feat(ai-proxy): strip dynamic cch field from billing header to enable caching (#3518)

plugins/wasm-go/extensions/ai-proxy/provider/claude_to_openai.go

@@ -145,7 +145,8 @@ func (c *ClaudeToOpenAIConverter) ConvertClaudeRequestToOpenAI(body []byte) ([]b
 	if claudeRequest.System != nil {
 		systemMsg := chatMessage{Role: roleSystem}
 		if !claudeRequest.System.IsArray {
-			systemMsg.Content = claudeRequest.System.StringValue
+			// Strip dynamic cch field from billing header to enable caching
+			systemMsg.Content = stripCchFromBillingHeader(claudeRequest.System.StringValue)
 		} else {
 			conversionResult := c.convertContentArray(claudeRequest.System.ArrayValue)
 			systemMsg.Content = conversionResult.openaiContents
@@ -832,10 +833,12 @@ func (c *ClaudeToOpenAIConverter) convertContentArray(claudeContents []claudeCha
 		switch claudeContent.Type {
 		case "text":
 			if claudeContent.Text != "" {
-				result.textParts = append(result.textParts, claudeContent.Text)
+				// Strip dynamic cch field from billing header to enable caching
+				processedText := stripCchFromBillingHeader(claudeContent.Text)
+				result.textParts = append(result.textParts, processedText)
 				result.openaiContents = append(result.openaiContents, chatMessageContent{
 					Type:         contentTypeText,
-					Text:         claudeContent.Text,
+					Text:         processedText,
 					CacheControl: claudeContent.CacheControl,
 				})
 			}
@@ -954,3 +957,42 @@ func (c *ClaudeToOpenAIConverter) startToolCall(toolState *toolCallInfo) []*clau
 
 	return responses
 }
+
+// stripCchFromBillingHeader removes the dynamic cch field from x-anthropic-billing-header text
+// to enable caching. The cch value changes on every request, which would break prompt caching.
+// Example input:  "x-anthropic-billing-header: cc_version=2.1.37.3a3; cc_entrypoint=claude-vscode; cch=abc123;"
+// Example output: "x-anthropic-billing-header: cc_version=2.1.37.3a3; cc_entrypoint=claude-vscode;"
+func stripCchFromBillingHeader(text string) string {
+	const billingHeaderPrefix = "x-anthropic-billing-header:"
+
+	// Check if this is a billing header
+	if !strings.HasPrefix(text, billingHeaderPrefix) {
+		return text
+	}
+
+	// Remove cch=xxx pattern (may appear with or without trailing semicolon)
+	// Pattern: ; cch=<any-non-semicolon-chars> followed by ; or end of string
+	result := text
+
+	// Try to find and remove ; cch=... pattern
+	// We need to handle both "; cch=xxx;" and "; cch=xxx" (at end)
+	for {
+		cchIdx := strings.Index(result, "; cch=")
+		if cchIdx == -1 {
+			break
+		}
+
+		// Find the end of cch value (next semicolon or end of string)
+		start := cchIdx + 2 // skip "; "
+		end := strings.Index(result[start:], ";")
+		if end == -1 {
+			// cch is at the end, remove from "; cch=" to end
+			result = result[:cchIdx]
+		} else {
+			// cch is followed by more content, remove "; cch=xxx" part
+			result = result[:cchIdx] + result[start+end:]
+		}
+	}
+
+	return result
+}

