsite
Theft Prevention:
query_limits: 10000/day per user
output_perturbation: enabled
watermarking: model and output
access_logging: all queries
class ModelProtection:
def __init__(self, model):
self.model = model
self.watermark = self._generate_watermark()
def protected_inference(self, input_data, user_id):
# Log the query
self.log_query(user_id, input_data)
# Check query limits
if self.exceeds_limit(user_id):
raise RateLimitError("Query limit exceeded")
# Run inference
output = self.model(input_data)
# Add output perturbation (anti-extraction)
output = self.add_perturbation(output)
# Apply watermark
output = self.apply_watermark(output)
return output
Network Configuration:
internal_only: true
vpc_isolation: enabled
firewall_rules:
- allow: internal_services
- deny: all_external (except API gateway)
TLS Configuration:
version: "1.3"
cipher_suites: [TLS_AES_256_GCM_SHA384]
certificate_rotation: 90 days
mtls: service_to_service
Container Security:
base_image: distroless
user: non-root
filesystem: read-only
capabilities: minimal
seccomp: enabled
Resource Limits:
cpu: 4 cores max
memory: 16GB max
gpu_memory: 24GB max
disk: ephemeral only
Isolation:
runtime: gvisor
network: namespace isolated
secrets: mounted, not in env
API Layer:
- [ ] Strong authentication (OAuth2/mTLS)
- [ ] Rate limiting implemented
- [ ] Input validation enabled
- [ ] Error messages sanitized
- [ ] Logging comprehensive
Storage Layer:
- [ ] Encryption at rest
- [ ] Access controls configured
- [ ] Audit logging enabled
- [ ] Backup encryption
Network Layer:
- [ ] TLS 1.3 enforced
- [ ] Internal VPC only
- [ ] Firewall rules configured
- [ ] DDoS protection enabled
Compute Layer:
- [ ] Non-root containers
- [ ] Resource limits set
- [ ] Secrets in vault
- [ ] Immutable infrastructure
class InfrastructureSecurityTester:
def test_api_security(self, endpoint):
results = []
# Test authentication bypass
results.append(self.test_auth_bypass(endpoint))
# Test rate limiting
results.append(self.test_rate_limits(endpoint))
# Test input validation
results.append(self.test_input_validation(endpoint))
# Test error handling
results.append(self.test_error_disclosure(endpoint))
return results
def test_auth_bypass(self, endpoint):
payloads = [
{'Authorization': ''},
{'Authorization': 'Bearer invalid'},
{'Authorization': 'Bearer ' + 'a' * 1000},
]
for payload in payloads:
response = requests.get(endpoint, headers=payload)
if response.status_code != 401:
return Finding("auth_bypass", "CRITICAL")
return None
CRITICAL:
- Authentication bypass
- Model theft possible
- Data exposure
HIGH:
- Rate limiting bypassable
- Weak encryption
- Insufficient logging
MEDIUM:
- Missing input validation
- Verbose error messages
- Outdated dependencies
LOW:
- Non-optimal configurations
- Minor policy gaps
Issue: API rate limiting not effective
Solution: Implement token-based limits, add IP reputation
Issue: Model extraction detected
Solution: Lower query limits, add output perturbation
Issue: High latency from security layers
Solution: Optimize validation, use caching, async logging
| Component | Purpose |
|---|---|
| Agent 06 | Security testing |
| Agent 08 | CI/CD security gates |
| /test api | Security scanning |
| SIEM | Security monitoring |
Protect AI infrastructure with defense-in-depth security.