Merge pull request 'fix(auth): make the AUTH_ENFORCE default fail-closed -- 0pt.app was answering anonymously' (#4) from fix/fail-closed-auth-default into main

.env.example

@@ -39,11 +39,13 @@ INTERNAL_TOKEN=change-me-internal-token
 #
 # Fleet Zitadel issuer (OIDC). Default when unset: https://id-customers.siax.io
 ZITADEL_ISSUER=https://id-customers.siax.io
-# Flag-gated enforcement. When != "true" (DEFAULT) auth is non-breaking:
-# a present Bearer token is best-effort verified and attached, but never
-# rejected. Set to "true" to REQUIRE a valid token on /api/0pt/* (401 on
-# missing/invalid, 403 when X-Tenant-Id disagrees with the token tenant).
-AUTH_ENFORCE=false
+# Enforcement is FAIL-CLOSED. Leave this UNSET (or set it to "true") and a
+# valid, verified Bearer token is REQUIRED on /api/0pt/* -- 401 on
+# missing/invalid, 403 when X-Tenant-Id disagrees with the token tenant.
+# ONLY the literal value "false" disables authentication, and that value is
+# REFUSED when NODE_ENV=production: the app throws at startup instead of
+# serving openly. Use "false" for local development only.
+AUTH_ENFORCE=true
 # Optional override for the claim key that carries the SIAX tenant id. When
 # unset, tenant derivation uses the documented precedence:
 # urn:zitadel:iam:org:id -> urn:zitadel:iam:user:resourceowner:id -> tenant_id -> tenant.

README.md

@@ -46,6 +46,10 @@ cd services/api && pip install -r requirements.txt && uvicorn orchestrator.main:
 - **Frontend** → Vercel (project `0pt`, scope *Simon Axelsson's projects*), domain `0pt.app`.
 - **Engine + DB + embeds** → one VPS via [`infra/docker-compose.yml`](infra/docker-compose.yml) (Postgres+pgvector, Redis, api, worker, NocoDB, Metabase, Caddy).
 - Env inventory: [`.env.example`](.env.example).
+- **Auth on `/api/0pt/*` is fail-closed**: with `AUTH_ENFORCE` unset the middleware
+  REQUIRES a verified Zitadel Bearer token (401 without one). Only the literal
+  `AUTH_ENFORCE=false` disables it, and that is refused when `NODE_ENV=production`
+  (the process throws at startup rather than serving anonymously).
 
 ## Status
 Frontend, full order lifecycle, customer + admin portals, edge API, payment/auth/embed

middleware.ts

@@ -2,23 +2,30 @@
  * `middleware.ts` -- Next.js request middleware = 0pt's framework-level auth
  * seam (the Next analog of a Fastify `onRequest` hook).
  *
- * Wave 2 hardening: this runs the flag-gated Zitadel verification
- * (src/auth/context.ts -> src/auth/zitadel.ts) in front of every 0pt
- * capability route (/api/0pt/*). It does NOT touch the v0-generated UI routes
- * or the unrelated payment/order APIs.
+ * This runs the Zitadel verification (src/auth/context.ts ->
+ * src/auth/zitadel.ts) in front of every 0pt capability route (/api/0pt/*).
+ * It does NOT touch the v0-generated UI routes or the unrelated
+ * payment/order APIs (see the matcher below and the note in the PR that
+ * introduced the fail-closed default).
  *
- *   AUTH_ENFORCE != "true" (default): never rejects; if a verified token is
- *     present the verified principal is forwarded to the handler as
- *     `x-0pt-verified-sub` / `x-0pt-verified-tenant` request headers.
- *   AUTH_ENFORCE == "true": a valid verified token is required; 401 on
- *     missing/invalid, 403 when X-Tenant-Id disagrees with the token tenant.
+ *   AUTH_ENFORCE unset, or anything except "false" (DEFAULT, enforced):
+ *     a valid verified token is required; 401 on missing/invalid, 403 when
+ *     X-Tenant-Id disagrees with the token tenant. On success the verified
+ *     principal is forwarded to the handler as `x-0pt-verified-sub` /
+ *     `x-0pt-verified-tenant` request headers.
+ *   AUTH_ENFORCE == "false" (explicit local-dev escape hatch): never rejects.
+ *     REFUSED in production -- `assertAuthConfigSafe()` below throws at module
+ *     load, so a production instance cannot boot with auth switched off.
  *
  * Health/metrics are exempt (they are also outside the matcher below, but the
  * exemption is kept explicit for robustness).
  */
 
