Skip to main content
Version: 4.1

Nauthilus Operational Security Checklist

This checklist is intended for release readiness and recurring security operations in Production and Staging.

Usage​

  • Mark each item as done for both environments.
  • Attach evidence (ticket, screenshot, log snippet, config diff, command output).
  • Re-run this checklist before each major release and after security-relevant changes.

1. Backchannel Access Control​

  • Production: At least one backchannel auth method is enabled:
    • auth.backchannel.basic_auth.enabled=true or auth.backchannel.oidc_bearer.enabled=true
  • Staging: At least one backchannel auth method is enabled.
  • Production: /api/v1/* is reachable only from trusted internal networks.
  • Backchannel listeners (/api/v1/* including the Policy API, and the gRPC authority listener) are not reachable from untrusted networks. A rejected caller is delayed and logged; with the HTTP rate middleware enabled, an address that exhausts its failure budget is refused on /api/v1 until it refills.
  • Every backchannel caller holds only the scopes of the routes it uses. Each route family answers to its own scopes (nauthilus:authenticate for /api/v1/auth/* and the management OpenAPI documents, nauthilus:security_read for brute-force and OIDC-session listings, nauthilus:security_write for cache and brute-force flushes and OIDC-session deletion, either of them for the status of an asynchronous job, the MFA and WebAuthn scopes for /api/v1/mfa-backchannel/*, the support scopes for /api/v1/support/*); no route asks for a scope of another family. nauthilus:security includes both security scopes, and nauthilus:admin is accepted on the security and MFA routes as well; neither opens the authentication endpoints or the support routes. Grant these two only where the narrower scope does not suffice. See OIDC Bearer Access.
  • Basic callers use scoped credentials under auth.backchannel.basic_auth.users, one per person or tool. The unrestricted credential (auth.backchannel.basic_auth.username and password) grants full management access and is announced with a warning at startup; remove it where scoped credentials or Bearer tokens suffice. See Backchannel Basic Credentials.
  • Backchannel credentials (Basic passwords, Policy-Basic passwords, and OIDC client_credentials client secrets or keys) are high-entropy.
  • Alerts watch backchannel_caller_auth_total{outcome=~"rejected|unavailable"}.
  • auth.pipeline.max_concurrent_requests is sized for the backchannel peak. On the caller-authenticated backchannel API (/api/v1/auth/*, /api/v1/cache/*, /api/v1/bruteforce/*, the management OpenAPI documents, and the other routes behind auth.backchannel), the per-client-IP HTTP rate limit (runtime.servers.http.middlewares.rate with runtime.servers.http.rate_limit.per_second and burst) is a failure budget: requests whose Basic or Bearer caller authentication succeeds never consume it, because a few proxy or load balancer addresses carry all backchannel traffic. The shared HTTP/gRPC request budget is the process-wide bound for these callers.
  • Every failed caller authentication on these routes (missing or wrong credentials, a caller without nauthilus:authenticate on the routes that require it, missing auth.backchannel configuration) and every request to the auth.basic endpoint consumes one token of the client address. An undecided token validation (503) does not. Within the budget a failure is answered with 401 or 403 after the rejection delay. Once the address has no token left, every request from it, including requests with valid credentials, is answered with 429 (scope: rate) immediately and without evaluating credentials, until the budget refills at per_second. Credential guessing is therefore bounded per address to burst attempts plus per_second attempts per second on average; concurrent failures that pass the check together are all charged and delay the refill.
  • Callers that share an address with a misconfigured or hostile client lose backchannel access while that client exhausts the budget. Watch backchannel_caller_auth_total{outcome="rejected"} and 429 answers with ratelimit_reason=rate on /api/v1, and fix failing callers promptly. Routes without backchannel caller authentication, such as the IdP, the frontend, the Policy API, custom hooks, and backchannel routes of a developer-mode server without auth.backchannel, keep the per-client-IP limit for every request.
  • /oidc/token and /oidc/introspect have no per-address brake for client-secret guessing; guessing there is limited by network exposure and upstream rate limits.
  • With runtime.servers.http.haproxy_v2 enabled, the listener accepts PROXY headers from every TCP peer and logs use the PROXY source, so the listener is reachable only from the PROXY-speaking load balancers.
  • Staging: /api/v1/* is not publicly exposed.

