fix(registry): keep the catalog rate limiter on catalog routes only

backend/__tests__/unit/routes/registry.catalog-rate-limit-scope.test.js

@@ -0,0 +1,44 @@
+/**
+ * D175 — the catalog's IP limiter must stay on the catalog's own routes.
+ *
+ * registry.ts mounts catalogRouter ahead of pod-agents, files, install and
+ * the rest. A `catalogRouter.use(limiter)` therefore ran for EVERY
+ * /api/registry/* request that reached it, and /v2/agents (one
+ * `/pods/:podId/agents` read per pod) hit 429 on all 62 calls from its third
+ * load on a 62-pod account (prod, 10-28). The limiter skips itself under
+ * NODE_ENV=test, so this guards the shape instead: catalogRouter carries no
+ * router-level middleware, and every route it declares runs the limiter.
+ */
+
+jest.mock('../../../middleware/auth', () => jest.fn((req, res, next) => next()));
+jest.mock('../../../models/Pod', () => ({}));
+jest.mock('../../../models/AgentRegistry', () => ({
+  AgentRegistry: { search: jest.fn(), getByName: jest.fn() },
+  AgentInstallation: { find: jest.fn(), getInstalledAgents: jest.fn() },
+}));
+
+const catalogRouter = require('../../../routes/registry/catalog');
+
+describe('catalog rate limiter scope (D175)', () => {
+  const layers = catalogRouter.stack;
+
+  it('has no router-level middleware that would run for later registry routers', () => {
+    const routerLevel = layers.filter((layer) => !layer.route);
+    expect(routerLevel).toHaveLength(0);
+  });
+
+  it('runs the same limiter on every catalog route, ahead of auth', () => {
+    const routes = layers.filter((layer) => layer.route);
+    expect(routes.length).toBeGreaterThan(0);
+    const firstHandlers = routes.map((layer) => layer.route.stack[0].handle);
+    const limiter = firstHandlers[0];
+    expect(firstHandlers.every((handle) => handle === limiter)).toBe(true);
+    // express-rate-limit's middleware exposes resetKey; auth does not.
+    expect(typeof limiter.resetKey).toBe('function');
+  });
+
+  it('does not declare /pods/:podId/agents, which belongs to pod-agents', () => {
+    const paths = layers.filter((layer) => layer.route).map((layer) => layer.route.path);
+    expect(paths).not.toContain('/pods/:podId/agents');
+  });
+});