 import { NextResponse, type NextRequest } from 'next/server';
-import { resolveAuthContext } from '@/src/auth/context';
+import { assertAuthConfigSafe, resolveAuthContext } from '@/src/auth/context';
+
+// STARTUP GATE: fail to boot rather than serve /api/0pt/* openly in production.
+assertAuthConfigSafe();
 
 // Only guard the 0pt capability surface. The matcher keeps this middleware off
 // the UI, auth, and payment routes entirely.

src/auth/context.ts

@@ -3,26 +3,31 @@
  * seam the framework layer (Next.js middleware.ts) calls.
  *
  * ==========================================================================
- * WAVE 2 HARDENING STATUS
+ * AUTH DEFAULT: FAIL-CLOSED
  *
- * 0pt now performs REAL, cryptographic Zitadel (OIDC) Bearer-token
- * verification (src/auth/zitadel.ts), gated behind the AUTH_ENFORCE env flag
- * so it can ship to the deployed console WITHOUT breaking existing callers:
+ * 0pt performs REAL, cryptographic Zitadel (OIDC) Bearer-token verification
+ * (src/auth/zitadel.ts). Enforcement is controlled by AUTH_ENFORCE, and the
+ * DEFAULT IS ENFORCED -- an absent, empty, or misspelled value authenticates:
  *
- *   AUTH_ENFORCE != "true"  (DEFAULT, non-breaking):
- *     Preserves the historical behavior (no token was ever required; tenant
- *     came from the X-Tenant-Id header / request body). ADDITIONALLY, if a
- *     Bearer token is present and JWT-shaped, we ATTEMPT `verifyBearer` and
- *     attach the verified principal to the resolved context -- but we NEVER
- *     reject on failure in this mode (a warning is logged). This lets tokens
- *     start flowing and being verified with zero caller migration.
- *
- *   AUTH_ENFORCE == "true"  (enforced):
+ *   AUTH_ENFORCE unset, or any value except "false"  (DEFAULT, enforced):
  *     REQUIRES a valid, verified token on every non-exempt route. The tenant
  *     is derived from the TOKEN (not the X-Tenant-Id header). Missing/invalid
  *     token -> 401 (OPT_UNAUTHENTICATED). If an X-Tenant-Id header is present
  *     and disagrees with the token's tenant -> 403 (OPT_ACCESS_DENIED).
  *
+ *   AUTH_ENFORCE == "false"  (explicit local-development escape hatch):
+ *     No token is required; the tenant comes from the X-Tenant-Id header /
+ *     request body. A present, JWT-shaped Bearer token is still best-effort
+ *     verified and attached, but never rejected. This mode is REFUSED in
+ *     production: {@link assertAuthConfigSafe} throws when NODE_ENV=production,
+ *     so an instance that loses (or mis-sets) the variable cannot serve
+ *     openly -- it fails to start, and fails the request, instead.
+ *
+ * WHY THE INVERSION: the predicate previously read `AUTH_ENFORCE === 'true'`,
+ * so every fresh container, local run, DR restore, or redeploy that dropped
+ * the variable ran with NO authentication at all, and `.env.example` shipped
+ * `AUTH_ENFORCE=false`. Absence now means enforced.
+ *
  * This module is framework-neutral: it takes a minimal header-lookup (the Web
  * `Headers` interface satisfies it, as does Next's `NextRequest.headers`) and
  * returns a decision object. It does NOT import next/*, so it is fully
@@ -38,7 +43,7 @@ export interface HeaderLookup {
 }
 
 export interface OptRequestContext {
-  /** Tenant id: from the token in enforce mode, else from X-Tenant-Id header. */
+  /** Tenant id: from the token when enforced, else from the X-Tenant-Id header. */
   tenantId: string;
   /** Actor id: from X-Actor-Id header, falling back to the token subject. */
   actorId: string;
@@ -61,9 +66,57 @@ export type AuthDecision =
       ctx: OptRequestContext;
     };
 
