use anthropic beta client (#2235)

skyvern/forge/agent.py

@@ -1433,7 +1433,6 @@ class ForgeAgent:
                 raw_response=True,
                 betas=["computer-use-2025-01-24"],
             )
-        LOG.info("Anthropic response", llm_response=llm_response)
         assistant_content = llm_response["content"]
         llm_caller.message_history.append({"role": "assistant", "content": assistant_content})
 

skyvern/forge/sdk/api/llm/api_handler_factory.py

@@ -6,7 +6,7 @@ from typing import Any
 
 import litellm
 import structlog
-from anthropic.types.message import Message as AnthropicMessage
+from anthropic.types.beta.beta_message import BetaMessage as AnthropicMessage
 from jinja2 import Template
 from litellm.utils import CustomStreamWrapper, ModelResponse
 
@@ -678,7 +678,7 @@ class LLMCaller:
     ) -> AnthropicMessage:
         max_tokens = active_parameters.get("max_completion_tokens") or active_parameters.get("max_tokens") or 4096
         model_name = self.llm_config.model_name.replace("bedrock/", "").replace("anthropic/", "")
-        return await app.ANTHROPIC_CLIENT.messages.create(
+        response = await app.ANTHROPIC_CLIENT.beta.messages.create(
             max_tokens=max_tokens,
             messages=messages,
             model=model_name,
@@ -686,6 +686,8 @@ class LLMCaller:
             timeout=timeout,
             betas=active_parameters.get("betas", None),
         )
+        LOG.info("Anthropic response", response=response)
+        return response
 
 
 class LLMCallerManager:

skyvern/webeye/actions/parse_actions.py

@@ -482,7 +482,6 @@ async def parse_anthropic_actions(
                     tool_call_id=tool_call_id,
                 )
             )
-            idx += 1
         elif action == "left_click":
             if idx - 1 >= 0:
                 prev_tool_call = tool_calls[idx - 1]
@@ -494,12 +493,12 @@ async def parse_anthropic_actions(
             else:
                 coordinate = parsed_args.get("coordinate")
 
-            idx += 1
             if not coordinate:
                 LOG.warning(
                     "Left click action has no coordinate and it doesn't have mouse_move before it",
                     tool_call=tool_call,
                 )
+                idx += 1
                 continue
             x, y = coordinate
             actions.append(
@@ -519,12 +518,12 @@ async def parse_anthropic_actions(
             )
         elif action == "type":
             text = parsed_args.get("text")
-            idx += 1
             if not text:
                 LOG.warning(
                     "Type action has no text",
                     tool_call=tool_call,
                 )
+                idx += 1
                 continue
             actions.append(
                 InputTextAction(
@@ -541,12 +540,12 @@ async def parse_anthropic_actions(
             )
         elif action == "key":
             text = parsed_args.get("text")
-            idx += 1
             if not text:
                 LOG.warning(
                     "Key action has no text",
                     tool_call=tool_call,
                 )
+                idx += 1
                 continue
             actions.append(
                 KeypressAction(
@@ -561,12 +560,24 @@ async def parse_anthropic_actions(
                     tool_call_id=tool_call_id,
                 )
             )
+        elif action == "screenshot":
+            actions.append(
+                NullAction(
+                    organization_id=task.organization_id,
+                    workflow_run_id=task.workflow_run_id,
+                    task_id=task.task_id,
+                    step_id=step.step_id,
+                    step_order=step.order,
+                    action_order=idx,
+                    tool_call_id=tool_call_id,
+                )
+            )
         else:
             LOG.error(
                 "Unsupported action",
                 tool_call=tool_call,
             )
-            idx += 1
+        idx += 1
     return actions