القائمة
Discover and inventory shadow API endpoints that operate outside documented specifications using traffic analysis, code scanning, and API discovery platforms.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Identifying and exploiting dangling DNS records pointing to unclaimed cloud services, enabling subdomain takeover for phishing, cookie stealing, and authentication bypass during authorized penetration tests.
Detects prompt injection attacks targeting LLM-based applications using a multi-layered defense combining regex pattern matching for known attack signatures, heuristic scoring for structural anomalies, and transformer-based classification with DeBERTa models. The detector analyzes user inputs before they reach the LLM, flagging direct injections (system prompt overrides, role-play escapes, instruction hijacking) and indirect injections (encoded payloads, multi-language obfuscation, delimiter-based escapes). Based on the OWASP LLM Top 10 (LLM01:2025 Prompt Injection) and Simon Willison's prompt injection taxonomy. Activates for requests involving prompt injection detection, LLM input sanitization, AI security scanning, or prompt attack classification.
Testing machine-learning model files and model-loading services for remote code execution caused by insecure deserialization (pickle/PyTorch), unsafe config instantiation (Hydra), archive path traversal, and dangerous layer types during authorized penetration tests of AI/ML pipelines.
Implements input and output validation guardrails for LLM-powered applications to prevent prompt injection, data leakage, toxic content generation, and hallucinated outputs. Builds a security validation pipeline using NVIDIA NeMo Guardrails Colang definitions, custom Python validators for PII detection and content policy enforcement, and the Guardrails AI framework for structured output validation. The guardrails system intercepts both user inputs (blocking injection attempts, stripping PII, enforcing topic boundaries) and model outputs (detecting hallucinations, filtering toxic content, validating JSON schema compliance). Activates for requests involving LLM output validation, AI content filtering, guardrail implementation, or LLM safety enforcement.
Using LLMs to accelerate vulnerability research and pentest workflows — generating syntax-valid fuzzing seeds and evolving grammars, fine-tuned mutation dictionaries, parallel agent-based proof-of-vulnerability generation, and evidence-driven passive analysis of real HTTP traffic via the Burp MCP server. Covers concrete prompts, AFL++/ libFuzzer wiring, and Burp+Codex/Gemini/Ollama MCP setup.
Testing LLM-backed applications, chatbots, and AI agents for direct and indirect prompt injection, jailbreaks, system-prompt leakage, and tool/agent abuse during authorized penetration tests, using structured payload families and reliable confirmation signals.
| name | detecting-shadow-api-endpoints |
| description | Discover and inventory shadow API endpoints that operate outside documented specifications using traffic analysis, code scanning, and API discovery platforms. |
| domain | cybersecurity |
| subdomain | api-security |
| tags | ["api-security","shadow-apis","api-discovery","undocumented-apis","zombie-apis","api-inventory","attack-surface-management","api-governance"] |
| version | 1.0 |
| author | mahipal |
| license | Apache-2.0 |
| nist_csf | ["PR.PS-01","ID.RA-01","PR.DS-10","DE.CM-01"] |
Shadow APIs are API endpoints operating within an organization's environment that are not tracked, documented, or secured. They emerge from rapid development cycles, forgotten test environments, deprecated API versions left running, third-party integrations, or developer side projects deployed without governance. Shadow APIs bypass authentication and monitoring controls, creating hidden entry points for attackers. Studies show that up to 30% of API endpoints in large organizations are undocumented, making shadow API detection a critical component of API security posture management.
/users/{id} vs observed /users/{uuid}) make documented endpoints look undocumented (false positive) or hide real shadows; verify your PARAM_PATTERNS cover hashids, ULIDs, and multi-segment IDs.startswith('/api')//v drops shadow endpoints under /internal, /graphql, /_next, /actuator, host-based vhosts, and non-standard ports - widen capture before concluding "no shadows."v0, v1 left running after v2 launch rarely appear in current traffic; enumerate from source/CI and cloud (Lambda URLs, ALB rules) too, not just logs.How to validate the detection fires: stand up a deliberately undocumented test route, send traffic, and confirm it surfaces as shadow with correct risk scoring. Tune false positives by reconciling against every OpenAPI/Swagger source (not one file) and excluding health/metrics probes that are known-and-accepted but undocumented.
Compare live API traffic against documented OpenAPI specifications to identify undocumented endpoints:
#!/usr/bin/env python3
"""Shadow API Endpoint Detector
Compares observed API traffic patterns against documented
OpenAPI specifications to identify undocumented (shadow) endpoints.