-/** True when hard auth enforcement is switched on (AUTH_ENFORCE=="true"). */
-export function authEnforced(): boolean {
-  return process.env.AUTH_ENFORCE === 'true';
+/** Name of the environment variable that can switch enforcement off. */
+export const AUTH_ENFORCE_VAR = 'AUTH_ENFORCE';
+
+/** The only two environment values this module reads; injectable for tests. */
+export interface AuthEnvView {
+  AUTH_ENFORCE?: string;
+  NODE_ENV?: string;
+}
+
+/**
+ * Snapshot the auth-relevant environment. `process.env.NODE_ENV` is written
+ * literally so bundlers (Next's edge compiler) inline it.
+ */
+function currentEnv(): AuthEnvView {
+  return { AUTH_ENFORCE: process.env.AUTH_ENFORCE, NODE_ENV: process.env.NODE_ENV };
+}
+
+/**
+ * True unless enforcement was EXPLICITLY disabled with AUTH_ENFORCE="false".
+ *
+ * FAIL-CLOSED: unset, empty, "flase", "0", or any other value => ENFORCED.
+ */
+export function authEnforced(env: AuthEnvView = currentEnv()): boolean {
+  return env.AUTH_ENFORCE?.trim().toLowerCase() !== 'false';
+}
+
+/** Thrown when the process is configured to run without auth in production. */
+export class AuthMisconfigurationError extends Error {
+  constructor(message: string) {
+    super(message);
+    this.name = 'AuthMisconfigurationError';
+  }
+}
+
+/**
+ * Refuse to run open in production. Throws {@link AuthMisconfigurationError}
+ * when NODE_ENV=production and enforcement resolved to OFF.
+ *
+ * Called at module scope by middleware.ts (startup) AND on the env-driven
+ * request path below, so a misconfigured production instance can neither boot
+ * nor serve a single unauthenticated request.
+ */
+export function assertAuthConfigSafe(env: AuthEnvView = currentEnv()): void {
+  if (env.NODE_ENV === 'production' && !authEnforced(env)) {
+    throw new AuthMisconfigurationError(
+      `0pt refuses to run with authentication disabled in production: ` +
+        `${AUTH_ENFORCE_VAR} is explicitly set to "${env.AUTH_ENFORCE}" while NODE_ENV=production. ` +
+        `Unset ${AUTH_ENFORCE_VAR} (absent = enforced) or set ${AUTH_ENFORCE_VAR}=true. ` +
+        `${AUTH_ENFORCE_VAR}=false is a local-development escape hatch only.`,
+    );
+  }
 }
 
 function firstHeader(headers: HeaderLookup, name: string): string | undefined {
@@ -111,23 +164,34 @@ export interface ResolveAuthOptions {
 }
 
 /**
- * Resolve and (in enforce mode) enforce the request auth context. Returns an
- * {@link AuthDecision}; the caller (middleware) turns a `!ok` decision into an
- * HTTP response and an `ok` decision into forwarded principal headers.
+ * Resolve and (when enforced -- the default) enforce the request auth context.
+ * Returns an {@link AuthDecision}; the caller (middleware) turns a `!ok`
+ * decision into an HTTP response and an `ok` decision into forwarded principal
+ * headers.
+ *
+ * Throws {@link AuthMisconfigurationError} when the enforcement decision comes
+ * from the environment and that environment is "production with auth off".
  */
 export async function resolveAuthContext(
   headers: HeaderLookup,
   opts: ResolveAuthOptions = {},
 ): Promise<AuthDecision> {
   const verify = opts.verify ?? verifyBearer;
-  const enforce = opts.enforce ?? authEnforced();
+  let enforce: boolean;
+  if (opts.enforce !== undefined) {
+    enforce = opts.enforce;
+  } else {
+    // Never serve openly in production, even if the startup check was bypassed.
+    assertAuthConfigSafe();
+    enforce = authEnforced();
+  }
   const logger: AuthLogger = opts.logger ?? {
     warn: (o, m) => console.warn(m ?? 'auth warning', o),
   };
   const ctx = readRequestContext(headers);
 
   if (enforce) {
-    // --- ENFORCED: a valid verified token is mandatory ----------------------
+    // --- ENFORCED (DEFAULT): a valid verified token is mandatory ------------
     if (!ctx.bearerToken) {
       return {
         ok: false,
@@ -168,19 +232,21 @@ export async function resolveAuthContext(
     return { ok: true, ctx };
   }
 
