From 8ea6f5206c8f4237525f1708e54753ad40c0ebf7 Mon Sep 17 00:00:00 2001
From: kiranvk2011 <kiranvk2011@gmail.com>
Date: Wed, 18 Mar 2026 15:36:32 +0000
Subject: [PATCH] fix: update cooldownReason on active-window failures + clear
 model scope for non-rate-limit

---
 src/agents/auth-profiles/usage.test.ts | 47 ++++++++++++++++++++++++++
 src/agents/auth-profiles/usage.ts      | 10 +++++-
 2 files changed, 56 insertions(+), 1 deletion(-)

diff --git a/src/agents/auth-profiles/usage.test.ts b/src/agents/auth-profiles/usage.test.ts
index 8d633a7d89b..69d3d4ca200 100644
--- a/src/agents/auth-profiles/usage.test.ts
+++ b/src/agents/auth-profiles/usage.test.ts
@@ -796,4 +796,51 @@ describe("markAuthProfileFailure — per-model cooldown metadata", () => {
     const stats = store.usageStats?.["github-copilot:github"];
     expect(stats?.cooldownModel).toBe("claude-sonnet-4.6");
   });
+
+  it("updates cooldownReason when auth failure occurs during active rate_limit window", async () => {
+    const now = 1_000_000;
+    const store = makeStoreWithCopilot({
+      "github-copilot:github": {
+        cooldownUntil: now + 30_000,
+        cooldownReason: "rate_limit",
+        cooldownModel: "claude-sonnet-4.6",
+        errorCount: 1,
+        lastFailureAt: now - 1000,
+      },
+    });
+    await markAuthProfileFailure({
+      store,
+      profileId: "github-copilot:github",
+      reason: "auth",
+      modelId: "claude-opus-4.6",
+    });
+    const stats = store.usageStats?.["github-copilot:github"];
+    // Reason should update to the new failure type, not stay as rate_limit
+    expect(stats?.cooldownReason).toBe("auth");
+    // Model scope should be cleared — auth failures are profile-wide
+    expect(stats?.cooldownModel).toBeUndefined();
+  });
+
+  it("clears cooldownModel when non-rate_limit failure hits same model during active window", async () => {
+    const now = 1_000_000;
+    const store = makeStoreWithCopilot({
+      "github-copilot:github": {
+        cooldownUntil: now + 30_000,
+        cooldownReason: "rate_limit",
+        cooldownModel: "claude-sonnet-4.6",
+        errorCount: 1,
+        lastFailureAt: now - 1000,
+      },
+    });
+    await markAuthProfileFailure({
+      store,
+      profileId: "github-copilot:github",
+      reason: "auth",
+      modelId: "claude-sonnet-4.6",
+    });
+    const stats = store.usageStats?.["github-copilot:github"];
+    // Even same-model auth failure should clear model scope (auth is profile-wide)
+    expect(stats?.cooldownReason).toBe("auth");
+    expect(stats?.cooldownModel).toBeUndefined();
+  });
 });
diff --git a/src/agents/auth-profiles/usage.ts b/src/agents/auth-profiles/usage.ts
index f9b7cfc975f..84f9bdaa83b 100644
--- a/src/agents/auth-profiles/usage.ts
+++ b/src/agents/auth-profiles/usage.ts
@@ -479,7 +479,11 @@ function computeNextProfileUsageStats(params: {
       typeof params.existing.cooldownUntil === "number" &&
       params.existing.cooldownUntil > params.now;
     if (existingCooldownActive) {
-      updatedStats.cooldownReason = params.existing.cooldownReason;
+      // Always use the latest failure reason so that downstream consumers
+      // (e.g. isProfileInCooldown model-bypass) see the most recent signal.
+      // A non-rate_limit failure (auth, billing, …) is profile-wide, so
+      // upgrading from rate_limit → auth correctly blocks all models.
+      updatedStats.cooldownReason = params.reason;
       // If a different model fails during an active window, widen the scope
       // to all models (undefined) so neither model bypasses the cooldown.
       if (
@@ -488,6 +492,10 @@ function computeNextProfileUsageStats(params: {
         params.existing.cooldownModel !== params.modelId
       ) {
         updatedStats.cooldownModel = undefined;
+      } else if (params.reason !== "rate_limit") {
+        // Non-rate-limit failures are profile-wide — clear model scope even
+        // when the same model fails, so that no model can bypass.
+        updatedStats.cooldownModel = undefined;
       } else {
         updatedStats.cooldownModel = params.existing.cooldownModel;
       }