backend/__tests__/unit/routes/registry.rate-limit-coverage.test.js

@@ -0,0 +1,73 @@
+/**
+ * D175 follow-up: scoping the catalog limiter to catalog routes must not strip
+ * rate limiting from the routers mounted after it. They used to inherit the
+ * catalog's router-level limiter; they now share a registry-wide IP limiter
+ * mounted right after the catalog, and generate-avatar (which calls a paid
+ * image provider per request) also carries its own per-caller limiter.
+ * Limiters skip themselves under NODE_ENV=test, so these guard the shape.
+ */
+const fs = require('fs');
+const path = require('path');
+
+jest.mock('../../../middleware/auth', () => jest.fn((req, res, next) => next()));
+jest.mock('../../../models/AgentRegistry', () => ({ AgentRegistry: {} }));
+jest.mock('../../../services/agentIdentityService', () => ({}));
+jest.mock('../../../routes/registry/helpers', () => ({ getUserId: jest.fn(), resolveUsername: jest.fn() }));
+jest.mock('../../../utils/agentManifestRegistry', () => ({
+  ManifestValidationError: class extends Error {},
+  normalizePublishPayload: jest.fn(),
+}));
+
+describe('registry rate-limit coverage (D175)', () => {
+  it('mounts a registry-wide limiter directly after the catalog router', () => {
+    const src = fs.readFileSync(path.join(__dirname, '../../../routes/registry.ts'), 'utf8');
+    const mounts = [...src.matchAll(/router\.use\((require\('\.\/registry\/([a-z-]+)'\)|rateLimit\()/g)]
+      .map((m) => m[2] || 'LIMITER');
+    const catalogAt = mounts.indexOf('catalog');
+    expect(catalogAt).toBeGreaterThan(-1);
+    expect(mounts[catalogAt + 1]).toBe('LIMITER');
+    expect(mounts.filter((m) => m === 'LIMITER')).toHaveLength(1);
+  });
+
+  it('runs a limiter before auth on POST /generate-avatar', () => {
+    const publishRouter = require('../../../routes/registry/publish');
+    const layer = publishRouter.stack.find((l) => l.route && l.route.path === '/generate-avatar');
+    expect(layer).toBeDefined();
+    expect(typeof layer.route.stack[0].handle.resetKey).toBe('function');
+  });
+
+  it('gives every spelling of one token one shared avatar budget, through the route', async () => {
+    // The limiter skips itself under NODE_ENV=test and reads it per request,
+    // so lift that for this test only. auth is mocked to pass; the limiter
+    // runs ahead of it, so the budget it enforces is the one under test.
+    const express = require('express');
+    const request = require('supertest');
+    const publishRouter = require('../../../routes/registry/publish');
+    const app = express();
+    app.use(express.json());
+    app.use(publishRouter);
+    const spellings = [
+      { Authorization: 'Bearer cm_avatar_budget' },
+      { Authorization: 'cm_avatar_budget' },
+      { 'x-auth-token': 'cm_avatar_budget' },
+    ];
+    const saved = process.env.NODE_ENV;
+    process.env.NODE_ENV = 'production';
+    try {
+      const statuses = [];
+      for (let i = 0; i < 11; i += 1) {
+        // An empty body is answered 400 by the handler, after the limiter has counted it.
+        // eslint-disable-next-line no-await-in-loop
+        const res = await request(app).post('/generate-avatar').set(spellings[i % 3]).send({});
+        statuses.push(res.status);
+      }
+      expect(statuses.slice(0, 10).every((status) => status === 400)).toBe(true);
+      expect(statuses[10]).toBe(429);
+
+      const other = await request(app).post('/generate-avatar').set({ Authorization: 'Bearer cm_someone_else' }).send({});
+      expect(other.status).toBe(400);
+    } finally {
+      process.env.NODE_ENV = saved;
+    }
+  });
+});

backend/__tests__/unit/routes/routeRateLimitGuard.baseline.json

@@ -110,7 +110,6 @@
   "routes/registry/pod-agents.ts GET /pods/:podId/agents [unlimited]",
   "routes/registry/pod-agents.ts GET /pods/:podId/agents/:name [unlimited]",
   "routes/registry/presets-router.ts GET /presets [unlimited]",
-  "routes/registry/publish.ts POST /generate-avatar [unlimited]",
   "routes/registry/publish.ts POST /publish [unlimited]",
   "routes/registry/publish.ts POST /seed [unlimited]",
   "routes/registry/templates.ts DELETE /templates/:id [unlimited]",

backend/routes/registry.ts

@@ -8,6 +8,11 @@
 // eslint-disable-next-line global-require
 const express = require('express');
 
+// eslint-disable-next-line @typescript-eslint/no-require-imports, global-require
+const { rateLimit } = require('express-rate-limit');
+// eslint-disable-next-line @typescript-eslint/no-require-imports, global-require
+const { cloudflareIpRateLimitKeyGenerator } = require('../middleware/ipRateLimit');
+
 const router: ReturnType<typeof express.Router> = express.Router();
 
 // eslint-disable-next-line global-require
@@ -18,6 +23,20 @@ router.use(require('./registry/manage-grants'));
 router.use(require('./registry/templates'));
 // eslint-disable-next-line global-require
 router.use(require('./registry/catalog'));
+// Every router mounted below shares this per-IP budget (D175). It used to
+// share the catalog's 120/min, which a per-pod read like /v2/agents spends in
+// two page loads. Routers above it (runtime, grants, templates) are not
+// covered, as before. Sized for fan-out: a 62-pod account is ~125 requests
+// per load.
+router.use(rateLimit({
+  windowMs: 60_000,
+  max: 1200,
+  standardHeaders: true,
+  legacyHeaders: false,
+  skip: () => process.env.NODE_ENV === 'test',
+  keyGenerator: (req: any) => cloudflareIpRateLimitKeyGenerator(req),
+  handler: (_req: any, res: any) => res.status(429).json({ code: 'rate_limited' }),
+}));
 // eslint-disable-next-line global-require
 router.use(require('./registry/admin'));
 // eslint-disable-next-line global-require

