feat: show token/elapsed summary after one-shot text runs (#288)

Text-mode --print previously discarded the stream usage event. Capture the
final usage and render a muted token/elapsed footer to stderr on completion,
so interactive one-shot runs report cost while stdout stays pure for scripts.
JSON/stream-json output is unchanged.

Author: Patel230

cmd/chat_print.go

@@ -83,6 +83,8 @@ func runPrint(text string) error {
 
 	var printed strings.Builder
 	var countdownShown bool
+	var lastUsage *engine.StreamUsage
+	started := time.Now()
 	for ev := range ch {
 		switch ev.Type {
 		case "content":
@@ -118,6 +120,9 @@ func runPrint(text string) error {
 				_, _ = fmt.Fprintf(os.Stderr, "%s %s\n", auditTint("["+ev.ToolName+"]", infoSky), content)
 			}
 		case "usage":
+			if ev.Usage != nil {
+				lastUsage = ev.Usage
+			}
 			if outputFormat == "stream-json" && ev.Usage != nil {
 				writePrintUsageEvent(sessionID, ev.Usage)
 			}
@@ -132,6 +137,7 @@ func runPrint(text string) error {
 				if !strings.HasSuffix(printed.String(), "\n") {
 					fmt.Println()
 				}
+				printTextUsageFooter(lastUsage, started)
 			case "json":
 				writePrintResult(printed.String(), sessionID, sess, false, nil)
 			case "stream-json":
@@ -148,6 +154,7 @@ func runPrint(text string) error {
 		if !strings.HasSuffix(printed.String(), "\n") {
 			fmt.Println()
 		}
+		printTextUsageFooter(lastUsage, started)
 	case "json":
 		writePrintResult(printed.String(), sessionID, sess, false, nil)
 	case "stream-json":
@@ -180,6 +187,21 @@ func writePrintUsageEvent(sessionID string, usage *engine.StreamUsage) {
 	fmt.Println(string(data))
 }
 
+// printTextUsageFooter renders a muted token/elapsed summary to stderr after a
+// one-shot text-mode run. It writes to stderr so stdout stays pure for scripts,
+// and is skipped entirely when no usage event was received.
+func printTextUsageFooter(usage *engine.StreamUsage, started time.Time) {
+	if usage == nil {
+		return
+	}
+	parts := []string{fmt.Sprintf("%d in · %d out", usage.PromptTokens, usage.CompletionTokens)}
+	if usage.CacheReadTokens > 0 || usage.CacheWriteTokens > 0 {
+		parts = append(parts, fmt.Sprintf("cache %d read · %d write", usage.CacheReadTokens, usage.CacheWriteTokens))
+	}
+	parts = append(parts, time.Since(started).Round(time.Second).String())
+	_, _ = fmt.Fprintf(os.Stderr, "%s\n", auditTint("tokens: "+strings.Join(parts, " · "), textMuted))
+}
+
 func writePrintResult(result, sessionID string, sess *engine.Session, isError bool, errors []string) {
 	event := map[string]interface{}{
 		"type":           "result",

cmd/chat_print_usage_test.go

@@ -0,0 +1,88 @@
+package cmd
+
+import (
+	"os"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/GrayCodeAI/graycode-cli/internal/engine"
+)
+
+// captureStderr runs fn with os.Stderr redirected to a pipe and returns the
+// captured bytes.
+func captureStderr(t *testing.T, fn func()) string {
+	t.Helper()
+	old := os.Stderr
+	r, w, err := os.Pipe()
+	if err != nil {
+		t.Fatalf("pipe: %v", err)
+	}
+	os.Stderr = w
+	defer func() { os.Stderr = old }()
+
+	fn()
+	_ = w.Close()
+
+	buf := make([]byte, 4096)
+	n, _ := r.Read(buf)
+	_ = r.Close()
+	return string(buf[:n])
+}
+
+func TestPrintTextUsageFooter(t *testing.T) {
+	started := time.Now().Add(-2 * time.Second)
+
+	t.Run("renders token and elapsed summary", func(t *testing.T) {
+		got := captureStderr(t, func() {
+			printTextUsageFooter(&engine.StreamUsage{
+				PromptTokens:     100,
+				CompletionTokens: 50,
+			}, started)
+		})
+		if !strings.Contains(got, "100 in · 50 out") {
+			t.Errorf("footer missing token counts: %q", got)
+		}
+		if !strings.Contains(got, "tokens:") {
+			t.Errorf("footer missing prefix: %q", got)
+		}
+		if !strings.Contains(got, "2s") {
+			t.Errorf("footer missing elapsed: %q", got)
+		}
+	})
+
+	t.Run("includes cache when nonzero", func(t *testing.T) {
+		got := captureStderr(t, func() {
+			printTextUsageFooter(&engine.StreamUsage{
+				PromptTokens:     10,
+				CompletionTokens: 5,
+				CacheReadTokens:  90,
+				CacheWriteTokens: 7,
+			}, started)
+		})
+		if !strings.Contains(got, "cache 90 read · 7 write") {
+			t.Errorf("footer missing cache summary: %q", got)
+		}
+	})
+
+	t.Run("omits cache when zero", func(t *testing.T) {
+		got := captureStderr(t, func() {
+			printTextUsageFooter(&engine.StreamUsage{
+				PromptTokens:     10,
+				CompletionTokens: 5,
+			}, started)
+		})
+		if strings.Contains(got, "cache") {
+			t.Errorf("footer should omit zero cache: %q", got)
+		}
+	})
+
+	t.Run("skips output when usage is nil", func(t *testing.T) {
+		got := captureStderr(t, func() {
+			printTextUsageFooter(nil, started)
+		})
+		if got != "" {
+			t.Errorf("expected no output for nil usage, got: %q", got)
+		}
+	})
+}