Security Deep Dive

Security Deep Dive

Think like an attacker. What would you try to break?

Progress Checklist

• Define scope and threat model
• Map the attack surface
• Analyze auth flows
• Audit LLM/GenAI threat surface (if applicable)
• Audit dependency chain
• Review infrastructure config
• Attempt exploitation paths
• Report findings with severity

Step 0: Scope and Threat Model

Before diving in, understand what we're protecting:

• What data is sensitive? (PII, credentials, financial, health)
• Who are the threat actors? (anonymous users, authenticated users, insiders, automated attacks)
• What's the impact of a breach? (data leak, financial loss, reputation, compliance violation)
• What's already in place? (auth, encryption, monitoring, WAF)

Step 1: Attack Surface Mapping

List every entry point:

# Find API endpoints
grep -rn "router\.\|app\.\(get\|post\|put\|delete\|patch\)" --include="*.ts" --include="*.js" --include="*.go"
grep -rn "\[Http\(Get\|Post\|Put\|Delete\|Patch\)\]" --include="*.cs"
grep -rn "@app\.\(route\|get\|post\)" --include="*.py"

# Find forms and user input
grep -rn "<form\|<input\|<textarea\|<select" --include="*.html" --include="*.tsx" --include="*.vue"

# Find file upload handlers
grep -rn "upload\|multipart\|formData\|IFormFile" -l

# Find webhook/queue consumers
grep -rn "webhook\|queue\|consumer\|subscriber" -l

For each entry point, note: authentication requirement, input sources, data sensitivity.

Step 2: Auth Flow Analysis

Trace the full authentication and authorization path:

1. How are credentials submitted? (form, header, cookie)
2. How are they validated? (database lookup, JWT verification, OAuth flow)
3. How is the session maintained? (JWT, session cookie, token refresh)
4. How is authorization checked? (middleware, per-endpoint, attribute-based)
5. What happens on auth failure? (error message, rate limiting, lockout)

Look for:

• Missing auth on endpoints that should require it
• Authorization checks that only check authentication ("is logged in" but not "can access this resource")
• IDOR vulnerabilities (can user A access user B's data by changing an ID?)
• Token handling issues (stored in localStorage? No expiry? No rotation?)

Step 2b: LLM/GenAI Threat Surface

Check if the codebase integrates LLMs:

rg -l "openai|anthropic|langchain|@ai-sdk|ChatCompletion|messages\.create|semantic[.-]kernel|Microsoft\.SemanticKernel|GenerateContent" --type-add 'code:*.{ts,js,py,cs,go}' -t code

If matches found, load ../_shared/owasp-llm-top-10.md and check:

1. Prompt construction - How are prompts built? Is user input concatenated directly, or separated into distinct message roles?
2. Output consumption - Where does LLM output go? Is it rendered as HTML, used in SQL, passed to eval/exec, or used in file paths?
3. Tool/function calling - What tools does the LLM have access to? Can it execute shell commands, run arbitrary SQL, or fetch arbitrary URLs?
4. RAG pipeline - Is there a vector store? Are queries filtered by tenant? Can users inject content into the knowledge base?
5. System prompts - Do they contain secrets, internal URLs, or security-sensitive logic?
6. Rate limiting - Are LLM endpoints rate-limited? Are there token budgets and timeouts?
7. Human-in-the-loop - Do destructive actions triggered by the LLM require user approval?

For each finding, assess against the OWASP LLM Top 10 severity levels.

Step 3: Dependency Chain Audit

# JavaScript
npm audit 2>/dev/null
npx better-npm-audit audit 2>/dev/null

# .NET
dotnet list package --vulnerable 2>/dev/null

# Go
govulncheck ./... 2>/dev/null

# Check for abandoned packages
# Look for: last publish date, open issues, maintainer activity

Beyond known CVEs:

• Any dependencies pulling in unexpected transitive dependencies?
• Any dependencies with excessive permissions (file system, network)?
• Supply chain risk: are dependencies from trusted sources?

Step 4: Infrastructure Config

# Docker
grep -r "USER\|EXPOSE\|ENV\|ARG.*secret\|COPY.*\.env" Dockerfile* 2>/dev/null

# Kubernetes
grep -rn "securityContext\|privileged\|runAsRoot\|capabilities" k8s/ 2>/dev/null
grep -rn "kind: Secret" k8s/ 2>/dev/null

# Environment and secrets
grep -rn "password\|secret\|key\|token" .env* docker-compose* 2>/dev/null

Check for:

• Containers running as root
• Exposed ports that shouldn't be public
• Default credentials
• Secrets in plaintext config
• Missing network policies
• Overly permissive CORS

Step 5: Exploitation Paths

For each finding, think through the attack:

1. How would you exploit this? Be specific: what request, what payload, what sequence
2. What's the impact? Data exposure, privilege escalation, denial of service, RCE
3. What's the likelihood? Does the attacker need to be authenticated? Special knowledge?
4. How easy is it? Script kiddie vs. targeted attack

Step 6: Report

For each finding:

1. What the vulnerability is
2. Where it exists (file, line, endpoint)
3. How it could be exploited
4. Recommended fix
5. Severity with justification

See ../_shared/security-checklist.md for the detailed checklist, ../codebase-audit/references/owasp-top-10.md for OWASP patterns, and ../_shared/owasp-llm-top-10.md for LLM-specific risks.