Add tools/ folder, fix agent answer format and Wikipedia proxy

- Add tools/ with web_search (Serper), wikipedia_search, python_repl, file_handler
- Fix Wikipedia 403 by adding User-Agent header
- Fix answer format: enforce ENTIRE response = only final answer
- Fix list output: comma-separated, no Python brackets
- Add [Task ID] prefix to questions so agent uses correct task_id
- Increase max iterations from 10 to 15
- Add .gitignore to protect .env, data/, __pycache__

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

.gitignore

@@ -0,0 +1,6 @@
+.env
+__pycache__/
+*.pyc
+*.pyo
+data/
+记忆.md

app.py

@@ -1,23 +1,150 @@
 import os
+import json
+from dotenv import load_dotenv
+load_dotenv()
 import gradio as gr
 import requests
 import inspect
 import pandas as pd
+from tools import web_search, wikipedia_search, python_repl, download_and_read_file
+from tools.file_handler import prefetch_file
 
 # (Keep Constants as is)
 # --- Constants ---
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
+TOOL_FUNCTIONS = {
+    "web_search": web_search,
+    "wikipedia_search": wikipedia_search,
+    "python_repl": python_repl,
+    "download_and_read_file": download_and_read_file,
+}
+
+TOOL_SCHEMAS = [
+    {
+        "type": "function",
+        "function": {
+            "name": "web_search",
+            "description": "Search Google for current information. Use for factual questions, recent events, or any topic requiring web search.",
+            "parameters": {
+                "type": "object",
+                "properties": {"query": {"type": "string", "description": "Search query"}},
+                "required": ["query"],
+            },
+        },
+    },
+    {
+        "type": "function",
+        "function": {
+            "name": "wikipedia_search",
+            "description": "Search Wikipedia for encyclopedic or historical information about a topic.",
+            "parameters": {
+                "type": "object",
+                "properties": {"query": {"type": "string", "description": "Topic to search"}},
+                "required": ["query"],
+            },
+        },
+    },
+    {
+        "type": "function",
+        "function": {
+            "name": "python_repl",
+            "description": "Execute Python code for math, calculations, data analysis, or logic. Use print() to output results.",
+            "parameters": {
+                "type": "object",
+                "properties": {"code": {"type": "string", "description": "Python code to execute"}},
+                "required": ["code"],
+            },
+        },
+    },
+    {
+        "type": "function",
+        "function": {
+            "name": "download_and_read_file",
+            "description": "Download and read a file attachment (CSV, Excel, text) for a given task_id from the question.",
+            "parameters": {
+                "type": "object",
+                "properties": {"task_id": {"type": "string", "description": "The task_id of the question"}},
+                "required": ["task_id"],
+            },
+        },
+    },
+]
+
+SYSTEM_PROMPT = (
+    "You are a precise research assistant. Solve each question step by step using tools.\n\n"
+
+    "STRICT RULES:\n"
+    "1. NEVER search for 'GAIA benchmark', 'GAIA answer', 'HuggingFace discussion', or any meta-search for pre-solved answers. Solve the problem yourself.\n"
+    "2. For ANY text manipulation (reversing, encoding, counting characters, etc.), ALWAYS use python_repl — never guess by eye.\n"
+    "3. Keep search queries SHORT and targeted (under 8 words). Never enumerate values (e.g. years) in one query.\n"
+    "4. If you have an [Attached file content] section, read it directly — do NOT call download_and_read_file again.\n"
+    "5. OUTPUT FORMAT: Your ENTIRE response must be ONLY the final answer — no explanation, no reasoning, no 'The answer is', no preamble. A single word, number, or comma-separated list. Nothing else.\n"
+    "6. Numbers: digits only (e.g. '42', not 'forty-two'). Names: as they appear in the source.\n"
+    "7. If a question involves reversing or encoding text, use python_repl to decode it first before reasoning.\n"
+    "8. Lists: output as plain comma-separated values (e.g. 'apple, banana, cherry') — NO brackets, NO quotes, NO Python syntax.\n"
+    "9. If the question has a [Task ID: xxx] prefix, use that exact value when calling download_and_read_file.\n"
+    "10. If an attached file is audio/image/video (marked [UNSUPPORTED]), do NOT call download_and_read_file — use web_search to find the answer instead.\n"
+)
+
 # --- Basic Agent Definition ---
 # ----- THIS IS WERE YOU CAN BUILD WHAT YOU WANT ------
 class BasicAgent:
     def __init__(self):
-        print("BasicAgent initialized.")
+        self.api_key = os.getenv("DEEPSEEK_API_KEY")
+        self.model = os.getenv("DEEPSEEK_MODEL", "deepseek-chat")
+        self.api_url = "https://www.ggwk1.online/v1/chat/completions"
+        print("BasicAgent initialized with DeepSeek (native HTTP).")
+
+    def _call_llm(self, messages: list) -> dict:
+        headers = {
+            "Authorization": f"Bearer {self.api_key}",
+            "Content-Type": "application/json",
+        }
+        payload = {
+            "model": self.model,
+            "messages": messages,
+            "tools": TOOL_SCHEMAS,
+            "tool_choice": "auto",
+            "temperature": 0,
+        }
+        response = requests.post(
+            self.api_url, headers=headers, json=payload, timeout=60,
+            proxies={"http": None, "https": None}
+        )
+        response.raise_for_status()
+        return response.json()
+
     def __call__(self, question: str) -> str:
         print(f"Agent received question (first 50 chars): {question[:50]}...")
-        fixed_answer = "This is a default answer."
-        print(f"Agent returning fixed answer: {fixed_answer}")
-        return fixed_answer
+        messages = [
+            {"role": "system", "content": SYSTEM_PROMPT},
+            {"role": "user", "content": question},
+        ]
+        for _ in range(15):  # max iterations
+            result = self._call_llm(messages)
+            choice = result["choices"][0]
+            message = choice["message"]
+            messages.append(message)
+
+            if choice["finish_reason"] == "tool_calls":
+                for tool_call in message.get("tool_calls", []):
+                    fn_name = tool_call["function"]["name"]
+                    fn_args = json.loads(tool_call["function"]["arguments"])
+                    print(f"  -> Tool: {fn_name}({fn_args})")
+                    tool_result = TOOL_FUNCTIONS[fn_name](**fn_args)
+                    print(f"  <- Result (first 200): {str(tool_result)[:200]}")
+                    messages.append({
+                        "role": "tool",
+                        "tool_call_id": tool_call["id"],
+                        "content": str(tool_result),
+                    })
+            else:
+                answer = message.get("content", "")
+                print(f"Agent answer: {answer}")
+                return answer
+
+        return "Max iterations reached without a final answer."
 
