[perf] update webshop modular rollout prompt

Author: realtmxi

openmanus_rl/environments/prompts/webshop.py

@@ -18,58 +18,59 @@
 You are an expert agent operating in the WebShop e-commerce environment.
 Your task is: {task_description}
 Your current observation is: {current_observation}
-Your admissible actions of the current situation are: [
-{available_actions}
-].
+Your admissible actions of the current situation are: {available_actions}.
 
 Please begin by analyzing the situation and planning your approach:
 
 <plan>
-Analyze the current shopping situation and devise a plan to accomplish the task: 
-What are the key steps needed to complete this task (e.g., search with the right keywords, open a relevant product, compare options, select attributes, finalize)?
-How to advance our plan toward completing the shopping goal in immediate next step?
-Based on the current observation, what should be my immediate next step?
+Plan the next step:
+- Given what I've learned, what should I do next?
+- Please explain why this plan is helpful for the next action?
+- What do I expect this action to achieve?
 </plan>
 
-Finally, choose ONE admissible action for the current step and present it within <action> </action> tags.
+<action>
+Finally, choose ONE admissible action for the current step and choose it within {available_actions}.
+</action>
 """
 
 WEBSHOP_TEMPLATE = """
 You are an expert agent operating in the WebShop e-commerce environment.
-Your task is: {task_description}
+Your task is to: {task_description}
 Prior to this step, you have already taken {step_count} step(s). Below are the most recent {history_length} observations and the corresponding actions you took: {action_history}
 You are now at step {current_step} and your current observation is: {current_observation}
-Your admissible actions of the current situation are: [{available_actions}].
+Your admissible actions of the current situation are: {available_actions}.
 
 Now it's your turn to take an action.
 
-You should first recall relevant past experience and reason from our conversation history, then MUST summarize within <memory_analysis> </memory_analysis> tags like this:
+You should first recall relevant past experience and reason from our conversation history, then MUST summarize within <memory> </memory> tags like this:
 
-<memory_analysis>
-[Recall relevant past experiences and reason from our conversation history]
-- Please summarize the most relavent memory for this step.
-- Please explain why this memory is helpful for the next reflection and planning.
-</memory_analysis>
+<memory>
+Look at the past observations and actions from our conversation history.
+- Please retrieve the most relavent memory for this step including the relevant observation and action in a RAG style along with the step number.
+- These memory should be helpful milestones to solve this task.
+</memory>
 
 After that, you should reflect on the last action and its outcome, then MUST summarize within <reflection> </reflection> tags like this:
 
 <reflection>
-[Reflect on the last action and its outcome]
-- What did my last action accomplish?
-- Was it successful or did it encounter issues?
-- How does this outcome affect my plan?
+Reflect on the last action and its outcome
+- Did I complete the task goal?
+- Was last action successful or did it encounter issues?
 - Am I making progress toward the task goal?
+- If the action did not go as expected and did not result in progress, provide constructive feedback to guide the next planning step.
 </reflection>
 
 After that, you should plan the next step based on memory and reflection, then MUST summarize within <plan> </plan> tags like this:
 
 <plan>
-[Plan the next step based on memory and reflection]
+Plan the next step based on memory and reflection
 - Given what I've learned, what should I do next?
 - Please explain why this plan is helpful for the next action?
-- How does this action fit into my overall shopping strategy?
-- What do I expect this action to achieve now?
+- What do I expect this action to achieve?
 </plan>
 
-Finally, choose ONE admissible action for the current step and present it within <action> </action> tags.
+<action>
+Finally, choose ONE admissible action for the current step and choose it within {available_actions}.
+</action>
 """