feat(tools): stream shell_run output live into the TUI

internal/agent/agent_test.go

@@ -87,6 +87,86 @@ func TestAgentRunsToolLoop(t *testing.T) {
 	}
 }
 
+type streamToolProvider struct {
+	calls int
+	t     *testing.T
+}
+
+func (p *streamToolProvider) ChatStream(ctx context.Context, req llm.ChatRequest, emit func(llm.StreamEvent) error) error {
+	p.calls++
+	switch p.calls {
+	case 1:
+		return emit(llm.StreamEvent{ToolCalls: []llm.ToolCallDelta{
+			{Index: 0, ID: "call_1", Type: "function", Name: "test_stream", Arguments: `{}`},
+		}})
+	case 2:
+		last := req.Messages[len(req.Messages)-1]
+		if last.Role != llm.RoleTool || last.Content != "full result" {
+			p.t.Fatalf("last message = %#v", last)
+		}
+		return emit(llm.StreamEvent{Content: "done"})
+	default:
+		p.t.Fatalf("unexpected call count %d", p.calls)
+		return nil
+	}
+}
+
+type streamEchoTool struct{}
+
+func (streamEchoTool) Definition() llm.Tool {
+	return llm.Tool{
+		Type: "function",
+		Function: llm.ToolFunction{
+			Name:        "test_stream",
+			Description: "streaming test",
+			Parameters:  json.RawMessage(`{"type":"object"}`),
+		},
+	}
+}
+
+func (streamEchoTool) Execute(ctx context.Context, raw json.RawMessage) (string, error) {
+	return "full result", nil
+}
+
+func (streamEchoTool) ExecuteStream(ctx context.Context, raw json.RawMessage, emit func(string) error) (string, error) {
+	if err := emit("one\n"); err != nil {
+		return "", err
+	}
+	if err := emit("two\n"); err != nil {
+		return "", err
+	}
+	return "full result", nil
+}
+
+func TestAgentStreamsToolOutputToLogs(t *testing.T) {
+	provider := &streamToolProvider{t: t}
+	a := New(Options{
+		Provider:     provider,
+		Model:        "test",
+		SystemPrompt: "system",
+		ToolRegistry: tools.NewRegistry(streamEchoTool{}),
+	})
+
+	var out strings.Builder
+	var logs strings.Builder
+	if err := a.RunTurn(context.Background(), "stream", &out, &logs); err != nil {
+		t.Fatal(err)
+	}
+	if out.String() != "done" {
+		t.Fatalf("out = %q", out.String())
+	}
+	logText := logs.String()
+	if strings.Count(logText, "[output]") != 2 {
+		t.Fatalf("logs = %q", logText)
+	}
+	if !strings.Contains(logText, "one\n") || !strings.Contains(logText, "two\n") {
+		t.Fatalf("logs missing streamed chunks: %q", logText)
+	}
+	if strings.Contains(logText, "[output]\nfull result") {
+		t.Fatalf("logs contain duplicate full output: %q", logText)
+	}
+}
+
 type contentProvider struct {
 	called bool
 	text   string