 def run_and_submit_all( profile: gr.OAuthProfile | None):
     """
@@ -51,7 +178,7 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
     # 2. Fetch Questions
     print(f"Fetching questions from: {questions_url}")
     try:
-        response = requests.get(questions_url, timeout=15)
+        response = requests.get(questions_url, timeout=15, proxies={"http": None, "https": None})
         response.raise_for_status()
         questions_data = response.json()
         if not questions_data:
@@ -80,7 +207,13 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
             print(f"Skipping item with missing task_id or question: {item}")
             continue
         try:
-            submitted_answer = agent(question_text)
+            # Pre-fetch file attachment and embed content directly in question
+            file_content = prefetch_file(task_id)
+            if file_content:
+                full_question = f"[Task ID: {task_id}]\n{question_text}\n\n[Attached file content]:\n{file_content}"
+            else:
+                full_question = f"[Task ID: {task_id}]\n{question_text}"
+            submitted_answer = agent(full_question)
             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})
         except Exception as e:
@@ -99,7 +232,7 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
     # 5. Submit
     print(f"Submitting {len(answers_payload)} answers to: {submit_url}")
     try:
-        response = requests.post(submit_url, json=submission_data, timeout=60)
+        response = requests.post(submit_url, json=submission_data, timeout=60, proxies={"http": None, "https": None})
         response.raise_for_status()
         result_data = response.json()
         final_status = (

requirements.txt

@@ -1,2 +1,5 @@
 gradio
-requests
+requests
+python-dotenv
+pandas
+openpyxl

test_agent.py

@@ -0,0 +1,30 @@
+"""
+Quick test script to debug a single question without launching Gradio.
+Usage: python test_agent.py
+"""
+import os
+import json
+from dotenv import load_dotenv
+load_dotenv()
+
+from app import BasicAgent
+from tools.file_handler import prefetch_file
+
+agent = BasicAgent()
+
+# --- Edit these to test any question ---
+task_id = "test-task-id"   # replace with real task_id if needed
+question = 'If we reverse the word "tfel", what is the antonym of the result?'
+# ----------------------------------------
+
+file_content = prefetch_file(task_id)
+if file_content:
+    full_question = f"{question}\n\n[Attached file content]:\n{file_content}"
+    print(f"[File found and attached, length={len(file_content)}]")
+else:
+    full_question = question
+    print("[No file attachment]")
+
+print(f"\nQuestion: {full_question[:200]}\n")
+answer = agent(full_question)
+print(f"\n=== Final Answer ===\n{answer}")

tools/__init__.py

@@ -0,0 +1,6 @@
+from .search import web_search
+from .wikipedia import wikipedia_search
+from .calculator import python_repl
+from .file_handler import download_and_read_file
+
+__all__ = ["web_search", "wikipedia_search", "python_repl", "download_and_read_file"]

