feat: updated app for tool-use agent

app.py

@@ -5,33 +5,62 @@ import inspect
 import pandas as pd
 import json
 import copy
-from basic_agent import BasicOpenAIAgentWorkflow
+
+from basic_agent import ToolAgent
+from tools import (
+    search_and_extract,
+    load_youtube_transcript,
+    search_and_extract_from_wikipedia,
+    image_query_tool,
+)
 
 
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
+TOOL_USE_SYS_PROMPT = """
+You are a helpful AI assistant operating in a structured reasoning and action loop using the ReAct pattern.
 
-class BasicAgent:
-    def __init__(self):
-        self.agent = BasicOpenAIAgentWorkflow(
-            tools=[]  # search_web_extract_info
-            )
-        self.agent.create_basic_tool_use_agent_state_graph()
-        print("BasicAgent initialized.")
-    def __call__(self, question: str) -> str:
-        print(f"Agent received question (first 50 chars): {question[:50]}...")
-        answer = self.agent.chat(
-            question, 
-            verbose=1,
-            only_final_answer=True
-        )
-        print(f"Agent returning answer: {answer}")
-        return answer
+Your reasoning loop consists of:
+  - Question: the input task you must solve
+  - Thought: Reflect on the task and decide what to do next.
+  - Action: Choose one of the following actions:
+      - Solve it directly using your own knowledge
+      - Break the problem into smaller steps
+      - Use a tool to get more information
+  - Action Input: Provide input for the selected action
+  - Observation: Record the result of the action and/or aggregate information from previous observations (summarize, count, analyse, ...).
+  (Repeat Thought/Action/Action Input/Observation as needed)
+
+Terminate your loop with:
+  - Thought: I now know the final answer
+  - Final Answer: [your best answer to the original question]
+
+**General Execution Rules:**
+- If you can answer using only your trained knowledge, do so directly without using tools.
+- If the question involves image content, use the `image_query_tool`:
+    - Action: image_query_tool
+    - Action Input:  'image_path': [image_path], 'question': [user's question about the image]
+
+**Tool Use Constraints:**
+- Never use any tool more than **2 consecutive times** without either:
+    - Reasoning about the information received so far: aggregate and analyze the tool outputs to answer the question.
+    - If you need more information, use a different tool or break the problem down further, but do not return a final answer yet.
+- Do not exceed **3 total calls** to *search-type tools* per query (e.g. `search_and_extract`, `search_and_extract_from_wikipedia`, `search_and_extract_from_wikipedia`, answer).
+- Do not ask the user for additional clarification or input. Work with only what is already provided.
+
+**If you are unable to answer:**
+- If neither your knowledge nor tool outputs yield useful information, say:
+    > Final Answer: I could not find any useful information to answer your query.
+- If the question is unanswerable due to lack of input (e.g., missing attachment) or is fundamentally outside your scope, say:
+    > Final Answer: I don't have the ability to answer this query: [brief reason]
+
+Always aim to provide the **best and most complete** answer based on your trained knowledge and the tools available.
+"""
 
 
 def run_and_submit_all( profile: gr.OAuthProfile | None):
     """
-    Fetches all questions, runs the BasicAgent on them, submits all answers,
+    Fetches all questions, runs the ToolAgent on them, submits all answers,
     and displays the results.
     """
 
@@ -51,7 +80,16 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
     
     # 1. Instantiate Agent ( modify this part to create your agent)
     try:
-        agent = BasicAgent()
+        agent = ToolAgent(
+            tools=[
+                search_and_extract,
+                load_youtube_transcript,
+                search_and_extract_from_wikipedia,
+                image_query_tool,
+            ]
+            backstory=TOOL_USE_SYS_PROMPT
+            )
+        agent.agent.initialize()
         
     except Exception as e:
         print(f"Error instantiating agent: {e}")
@@ -107,7 +145,12 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
             continue
             
         try:
-            submitted_answer = agent(question_text)   # todo: send more data (files)
+            image_path = question_file_name if ".png" in question_file_name else ""
+            metadata = {}
+            if image_path:
+                metadata = {'image_path': image_path}
+            q_data {'query': question_text, 'metadata': metadata}
+            submitted_answer = agent(q_data)   # todo: send more data (files)
             answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
             results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})