"""
import json
import re
import yaml
import sys
from collections import defaultdict
from datetime import datetime
from typing import Dict, List, Set, Tuple, Optional
from dataclasses import dataclass, field
@dataclass
class DiscoveredEndpoint:
method: str
path_pattern: str
first_seen: str
last_seen: str
request_count: int
source_ips: Set[str] = field(default_factory=set)
status_codes: Set[int] = field(default_factory=set)
has_auth_header: bool = False
documented: bool = False
class ShadowAPIDetector:
# Common patterns for parameterized path segments
PARAM_PATTERNS = [
(re.compile(r'/\d+'), '/{id}'),
(re.compile(r'/[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}'), '/{uuid}'),
(re.compile(r'/[a-zA-Z0-9]{20,40}'), '/{token}'),
]
def __init__(self):
self.documented_endpoints: Set[Tuple[str, str]] = set()
self.discovered_endpoints: Dict[Tuple[str, str], DiscoveredEndpoint] = {}
def load_openapi_spec(self, spec_path: str):
"""Load documented endpoints from OpenAPI specification."""
with open(spec_path, 'r') as f:
if spec_path.endswith('.json'):
spec = json.load(f)
else:
spec = yaml.safe_load(f)
paths = spec.get('paths', {})
for path, methods in paths.items():
# Normalize OpenAPI path parameters
normalized_path = re.sub(r'\{[^}]+\}', '{id}', path)
for method in methods:
if method.upper() in ('GET', 'POST', 'PUT', 'DELETE', 'PATCH', 'HEAD', 'OPTIONS'):
self.documented_endpoints.add((method.upper(), normalized_path))
print(f"Loaded {len(self.documented_endpoints)} documented endpoints from {spec_path}")
def normalize_path(self, path: str) -> str:
"""Normalize an observed path by replacing dynamic segments with placeholders."""
# Remove query string
path = path.split('?')[0]
for pattern, replacement in self.PARAM_PATTERNS:
path = pattern.sub(replacement, path)
return path
def process_access_log(self, log_file: str, log_format: str = "common"):
"""Process API access logs to discover endpoints."""
patterns = {
"common": re.compile(
r'(?P<ip>[\d.]+)\s+\S+\s+\S+\s+\[(?P<time>[^\]]+)\]\s+'
r'"(?P<method>\w+)\s+(?P<path>\S+)\s+\S+"\s+(?P<status>\d+)'
),
"json": None # Handle JSON logs separately
}
with open(log_file, 'r') as f:
for line in f:
if log_format == "json":
try:
entry = json.loads(line)
method = entry.get('method', entry.get('http_method', ''))
path = entry.get('path', entry.get('uri', ''))
status = int(entry.get('status', entry.get('status_code', 0)))
ip = entry.get('remote_addr', entry.get('client_ip', ''))
timestamp = entry.get('timestamp', entry.get('@timestamp', ''))
has_auth = bool(entry.get('authorization', entry.get('auth_header', '')))
except json.JSONDecodeError:
continue
else:
match = patterns[log_format].match(line)
if not match:
continue
method = match.group('method')
path = match.group('path')
status = int(match.group('status'))
ip = match.group('ip')
timestamp = match.group('time')
has_auth = 'Authorization' in line
# Only process API paths
if not path.startswith('/api') and not path.startswith('/v'):
continue
normalized = self.normalize_path(path)
key = (method.upper(), normalized)
if key not in self.discovered_endpoints:
self.discovered_endpoints[key] = DiscoveredEndpoint(
method=method.upper(),
path_pattern=normalized,
first_seen=timestamp,
last_seen=timestamp,
request_count=0,
documented=(key in self.documented_endpoints)
)
endpoint = self.discovered_endpoints[key]
endpoint.request_count += 1
endpoint.last_seen = timestamp
endpoint.source_ips.add(ip)
endpoint.status_codes.add(status)
if has_auth:
endpoint.has_auth_header = True
def identify_shadow_apis(self) -> List[DiscoveredEndpoint]:
"""Identify endpoints that are not in the documented specification."""
shadows = []
for key, endpoint in self.discovered_endpoints.items():
if not endpoint.documented:
shadows.append(endpoint)
# Sort by request count descending (most active shadows first)
shadows.sort(key=lambda e: e.request_count, reverse=True)
return shadows
def classify_risk(self, endpoint: DiscoveredEndpoint) -> str:
"""Classify the risk level of a shadow endpoint."""
risk_score = 0
# No authentication observed
if not endpoint.has_auth_header:
risk_score += 3
# High traffic volume
if endpoint.request_count > 1000:
risk_score += 2
elif endpoint.request_count > 100:
risk_score += 1
# Multiple source IPs (wider exposure)
if len(endpoint.source_ips) > 10:
risk_score += 2
# Successful responses (endpoint is functional)
if 200 in endpoint.status_codes or 201 in endpoint.status_codes:
risk_score += 1
# Write operations are higher risk
if endpoint.method in ('POST', 'PUT', 'DELETE', 'PATCH'):
risk_score += 2
# Sensitive path patterns
sensitive_patterns = ['admin', 'internal', 'debug', 'test', 'backup',
'config', 'health', 'metrics', 'graphql', 'console']
for pattern in sensitive_patterns:
if pattern in endpoint.path_pattern.lower():
risk_score += 3
break
if risk_score >= 8:
return "CRITICAL"
elif risk_score >= 5:
return "HIGH"
elif risk_score >= 3:
return "MEDIUM"
return "LOW"
def generate_report(self) -> dict:
"""Generate a comprehensive shadow API discovery report."""
