feat: replace claude sdk

agentscope-core/pom.xml

@@ -113,12 +113,7 @@
             <artifactId>google-genai</artifactId>
         </dependency>
 
-        <!-- Anthropic Java SDK -->
-        <dependency>
-            <groupId>com.anthropic</groupId>
-            <artifactId>anthropic-java</artifactId>
-        </dependency>
-
+
         <!-- Model Context Protocol (MCP) SDK -->
         <dependency>
             <groupId>io.modelcontextprotocol.sdk</groupId>

agentscope-core/src/main/java/io/agentscope/core/formatter/anthropic/AnthropicBaseFormatter.java

@@ -15,92 +15,105 @@
  */
 package io.agentscope.core.formatter.anthropic;
 
-import com.anthropic.models.messages.MessageCreateParams;
-import com.anthropic.models.messages.MessageParam;
 import io.agentscope.core.formatter.AbstractBaseFormatter;
+import io.agentscope.core.formatter.anthropic.dto.AnthropicMessage;
+import io.agentscope.core.formatter.anthropic.dto.AnthropicRequest;
+import io.agentscope.core.formatter.anthropic.dto.AnthropicResponse;
 import io.agentscope.core.message.Msg;
 import io.agentscope.core.model.GenerateOptions;
 import io.agentscope.core.model.ToolSchema;
 import java.util.List;
 
 /**
- * Abstract base formatter for Anthropic API with shared logic for handling Anthropic-specific
+ * Abstract base formatter for Anthropic API with shared logic for handling
+ * Anthropic-specific
  * requirements.
  *
- * <p>This class handles:
+ * <p>
+ * This class handles:
  *
  * <ul>
- *   <li>System message extraction and application (Anthropic requires system via system parameter)
- *   <li>Tool choice configuration with GenerateOptions
+ * <li>System message extraction and application (Anthropic requires system via
+ * system parameter)
+ * <li>Tool choice configuration with GenerateOptions
  * </ul>
  */
 public abstract class AnthropicBaseFormatter
-        extends AbstractBaseFormatter<MessageParam, Object, MessageCreateParams.Builder> {
+        extends AbstractBaseFormatter<AnthropicMessage, AnthropicResponse, AnthropicRequest> {
 
     protected final AnthropicMessageConverter messageConverter;
 
-    /** Thread-local storage for generation options (passed from applyOptions to applyTools). */
+    /**
+     * Thread-local storage for generation options (passed from applyOptions to
+     * applyTools).
+     */
     private final ThreadLocal<GenerateOptions> currentOptions = new ThreadLocal<>();
 
     protected AnthropicBaseFormatter() {
         this.messageConverter = new AnthropicMessageConverter(this::convertToolResultToString);
     }
 
+    protected AnthropicBaseFormatter(AnthropicMessageConverter messageConverter) {
+        this.messageConverter = messageConverter;
+    }
+
     /**
      * Apply generation options to Anthropic request parameters.
      *
-     * @param paramsBuilder Anthropic request parameters builder
-     * @param options Generation options to apply
+     * @param request        Anthropic request
+     * @param options        Generation options to apply
      * @param defaultOptions Default options to use if options parameter is null
      */
     @Override
     public void applyOptions(
-            MessageCreateParams.Builder paramsBuilder,
-            GenerateOptions options,
-            GenerateOptions defaultOptions) {
+            AnthropicRequest request, GenerateOptions options, GenerateOptions defaultOptions) {
         // Save options for applyTools
         currentOptions.set(options);
 
         // Apply other options
-        AnthropicToolsHelper.applyOptions(paramsBuilder, options, defaultOptions);
+        AnthropicToolsHelper.applyOptions(request, options, defaultOptions);
     }
 
