Merge pull request #2985 from chengzhichao-xydt/codex/context-show-summarize-threshold

fix(context): show both summarize and compress thresholds in /context

pkg/agent/agent_command.go

@@ -350,11 +350,13 @@ func (al *AgentLoop) buildCommandsRuntime(
 			}
 			history := agent.Sessions.GetHistory(opts.SessionKey)
 			return &commands.ContextStats{
-				UsedTokens:       usage.UsedTokens,
-				TotalTokens:      usage.TotalTokens,
-				CompressAtTokens: usage.CompressAtTokens,
-				UsedPercent:      usage.UsedPercent,
-				MessageCount:     len(history),
+				UsedTokens:        usage.UsedTokens,
+				TotalTokens:       usage.TotalTokens,
+				HistoryTokens:     usage.HistoryTokens,
+				CompressAtTokens:  usage.CompressAtTokens,
+				SummarizeAtTokens: usage.SummarizeAtTokens,
+				UsedPercent:       usage.UsedPercent,
+				MessageCount:      len(history),
 			}
 		}
 	}

pkg/agent/context_usage.go

@@ -61,6 +61,17 @@ func computeContextUsage(agent *AgentInstance, sessionKey string) *bus.ContextUs
 	// proactive trigger (msgTokens + toolTokens + maxTokens > contextWindow).
 	compressAt := effectiveWindow
 
+	// summarizeAt = soft summarization trigger: matches maybeSummarize's
+	// threshold (contextWindow * SummarizeTokenPercent / 100).
+	//
+	// The engine compares this against history-message tokens ONLY (not
+	// UsedTokens).  HistoryTokens is exposed alongside UsedTokens so the
+	// UI can show both values and avoid user confusion.
+	summarizeAt := contextWindow * agent.SummarizeTokenPercent / 100
+	if summarizeAt <= 0 {
+		summarizeAt = compressAt
+	}
+
 	usedPercent := 0
 	if compressAt > 0 {
 		usedPercent = usedTokens * 100 / compressAt
@@ -70,9 +81,11 @@ func computeContextUsage(agent *AgentInstance, sessionKey string) *bus.ContextUs
 	}
 
 	return &bus.ContextUsage{
-		UsedTokens:       usedTokens,
-		TotalTokens:      contextWindow,
-		CompressAtTokens: compressAt,
-		UsedPercent:      usedPercent,
+		UsedTokens:        usedTokens,
+		TotalTokens:       contextWindow,
+		HistoryTokens:     historyTokens,
+		CompressAtTokens:  compressAt,
+		SummarizeAtTokens: summarizeAt,
+		UsedPercent:       usedPercent,
 	}
 }

pkg/bus/types.go

@@ -64,10 +64,12 @@ type OutboundScope struct {
 // ContextUsage describes how much of the model's context window the current
 // session consumes, and how far it is from triggering compression.
 type ContextUsage struct {
-	UsedTokens       int `json:"used_tokens"`
-	TotalTokens      int `json:"total_tokens"`       // model context window
-	CompressAtTokens int `json:"compress_at_tokens"` // threshold that triggers compression
-	UsedPercent      int `json:"used_percent"`       // 0-100
+	UsedTokens        int `json:"used_tokens"`
+	TotalTokens       int `json:"total_tokens"`        // model context window
+	HistoryTokens     int `json:"history_tokens"`      // history-message tokens only (what maybeSummarize checks)
+	CompressAtTokens  int `json:"compress_at_tokens"`  // hard budget compression threshold (contextWindow - maxTokens)
+	SummarizeAtTokens int `json:"summarize_at_tokens"` // soft summarization trigger (vs history tokens)
+	UsedPercent       int `json:"used_percent"`        // 0-100, relative to compressAt
 }
 
