feat(orchestrator): Implement high-level task decomposition in Planner node

backend/src/ea_chatbot/graph/prompts/planner.py

@@ -1,41 +1,29 @@
 from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
 
-PLANNER_SYSTEM = """You are a Research Specialist and your job is to find answers and educate the user. 
-Provide factual information responding directly to the user's question. Include key details and context to ensure your response comprehensively answers their query.
+PLANNER_SYSTEM = """You are a Lead Orchestrator for an Election Analytics Chatbot.
+Your job is to decompose complex user queries into a high-level checklist of tasks.
+
+**Specialized Workers:**
+1. `data_analyst`: Handles SQL queries, Python data analysis, and plotting. Use this when the user needs numbers, trends, or charts from the internal database.
+2. `researcher`: Performs web searches for current news, facts, or external data not in the primary database.
+
+**Orchestration Strategy:**
+- Analyze the user's question and the available data summary.
+- Create a logical sequence of tasks (checklist) for these workers.
+- Be specific in the task description for the worker (e.g., "Find the total votes in Florida 2020").
+- If the query is ambiguous, the Orchestrator loop will later handle clarification, but for now, make the best plan possible.
 
 Today's Date is: {date}"""
 
 PLANNER_USER = """Conversation Summary: {summary}
 
-TASK:
+USER QUESTION:
 {question}
 
-AVAILABLE DATA SUMMARY (Use only if relevant to the task):
+AVAILABLE DATABASE SUMMARY:
 {database_description}
 
-First: Evaluate whether you have all necessary and requested information to provide a solution. 
-Use the dataset description above to determine what data and in what format you have available to you.
-You are able to search internet if the user asks for it, or you require any information that you can not derive from the given dataset or the instruction.
-
-Second: Incorporate any additional relevant context, reasoning, or details from previous interactions or internal chain-of-thought that may impact the solution.
-Ensure that all such information is fully included in your response rather than referring to previous answers indirectly.
-
-Third: Reflect on the problem and briefly describe it, while addressing the problem goal, inputs, outputs,
-rules, constraints, and other relevant details that appear in the problem description.
-
-Fourth: Based on the preceding steps, formulate your response as an algorithm, breaking the solution in up to eight simple concise yet descriptive, clear English steps. 
-You MUST Include all values or instructions as described in the above task, or retrieved using internet search!
-If fewer steps suffice, that's acceptable. If more are needed, please include them.
-Remember to explain steps rather than write code.
-
-This algorithm will be later converted to Python code.
-If a dataframe is required, assume it is named 'df' and is already defined/populated based on the data summary.
-
-There is a list variable called `plots` that you need to use to store any plots you generate. Do not use `plt.show()` as it will render the plot and cause an error.
-
-Output the algorithm as a YAML string. Always enclose the YAML string within ```yaml tags.
-
-**Note: Ensure that any necessary context from prior interactions is fully embedded in the plan. Do not use phrases like "refer to previous answer"; instead, provide complete details inline.**
+Decompose the question into a strategic checklist. For each task, specify which worker should handle it.
 
 {example_plan}"""