Merge pull request #2291 from docker/board/fix-message-in-docker-agent-pr-2239-4d14d575

tui: show elapsed time and warning for long-running tool calls

Author: dgageot

pkg/tui/components/messages/messages.go

@@ -5,6 +5,7 @@ import (
 	"strconv"
 	"strings"
 	"sync/atomic"
+	"time"
 
 	"charm.land/bubbles/v2/help"
 	"charm.land/bubbles/v2/key"
@@ -1333,6 +1334,10 @@ func (m *model) AddOrUpdateToolCall(agentName string, toolCall tools.ToolCall, t
 		msg := m.messages[i]
 		if msg.Type == types.MessageTypeToolCall && msg.ToolCall.ID == toolCall.ID {
 			msg.ToolStatus = status
+			if status == types.ToolStatusRunning && msg.StartedAt == nil {
+				now := time.Now()
+				msg.StartedAt = &now
+			}
 			if toolCall.Function.Arguments != "" {
 				if status == types.ToolStatusPending {
 					msg.ToolCall.Function.Arguments += toolCall.Function.Arguments

pkg/tui/components/reasoningblock/reasoningblock.go

@@ -205,6 +205,10 @@ func (m *Model) UpdateToolCall(toolCallID string, status types.ToolStatus, args
 			continue
 		}
 		entry.msg.ToolStatus = status
+		if status == types.ToolStatusRunning && entry.msg.StartedAt == nil {
+			now := time.Now()
+			entry.msg.StartedAt = &now
+		}
 		if args != "" {
 			if status == types.ToolStatusPending {
 				entry.msg.ToolCall.Function.Arguments += args

pkg/tui/components/toolcommon/common.go

@@ -4,6 +4,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"strings"
+	"time"
 
 	"charm.land/lipgloss/v2"
 
@@ -101,13 +102,24 @@ func ExtractField[T any](field func(T) string) func(string) string {
 	}
 }
 
+// LongRunningThreshold is the duration after which a running tool call
+// displays a warning hint that it may be blocked on external input.
+const LongRunningThreshold = 60 * time.Second
+
 func Icon(msg *types.Message, inProgress spinner.Spinner) string {
 	switch msg.ToolStatus {
 	case types.ToolStatusRunning, types.ToolStatusPending:
 		// Animated spinner for both executing and streaming tool calls.
 		// With centralized animation ticks, all spinners share a single tick
 		// so there's no performance penalty for multiple animated spinners.
-		return styles.NoStyle.MarginLeft(2).Render(inProgress.View())
+		icon := styles.NoStyle.MarginLeft(2).Render(inProgress.View())
+		if msg.StartedAt != nil {
+			elapsed := time.Since(*msg.StartedAt)
+			if elapsed >= time.Second {
+				icon += " " + styles.ToolMessageStyle.Render(formatDuration(elapsed))
+			}
+		}
+		return icon
 	case types.ToolStatusCompleted:
 		return styles.ToolCompletedIcon.Render("✓")
 	case types.ToolStatusError:
@@ -119,6 +131,35 @@ func Icon(msg *types.Message, inProgress spinner.Spinner) string {
 	}
 }
 
+// LongRunningWarning returns a warning string if the tool call has been
+// running longer than LongRunningThreshold, or empty string otherwise.
+func LongRunningWarning(msg *types.Message) string {
+	if msg.StartedAt == nil {
+		return ""
+	}
+	if msg.ToolStatus != types.ToolStatusRunning {
+		return ""
+	}
+	if time.Since(*msg.StartedAt) < LongRunningThreshold {
+		return ""
+	}
+	return "⚠ Tool call running for over 60s. The tool may be waiting for external input. Press Esc to cancel."
+}
+
+// formatDuration formats a duration as a human-readable string like "5s", "1m30s", "2m15s".
+func formatDuration(d time.Duration) string {
+	d = d.Truncate(time.Second)
+	if d < time.Minute {
+		return fmt.Sprintf("%ds", int(d.Seconds()))
+	}
+	m := int(d.Minutes())
+	s := int(d.Seconds()) % 60
+	if s == 0 {
+		return fmt.Sprintf("%dm", m)
+	}
+	return fmt.Sprintf("%dm%02ds", m, s)
+}
+
 func FormatToolResult(content string, width int) string {
 	var formattedContent string
 	var m map[string]any
@@ -153,6 +194,8 @@ func RenderTool(msg *types.Message, inProgress spinner.Spinner, args, result str
 	icon := Icon(msg, inProgress)
 	name := nameStyle.Render(msg.ToolDefinition.DisplayName())
 