Evidence:

  • Config snapshot attached
  • Network policy / firewall rule attached

2. Metrics Endpoint Access Control​

  • Production: If /metrics is exposed beyond a private scrape network, dedicated metrics Basic Auth is enabled:
    • observability.metrics.endpoint_auth.basic.enabled=true
  • Staging: /metrics exposure is documented and protected when it crosses trust boundaries.
  • Prometheus uses dedicated metrics credentials, not backchannel API credentials.
  • Bearer/OIDC authentication is not used for /metrics.

Evidence:

  • Metrics scrape config attached
  • Network policy / firewall rule attached

3. OIDC Token Endpoint Hardening​

  • Production: identity.oidc.tokens.token_endpoint_allow_get=false
  • Staging: identity.oidc.tokens.token_endpoint_allow_get=false
  • Any temporary GET enablement has a documented exception owner and expiration date.

Evidence:

  • Config diff attached
  • Exception ticket (if applicable)

4. Configuration Endpoint Exposure​

  • Production: runtime.servers.http.disabled_endpoints.configuration=true unless explicitly required.
  • Staging: Configuration endpoint exposure is justified and documented.
  • If enabled, access is restricted and audited.

Evidence:

  • Endpoint accessibility test attached
  • Audit log sample attached

5. CSP and Security Headers​

  • Production: CSP keeps default form-action 'self' https: unless form_action_optional_uris is intentionally used.
  • Staging: Any CSP widening (for redirects/dev compatibility) is explicitly documented.
  • Security headers are enabled under identity.frontend.security_headers.

Evidence:

  • Response header capture attached
  • Config snippet attached

6. Developer Mode Controls​

  • Production: NAUTHILUS_DEVELOPER_MODE=false
  • Staging: NAUTHILUS_DEVELOPER_MODE=false unless explicitly needed for a short test window.
  • Startup/runtime guardrails prevent accidental non-loopback developer mode usage.

Evidence:

  • Deployment env vars attached
  • Startup log excerpt attached

7. Network and Redis Hardening​

  • Redis is not publicly reachable.
  • Redis authentication and TLS are configured where applicable.
  • Redis ACLs follow least privilege.
  • runtime.servers.http.trusted_proxies is explicitly configured (no broad trust).

Evidence:

  • Redis bind/ACL config attached
  • Proxy trust config attached

8. Secrets and Key Management​

  • No secrets in repository or plaintext deployment artifacts.
  • OIDC signing keys have a rotation process and owner.
  • Client secrets have rotation and revocation procedures.

Evidence:

  • Secret management policy attached
  • Rotation record attached

9. Runtime Hardening​

  • Service runs non-root (run_as_user, run_as_group) where supported.
  • Optional debug endpoints (for example pprof) are disabled in production.
  • Only required endpoints are enabled.

Evidence:

  • Runtime/service config attached
  • Endpoint inventory attached

10. Logging, Detection, and Alerting​

  • Security-relevant logs are centralized.
  • Alerts exist for:
    • repeated auth failures / brute-force patterns
    • unusual 401/403 spikes
    • token validation and scope-denial anomalies
  • Alert ownership and on-call routing are documented.

Evidence:

  • Alert rule export attached
  • Recent alert test attached

11. CI/CD Security Gates​

  • govulncheck is mandatory for merges to main.
  • Dependency update policy is active (scheduled updates, review owner).
  • SBOM generation/verification is part of release flow.

Evidence:

  • CI workflow link attached
  • Last successful run attached

12. Backup and Recovery​

  • Backup/restore procedures exist for config, keys, and stateful dependencies.
  • Restore drills are executed on a schedule.
  • Security incident rollback procedure is tested.

Evidence:

  • Drill report attached
  • Recovery runbook attached

13. Independent Validation​

  • External security assessment (pentest/blackbox) is planned or completed.
  • Findings are tracked to closure.
  • High-severity findings are converted into automated regression tests.

Evidence:

  • Assessment report attached
  • Tracking ticket list attached

Sign-off​

  • Release/Change ID:
  • Environment:
  • Reviewer:
  • Date:
  • Result: PASS / PASS WITH EXCEPTIONS / FAIL