security-owasp

OWASP Top 10:2025 — Code-Level Security

Tactical security patterns for Kotlin, Go, Java, and Node.js on NAIS, aligned with the 2025 OWASP Top 10.

Complements @security-champion-agent (architecture-level threat modeling) and the security-review skill (scanning tools).

Full code examples for each category: see examples.md in this skill directory.

A01: Broken Access Control (incl. SSRF)

// ❌ IDOR — trusts user-supplied ID without ownership check
get("/api/vedtak/{id}") {
    val vedtak = vedtakRepository.findById(call.parameters["id"]!!.toLong())
    call.respond(vedtak)
}

// ✅ Verify ownership before returning resource
get("/api/vedtak/{id}") {
    val bruker = call.hentBruker()
    val vedtak = vedtakRepository.findById(call.parameters["id"]!!.toLong())
        ?: return@get call.respond(HttpStatusCode.NotFound)
    if (vedtak.brukerId != bruker.id) return@get call.respond(HttpStatusCode.Forbidden)
    call.respond(vedtak.toDTO())
}

// ✅ SSRF prevention — validate outbound URL against allowlist
func fetchExternal(targetURL string) error {
    parsed, err := url.Parse(targetURL)
    if err != nil { return err }
    if !isAllowedHost(parsed.Host) { return fmt.Errorf("host not allowed: %s", parsed.Host) }
    // proceed with request
}

• Deny by default — require explicit grants, not explicit denials
• Resource-level checks — not just "is authenticated" but "owns this resource"
• M2M tokens — validate azp claim against AZURE_APP_PRE_AUTHORIZED_APPS
• SSRF — validate outbound URLs; use Nais accessPolicy.outbound as defense-in-depth

A02: Security Misconfiguration

// ❌ Open CORS
install(CORS) { anyHost() }

// ✅ Restrict to known origins
install(CORS) { allowHost("my-app.intern.nav.no", schemes = listOf("https")) }

// ❌ Debug endpoint exposed on public ingress
mux.HandleFunc("/debug/pprof/", pprof.Index)

// ✅ Debug endpoints on separate internal-only port (Nais handles this)
internalMux := http.NewServeMux()
internalMux.HandleFunc("/debug/pprof/", pprof.Index)
go http.ListenAndServe(":9090", internalMux) // not exposed via ingress

• CORS restricted to known origins — never * or anyHost()
• Debug/admin endpoints not on public ingress
• Error responses sanitized — no stack traces, SQL errors, or file paths to client
• Default-deny Nais accessPolicy — explicit inbound/outbound only

A03: Software Supply Chain Failures (NEW in 2025)

// go.sum provides integrity verification — always commit it
// Use govulncheck for known vulnerabilities
// $ govulncheck ./...

// ✅ Pin dependencies to exact versions in go.mod
require (
    golang.org/x/crypto v0.31.0
    github.com/jackc/pgx/v5 v5.7.2
)

// build.gradle.kts — use dependency locking and BOM
dependencyLocking { lockAllConfigurations() }
dependencyManagement {
    imports { mavenBom("org.springframework.boot:spring-boot-dependencies:3.4.1") }
}
// Run: ./gradlew dependencies --write-locks

# ✅ GitHub Actions — pin to full SHA, never @main or floating tags
- uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
# ❌ Vulnerable to supply chain attack
- uses: actions/checkout@main

• Pin all dependencies to exact versions; use lockfiles
• Scan dependencies: govulncheck ./..., trivy repo ., ./gradlew dependencyCheckAnalyze
• GitHub Actions pinned to full commit SHA (not tags)
• Generate SBOM for production artifacts when possible
• Prefer well-maintained, first-party packages

A04: Cryptographic Failures

// ❌ Disabling TLS verification
client := &http.Client{
    Transport: &http.Transport{
        TLSClientConfig: &tls.Config{InsecureSkipVerify: true},
    },
}

// ✅ Default TLS config (Go enforces TLS 1.2+ by default)
client := &http.Client{}

// ❌ Weak hashing for passwords
val hash = MessageDigest.getInstance("MD5").digest(password.toByteArray())

// ✅ bcrypt for password hashing
val hashed = BCrypt.hashpw(password, BCrypt.gensalt(12))

• Passwords: bcrypt (cost ≥ 12) or argon2id — never MD5/SHA-1/SHA-256
• Secrets: always from Nais environment variables or Secret resources — never hardcoded
• TLS 1.2+: never set InsecureSkipVerify: true
• Encryption: AES-256-GCM; rotate keys periodically

