fix: update ACP SDK large frame handling (#1496)

apps/backend/go.mod

@@ -116,4 +116,4 @@ require (
 	gotest.tools/v3 v3.5.2 // indirect
 )
 
-replace github.com/coder/acp-go-sdk => github.com/kdlbs/acp-go-sdk v0.13.6-0.20260609161428-91d5e43d6599
+replace github.com/coder/acp-go-sdk => github.com/kdlbs/acp-go-sdk v0.13.6-0.20260625174839-e37105a28386

apps/backend/go.sum

@@ -217,8 +217,8 @@ github.com/json-iterator/go v1.1.12 h1:PV8peI4a0ysnczrg+LtxykD8LfKY9ML6u2jnxaEnr
 github.com/json-iterator/go v1.1.12/go.mod h1:e30LSqwooZae/UwlEbR2852Gd8hjQvJoHmT4TnhNGBo=
 github.com/jstemmer/go-junit-report v0.0.0-20190106144839-af01ea7f8024/go.mod h1:6v2b51hI/fHJwM22ozAgKL4VKDeJcHhJFhtBdhmNjmU=
 github.com/jstemmer/go-junit-report v0.9.1/go.mod h1:Brl9GWCQeLvo8nXZwPNNblvFj/XSXhF0NWZEnDohbsk=
-github.com/kdlbs/acp-go-sdk v0.13.6-0.20260609161428-91d5e43d6599 h1:VeNo6BcOkY18Hj9HopG7/3t5PpD4ocmuxMIuGSobzLQ=
-github.com/kdlbs/acp-go-sdk v0.13.6-0.20260609161428-91d5e43d6599/go.mod h1:yKzM/3R9uELp4+nBAwwtkS0aN1FOFjo11CNPy37yFko=
+github.com/kdlbs/acp-go-sdk v0.13.6-0.20260625174839-e37105a28386 h1:zUFkN1/+o3vEYqrkshpykUBAstk6NqiitMuh+dR+Tik=
+github.com/kdlbs/acp-go-sdk v0.13.6-0.20260625174839-e37105a28386/go.mod h1:yKzM/3R9uELp4+nBAwwtkS0aN1FOFjo11CNPy37yFko=
 github.com/kevinburke/ssh_config v1.6.0 h1:J1FBfmuVosPHf5GRdltRLhPJtJpTlMdKTBjRgTaQBFY=
 github.com/kevinburke/ssh_config v1.6.0/go.mod h1:q2RIzfka+BXARoNexmF9gkxEX7DmvbW9P4hIVx2Kg4M=
 github.com/kisielk/gotool v1.0.0/go.mod h1:XhKaO+MFFWcvkIS/tQcRk01m1F5IRFswLeQ+oQHNcck=

apps/backend/internal/agentctl/server/adapter/transport/acp/adapter.go

@@ -59,11 +59,11 @@ const wakeupPromptTimeout = 30 * time.Minute
 const notifQueueCapacity = 4096
 
 // acpNotifQueueDefault is the per-connection capacity passed to the SDK's
-// inbound notification queue. Larger than the SDK's own default (1024) so
-// long session/load replays (auggie 500+ exchanges, claude with heavy tool
-// use) don't overflow and tear the connection down. The SDK still bounds
-// memory to (capacity * avg notification size).
-const acpNotifQueueDefault = 16384
+// inbound notification queue. Long session/load replays can emit tens of
+// thousands of notifications before the response, so default to the configured
+// ceiling instead of requiring operators to know about KANDEV_ACP_NOTIF_QUEUE.
+// The SDK still bounds memory to (capacity * avg notification size).
+const acpNotifQueueDefault = 131072
 
 // acpNotifQueueMin / acpNotifQueueMax clamp KANDEV_ACP_NOTIF_QUEUE so a
 // misconfigured value can't either re-introduce the overflow (too low) or

apps/backend/internal/agentctl/server/adapter/transport/acp/adapter_queue_test.go

@@ -0,0 +1,56 @@
+package acp
+
+import "testing"
+
+func TestACPNotifQueueCapacity(t *testing.T) {
+	t.Run("default uses max capacity for long session replays", func(t *testing.T) {
+		t.Setenv("KANDEV_ACP_NOTIF_QUEUE", "")
+
+		if got := acpNotifQueueCapacity(); got != acpNotifQueueDefault {
+			t.Fatalf("default queue capacity = %d, want %d", got, acpNotifQueueDefault)
+		}
+		if got := acpNotifQueueDefault; got != acpNotifQueueMax {
+			t.Fatalf("default queue capacity = %d, want max %d", got, acpNotifQueueMax)
+		}
+	})
+
+	t.Run("invalid env falls back to default", func(t *testing.T) {
+		t.Setenv("KANDEV_ACP_NOTIF_QUEUE", "not-a-number")
+
+		if got := acpNotifQueueCapacity(); got != acpNotifQueueDefault {
+			t.Fatalf("invalid env queue capacity = %d, want %d", got, acpNotifQueueDefault)
+		}
+	})
+
+	t.Run("non-positive env falls back to default", func(t *testing.T) {
+		for _, value := range []string{"0", "-5"} {
+			t.Run(value, func(t *testing.T) {
+				t.Setenv("KANDEV_ACP_NOTIF_QUEUE", value)
+
+				if got := acpNotifQueueCapacity(); got != acpNotifQueueDefault {
+					t.Fatalf("non-positive env queue capacity = %d, want %d", got, acpNotifQueueDefault)
+				}
+			})
+		}
+	})
+
+	t.Run("in-range env is honored", func(t *testing.T) {
+		t.Setenv("KANDEV_ACP_NOTIF_QUEUE", "4096")
+
+		if got := acpNotifQueueCapacity(); got != 4096 {
+			t.Fatalf("in-range env queue capacity = %d, want 4096", got)
+		}
+	})
+
+	t.Run("env is clamped", func(t *testing.T) {
+		t.Setenv("KANDEV_ACP_NOTIF_QUEUE", "1")
+		if got := acpNotifQueueCapacity(); got != acpNotifQueueMin {
+			t.Fatalf("low env queue capacity = %d, want %d", got, acpNotifQueueMin)
+		}
+
+		t.Setenv("KANDEV_ACP_NOTIF_QUEUE", "999999")
+		if got := acpNotifQueueCapacity(); got != acpNotifQueueMax {
+			t.Fatalf("high env queue capacity = %d, want %d", got, acpNotifQueueMax)
+		}
+	})
+}