-  // --- NON-ENFORCE (default): never reject, best-effort verify --------------
+  // --- EXPLICIT OPT-OUT (AUTH_ENFORCE=false, non-production only) -----------
+  // Never rejects; a present token is best-effort verified so tokens can start
+  // flowing in local development.
   if (ctx.bearerToken && looksLikeJwt(ctx.bearerToken)) {
     try {
       const principal = await verify(ctx.bearerToken);
       ctx.principal = principal;
-      // Do NOT overwrite the header/body tenant in non-enforce mode; only
-      // fill it in when the caller supplied none, to stay non-breaking.
+      // Do NOT overwrite the header/body tenant in this mode; only fill it in
+      // when the caller supplied none.
       if (!ctx.tenantId) ctx.tenantId = principal.tenantId;
       if (!ctx.actorId) ctx.actorId = principal.sub;
     } catch (err) {
       logger.warn(
         { err: err instanceof Error ? err.message : String(err) },
-        '0pt: bearer verification failed (AUTH_ENFORCE off, ignoring)',
+        `0pt: bearer verification failed (${AUTH_ENFORCE_VAR}=false, ignoring)`,
       );
     }
   }

tests/auth/context.test.ts

@@ -1,12 +1,24 @@
 /**
- * Unit tests for src/auth/context.ts `resolveAuthContext` -- the flag-gated
- * auth seam. Verification is injected (opts.verify) so these tests are pure
- * and network-free; they exercise the decision logic in both AUTH_ENFORCE
- * modes.
+ * Unit tests for src/auth/context.ts -- the auth seam. Verification is
+ * injected (opts.verify) so these tests are pure and network-free.
+ *
+ * Three groups:
+ *   1. the ENV DEFAULT (the fail-closed contract: absent AUTH_ENFORCE means
+ *      an unauthenticated request is REJECTED, and production refuses to run
+ *      with auth switched off),
+ *   2. enforced-mode decision logic,
+ *   3. the explicit `AUTH_ENFORCE=false` local-development escape hatch.
  */
 
-import { describe, it, expect, vi } from 'vitest';
-import { resolveAuthContext, type OptRequestContext } from '@/src/auth/context';
+import { describe, it, expect, vi, beforeEach, afterEach } from 'vitest';
+import {
+  resolveAuthContext,
+  authEnforced,
+  assertAuthConfigSafe,
+  AuthMisconfigurationError,
+  AUTH_ENFORCE_VAR,
+  type OptRequestContext,
+} from '@/src/auth/context';
 import type { VerifiedPrincipal } from '@/src/auth/zitadel';
 
 const principal: VerifiedPrincipal = {
@@ -24,7 +36,77 @@ function h(init: Record<string, string>): Headers {
   return new Headers(init);
 }
 
