Skip to content

OWASP ASVS level 1 review

FSD §18 asks for an OWASP ASVS 4.0 level 1 checklist before the pilot. This review covers each chapter against the code on main as of 26 Sep 2026. It names the test or file that shows each control, and lists what is still open.

Status: ready for a pilot inside a customer network. Before the public launch, repeat the review with an outside tester.

V2 Authentication

Requirement Status Evidence
Passwords of 12+ characters, no maximum below 64, all characters allowed (2.1.1–2.1.4) Done auth.CheckPolicy, TestCheckPolicy, TestSetupRejectsWeakPasswords
Checked against breached or common passwords (2.1.7) Done 100,000-entry SecLists list, embedded (auth/common-passwords.txt)
Password change needs the current password (2.1.6) Done TestChangingPasswordNeedsTheCurrentOne
Brute-force protection (2.2.1) Done 5 failures lock an account for 15 minutes, and 20 attempts per IP per minute are rate-limited: TestFiveFailuresLockTheAccount, TestTwentyOneAttemptsFromOneIPAreRateLimited
Passwords stored with a slow hash (2.4.1) Done argon2id, TestHashAndCheckPassword
Initial and reset secrets random, short-lived and single-use (2.3.1, 2.5.x) Done 256-bit setup links, 72 hours, stored hashed: TestExpiredSetupLinkIsRefused, TestResetPasswordEndsSessionsAndIssuesANewLink
No default accounts (2.5.4) Done The installer creates the first admin with a setup link; no built-in credentials
Unknown and known emails answer alike (2.2.x) Done dummyHash makes sign-in for an unknown email as slow as for a known one

V3 Session management

Requirement Status Evidence
256-bit random tokens, stored hashed (3.2.2) Done auth.NewToken, TestNewToken
Cookie HttpOnly, SameSite=Lax, Secure over HTTPS (3.4.1–3.4.3) Done TestSessionCookieFlags
New token at sign-in (3.2.1) Done TestLoginStartsASessionAndIsAudited
Sign-out, disable and password reset end sessions (3.3.1, 3.3.3) Done TestLogoutRevokesTheSession, TestDisablingAUserSignsThemOut, TestResetPasswordEndsSessionsAndIssuesANewLink
Idle and absolute timeouts (3.3.2) Done 12 hours idle, 7 days absolute: TestIdleSessionExpires

V4 Access control

Requirement Status Evidence
Deny by default; checks on the server (4.1.1–4.1.3) Done The visibility predicate lives in SQL; there are no Go-side filters (§5.3)
Row-level access across projects and clients (4.2.1) Done The permission suite runs a seeded role and client matrix on every list, read, search and Ask path: TestPermissionSuiteReads, TestPermissionSuiteWrites, TestPermissionSuiteAsk. It is merge-blocking in CI.
Hidden and missing look the same Done 404 for both, in every endpoint test
CSRF (4.2.2) Done Writes need the app's Origin (TestWritesNeedTheAppOrigin), plus SameSite=Lax cookies
Admin functions need the admin flag (4.3.1) Done requireAdmin; TestNonAdminsCannotManageUsers, and the Ask log, audit, backups and status tests

V5 Validation, sanitisation and encoding

Requirement Status Evidence
Input validated by type, length and format (5.1.x) Done OpenAPI-generated types, then per-field checks with 422 problem details
No raw HTML from users (5.2.x, 5.3.x) Done Markdown through rehype-sanitize, and images only from the app's own attachment URLs
SQL injection (5.3.4) Done All SQL through sqlc and pgx parameters
Prompt injection (Ask) Mitigated Evidence is fenced as data, output is schema-constrained, the model has no tools, and every citation is checked against the evidence (§11.5)

V7 Error handling and logging

Requirement Status Evidence
No stack traces or internals in responses (7.4.1) Done RFC 9457 problems with codes; s.fail logs the error and returns a generic 500
Security events logged (7.1.x, 7.2.x) Done audit_events is append-only (UPDATE and DELETE revoked from the app role); sign-in is audited
No secrets or personal text in logs (7.1.1) Done Logs carry IDs; question text lives only in the Ask log (§18.2)
Log access for admins (7.3.x) Done Admin → Audit log with CSV export, and Admin → Ask log

V8 Data protection

Requirement Status Evidence
Sensitive data not cached by browsers (8.2.1) Done Cache-Control: no-store on every API response (requestContext), asserted in TestSecurityHeaders
Secrets at rest encrypted (8.3.x) Done AI API keys sealed with AES-GCM under APP_SECRET_KEY, never returned (R-AI-3)
Data leaves only by admin choice Done Local and off modes make no outbound calls; BYOK needs the audited acknowledgement (R-AI-1)

V9 Communications

Requirement Status Evidence
TLS for all traffic (9.1.1) Done in deploy Caddy with the customer's certificate or its internal CA; HSTS for a year
Internal services not exposed (9.2.x) Done Only Caddy publishes ports; app, web, db, backup and model sit on an internal network

V12 Files and resources

Requirement Status Evidence
Upload size limits (12.1.1) Done 25 MB per file: TestAttachmentLimits
Downloads cannot run as content (12.5.x) Done nosniff, attachment disposition except for images, and a closed CSP on API responses
Files reachable only through the ticket's access (12.4.x) Done TestAttachmentAccessFollowsTheTicket
Uploads named by hash, not by user input (12.3.x) Done Stored as sha256 under the attachments volume

V13 API

Requirement Status Evidence
Same checks on every API endpoint (13.1.x) Done One middleware chain, with the auth and origin checks, for /api/v1
Content types enforced (13.2.5) Done JSON decoding refuses unknown fields; uploads are multipart only
Rate limits on expensive calls Done Sign-in per IP; Ask 10 a minute per user

V14 Configuration

Requirement Status Evidence
Security headers (14.4.x) Done Nonce CSP on pages, a closed policy on the API, nosniff, Referrer-Policy, frame-ancestors 'none', HSTS: TestSecurityHeaders and web/e2e/security.spec.ts
Dependencies from trusted sources, licences checked (14.2.x) Done Pinned versions and lockfiles; the licence scan runs in CI
Dependency vulnerabilities (14.2.1) Done govulncheck and npm audit --omit=dev --audit-level=high run in CI
Secrets generated per install (14.1.x) Done install.sh writes new database passwords and APP_SECRET_KEY into .env (mode 600)
Server banners removed (14.3.x) Done Caddy's Server header dropped; Next's X-Powered-By off

Open items

  1. An outside penetration test before the public launch.