plugins/wasm-go/extensions/ai-proxy/provider/claude_to_openai_test.go

@@ -859,3 +859,125 @@ func TestClaudeToOpenAIConverter_ConvertReasoningResponseToClaude(t *testing.T)
 		})
 	}
 }
+
+func TestClaudeToOpenAIConverter_StripCchFromSystemMessage(t *testing.T) {
+	converter := &ClaudeToOpenAIConverter{}
+
+	t.Run("string_system_with_billing_header", func(t *testing.T) {
+		// Test that cch field is stripped from string format system message
+		claudeRequest := `{
+			"model": "claude-sonnet-4",
+			"max_tokens": 1024,
+			"system": [
+				{
+					"type": "text",
+					"text": "x-anthropic-billing-header: cc_version=2.1.37.3a3; cc_entrypoint=claude-vscode; cch=abc123;"
+				}
+			],
+			"messages": [{
+				"role": "user",
+				"content": "Hello"
+			}]
+		}`
+
+		result, err := converter.ConvertClaudeRequestToOpenAI([]byte(claudeRequest))
+		require.NoError(t, err)
+
+		var openaiRequest chatCompletionRequest
+		err = json.Unmarshal(result, &openaiRequest)
+		require.NoError(t, err)
+
+		require.Len(t, openaiRequest.Messages, 2)
+
+		// First message should be system with cch stripped
+		systemMsg := openaiRequest.Messages[0]
+		assert.Equal(t, "system", systemMsg.Role)
+
+		// The system content should have cch removed
+		contentArray, ok := systemMsg.Content.([]interface{})
+		require.True(t, ok, "System content should be an array")
+		require.Len(t, contentArray, 1)
+
+		contentMap, ok := contentArray[0].(map[string]interface{})
+		require.True(t, ok)
+		assert.Equal(t, "text", contentMap["type"])
+		assert.Equal(t, "x-anthropic-billing-header: cc_version=2.1.37.3a3; cc_entrypoint=claude-vscode;", contentMap["text"])
+		assert.NotContains(t, contentMap["text"], "cch=")
+	})
+
+	t.Run("plain_string_system_unchanged", func(t *testing.T) {
+		// Test that normal system messages are not modified
+		claudeRequest := `{
+			"model": "claude-sonnet-4",
+			"max_tokens": 1024,
+			"system": "You are a helpful assistant.",
+			"messages": [{
+				"role": "user",
+				"content": "Hello"
+			}]
+		}`
+
+		result, err := converter.ConvertClaudeRequestToOpenAI([]byte(claudeRequest))
+		require.NoError(t, err)
+
+		var openaiRequest chatCompletionRequest
+		err = json.Unmarshal(result, &openaiRequest)
+		require.NoError(t, err)
+
+		// First message should be system with original content
+		systemMsg := openaiRequest.Messages[0]
+		assert.Equal(t, "system", systemMsg.Role)
+		assert.Equal(t, "You are a helpful assistant.", systemMsg.Content)
+	})
+}
+
+func TestStripCchFromBillingHeader(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    string
+		expected string
+	}{
+		{
+			name:     "billing header with cch at end",
+			input:    "x-anthropic-billing-header: cc_version=2.1.37.3a3; cc_entrypoint=claude-vscode; cch=abc123;",
+			expected: "x-anthropic-billing-header: cc_version=2.1.37.3a3; cc_entrypoint=claude-vscode;",
+		},
+		{
+			name:     "billing header with cch at end without trailing semicolon",
+			input:    "x-anthropic-billing-header: cc_version=2.1.37.3a3; cc_entrypoint=claude-vscode; cch=abc123",
+			expected: "x-anthropic-billing-header: cc_version=2.1.37.3a3; cc_entrypoint=claude-vscode",
+		},
+		{
+			name:     "billing header with cch in middle",
+			input:    "x-anthropic-billing-header: cc_version=2.1.37.3a3; cch=abc123; cc_entrypoint=claude-vscode;",
+			expected: "x-anthropic-billing-header: cc_version=2.1.37.3a3; cc_entrypoint=claude-vscode;",
+		},
+		{
+			name:     "billing header without cch",
+			input:    "x-anthropic-billing-header: cc_version=2.1.37.3a3; cc_entrypoint=claude-vscode;",
+			expected: "x-anthropic-billing-header: cc_version=2.1.37.3a3; cc_entrypoint=claude-vscode;",
+		},
+		{
+			name:     "non-billing header text unchanged",
+			input:    "This is a normal system prompt",
+			expected: "This is a normal system prompt",
+		},
+		{
+			name:     "empty string unchanged",
+			input:    "",
+			expected: "",
+		},
+		{
+			name:     "billing header with multiple cch fields",
+			input:    "x-anthropic-billing-header: cc_version=2.1.37.3a3; cch=first; cc_entrypoint=claude-vscode; cch=second;",
+			expected: "x-anthropic-billing-header: cc_version=2.1.37.3a3; cc_entrypoint=claude-vscode;",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := stripCchFromBillingHeader(tt.input)
+			assert.Equal(t, tt.expected, result)
+		})
+	}
+}