     /**
-     * Apply tool schemas to Anthropic request parameters. This method uses the options saved from
+     * Apply tool schemas to Anthropic request parameters. This method uses the
+     * options saved from
      * applyOptions to apply tool choice configuration.
      *
-     * @param paramsBuilder Anthropic request parameters builder
-     * @param tools List of tool schemas to apply (may be null or empty)
+     * @param request Anthropic request
+     * @param tools   List of tool schemas to apply (may be null or empty)
      */
     @Override
-    public void applyTools(MessageCreateParams.Builder paramsBuilder, List<ToolSchema> tools) {
+    public void applyTools(AnthropicRequest request, List<ToolSchema> tools) {
         if (tools == null || tools.isEmpty()) {
             currentOptions.remove();
             return;
         }
 
         // Use saved options to apply tools with tool choice
         GenerateOptions options = currentOptions.get();
-        AnthropicToolsHelper.applyTools(paramsBuilder, tools, options);
+        AnthropicToolsHelper.applyTools(request, tools, options);
 
         // Clean up thread-local storage
         currentOptions.remove();
     }
 
     /**
-     * Extract and apply system message if present. Anthropic API requires system message to be set
+     * Extract and apply system message if present. Anthropic API requires system
+     * message to be set
      * via the system parameter, not as a message.
      *
-     * <p>This method is called by Model to extract the first system message from the messages list
+     * <p>
+     * This method is called by Model to extract the first system message from the
+     * messages list
      * and apply it to the system parameter.
      *
-     * @param paramsBuilder Anthropic request parameters builder
+     * @param request  Anthropic request
      * @param messages All messages including potential system message
      */
-    public void applySystemMessage(MessageCreateParams.Builder paramsBuilder, List<Msg> messages) {
+    public void applySystemMessage(AnthropicRequest request, List<Msg> messages) {
         String systemMessage = messageConverter.extractSystemMessage(messages);
         if (systemMessage != null && !systemMessage.isEmpty()) {
-            paramsBuilder.system(systemMessage);
+            request.setSystem(systemMessage);
         }
     }
 }

agentscope-core/src/main/java/io/agentscope/core/formatter/anthropic/AnthropicChatFormatter.java

@@ -15,38 +15,238 @@
  */
 package io.agentscope.core.formatter.anthropic;
 
-import com.anthropic.models.messages.Message;
-import com.anthropic.models.messages.MessageParam;
+import io.agentscope.core.formatter.anthropic.dto.AnthropicContent;
+import io.agentscope.core.formatter.anthropic.dto.AnthropicMessage;
+import io.agentscope.core.formatter.anthropic.dto.AnthropicResponse;
+import io.agentscope.core.message.ImageBlock;
 import io.agentscope.core.message.Msg;
+import io.agentscope.core.message.MsgRole;
+import io.agentscope.core.message.ToolResultBlock;
+import io.agentscope.core.message.ToolUseBlock;
 import io.agentscope.core.model.ChatResponse;
 import java.time.Instant;
+import java.util.ArrayList;
+import java.util.HashSet;
 import java.util.List;
+import java.util.Set;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
 
 /**
- * Formatter for Anthropic Messages API. Converts between AgentScope Msg objects and Anthropic SDK
- * types.
+ * Formatter for Anthropic Messages API. Converts between AgentScope Msg objects
+ * and Anthropic DTO types.
  *
  * <p>Important: Anthropic API has special requirements:
- *
  * <ul>
  *   <li>Only the first message can be a system message (handled via system parameter)
  *   <li>Tool results must be in separate user messages
  *   <li>Supports thinking blocks natively (extended thinking feature)
+ *   <li>Automatic multi-agent conversation handling for MsgHub scenarios
  * </ul>
+ *
+ * <p><b>Multi-Agent Detection:</b> This formatter automatically detects multi-agent scenarios
+ * (e.g., MsgHub conversations) by checking for multiple messages with different names but
+ * the same role. When detected, it uses {@link AnthropicConversationMerger} to consolidate the
+ * conversation into a format compatible with Anthropic's API.
  */
 public class AnthropicChatFormatter extends AnthropicBaseFormatter {
 