backend/routes/registry/catalog.ts

@@ -25,18 +25,24 @@ const catalogRouter = express.Router();
 // ~120 req/min/IP, same shape as agentProfile's limiter: generous for a
 // human browsing the catalog, low enough to blunt scrapers now that every
 // listing request may also cost a role lookup. Skipped in tests.
-catalogRouter.use(rateLimit({
+//
+// Attached per route, never via `catalogRouter.use`: registry.ts mounts this
+// router ahead of pod-agents, files, install and the rest, so a router-level
+// limiter ran for every /api/registry/* request and spent the catalog budget
+// on them. /v2/agents reads `/pods/:podId/agents` once per pod, and a 62-pod
+// account hit 429 on every call from its third load (D175, 10-28).
+const catalogRateLimit = rateLimit({
   windowMs: 60_000,
   max: 120,
   standardHeaders: true,
   legacyHeaders: false,
   skip: () => process.env.NODE_ENV === 'test',
   keyGenerator: (req: any) => cloudflareIpRateLimitKeyGenerator(req),
   handler: (_req: any, res: any) => res.status(429).json({ code: 'rate_limited' }),
-}));
+});
 
 /** Suggest a free BYO name across the caller's installable pods. */
-catalogRouter.get('/agents/next-free-name', auth, async (req: any, res: any) => {
+catalogRouter.get('/agents/next-free-name', catalogRateLimit, auth, async (req: any, res: any) => {
   try {
     const userId = getUserId(req);
     if (!userId || !mongoose.Types.ObjectId.isValid(String(userId))) {
@@ -96,7 +102,7 @@ const findExistingAgentInstance = async (agentName: any, instanceId: any) => {
  * GET /api/registry/agents
  * List available agents in the registry
  */
-catalogRouter.get('/agents', auth, async (req: any, res: any) => {
+catalogRouter.get('/agents', catalogRateLimit, auth, async (req: any, res: any) => {
   try {
     const {
       q, category, verified, registry, limit = 20, offset = 0,
@@ -161,7 +167,7 @@ catalogRouter.get('/agents', auth, async (req: any, res: any) => {
  * GET /api/registry/agents/:name
  * Get agent details
  */
-catalogRouter.get('/agents/:name', auth, async (req: any, res: any) => {
+catalogRouter.get('/agents/:name', catalogRateLimit, auth, async (req: any, res: any) => {
   try {
     const agent = await AgentRegistry.getByName(req.params.name);
 
