plan_react agent with function calling and sonnet llm

Author: ollmer

src/agentlab/agents/tapeagent/conf/agent/plan_react_fcall.yaml

@@ -0,0 +1,57 @@
+_target_: tapeagents.agent.Agent
+name : web_agent
+max_iterations: 2
+llms:
+  default: ${llm}
+templates:
+  system_prompt: |
+    You are an expert AI Agent trained to assist users with complex web tasks.
+    Your role is to understand user queries, perform actions and respond in a helpful and accurate manner.
+    Keep your replies concise and direct. Prioritize clarity and avoid over-elaboration.
+    Do not express emotions or opinions about user questions.
+  allowed_tools: |
+    You have access to the following tools:
+    {tools_description}
+  thought_format: |
+    Important! Respond with the plain text, do not include any JSON or code.
+    Do not output anything besides what I asked in this message.
+
+nodes:
+  - _target_: tapeagents.nodes.StandardNode
+    name: plan
+    system_prompt: ${agent.templates.system_prompt}
+    guidance: |
+      Write a concise multi-step plan explaining which steps should be performed to find the answer for the given task.
+      Be specific about how each step should be performed. Only describe the intended actions here, do not perform them yet.
+      Consider that next steps may depend on results of previous steps, so include conditional branching using "if" statements where needed.
+      Start with the title "Plan".
+      ${agent.templates.thought_format}
+    steps_prompt: ${agent.templates.allowed_tools}
+
+  - _target_: tapeagents.nodes.StandardNode
+    name: reflect
+    system_prompt: ${agent.templates.system_prompt}
+    guidance: |
+      Produce the reasoning with a bullet-point list of thoughts strictly following the rules:
+        1. Summarize the last observation and describe any webpage interactions/effects.
+        2. Evaluate action success, explain impact on task/plan, and describe any errors with solutions.
+        3. If the last action was not successful, ask yourself about the reasons for failure.
+        4. List next steps to accomplish current plan step and propose next immediate action.
+
+      Additional notes for web page observations:
+        - Accept cookie consents first
+        - Quote relevant observation parts verbatim
+        - Close popups before interacting
+        - If last action was not successful, check if the target element is visible, use scrolling if its not.
+      ${agent.templates.thought_format}
+    steps_prompt: ${agent.templates.allowed_tools}
+
+  - _target_: tapeagents.nodes.StandardNode
+    name: act
+    system_prompt: ${agent.templates.system_prompt}
+    guidance: Produce an function call that performs the proposed step, if the task is complete, produce the final step.
+    steps:
+      - tapeagents.core.FinalStep
+    use_known_actions: true
+    use_function_calls: true
+    next_node: reflect

src/agentlab/agents/tapeagent/conf/llm/sonnet.yaml

@@ -0,0 +1,6 @@
+_target_: tapeagents.llms.Claude
+model_name: claude-sonnet-4-5-20250929
+use_cache: false
+context_size: 128000
+parameters:
+  temperature: 0.1

src/agentlab/agents/tapeagent/conf/miniwob.yaml

@@ -1,6 +1,6 @@
 defaults:
-  - llm: gpt5-mini
-  - agent: plan_act
+  - llm: sonnet
+  - agent: plan_react_fcall
   - _self_
 
 name: miniwob_tapeagent