shadows = self.identify_shadow_apis()
total_documented = len(self.documented_endpoints)
total_discovered = len(self.discovered_endpoints)
report = {
"scan_date": datetime.now().isoformat(),
"summary": {
"documented_endpoints": total_documented,
"total_discovered_endpoints": total_discovered,
"shadow_endpoints": len(shadows),
"shadow_ratio": f"{len(shadows)/max(total_discovered,1)*100:.1f}%",
},
"shadow_endpoints": []
}
for endpoint in shadows:
risk = self.classify_risk(endpoint)
report["shadow_endpoints"].append({
"method": endpoint.method,
"path": endpoint.path_pattern,
"risk_level": risk,
"request_count": endpoint.request_count,
"unique_sources": len(endpoint.source_ips),
"authenticated": endpoint.has_auth_header,
"status_codes": sorted(endpoint.status_codes),
"first_seen": endpoint.first_seen,
"last_seen": endpoint.last_seen,
})
return report
def main():
detector = ShadowAPIDetector()
# Load documented API specifications
spec_files = sys.argv[1:] if len(sys.argv) > 1 else ["openapi.yaml"]
for spec in spec_files:
if spec.endswith(('.yaml', '.yml', '.json')):
detector.load_openapi_spec(spec)
# Process access logs
detector.process_access_log("/var/log/api/access.log")
report = detector.generate_report()
print(f"\n{'='*60}")
print(f"SHADOW API DISCOVERY REPORT")
print(f"{'='*60}")
print(f"Documented: {report['summary']['documented_endpoints']}")
print(f"Discovered: {report['summary']['total_discovered_endpoints']}")
print(f"Shadow: {report['summary']['shadow_endpoints']} ({report['summary']['shadow_ratio']})")
print()
for ep in report["shadow_endpoints"]:
risk_marker = {"CRITICAL": "[!!!]", "HIGH": "[!!]", "MEDIUM": "[!]", "LOW": "[.]"}
print(f" {risk_marker.get(ep['risk_level'], '[?]')} {ep['method']} {ep['path']}")
print(f" Risk: {ep['risk_level']} | Requests: {ep['request_count']} | Auth: {ep['authenticated']}")
# Save full report
with open("shadow_api_report.json", "w") as f:
json.dump(report, f, indent=2, default=str)
print(f"\nFull report saved to shadow_api_report.json")
if __name__ == "__main__":
main()
# AWS: Discover API Gateway endpoints not in documentation
aws apigateway get-rest-apis --query 'items[*].[name,id]' --output table
# List all routes for each API
aws apigatewayv2 get-apis --query 'Items[*].[Name,ApiId,ProtocolType]' --output table
# AWS Lambda function URLs (potential shadow APIs)
aws lambda list-function-url-configs --function-name "*" 2>/dev/null
# Find ALB listener rules routing to undocumented backends
aws elbv2 describe-rules --listener-arn $LISTENER_ARN \
--query 'Rules[*].[Priority,Conditions[0].Values[0],Actions[0].TargetGroupArn]'
# Search for undocumented route definitions in source code
# Express.js routes
grep -rn "app\.\(get\|post\|put\|delete\|patch\)" --include="*.js" --include="*.ts" src/
# Flask/Django routes
grep -rn "@app\.route\|@api\.route\|path(" --include="*.py" src/
# Spring Boot endpoints
grep -rn "@\(Get\|Post\|Put\|Delete\|Patch\)Mapping\|@RequestMapping" --include="*.java" src/
# Compare found routes against OpenAPI specification
diff <(grep -roh "'/api/[^']*'" src/ | sort -u) \
<(yq '.paths | keys[]' openapi.yaml | sort -u)
# Kong plugin configuration - reject unregistered routes
plugins:
- name: request-validator
config:
allowed_content_types:
- application/json
body_schema: null
- name: pre-function
config:
access:
- |
-- Block requests to unregistered endpoints
local registered = kong.cache:get("registered_endpoints")
local path = kong.request.get_path()
local method = kong.request.get_method()
local key = method .. ":" .. path
if not registered[key] then
kong.log.warn("Shadow API access attempt: ", key)
return kong.response.exit(404, {error = "Endpoint not registered"})
end