added claude api dependency

requirements-space.txt

@@ -3,6 +3,7 @@
 # They are removed from this list as they are not available on PyPI
 
 beautifulsoup4>=4.14.2
+claude-agent-sdk>=0.1.35
 ddgs>=9.6.1
 feedparser>=6.0.12
 filetype>=1.2.0

scievo/agents/experiment_agent/exec_subagent/execute.py

@@ -282,24 +282,55 @@ def summary_node(agent_state: ExecAgentState) -> ExecAgentState:
     agent_state.add_message(summary_prompt)
 
     # Get summary from LLM
-    msg = ModelRegistry.completion(
-        LLM_NAME,
-        agent_state.patched_history,
-        system_prompt=(
-            Message(
-                role="system",
-                content=PROMPTS.experiment_exec.summary_system_prompt.render(),
-            )
-            .with_log(cond=constant.LOG_SYSTEM_PROMPT)
-            .content
-        ),
-        agent_sender=AGENT_NAME,
-        tools=None,  # No tools needed for summary
-    ).with_log()
+    try:
+        msg = ModelRegistry.completion(
+            LLM_NAME,
+            agent_state.patched_history,
+            system_prompt=(
+                Message(
+                    role="system",
+                    content=PROMPTS.experiment_exec.summary_system_prompt.render(),
+                )
+                .with_log(cond=constant.LOG_SYSTEM_PROMPT)
+                .content
+            ),
+            agent_sender=AGENT_NAME,
+            tools=None,  # No tools needed for summary
+        ).with_log()
 
-    # Store the summary text
-    agent_state.execution_summary = msg.content or ""
-    agent_state.add_message(msg)
+        # Store the summary text
+        agent_state.execution_summary = msg.content or ""
+        agent_state.add_message(msg)
+    except Exception as e:
+        logger.exception("Failed to generate execution summary")
+        # Create a fallback summary based on execution history
+        fallback_summary = (
+            "Execution completed. Unable to generate detailed summary due to API error."
+        )
+        if agent_state.history:
+            # Try to extract information from history
+            last_messages = agent_state.history[-5:]
+            execution_info = []
+            for msg in last_messages:
+                if msg.role == "tool" and msg.content:
+                    execution_info.append(f"Tool output: {msg.content[:200]}")
+                elif msg.role == "assistant" and msg.content:
+                    execution_info.append(f"Assistant: {msg.content[:200]}")
+            if execution_info:
+                fallback_summary = (
+                    "Execution completed. Summary generation failed, but execution appears to have run.\n\n"
+                    + "\n".join(execution_info[:3])
+                )
+
+        agent_state.execution_summary = fallback_summary
+        agent_state.add_message(
+            Message(
+                role="assistant",
+                content=fallback_summary,
+                agent_sender=AGENT_NAME,
+            ).with_log()
+        )
+        logger.warning(f"Using fallback summary due to error: {e}")
 
     # Parse JSON summary from the response
     try:

scievo/core/llms.py

@@ -238,7 +238,18 @@ class ModelRegistry:
 
             response = ll_completion(**params)
             if response.choices is None or len(response.choices) == 0:
-                raise ZeroChoiceError("No choices returned from completion API")
+                # Log detailed error information for debugging
+                logger.error(f"ZeroChoiceError for model '{name}': No choices returned")
+                logger.error(f"Model params: {model_params}")
+                logger.error(f"Response object: {response}")
+                if hasattr(response, "error"):
+                    logger.error(f"API error: {response.error}")
+                if hasattr(response, "model"):
+                    logger.error(f"Model used: {response.model}")
+                raise ZeroChoiceError(
+                    f"No choices returned from completion API for model '{name}'. "
+                    f"This may indicate an API key issue, rate limit, or model unavailability."
+                )
             msg: Message = Message.from_ll_message(response.choices[0].message)  # type: ignore
             usage: Usage = response.usage  # type: ignore
             msg.llm_sender = name