fix: add image response support

Convert Google's inlineData format to Anthropic's image format:
- response-converter.js: Handle inlineData in non-streaming responses
- sse-parser.js: Parse inlineData for thinking models
- sse-streamer.js: Stream inlineData as image content blocks

Co-Authored-By: Claude <noreply@anthropic.com>

src/cloudcode/sse-parser.js

@@ -89,6 +89,11 @@ export async function parseThinkingSSEResponse(response, originalModel) {
                         if (!part.text) continue;
                         flushThinking();
                         accumulatedText += part.text;
+                    } else if (part.inlineData) {
+                        // Handle image content
+                        flushThinking();
+                        flushText();
+                        finalParts.push(part);
                     }
                 }
             } catch (e) {
@@ -105,7 +110,7 @@ export async function parseThinkingSSEResponse(response, originalModel) {
         usageMetadata
     };
 
-    const partTypes = finalParts.map(p => p.thought ? 'thought' : (p.functionCall ? 'functionCall' : 'text'));
+    const partTypes = finalParts.map(p => p.thought ? 'thought' : (p.functionCall ? 'functionCall' : (p.inlineData ? 'inlineData' : 'text')));
     logger.debug('[CloudCode] Response received (SSE), part types:', partTypes);
     if (finalParts.some(p => p.thought)) {
         const thinkingPart = finalParts.find(p => p.thought);

src/cloudcode/sse-streamer.js

@@ -209,6 +209,39 @@ export async function* streamSSEResponse(response, originalModel) {
                                 partial_json: JSON.stringify(part.functionCall.args || {})
                             }
                         };
+                    } else if (part.inlineData) {
+                        // Handle image content from Google format
+                        if (currentBlockType === 'thinking' && currentThinkingSignature) {
+                            yield {
+                                type: 'content_block_delta',
+                                index: blockIndex,
+                                delta: { type: 'signature_delta', signature: currentThinkingSignature }
+                            };
+                            currentThinkingSignature = '';
+                        }
+                        if (currentBlockType !== null) {
+                            yield { type: 'content_block_stop', index: blockIndex };
+                            blockIndex++;
+                        }
+                        currentBlockType = 'image';
+
+                        // Emit image block as a complete block
+                        yield {
+                            type: 'content_block_start',
+                            index: blockIndex,
+                            content_block: {
+                                type: 'image',
+                                source: {
+                                    type: 'base64',
+                                    media_type: part.inlineData.mimeType,
+                                    data: part.inlineData.data
+                                }
+                            }
+                        };
+
+                        yield { type: 'content_block_stop', index: blockIndex };
+                        blockIndex++;
+                        currentBlockType = null;
                     }
                 }
 

src/format/response-converter.js

@@ -71,6 +71,16 @@ export function convertGoogleToAnthropic(googleResponse, model) {
 
             anthropicContent.push(toolUseBlock);
             hasToolCalls = true;
+        } else if (part.inlineData) {
+            // Handle image content from Google format
+            anthropicContent.push({
+                type: 'image',
+                source: {
+                    type: 'base64',
+                    media_type: part.inlineData.mimeType,
+                    data: part.inlineData.data
+                }
+            });
         }
     }