From 799b15bf8ba387586e2bdf73b15fa0d76649a15e Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Ernesto=20Mart=C3=ADnez?= <emagodev@gmail.com>
Date: Mon, 8 Dec 2025 00:46:03 +0100
Subject: [PATCH] fix: Ensure  event is emitted before any  events in Claude
 SSE responses.

---
 .../openai/claude/openai_claude_response.go   | 43 ++++++++++++++++---
 1 file changed, 36 insertions(+), 7 deletions(-)

diff --git a/pkg/llmproxy/translator/openai/claude/openai_claude_response.go b/pkg/llmproxy/translator/openai/claude/openai_claude_response.go
index 4f0c8d983e..e1f78fbc27 100644
--- a/pkg/llmproxy/translator/openai/claude/openai_claude_response.go
+++ b/pkg/llmproxy/translator/openai/claude/openai_claude_response.go
@@ -8,6 +8,7 @@ package claude
 import (
 	"bytes"
 	"context"
+	"encoding/json"
 	"fmt"
 	"strings"
 
@@ -132,16 +133,40 @@ func convertOpenAIStreamingChunkToAnthropic(rawJSON []byte, param *ConvertOpenAI
 		param.CreatedAt = root.Get("created").Int()
 	}
 
-	// Emit message_start on the very first chunk, regardless of whether it has a role field.
-	// Some providers (like Copilot) may send tool_calls in the first chunk without a role field.
+	// Helper to ensure message_start is sent before any content_block_start
+	// This is required by the Anthropic SSE protocol - message_start must come first.
+	// Some OpenAI-compatible providers (like GitHub Copilot) may not send role: "assistant"
+	// in the first chunk, so we need to emit message_start when we first see content.
+	ensureMessageStarted := func() {
+		if param.MessageStarted {
+			return
+		}
+		messageStart := map[string]interface{}{
+			"type": "message_start",
+			"message": map[string]interface{}{
+				"id":            param.MessageID,
+				"type":          "message",
+				"role":          "assistant",
+				"model":         param.Model,
+				"content":       []interface{}{},
+				"stop_reason":   nil,
+				"stop_sequence": nil,
+				"usage": map[string]interface{}{
+					"input_tokens":  0,
+					"output_tokens": 0,
+				},
+			},
+		}
+		messageStartJSON, _ := json.Marshal(messageStart)
+		results = append(results, "event: message_start\ndata: "+string(messageStartJSON)+"\n\n")
+		param.MessageStarted = true
+	}
+
+	// Check if this is the first chunk (has role)
 	if delta := root.Get("choices.0.delta"); delta.Exists() {
 		if !param.MessageStarted {
 			// Send message_start event
-			messageStartJSON := `{"type":"message_start","message":{"id":"","type":"message","role":"assistant","model":"","content":[],"stop_reason":null,"stop_sequence":null,"usage":{"input_tokens":0,"output_tokens":0}}}`
-			messageStartJSON, _ = sjson.Set(messageStartJSON, "message.id", param.MessageID)
-			messageStartJSON, _ = sjson.Set(messageStartJSON, "message.model", param.Model)
-			results = append(results, "event: message_start\ndata: "+messageStartJSON+"\n\n")
-			param.MessageStarted = true
+			ensureMessageStarted()
 
 			// Don't send content_block_start for text here - wait for actual content
 		}
@@ -154,6 +179,7 @@ func convertOpenAIStreamingChunkToAnthropic(rawJSON []byte, param *ConvertOpenAI
 				}
 				stopTextContentBlock(param, &results)
 				if !param.ThinkingContentBlockStarted {
+					ensureMessageStarted() // Must send message_start before content_block_start
 					if param.ThinkingContentBlockIndex == -1 {
 						param.ThinkingContentBlockIndex = param.NextContentBlockIndex
 						param.NextContentBlockIndex++
@@ -175,6 +201,7 @@ func convertOpenAIStreamingChunkToAnthropic(rawJSON []byte, param *ConvertOpenAI
 		if content := delta.Get("content"); content.Exists() && content.String() != "" {
 			// Send content_block_start for text if not already sent
 			if !param.TextContentBlockStarted {
+				ensureMessageStarted() // Must send message_start before content_block_start
 				stopThinkingContentBlock(param, &results)
 				if param.TextContentBlockIndex == -1 {
 					param.TextContentBlockIndex = param.NextContentBlockIndex
@@ -222,6 +249,8 @@ func convertOpenAIStreamingChunkToAnthropic(rawJSON []byte, param *ConvertOpenAI
 					if name := function.Get("name"); name.Exists() {
 						accumulator.Name = name.String()
 
+						ensureMessageStarted() // Must send message_start before content_block_start
+
 						stopThinkingContentBlock(param, &results)
 
 						stopTextContentBlock(param, &results)