Sast

Gate: Scanner found Critical/High findings? Jump to Phase 3 for those immediately. Otherwise continue.

Phase 2 - Manual Review (30 min max)

Review targets in this priority order. For each, grep for the vulnerable pattern, then trace from source to sink.

Priority 1: Auth and access control

# Find auth middleware, decorators, guards
grep -rn "login\|authenticate\|authorize\|@login_required\|isAuthenticated\|requireAuth\|checkPermission" --include="*.{py,js,ts,java,go,php,rb}" .

Look for: missing auth checks on sensitive endpoints, role checks that can be bypassed, session handling flaws.

Gate: Found unprotected endpoint? Trace it. Confirm it handles sensitive data. Write finding with curl PoC.

Priority 2: SQL/NoSQL injection

Gate: Found string concat/interpolation in a query? Trace the variable back to user input. Reachable? Write finding with SQLi payload.

Priority 3: Command injection

Gate: User input reaches shell execution? Write finding with command injection PoC.

Priority 4: Deserialization

Gate: Untrusted data reaches deserialization? Write finding. This is usually Critical (RCE).

Priority 5: File operations

• Path traversal: user input in open(), sendFile(), include(), require()
• File upload: no extension validation, no content-type check, writable to web root
• PHP file inclusion: include($_GET['page']) - check for LFI/RFI

Gate: Can read arbitrary files or upload executable content? Write finding.

Priority 6: SSRF

• requests.get(user_url), http.Get(user_url), HttpURLConnection(user_url) without URL validation
• Check if internal IPs are blocked (127.0.0.1, 169.254.169.254, 10.x, 172.16.x)

Gate: Can reach internal services or cloud metadata? Write finding.

Priority 7: Template injection (SSTI)

• Python: render_template_string(input), Template(input).render()
• Java: ${input} in JSP, #{input} in JSF, SpEL injection
• Ruby: ERB.new(input).result
• PHP: Twig/Blade with raw output of user input

Gate: User input rendered as template code? Write finding. Usually leads to RCE.

Priority 8: Framework-specific

• Flask/Django: debug=True in production, DEBUG=True, exposed admin panel
• Express: missing helmet(), no CSRF, trust proxy misconfiguration
• Rails: mass assignment without strong params, render file: with user input
• Spring: exposed actuator endpoints (/actuator/env, /actuator/heapdump)
• PHP: extract() from user input, type juggling with == instead of ===

Phase 3 - Verify and Prove (per finding)

For EACH finding from Phase 1 or Phase 2:

1. Is the input user-controlled? Trace the variable back to an HTTP parameter, header, cookie, or file upload. If it only comes from config or internal code, deprioritize.
2. Is the code reachable? Find the route/endpoint that triggers it. Is it behind auth? Is it dead code?
3. Is sanitization present? Check for validation, escaping, or parameterization between source and sink that the scanner missed.
4. Write PoC. A curl command, a test script, or exact reproduction steps. Source-only findings without proof of exploitability get rejected.

Gate:

• Reachable + no sanitization + PoC works? Report it.
• Reachable but sanitization present? Check if sanitization is bypassable. If not, discard.
• Not reachable from external input? Discard unless it is a hardcoded secret.

Secrets Detection Patterns

# API keys
grep -rn "AKIA[0-9A-Z]{16}" .                     # AWS Access Key
grep -rn "AIza[0-9A-Za-z\-_]{35}" .               # Google API Key
grep -rn "sk_live_[0-9a-zA-Z]{24}" .              # Stripe Secret Key
grep -rn "ghp_[a-zA-Z0-9]{36}" .                  # GitHub Token
grep -rn "glpat-[a-zA-Z0-9\-]{20}" .              # GitLab Token

# Passwords and connection strings
grep -rn "password\s*=\s*['\"]" .
grep -rn "mongodb://.*:.*@\|postgres://.*:.*@\|mysql://.*:.*@" .

# Private keys
grep -rn "BEGIN.*PRIVATE KEY" .

# JWTs (may contain secrets in payload)
grep -rn "eyJ[a-zA-Z0-9_-]*\.eyJ[a-zA-Z0-9_-]*\." .

Gate: Found a secret? Verify it is still active (try to use it). Active production secret? Report immediately.

Evidence Requirements

Every finding needs ALL of these:

1. File path and line number
2. Vulnerable code snippet with context
3. Data flow trace: source (user input) to sink (dangerous operation)
4. Proof the code path is reachable from an external request
5. Working PoC (curl command or script)
6. Impact statement
7. Suggested fix with corrected code