diff --git a/src/ui/transcript/tool_group_projection.zig b/src/ui/transcript/tool_group_projection.zig
index cfd2c4212..300317df3 100644
--- a/src/ui/transcript/tool_group_projection.zig
+++ b/src/ui/transcript/tool_group_projection.zig
@@ -151,6 +151,7 @@ const category_labels = [_][]const u8{
     "browser",
 };
 const command_category_index = 5;
+const subagent_category_index = 6;
 
 const Summary = struct {
     total: usize = 0,
@@ -182,7 +183,7 @@ fn categoryIndex(kind: types.ToolActivityKind) ?usize {
         .edit => 3,
         .open => 4,
         .command => command_category_index,
-        .subagent => 6,
+        .subagent => subagent_category_index,
         .ask => null,
     };
 }
@@ -531,10 +532,11 @@ fn formatGroupSummary(alloc: std.mem.Allocator, summary: Summary) ![]u8 {
         const index = next_index orelse break;
         emitted[index] = true;
         const count = summary.categories[index];
-        const label = if (index == command_category_index and count != 1)
-            "commands"
-        else
-            category_labels[index];
+        const label = if (count == 1) category_labels[index] else switch (index) {
+            command_category_index => "commands",
+            subagent_category_index => "subagents",
+            else => category_labels[index],
+        };
         try appendSegment(&out.writer, count, label);
     }
     try appendSegment(&out.writer, summary.completion_unreported, "unreported");
@@ -1490,7 +1492,7 @@ test "grouped subagent status keeps the vertical continuation for middle rows" {
     defer projection.deinit(alloc);
 
     try std.testing.expectEqualStrings(
-        "● 4 tool calls · 4 subagent\n" ++
+        "● 4 tool calls · 4 subagents\n" ++
             "├ Subagent working · inspect auth\n│ glm-5.3-flash · max\n" ++
             "├ Subagent working · inspect auth\n│ glm-5.3-flash · max\n" ++
             "├ Subagent working · inspect auth\n│ glm-5.3-flash · max\n" ++
@@ -1499,6 +1501,23 @@ test "grouped subagent status keeps the vertical continuation for middle rows" {
     );
 }
 
+test "group summary uses singular and plural subagent labels" {
+    const alloc = std.testing.allocator;
+    var summary: Summary = .{ .total = 2 };
+    summary.categories[command_category_index] = 1;
+    summary.categories[subagent_category_index] = 1;
+    const single = try formatGroupSummary(alloc, summary);
+    defer alloc.free(single);
+    try std.testing.expectEqualStrings("● 2 tool calls · 1 command · 1 subagent", single);
+
+    summary = .{ .total = 5 };
+    summary.categories[command_category_index] = 2;
+    summary.categories[subagent_category_index] = 3;
+    const plural = try formatGroupSummary(alloc, summary);
+    defer alloc.free(plural);
+    try std.testing.expectEqualStrings("● 5 tool calls · 3 subagents · 2 commands", plural);
+}
+
 test "small minimal tool groups surface canonical action targets" {
     const alloc = std.testing.allocator;
     const entries = [_]TranscriptEntry{
diff --git a/tests/e2e/tui-gateway-stream-lifecycle.test.ts b/tests/e2e/tui-gateway-stream-lifecycle.test.ts
index a21f4bec4..2a580e043 100644
--- a/tests/e2e/tui-gateway-stream-lifecycle.test.ts
+++ b/tests/e2e/tui-gateway-stream-lifecycle.test.ts
@@ -6369,7 +6369,7 @@ describe.skipIf(!tmuxAvailable())("TUI gateway stream lifecycle", () => {
         }
         await session.waitForText(finalText, TIMEOUT);
         const compact = await session.captureFullScrollback();
-        expect(compact).toContain("● 5 tool calls · 4 subagent · 1 read");
+        expect(compact).toContain("● 5 tool calls · 4 subagents · 1 read");
         expect(compact).toContain("Invalid Agent failed · Check failure visibility");
         expect(compact).not.toContain("Invalid Agent replied");
         for (let index = 0; index < tasks.length; index += 1) {
@@ -6400,7 +6400,7 @@ describe.skipIf(!tmuxAvailable())("TUI gateway stream lifecycle", () => {
         session = await TmuxSession.create({ cmd: `${FX_BIN} --resume-last`, cwd: workspace, env, width: 110, height: 35, stderrPath: join(root, "resumed-stderr.log") });
         await session.waitForText(finalText, TIMEOUT);
         const resumed = await session.captureFullScrollback();
-        expect(resumed).toContain("● 5 tool calls · 4 subagent · 1 read");
+        expect(resumed).toContain("● 5 tool calls · 4 subagents · 1 read");
         expect(resumed).toContain("Invalid Agent failed · Check failure visibility");
         expect(resumed).toContain("Subagent finished · Check one-off cleanup");
         expect(resumed).toContain("reviewer replied · Check replay again");