fix(server): a Claude account with no rate-limit data no longer stays unsupported

apps/server/src/provider/claudeUsageLimits.test.ts

@@ -95,6 +95,17 @@ describe("claudeUsageResponseToLimits", () => {
     ).toEqual({ checkedAt, windows: [], unavailable: { reason: "unsupported" } });
   });
 
+  it("reports a subscription account whose usage read came back empty as a failed probe", () => {
+    // The CLI sets `rate_limits` to null when its own usage fetch fails, while
+    // `rate_limits_available` stays true for an account that has limits.
+    const { limits, names } = claudeUsageResponseToLimits({
+      checkedAt,
+      response: { rate_limits_available: true, rate_limits: null },
+    });
+    expect(limits).toEqual({ checkedAt, windows: [], unavailable: { reason: "probeFailed" } });
+    expect(names).toEqual(noNames);
+  });
+
   it("skips a window the endpoint reports without a utilization", () => {
     expect(
       claudeUsageResponseToLimits({

apps/server/src/provider/claudeUsageLimits.ts

@@ -160,8 +160,15 @@ export function claudeUsageResponseToLimits(input: {
 }): { readonly limits: ServerProviderUsageLimits; readonly names: ClaudeScopedLimitNames } {
   const { response, checkedAt } = input;
   if (!response.rate_limits_available || !response.rate_limits) {
+    // `unsupported` is a lock: it hides the account from Limits and ignores
+    // streamed windows. Only `rate_limits_available: false` earns it. The CLI
+    // also sends `rate_limits: null` beside `true` when its own usage fetch
+    // failed, which is transient, so the next probe or event can recover.
     return {
-      limits: makeUnavailableUsageLimits({ checkedAt, reason: "unsupported" }),
+      limits: makeUnavailableUsageLimits({
+        checkedAt,
+        reason: response.rate_limits_available ? "probeFailed" : "unsupported",
+      }),
       names: { overageIncluded: undefined },
     };
   }