fix: allow for tokens to be unbounded by rate limiting

.changeset/unbounded-rate-limit-token.md

@@ -0,0 +1,6 @@
+---
+"@anticapture/gateful": minor
+"@anticapture/authful": minor
+---
+
+Support unbounded (rate-limit-exempt) tokens. A token with `rateLimitPerMin` set to `0` (the sentinel for any non-positive value) is now skipped entirely by Gateful's rate-limit middleware — it never touches Redis and is never throttled. Authful's mint endpoint accepts `0` accordingly (`rateLimitPerMin` validation relaxed from positive to non-negative).

apps/authful/README.md

@@ -42,7 +42,8 @@ curl -sX POST http://localhost:4002/tokens \
   -d '{"tenant": "acme", "name": "acme mcp prod", "rateLimitPerMin": 600}'
 ```
 
-`rateLimitPerMin` is optional (defaults to 600).
+`rateLimitPerMin` is optional (defaults to 600). Set it to `0` to make the
+token unbounded — Gateful exempts it from rate limiting entirely.
 
 ## Migrations
 

apps/authful/src/controllers/authful.integration.test.ts

@@ -130,6 +130,14 @@ describe("authful app", () => {
       expect(JSON.stringify(row)).not.toContain(body.token);
     });
 
+    it("accepts rateLimitPerMin 0 to mint an unbounded token", async () => {
+      const body = await mint({ rateLimitPerMin: 0 });
+      expect(body.rateLimitPerMin).toBe(0);
+
+      const [row] = await db.select().from(tokens);
+      expect(row!.rateLimitPerMin).toBe(0);
+    });
+
     it("rejects a body sent without a JSON content-type (no silent {})", async () => {
       // Regression: without `required: true`, @hono/zod-openapi skips
       // validation for a non-JSON content-type and substitutes {}, so mint

apps/authful/src/mappers/tokens/index.ts

@@ -20,7 +20,10 @@ export const MintTokenBodySchema = z
   .object({
     tenant: z.string().min(1),
     name: z.string().min(1),
-    rateLimitPerMin: z.number().int().positive().optional(),
+    rateLimitPerMin: z.number().int().nonnegative().optional().openapi({
+      description:
+        "Requests per minute. 0 means unbounded (no rate limiting). Omit to default to 600.",
+    }),
   })
   .openapi("MintTokenBody");
 

apps/gateful/src/auth/rate-limit.test.ts

@@ -63,6 +63,17 @@ describe("rateLimitMiddleware", () => {
     expect(retryAfter).toBeLessThanOrEqual(60);
   });
 
+  it("never limits an unbounded token (rateLimitPerMin 0) and skips the store", async () => {
+    const store = new FakeRateLimitStore();
+    const app = buildApp(store, { ...AUTH, rateLimitPerMin: 0 });
+
+    for (let i = 0; i < 10; i++) {
+      const res = await app.request("/test");
+      expect(res.status).toBe(200);
+    }
+    expect(store.counters.size).toBe(0);
+  });
+
   it("skips limiting when there is no auth context (public/legacy)", async () => {
     const store = new FakeRateLimitStore();
     const app = buildApp(store, null);

apps/gateful/src/auth/rate-limit.ts

@@ -15,11 +15,14 @@ export interface RateLimitStore {
  * Redis (or on Redis errors) requests pass — availability over enforcement.
  * Runs after tokenAuthMiddleware; requests without auth context (public
  * paths, auth disabled) are not limited.
+ *
+ * A non-positive limit (0 is the sentinel) means "unbounded": the token is
+ * exempt from rate limiting and never touches Redis.
  */
 export function rateLimitMiddleware(store?: RateLimitStore) {
   return async (c: Context, next: Next) => {
     const auth = c.get("auth");
-    if (!auth || !store) return next();
+    if (!auth || !store || auth.rateLimitPerMin <= 0) return next();
 
     const epochMinute = Math.floor(Date.now() / 60_000);
     const key = `rl:${auth.tokenId}:${epochMinute}`;