@@ -201,7 +207,7 @@ catalogRouter.get('/agents/:name', auth, async (req: any, res: any) => {
  * GET /api/registry/agents/:name/instances/:instanceId
  * Check if an agent instance exists globally (across all pods).
  */
-catalogRouter.get('/agents/:name/instances/:instanceId', auth, async (req: any, res: any) => {
+catalogRouter.get('/agents/:name/instances/:instanceId', catalogRateLimit, auth, async (req: any, res: any) => {
   try {
     const { name, instanceId } = req.params;
     const userId = getUserId(req);
@@ -242,7 +248,7 @@ catalogRouter.get('/agents/:name/instances/:instanceId', auth, async (req: any,
  * GET /api/registry/agents/:name/instances
  * List all instances of an agent type (for discovery).
  */
-catalogRouter.get('/agents/:name/instances', auth, async (req: any, res: any) => {
+catalogRouter.get('/agents/:name/instances', catalogRateLimit, auth, async (req: any, res: any) => {
   try {
     const { name } = req.params;
     const userId = getUserId(req);
@@ -291,7 +297,7 @@ catalogRouter.get('/agents/:name/instances', auth, async (req: any, res: any) =>
  * GET /api/registry/categories
  * List agent categories
  */
-catalogRouter.get('/categories', auth, async (req: any, res: any) => {
+catalogRouter.get('/categories', catalogRateLimit, auth, async (req: any, res: any) => {
   try {
     const categories = await AgentRegistry.distinct('categories');
     res.json({ categories });
@@ -305,7 +311,7 @@ catalogRouter.get('/categories', auth, async (req: any, res: any) => {
  * GET /api/registry/openclaw/bundled-skills
  * List bundled gateway skills available under /app/skills.
  */
-catalogRouter.get('/openclaw/bundled-skills', auth, async (req: any, res: any) => {
+catalogRouter.get('/openclaw/bundled-skills', catalogRateLimit, auth, async (req: any, res: any) => {
   try {
     const userId = getUserId(req);
     if (!userId) {
@@ -329,7 +335,7 @@ catalogRouter.get('/openclaw/bundled-skills', auth, async (req: any, res: any) =
  * GET /api/registry/agents/:name/installations
  * List pods where an agent instance is installed (user-visible only)
  */
-catalogRouter.get('/agents/:name/installations', auth, async (req: any, res: any) => {
+catalogRouter.get('/agents/:name/installations', catalogRateLimit, auth, async (req: any, res: any) => {
   try {
     const userId = getUserId(req);
     if (!userId) {

backend/routes/registry/publish.ts

@@ -1,7 +1,12 @@
 // Agent publish/seed/avatar routes — extracted from registry.js (GH#112)
 // Handles: POST /publish, POST /seed, POST /generate-avatar
+const { createHash } = require('crypto');
 const express = require('express');
+// eslint-disable-next-line @typescript-eslint/no-require-imports, global-require
+const { rateLimit } = require('express-rate-limit');
 const auth = require('../../middleware/auth');
+// eslint-disable-next-line @typescript-eslint/no-require-imports, global-require
+const { cloudflareIpRateLimitKeyGenerator } = require('../../middleware/ipRateLimit');
 const { AgentRegistry } = require('../../models/AgentRegistry');
 const AgentIdentityService = require('../../services/agentIdentityService');
 const { getUserId, resolveUsername } = require('./helpers');
@@ -12,6 +17,28 @@ const {
 
 const publishRouter = express.Router();
 
+// generate-avatar calls the configured image provider (Gemini or OpenAI) on
+// every request, so it gets its own per-caller budget on top of the
+// registry-wide IP limiter. It runs ahead of `auth` (which does a Mongo
+// lookup), so it derives the token exactly as `auth` does: every spelling that
+// `auth` accepts as one token (`Bearer cm_x`, `cm_x`, `x-auth-token: cm_x`)
+// shares one budget. Only a hash of the token is kept in limiter state.
+const avatarCallerKey = (req: any) => {
+  const token = req.header?.('Authorization')?.replace('Bearer ', '') || req.header?.('x-auth-token');
+  if (token) return `tok:${createHash('sha256').update(token).digest('hex').slice(0, 16)}`;
+  return cloudflareIpRateLimitKeyGenerator(req);
+};
+
+const avatarRateLimit = rateLimit({
+  windowMs: 60_000,
+  max: 10,
+  standardHeaders: true,
+  legacyHeaders: false,
+  skip: () => process.env.NODE_ENV === 'test',
+  keyGenerator: avatarCallerKey,
+  handler: (_req: any, res: any) => res.status(429).json({ code: 'rate_limited' }),
+});
+
 /**
  * POST /api/registry/publish
  * Publish a new agent to the registry (for developers)
@@ -199,7 +226,7 @@ publishRouter.post('/seed', auth, async (req: any, res: any) => {
  * POST /api/registry/generate-avatar
  * Generate AI avatar for an agent
  */
-publishRouter.post('/generate-avatar', auth, async (req: any, res: any) => {
+publishRouter.post('/generate-avatar', avatarRateLimit, auth, async (req: any, res: any) => {
   try {
     // eslint-disable-next-line global-require
     const AgentAvatarService = require('../../services/agentAvatarService');