+    private static final Logger log = LoggerFactory.getLogger(AnthropicChatFormatter.class);
+
+    private static final String DEFAULT_CONVERSATION_HISTORY_PROMPT =
+            "# Conversation History\n"
+                    + "The content between <history></history> tags contains your conversation"
+                    + " history\n";
+
+    private final AnthropicMediaConverter mediaConverter;
+
+    public AnthropicChatFormatter() {
+        super();
+        this.mediaConverter = new AnthropicMediaConverter();
+    }
+
+    public AnthropicChatFormatter(AnthropicMessageConverter messageConverter) {
+        super(messageConverter);
+        this.mediaConverter = new AnthropicMediaConverter();
+    }
+
     @Override
-    public List<MessageParam> doFormat(List<Msg> msgs) {
+    public List<AnthropicMessage> doFormat(List<Msg> msgs) {
+        // Detect multi-agent scenario (multiple messages with different names but same role)
+        boolean isMultiAgent = isMultiAgentConversation(msgs);
+        log.debug(
+                "doFormat: message count={}, isMultiAgent={}",
+                msgs != null ? msgs.size() : 0,
+                isMultiAgent);
+
+        if (isMultiAgent) {
+            log.info("Detected multi-agent conversation, using conversation merger");
+            return formatMultiAgentConversation(msgs);
+        }
+
+        // Single-agent or simple conversation - use standard formatting
+        log.debug("Using standard formatting for single-agent conversation");
         return messageConverter.convert(msgs);
     }
 