+	warning := LongRunningWarning(msg)
+
 	if header, ok := RenderFriendlyHeader(msg, inProgress); ok {
 		content := header
 		if args != "" {
@@ -173,6 +216,9 @@ func RenderTool(msg *types.Message, inProgress spinner.Spinner, args, result str
 				content += " " + renderedResult
 			}
 		}
+		if warning != "" {
+			content += "\n" + styles.WarningStyle.MarginLeft(styles.ToolCompletedIcon.GetMarginLeft()).Render(warning)
+		}
 		return styles.RenderComposite(styles.ToolMessageStyle.Width(width), content)
 	}
 
@@ -199,6 +245,9 @@ func RenderTool(msg *types.Message, inProgress spinner.Spinner, args, result str
 			content += " " + renderedResult
 		}
 	}
+	if warning != "" {
+		content += "\n" + styles.WarningStyle.MarginLeft(styles.ToolCompletedIcon.GetMarginLeft()).Render(warning)
+	}
 
 	return styles.RenderComposite(styles.ToolMessageStyle.Width(width), content)
 }

pkg/tui/components/toolcommon/common_test.go

@@ -2,9 +2,12 @@ package toolcommon
 
 import (
 	"testing"
+	"time"
 
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
+
+	"github.com/docker/docker-agent/pkg/tui/types"
 )
 
 func TestTryFixPartialJSON(t *testing.T) {
@@ -713,3 +716,57 @@ func BenchmarkRuneWidth(b *testing.B) {
 		}
 	})
 }
+
+func TestFormatDuration(t *testing.T) {
+	tests := []struct {
+		d    time.Duration
+		want string
+	}{
+		{0, "0s"},
+		{500 * time.Millisecond, "0s"},
+		{1 * time.Second, "1s"},
+		{45 * time.Second, "45s"},
+		{60 * time.Second, "1m"},
+		{90 * time.Second, "1m30s"},
+		{135 * time.Second, "2m15s"},
+		{5 * time.Minute, "5m"},
+	}
+	for _, tt := range tests {
+		t.Run(tt.want, func(t *testing.T) {
+			got := formatDuration(tt.d)
+			if got != tt.want {
+				t.Errorf("formatDuration(%v) = %q, want %q", tt.d, got, tt.want)
+			}
+		})
+	}
+}
+
+func TestLongRunningWarning(t *testing.T) {
+	t.Run("no StartedAt", func(t *testing.T) {
+		msg := &types.Message{ToolStatus: types.ToolStatusRunning}
+		if w := LongRunningWarning(msg); w != "" {
+			t.Errorf("expected empty warning, got %q", w)
+		}
+	})
+	t.Run("under threshold", func(t *testing.T) {
+		now := time.Now()
+		msg := &types.Message{ToolStatus: types.ToolStatusRunning, StartedAt: &now}
+		if w := LongRunningWarning(msg); w != "" {
+			t.Errorf("expected empty warning, got %q", w)
+		}
+	})
+	t.Run("over threshold", func(t *testing.T) {
+		past := time.Now().Add(-2 * time.Minute)
+		msg := &types.Message{ToolStatus: types.ToolStatusRunning, StartedAt: &past}
+		if w := LongRunningWarning(msg); w == "" {
+			t.Error("expected warning for long-running tool call")
+		}
+	})
+	t.Run("completed tool no warning", func(t *testing.T) {
+		past := time.Now().Add(-2 * time.Minute)
+		msg := &types.Message{ToolStatus: types.ToolStatusCompleted, StartedAt: &past}
+		if w := LongRunningWarning(msg); w != "" {
+			t.Errorf("expected no warning for completed tool, got %q", w)
+		}
+	})
+}

pkg/tui/types/types.go

@@ -2,6 +2,7 @@ package types
 
 import (
 	"strings"
+	"time"
 
 	"github.com/docker/docker-agent/pkg/tools"
 )
@@ -48,6 +49,9 @@ type Message struct {
 	ToolDefinition tools.Tool            // Definition of the tool being called
 	ToolStatus     ToolStatus            // Status for tool calls
 	ToolResult     *tools.ToolCallResult // Result of tool call (when completed)
+	// StartedAt records when a tool call entered ToolStatusRunning.
+	// Used to display elapsed time for long-running tool calls.
+	StartedAt *time.Time
 	// SessionPosition is the index of this message in session.Messages (when known).
 	// Used for operations like branching on edits.
 	SessionPosition *int
@@ -102,13 +106,18 @@ func Welcome(content string) *Message {
 }
 
 func ToolCallMessage(agentName string, toolCall tools.ToolCall, toolDef tools.Tool, status ToolStatus) *Message {
-	return &Message{
+	msg := &Message{
 		Type:           MessageTypeToolCall,
 		Sender:         agentName,
 		ToolCall:       toolCall,
 		ToolDefinition: toolDef,
 		ToolStatus:     status,
 	}
+	if status == ToolStatusRunning {
+		now := time.Now()
+		msg.StartedAt = &now
+	}
+	return msg
 }
 
 func Loading(description string) *Message {