From c6d867cde7d87696feaf3073f097ca15b0145273 Mon Sep 17 00:00:00 2001
From: weichen <weichen@home.local>
Date: Fri, 2 Oct 2026 02:11:43 +0000
Subject: [PATCH] Let length continuations and Anthropic output run unbounded

Why: A fixed continuation cap and a 4096-token Anthropic fallback stopped long replies before the agent finished.
Evidence: go test ./runtime/ ./runtime/llm/
Impact: MaxLengthContinuations of 0 keeps the old cap of 3; a negative value removes it. Unset Anthropic max_tokens defaults to 128000.
Refs: feature/stream-display-overhaul
Co-authored-by: Cursor <cursoragent@cursor.com>
---
 runtime/llm/providers.go     |  4 +++-
 runtime/loop.go              | 13 +++++++++---
 runtime/loop_harness_test.go | 40 ++++++++++++++++++++++++++++++++++++
 3 files changed, 53 insertions(+), 4 deletions(-)

diff --git a/runtime/llm/providers.go b/runtime/llm/providers.go
index 0e8dc9b..5121175 100644
--- a/runtime/llm/providers.go
+++ b/runtime/llm/providers.go
@@ -258,8 +258,10 @@ func BuildBody(protocol Protocol, model string, messages []Message, tools []Tool
 			"max_tokens": opts.MaxTokens,
 			"stream":     true,
 		}
+		// Anthropic 要求 max_tokens。0 表示调用方不限制：用足够大的默认值，
+		// 避免 4096 把长回复截断。真正的上限由调用方 StreamOptions.MaxTokens 决定。
 		if body["max_tokens"] == 0 {
-			body["max_tokens"] = 4096
+			body["max_tokens"] = 128000
 		}
 		if sys != "" {
 			body["system"] = sys
diff --git a/runtime/loop.go b/runtime/loop.go
index 5c560a8..529dc70 100644
--- a/runtime/loop.go
+++ b/runtime/loop.go
@@ -84,9 +84,12 @@ type Config struct {
 	// RepeatGuard 为真时，同名且参数按键排序后相同的调用连续 3 次，注入一条提醒。
 	// 每个 run 最多 3 条，不中断循环。
 	RepeatGuard bool
-	// ContinueOnLength 为真时，FinishReason=length 且没有工具调用就续写，最多 3 次。
-	// 续写提示不写进 Session。第 4 次截断返回错误。
+	// ContinueOnLength 为真时，FinishReason=length 且没有工具调用就续写。
+	// 续写提示不写进 Session。次数由 MaxLengthContinuations 决定。
 	ContinueOnLength bool
+	// MaxLengthContinuations 是 length 续写的次数上限。
+	// 0 表示默认 3 次；负数表示不限制。
+	MaxLengthContinuations int
 }
 
 // StopInfo 是 Stop 钩子看到的本 run 进度。
@@ -219,7 +222,11 @@ func Run(ctx context.Context, cfg Config, user llm.Message) (Result, error) {
 		}
 		emit(cfg, Event{Type: "turn_end", Message: last})
 		if cfg.ContinueOnLength && end.FinishReason == "length" && len(last.ToolCalls) == 0 {
-			if lengthContinues >= maxLengthContinuations {
+			lengthCap := cfg.MaxLengthContinuations
+			if lengthCap == 0 {
+				lengthCap = maxLengthContinuations
+			}
+			if lengthCap > 0 && lengthContinues >= lengthCap {
 				errCap := errString("length continuation exceeded")
 				cfg.logger().Error("runtime length continuation exceeded", errCap)
 				return Result{Text: last.Content, Messages: messages, StopReason: llm.StopError, Usage: usage, Err: errCap}, nil
diff --git a/runtime/loop_harness_test.go b/runtime/loop_harness_test.go
index 6c32c2d..3683693 100644
--- a/runtime/loop_harness_test.go
+++ b/runtime/loop_harness_test.go
@@ -248,6 +248,46 @@ func TestLoopContinueOnLengthCaps(t *testing.T) {
 	}
 }
 
+func TestLoopContinueOnLengthUnlimited(t *testing.T) {
+	var n int
+	stream := func(ctx context.Context, model string, messages []llm.Message, decls []llm.ToolDecl, opts llm.StreamOptions) <-chan llm.Event {
+		n++
+		ch := make(chan llm.Event, 1)
+		finish := "length"
+		text := "part"
+		if n >= 5 {
+			finish = "stop"
+			text = "done"
+		}
+		ch <- llm.Event{
+			Kind:         llm.KindMessageEnd,
+			FinishReason: finish,
+			Message:      llm.Message{Role: llm.RoleAssistant, Content: text, StopReason: llm.StopEnd},
+		}
+		close(ch)
+		return ch
+	}
+	res, err := Run(context.Background(), Config{
+		Model:                  "fake",
+		MaxTurns:               8,
+		ContinueOnLength:       true,
+		MaxLengthContinuations: -1,
+		Stream:                 stream,
+	}, llm.Message{Role: llm.RoleUser, Content: "go"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if res.Err != nil {
+		t.Fatalf("err = %v", res.Err)
+	}
+	if n != 5 {
+		t.Fatalf("model calls = %d, want 5", n)
+	}
+	if res.Text != "done" {
+		t.Fatalf("text = %q", res.Text)
+	}
+}
+
 func TestLoopHistorySkipsGeneratedSystem(t *testing.T) {
 	var first string
 	_, err := Run(context.Background(), Config{
-- 
2.54.0