-    @Override
-    public ChatResponse parseResponse(Object response, Instant startTime) {
-        if (response instanceof Message message) {
-            return AnthropicResponseParser.parseMessage(message, startTime);
-        } else {
-            throw new IllegalArgumentException("Unsupported response type: " + response.getClass());
+    /**
+     * Detects if the message list represents a multi-agent conversation.
+     *
+     * <p>A multi-agent conversation is detected when:
+     * <ul>
+     *   <li>There are at least 2 ASSISTANT role messages with different names</li>
+     *   <li>OR there are multiple ASSISTANT messages that would create consecutive
+     *       messages with the same role</li>
+     * </ul>
+     *
+     * @param msgs List of messages to check
+     * @return true if this appears to be a multi-agent conversation
+     */
+    private boolean isMultiAgentConversation(List<Msg> msgs) {
+        if (msgs == null || msgs.size() < 2) {
+            log.debug(
+                    "isMultiAgentConversation: too few messages (count={})",
+                    msgs != null ? msgs.size() : 0);
+            return false;
+        }
+
+        Set<String> assistantNames = new HashSet<>();
+        MsgRole lastRole = null;
+        boolean hasConsecutiveAssistant = false;
+        boolean hasSystemNamedUserMessage = false;
+
+        for (int i = 0; i < msgs.size(); i++) {
+            Msg msg = msgs.get(i);
+            MsgRole currentRole = msg.getRole();
+            String msgName = msg.getName();
+
+            log.trace("Message {}: role={}, name={}", i, currentRole, msgName);
+
+            // Check for consecutive ASSISTANT messages (without tool calls in between)
+            if (lastRole == MsgRole.ASSISTANT && currentRole == MsgRole.ASSISTANT) {
+                hasConsecutiveAssistant = true;
+                log.debug("Found consecutive ASSISTANT messages at index {}", i);
+            }
+
+            // Check if USER message has name="system" (indicates MsgHub announcement)
+            if (currentRole == MsgRole.USER && "system".equals(msgName)) {
+                hasSystemNamedUserMessage = true;
+                log.debug("Found USER message with name='system' (MsgHub announcement)");
+            }
+
+            // Collect ASSISTANT message names
+            if (currentRole == MsgRole.ASSISTANT && msgName != null) {
+                assistantNames.add(msgName);
+            }
+
+            lastRole = currentRole;
         }
+
+        // Multi-agent if:
+        // 1. Multiple assistant names (different agents), OR
+        // 2. Consecutive assistant messages, OR
+        // 3. System-named USER message (MsgHub announcement)
+        boolean result =
+                assistantNames.size() > 1 || hasConsecutiveAssistant || hasSystemNamedUserMessage;
+        log.debug(
+                "isMultiAgentConversation: assistantNames={}, hasConsecutive={}, "
+                        + "hasSystemNamedUserMessage={}, result={}",
+                assistantNames,
+                hasConsecutiveAssistant,
+                hasSystemNamedUserMessage,
+                result);
+        return result;
+    }
+
+    /**
+     * Formats a multi-agent conversation using the conversation merger.
+     * This consolidates multiple agent messages into a format compatible with Anthropic's API.
+     *
+     * @param msgs List of messages in the multi-agent conversation
+     * @return List of Anthropic-formatted messages
+     */
+    private List<AnthropicMessage> formatMultiAgentConversation(List<Msg> msgs) {
+        log.debug("formatMultiAgentConversation: processing {} messages", msgs.size());
+
+        // Separate messages into groups: SYSTEM, TOOL_SEQUENCE, AGENT_CONVERSATION
+        List<AnthropicMessage> result = new ArrayList<>();
+        List<Msg> systemMsgs = new ArrayList<>();
+        List<Msg> toolSequence = new ArrayList<>();
+        List<Msg> agentConversation = new ArrayList<>();
+
+        for (Msg msg : msgs) {
+            MsgRole role = msg.getRole();
+
+            if (role == MsgRole.SYSTEM) {
+                systemMsgs.add(msg);
+            } else if (msg.hasContentBlocks(ToolUseBlock.class)
+                    || msg.hasContentBlocks(ToolResultBlock.class)) {
+                // Tool-related messages: use standard converter
+                toolSequence.add(msg);
+            } else {
+                // Regular conversation messages (USER, ASSISTANT without tools)
+                agentConversation.add(msg);
+            }
+        }
+
+        // Add system messages using standard converter
+        for (Msg sysMsg : systemMsgs) {
+            List<AnthropicMessage> converted = messageConverter.convert(List.of(sysMsg));
+            result.addAll(converted);
+        }
+
+        // Add tool sequence using standard converter
+        if (!toolSequence.isEmpty()) {
+            List<AnthropicMessage> converted = messageConverter.convert(toolSequence);
+            result.addAll(converted);
+            log.debug("Added {} tool messages using standard converter", toolSequence.size());
+        }
+
+        // Merge agent conversation into a single user message
+        if (!agentConversation.isEmpty()) {
+            log.debug("Merging {} agent conversation messages", agentConversation.size());
+
+            List<Object> mergedContent =
+                    AnthropicConversationMerger.mergeConversation(
+                            agentConversation, DEFAULT_CONVERSATION_HISTORY_PROMPT);
+
+            List<AnthropicContent> contentBlocks = new ArrayList<>();
+
+            for (Object item : mergedContent) {
+                if (item instanceof String text) {
+                    contentBlocks.add(AnthropicContent.text(text));
+                    log.trace("Added text content block (length: {})", text.length());
+                } else if (item instanceof ImageBlock ib) {
+                    try {
+                        AnthropicContent.ImageSource imageSource =
+                                mediaConverter.convertImageBlock(ib);
+                        contentBlocks.add(
+                                AnthropicContent.image(
+                                        imageSource.getMediaType(), imageSource.getData()));
+                        log.trace("Added image content block");
+                    } catch (Exception e) {
+                        log.warn("Failed to convert image block: {}", e.getMessage());
+                        contentBlocks.add(AnthropicContent.text("[Image - conversion failed]"));
+                    }
+                }
+            }
+
+            if (!contentBlocks.isEmpty()) {
+                AnthropicMessage mergedMessage = new AnthropicMessage("user", contentBlocks);
+                result.add(mergedMessage);
+                log.debug(
+                        "Created merged user message with {} content blocks", contentBlocks.size());
+            } else {
+                log.warn("No content blocks created from merged agent conversation");
+            }
+        }
+
+        log.debug("formatMultiAgentConversation: returning {} messages", result.size());
+        return result;
+    }
+
+    @Override
+    public ChatResponse parseResponse(AnthropicResponse response, Instant startTime) {
+        return AnthropicResponseParser.parseMessage(response, startTime);
     }
 }