A05: Injection

// ❌ SQL injection via string template
session.run(queryOf("SELECT * FROM vedtak WHERE status = '$status'").map { ... }.asList)

// ✅ Parameterized query (Kotliquery)
session.run(queryOf("SELECT * FROM vedtak WHERE status = ?", status).map { ... }.asList)

// ❌ Command injection via shell
exec.Command("sh", "-c", fmt.Sprintf("process %s", userInput)).Run()

// ✅ Pass arguments directly (no shell interpretation)
exec.Command("process", userInput).Run()

• All SQL queries parameterized (? / $1) — never string concatenation
• No shell execution with user-controlled input
• Validate/sanitize all external input at service boundary

A09: Security Logging and Alerting Failures

// ✅ Structured logging with correlation ID, no PII
log.info("Vedtak opprettet", kv("vedtakId", vedtak.id), kv("sakId", sak.id))

// ❌ PII in logs — GDPR violation
log.info("Vedtak for bruker ${bruker.fnr}")

• No PII in logs (fnr, name, address, tokens)
• Audit trail for sensitive operations (vedtak, utbetaling, tilgang)
• Correlation IDs propagated across services (OpenTelemetry trace context)
• Alerting on anomalous patterns (auth failures, rate spikes)

A10: Mishandling of Exceptional Conditions (NEW in 2025)

// ❌ Panic leaks to caller, crashes service
func processRequest(data []byte) Result {
    var req Request
    json.Unmarshal(data, &req) // ignores error, req may be zero-value
    return handle(req)
}

// ✅ Handle errors explicitly, fail safely
func processRequest(data []byte) (Result, error) {
    var req Request
    if err := json.Unmarshal(data, &req); err != nil {
        return Result{}, fmt.Errorf("invalid request payload: %w", err)
    }
    return handle(req)
}

// ❌ Swallowing exceptions silently
fun process(data: String): Result {
    try { return parse(data) }
    catch (e: Exception) { return Result.empty() } // silent failure, no logging
}

// ✅ Log, wrap, and surface errors appropriately
fun process(data: String): Result {
    return try { parse(data) }
    catch (e: Exception) {
        log.error("Parsing failed", kv("error", e.message))
        throw ServiceException("Could not process input", e)
    }
}

• Always handle errors — never ignore returned errors in Go
• Recover from panics at HTTP handler boundaries (middleware)
• Fail securely: deny access by default when state is uncertain
• Sanitize error messages: internal details stay in logs, not in responses
• Centralized error handling via middleware/exception mappers

Quick Reference Checklist

• A01 — Resource-level access checks on every endpoint (not just auth)
• A01 — M2M tokens validate azp against pre-authorized apps
• A01 — Outbound URLs validated; Nais egress policy configured
• A02 — CORS restricted to known origins
• A02 — Debug endpoints not on public ingress
• A02 — Error responses sanitized (no stack traces to client)
• A03 — Dependencies pinned to exact versions with lockfiles
• A03 — govulncheck / trivy repo . pass without HIGH/CRITICAL
• A03 — GitHub Actions pinned to full SHA
• A04 — bcrypt/argon2id for passwords, never MD5/SHA-1
• A04 — TLS 1.2+ enforced, no InsecureSkipVerify
• A04 — Secrets from environment/Nais, never hardcoded
• A05 — All SQL queries parameterized (? / $1)
• A05 — No shell execution with user input
• A07 — JWT validates exp, iss, aud, and algorithm
• A08 — Deserialization into concrete types only
• A09 — No PII in logs (fnr, name, address)
• A09 — Audit trail for sensitive operations
• A10 — All errors handled (no ignored returns in Go)
• A10 — Panic recovery in HTTP handlers
• A10 — Error messages sanitized before client responses

Related

Resource	Use For
security-review skill	Pre-commit scanning (trivy, zizmor, govulncheck)
@security-champion-agent	Threat modeling, compliance, Nav security architecture
@auth-agent	JWT validation, TokenX, ID-porten implementation
threat-model skill	STRIDE-A analysis for new services
OWASP Top 10:2025	Official category descriptions
OWASP Go SCP	Go-specific secure coding guide
OWASP CI/CD Top 10	Pipeline security risks
sikkerhet.nav.no	Nav Golden Path

Boundaries

✅ Always

• Parameterized queries for all SQL
• Resource-level access checks on every data-returning endpoint
• Structured logging without PII
• SHA-pinned GitHub Actions
• Explicit error handling (no ignored errors)
• Dependencies scanned before release

⚠️ Ask First

• Custom cryptographic implementations
• Disabling security features for testing
• Changing authentication or authorizati

---

Content truncated.