Sentinel and OWASP ASVS 4.0.3

The OWASP Application Security Verification Standard (ASVS) lists what a secure application must do. Sentinel is a gateway, not the application itself, so for each requirement this table says what Sentinel does about it, and what is still the protected app’s job.

Requirement numbers and wording are from ASVS 4.0.3 (summarised).

Status:

  • Met: Sentinel does this for every app behind it.
  • Helps: Sentinel adds a layer of defence, but the requirement is about how the app itself is built, so the app still has to meet it.
  • App: only the app can do this. Sentinel can’t.
  • Partly: Sentinel does part of it. The note says what’s missing.
  • Pro: met by Sentinel Pro’s zero-trust features (logins for your apps, access policies, mTLS, risk-based step-up, tamper-evident audit log); the free edition leaves it to the app.

V2 Authentication

ASVS Requirement (summary) Status How Sentinel meets it
2.1.1 Passwords at least 12 characters Partly Sentinel’s own accounts are its dashboard administrators, created through the first-run setup. There is no self-service password change yet, so there is no length check at change time.
2.2.1 Anti-automation against brute force and credential stuffing Met Rate limit on the login API (/__sentinel/auth/=10/1m, counted per IP), flood bans, and a per-account limit on failed logins from any IP (password spraying). Five wrong 2FA codes end the login attempt.
2.4.1 Passwords stored resistant to offline attacks Met bcrypt hashes.
2.4.4 bcrypt work factor as high as practical (min. 10) Met Cost 12.
2.8.1 Time-based OTPs have a defined lifetime Met TOTP per RFC 6238: 30-second steps, ±1 step tolerance. The MFA token that waits for the code lasts 5 minutes.
2.8.3 Approved algorithms for OTPs Met HMAC-SHA1 as RFC 6238 specifies. Secrets come from a CSPRNG.
2.8.4 A TOTP code can be used only once Met The last accepted time step is stored per user, and replays are rejected.
2.8.5 A reused TOTP code is logged and rejected Met Rejected with “already used”, logged as mfa_failed.
2.10.1 Service-to-service auth doesn’t rely on unchanging passwords or API keys Pro mTLS for services (MTLS_ROUTES).
2.10.4 Secrets not in source code Partly All secrets come from environment variables, and the 2FA secrets in the database are encrypted with AES-256-GCM. The lab’s docker-compose.yml contains lab-only values, marked as such.

V3 Session management

ASVS Requirement (summary) Status How Sentinel meets it
3.3.1 Logout and expiry invalidate the session token Met Tokens expire after 15 minutes. Logout puts the token’s ID on a revocation list in PostgreSQL.
3.3.2 Periodic re-authentication Met 15-minute tokens. There are no refresh tokens, so users log in again.
3.5.2 Session tokens rather than static API secrets Met Short-lived signed JWTs for people, certificates for services.
3.5.3 Stateless tokens protected against tampering, null cipher, key substitution Met HMAC-SHA256 signatures. Only HS256 is accepted (alg: none and algorithm switching are refused). The signing key is derived with HKDF and kept separate from the encryption key.

V4 Access control

ASVS Requirement (summary) Status How Sentinel meets it
4.1.1 Access control enforced on a trusted server-side layer Pro Access policies checked in the gateway on every request. The free edition protects Sentinel’s own dashboard this way; your app’s access control is the app’s job.
4.1.2 Access control data can’t be manipulated by users Met Client-sent X-Sentinel-* headers are removed before the app sees them, so no client can forge an identity header.
4.1.3 Least privilege Met Rules give each role only the paths and methods it needs.
4.1.5 Access controls fail securely Partly Sentinel’s own checks fail securely (an invalid token is never treated as a login, schema checks don’t depend on Redis); only rate limiting fails open if Redis is down, which favours availability. Default-deny access policies for your app are in Pro.
4.2.1 Protection against IDOR (BOLA) Pro Owner rules such as /employees/{username}/salary only for that user, HR or admins.
4.2.2 Strong anti-CSRF App Sentinel uses bearer tokens in a header, which aren’t sent automatically. CSRF protection for the app’s own cookie sessions is the app’s job.
4.3.1 Admin interfaces use MFA Met require_mfa on admin routes, and the dashboard needs admin + 2FA.
4.3.3 Step-up or adaptive authentication Pro Risk-based step-up: medium risk needs a session that passed 2FA, and high risk is blocked.

V5 Validation, sanitization and encoding

ASVS Requirement (summary) Status How Sentinel meets it
5.1.1 HTTP parameter pollution Helps The WAF inspects every value of repeated parameters, and the schema validator rejects unexpected parameters.
5.1.2 Mass assignment Met Schema validation rejects fields the API spec doesn’t allow (additionalProperties: false).
5.1.3 Positive (allow-list) input validation Met OpenAPI schema validation on the APIs it covers (types, formats, lengths, patterns).
5.1.4 Structured data strongly typed and validated against a schema Met Same.
5.3.3 Output escaping against XSS Helps The WAF and the ML model block XSS payloads on the way in. Escaping output is the app’s job.
5.3.4 Parameterised database queries Helps SQL injection is blocked by the WAF and the ML model, and was tested with sqlmap (docs/evaluation.md). Sentinel’s own database access is parameterised.
5.3.7 LDAP injection Helps ML model (LDAP family in training).
5.3.8 OS command injection Helps WAF rule CMDI-001 and the ML model.
5.3.9 Local/remote file inclusion Helps WAF rules LFI-001/002 and NULL-001, and the ML model.
5.3.10 XPath/XML injection Helps ML model (XXE family).