-describe('resolveAuthContext -- enforce mode', () => {
+describe('auth default -- fail-closed (no AUTH_ENFORCE in the environment)', () => {
+  const saved = { auth: process.env.AUTH_ENFORCE, node: process.env.NODE_ENV };
+
+  beforeEach(() => {
+    delete process.env.AUTH_ENFORCE;
+  });
+
+  afterEach(() => {
+    if (saved.auth === undefined) delete process.env.AUTH_ENFORCE;
+    else process.env.AUTH_ENFORCE = saved.auth;
+    if (saved.node === undefined) delete process.env.NODE_ENV;
+    else process.env.NODE_ENV = saved.node;
+  });
+
+  it('REGRESSION: with the variable unset, an unauthenticated request is REJECTED (401)', async () => {
+    // No `enforce` option -> the decision comes from the environment, which is
+    // exactly the fresh-container / DR-restore / dropped-env-var case.
+    const d = await resolveAuthContext(h({ 'x-tenant-id': 'tenant_attacker' }), { verify: okVerify });
+    expect(d.ok).toBe(false);
+    if (!d.ok) {
+      expect(d.status).toBe(401);
+      expect(d.code).toBe('OPT_UNAUTHENTICATED');
+    }
+  });
+
+  it('with the variable unset, an invalid token is REJECTED (401) rather than ignored', async () => {
+    const d = await resolveAuthContext(h({ authorization: 'Bearer bad.tok.en' }), { verify: failVerify });
+    expect(d.ok).toBe(false);
+    if (!d.ok) expect(d.status).toBe(401);
+  });
+
+  it('authEnforced() is true when the variable is absent, empty, or misspelled', () => {
+    expect(authEnforced({})).toBe(true);
+    expect(authEnforced({ AUTH_ENFORCE: '' })).toBe(true);
+    expect(authEnforced({ AUTH_ENFORCE: 'flase' })).toBe(true);
+    expect(authEnforced({ AUTH_ENFORCE: '0' })).toBe(true);
+    expect(authEnforced({ AUTH_ENFORCE: 'no' })).toBe(true);
+    expect(authEnforced({ AUTH_ENFORCE: 'true' })).toBe(true);
+  });
+
+  it('only the literal string "false" opts out', () => {
+    expect(authEnforced({ AUTH_ENFORCE: 'false' })).toBe(false);
+    expect(authEnforced({ AUTH_ENFORCE: 'FALSE' })).toBe(false);
+    expect(authEnforced({ AUTH_ENFORCE: '  false  ' })).toBe(false);
+  });
+
+  it('assertAuthConfigSafe throws in production when enforcement is off, naming the variable', () => {
+    expect(() => assertAuthConfigSafe({ NODE_ENV: 'production', AUTH_ENFORCE: 'false' })).toThrow(
+      AuthMisconfigurationError,
+    );
+    expect(() => assertAuthConfigSafe({ NODE_ENV: 'production', AUTH_ENFORCE: 'false' })).toThrow(
+      new RegExp(AUTH_ENFORCE_VAR),
+    );
+  });
+
+  it('assertAuthConfigSafe is silent when production enforces, or when off outside production', () => {
+    expect(() => assertAuthConfigSafe({ NODE_ENV: 'production' })).not.toThrow();
+    expect(() => assertAuthConfigSafe({ NODE_ENV: 'production', AUTH_ENFORCE: 'true' })).not.toThrow();
+    expect(() => assertAuthConfigSafe({ NODE_ENV: 'development', AUTH_ENFORCE: 'false' })).not.toThrow();
+  });
+
+  it('a production process with AUTH_ENFORCE=false cannot serve a request at all (throws)', async () => {
+    process.env.NODE_ENV = 'production';
+    process.env.AUTH_ENFORCE = 'false';
+    await expect(
+      resolveAuthContext(h({ 'x-tenant-id': 'tenant_attacker' }), { verify: okVerify }),
+    ).rejects.toBeInstanceOf(AuthMisconfigurationError);
+  });
+});
+
+describe('resolveAuthContext -- enforced mode (the default)', () => {
   it('401s when no bearer token is present', async () => {
     const d = await resolveAuthContext(h({ 'x-tenant-id': 'tenant_x' }), { enforce: true, verify: okVerify });
     expect(d.ok).toBe(false);
@@ -78,7 +160,7 @@ describe('resolveAuthContext -- enforce mode', () => {
   });
 });
 
-describe('resolveAuthContext -- non-enforce (default) mode', () => {
+describe('resolveAuthContext -- explicit AUTH_ENFORCE=false escape hatch (local dev only)', () => {
   it('never rejects when no token is present', async () => {
     const d = await resolveAuthContext(h({ 'x-tenant-id': 'tenant_header' }), {
       enforce: false,
@@ -111,7 +193,7 @@ describe('resolveAuthContext -- non-enforce (default) mode', () => {
     expect(d.ok).toBe(true);
     if (d.ok) {
       expect(d.ctx.principal).toEqual(principal);
-      // non-breaking: caller-supplied header tenant is preserved
+      // the caller-supplied header tenant is preserved in this mode
       expect(d.ctx.tenantId).toBe('tenant_header');
     }
   });