 type OutboundMessage struct {

pkg/channels/pico/pico.go

@@ -1394,10 +1394,12 @@ func setContextUsagePayload(payload map[string]any, u *bus.ContextUsage) {
 		return
 	}
 	payload["context_usage"] = map[string]any{
-		"used_tokens":        u.UsedTokens,
-		"total_tokens":       u.TotalTokens,
-		"compress_at_tokens": u.CompressAtTokens,
-		"used_percent":       u.UsedPercent,
+		"used_tokens":         u.UsedTokens,
+		"total_tokens":        u.TotalTokens,
+		"history_tokens":      u.HistoryTokens,
+		"compress_at_tokens":  u.CompressAtTokens,
+		"summarize_at_tokens": u.SummarizeAtTokens,
+		"used_percent":        u.UsedPercent,
 	}
 }
 

pkg/channels/pico/pico_test.go

@@ -602,10 +602,12 @@ func TestBeginStream_FinalizeIncludesContextUsage(t *testing.T) {
 		t.Fatal("streamer should support FinalizeWithContext")
 	}
 	if err := contextStreamer.FinalizeWithContext(context.Background(), "final", &bus.ContextUsage{
-		UsedTokens:       10,
-		TotalTokens:      100,
-		CompressAtTokens: 80,
-		UsedPercent:      10,
+		UsedTokens:        10,
+		TotalTokens:       100,
+		HistoryTokens:     5,
+		CompressAtTokens:  80,
+		SummarizeAtTokens: 60,
+		UsedPercent:       10,
 	}); err != nil {
 		t.Fatalf("FinalizeWithContext() error = %v", err)
 	}
@@ -627,6 +629,12 @@ func TestBeginStream_FinalizeIncludesContextUsage(t *testing.T) {
 	if got := rawUsage["used_tokens"]; got != float64(10) {
 		t.Fatalf("used_tokens = %#v, want 10", got)
 	}
+	if got := rawUsage["history_tokens"]; got != float64(5) {
+		t.Fatalf("history_tokens = %#v, want 5", got)
+	}
+	if got := rawUsage["summarize_at_tokens"]; got != float64(60) {
+		t.Fatalf("summarize_at_tokens = %#v, want 60", got)
+	}
 }
 
 func TestCreateAndAddConnection_RespectsMaxConnectionsConcurrently(t *testing.T) {

pkg/commands/cmd_context.go

@@ -29,14 +29,17 @@ func formatContextStats(s *ContextStats) string {
 		remaining = 0
 	}
 	usedWindowPercent := s.UsedTokens * 100 / max(s.TotalTokens, 1)
-	return fmt.Sprintf(
-		"Context usage  \nMessages: %d  \nUsed: ~%d / %d tokens (%d%%)  \nCompress at: %d tokens  \nCompression progress: %d%%  \nRemaining: ~%d tokens",
+	msg := fmt.Sprintf(
+		"Context usage  \nMessages: %d  \nUsed: ~%d / %d tokens (%d%%)  \nHistory: ~%d tokens  \nCompress at: %d tokens  \nSummarize at: %d tokens  \nCompression progress: %d%%  \nRemaining: ~%d tokens",
 		s.MessageCount,
 		s.UsedTokens,
 		s.TotalTokens,
 		usedWindowPercent,
+		s.HistoryTokens,
 		s.CompressAtTokens,
+		s.SummarizeAtTokens,
 		s.UsedPercent,
 		remaining,
 	)
+	return msg
 }

pkg/commands/runtime.go

@@ -29,11 +29,13 @@ type MCPToolInfo struct {
 
 // ContextStats describes current session context window usage.
 type ContextStats struct {
-	UsedTokens       int
-	TotalTokens      int // model context window
-	CompressAtTokens int // compression threshold
-	UsedPercent      int // 0-100
-	MessageCount     int
+	UsedTokens        int
+	TotalTokens       int // model context window
+	HistoryTokens     int // history-only tokens (what maybeSummarize checks)
+	CompressAtTokens  int // hard budget compression threshold
+	SummarizeAtTokens int // soft summarization trigger
+	UsedPercent       int // 0-100
+	MessageCount      int
 }
 
 // StopResult describes the outcome of a stop request for the current session.