V7 Logging and error handling

ASVS Requirement (summary) Status How Sentinel meets it
7.1.1 No credentials or payment details in logs Met Passwords, tokens and 2FA codes are never logged. Login failures log only the username.
7.1.3 Security events logged (auth success and failure, access control failures) Met One access-log line per request, with Sentinel’s decision: attacks, refusals, bans, dashboard logins and dashboard actions.
7.1.4 Enough detail to investigate Met Time, request ID, client IP and country, rule, field, reason and matched text.
7.2.1 All authentication decisions logged Met login, login_failed, mfa_required, mfa_ok, mfa_failed.
7.2.2 All access control decisions can be logged Met Every request’s outcome is logged.
7.3.1 Encoding prevents log injection Met Structured JSON logs (log/slog).
7.3.3 Security logs protected from modification Pro Hash-chained audit log with an anchor outside the file, so editing, deleting, reordering or cutting off the end is detected.
7.4.1 Generic error messages Met Blocked requests get a generic message and a request ID. The rule that fired is only in the log.

V9 Communication

ASVS Requirement (summary) Status How Sentinel meets it
9.1.1 TLS for all client connections Partly HTTPS listener (TLS_LISTEN_ADDR). The plain HTTP port is still open for the lab, so in production either turn it off or put a TLS proxy in front.
9.1.3 Only recent TLS versions Met TLS 1.2 minimum.
9.2.1 Only trusted certificates, specific CAs for internal ones Pro mTLS trusts only the configured client CA.
9.2.2 Encrypt all server connections Partly Connections to the apps, PostgreSQL and Redis are plain inside the Docker network.

V11 Business logic

ASVS Requirement (summary) Status How Sentinel meets it
11.1.2 Steps processed in realistic human time Helps Rate limits per user or IP.
11.1.4 Anti-automation against excessive calls and mass exfiltration Met Rate limits per path prefix, counted per user when logged in, with 429 and Retry-After.
11.1.7 Monitor unusual activity Met Attack-burst and password-guessing alerts, the live console and the dashboard.
11.1.8 Alerting on automated attacks Partly The dashboard and live window show attacks as they happen. There are no push alerts (email, chat) yet.

V13 API and web services

ASVS Requirement (summary) Status How Sentinel meets it
13.1.3 API URLs don’t expose secrets Met Tokens are only accepted in the Authorization header, and the dashboard never puts them in URLs.
13.1.4 Authorization at URI and resource level Pro Access policies on path patterns with {username} owner checks.
13.1.5 Unexpected content types rejected Partly The schema validator checks bodies on the APIs it covers. Other routes pass through.
13.2.1 Enabled HTTP methods valid for the user and action Partly Unknown endpoints and methods are rejected by the schema validator (with an OpenAPI spec). Per-user method rules are in Pro.
13.2.2 JSON schema validation before accepting input Met OpenAPI validation.
13.2.5 Content-Type checked Partly See 13.1.5.

V14 Configuration

ASVS Requirement (summary) Status How Sentinel meets it
14.1.1 Secure, repeatable automated build and deployment Met GitHub Actions CI on every push, and a release workflow on version tags.
14.1.5 Integrity of security-relevant configuration can be verified Partly Configuration is in git. The dashboard shows the running configuration.
14.2.1 Components up to date, dependency checker in the build Met govulncheck in CI and the release, and Grype scans the image SBOM (the release fails on critical findings).
14.2.4 Components from trusted repositories Met Go modules from the official repositories, vendored and checksummed (go.sum). Base images are official (golang, distroless).
14.2.5 Software Bill of Materials Met The release publishes an SPDX SBOM for every image (docs/supply-chain.md).
14.2.6 Third-party libraries sandboxed or encapsulated Partly Only two direct Go dependencies (pgx, x/crypto). The runtime image is distroless, non-root, with no shell.
14.4.4 X-Content-Type-Options: nosniff Met Added to every response the app didn’t set it on (SECURITY_HEADERS).
14.4.5 Strict-Transport-Security Met On HTTPS responses.
14.4.6 Suitable Referrer-Policy Met strict-origin-when-cross-origin unless the app sets one.
14.4.7 No framing by third-party sites Met X-Frame-Options: SAMEORIGIN unless the app sets one. The dashboard uses DENY and CSP frame-ancestors 'none'.
14.4.3 Content Security Policy Partly The dashboard has a strict CSP. CSP for the protected app depends on the app’s own scripts, so it’s the app’s job.
14.5.1 Only the HTTP methods the app uses Met The schema validator rejects methods the API spec does not list.
14.5.4 Headers added by trusted proxies are authenticated Met X-Forwarded-For is only believed from TRUSTED_PROXIES, and identity headers are always overwritten.

Known gaps (roadmap)

  • No self-service password change or password policy (2.1.x). No breached-password check (2.1.7).
  • TLS between Sentinel and the apps, PostgreSQL and Redis (9.2.2).
  • Push alerts for attacks (11.1.8).

Sentinel Free, © 2026 Saif Zahid. PolyForm Shield 1.0.0.