tools/_session.py

@@ -0,0 +1,5 @@
+import requests
+
+# Shared session that ignores ALL system proxy environment variables
+_session = requests.Session()
+_session.trust_env = False

tools/calculator.py

@@ -0,0 +1,26 @@
+import sys
+import io
+import traceback
+
+
+def python_repl(code: str) -> str:
+    """Execute Python code for calculations or data processing."""
+    stdout_capture = io.StringIO()
+    stderr_capture = io.StringIO()
+    local_vars = {}
+    try:
+        sys.stdout = stdout_capture
+        sys.stderr = stderr_capture
+        exec(compile(code, "<string>", "exec"), {"__builtins__": __builtins__}, local_vars)
+    except Exception:
+        sys.stdout = sys.__stdout__
+        sys.stderr = sys.__stderr__
+        return f"Error:\n{traceback.format_exc()}"
+    finally:
+        sys.stdout = sys.__stdout__
+        sys.stderr = sys.__stderr__
+    output = stdout_capture.getvalue()
+    err = stderr_capture.getvalue()
+    if err:
+        return f"Stderr:\n{err}\nStdout:\n{output}"
+    return output if output else "Code executed successfully (no output)."

tools/file_handler.py

@@ -0,0 +1,90 @@
+import os
+import requests
+import pandas as pd
+from tools._session import _session
+
+DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
+DATA_DIR = os.path.join(os.path.dirname(os.path.dirname(__file__)), "data")
+os.makedirs(DATA_DIR, exist_ok=True)
+
+
+def _parse_file(file_path: str, content_bytes: bytes, ext: str) -> str:
+    """Parse file content based on extension and return as string."""
+    try:
+        if ext == ".csv":
+            df = pd.read_csv(file_path)
+            return f"CSV file ({len(df)} rows, {len(df.columns)} columns):\n{df.to_string(index=False)}"
+        elif ext in (".xlsx", ".xls"):
+            # Read all sheets
+            xl = pd.ExcelFile(file_path)
+            parts = []
+            for sheet in xl.sheet_names:
+                df = xl.parse(sheet)
+                parts.append(f"Sheet '{sheet}' ({len(df)} rows, {len(df.columns)} columns):\n{df.to_string(index=False)}")
+            return "\n\n".join(parts)
+        elif ext in (".py", ".txt", ".md", ".json", ".xml", ".html", ""):
+            return f"File contents:\n{content_bytes.decode('utf-8', errors='replace')[:5000]}"
+        else:
+            try:
+                return f"File contents:\n{content_bytes.decode('utf-8', errors='replace')[:5000]}"
+            except Exception:
+                return f"Binary file, cannot display as text. Size: {len(content_bytes)} bytes."
+    except Exception as e:
+        return f"Failed to parse file: {e}"
+
+
+def prefetch_file(task_id: str) -> str | None:
+    """
+    Try to download the file for a task_id.
+    Returns parsed file content string if found, None if no attachment exists.
+    Caches file to data/ directory.
+    """
+    # Check cache first
+    cached = [f for f in os.listdir(DATA_DIR) if f.startswith(task_id)]
+    if cached:
+        file_path = os.path.join(DATA_DIR, cached[0])
+        ext = os.path.splitext(cached[0])[-1].lower()
+        with open(file_path, "rb") as f:
+            content_bytes = f.read()
+        return _parse_file(file_path, content_bytes, ext)
+
+    file_url = f"{DEFAULT_API_URL}/files/{task_id}"
+    try:
+        response = _session.get(file_url, timeout=30)
+        if response.status_code == 404:
+            return None
+        response.raise_for_status()
+    except Exception:
+        return None
+
+    # Determine extension
+    ext = ""
+    cd = response.headers.get("content-disposition", "")
+    if "filename=" in cd:
+        fname = cd.split("filename=")[-1].strip().strip('"')
+        ext = os.path.splitext(fname)[-1].lower()
+    content_type = response.headers.get("content-type", "")
+    if not ext:
+        if "csv" in content_type:
+            ext = ".csv"
+        elif "excel" in content_type or "spreadsheet" in content_type or "openxmlformats" in content_type:
+            ext = ".xlsx"
+        elif "text" in content_type:
+            ext = ".txt"
+
+    # Save to data/
+    file_path = os.path.join(DATA_DIR, f"{task_id}{ext}")
+    with open(file_path, "wb") as f:
+        f.write(response.content)
+
+    return _parse_file(file_path, response.content, ext)
+
+
+def download_and_read_file(task_id: str) -> str:
+    """Download and read a file attachment for a given task_id.
+    Supports CSV, Excel (.xlsx/.xls), and plain text files.
+    """
+    result = prefetch_file(task_id)
+    if result is None:
+        return "No file attachment found for this task."
+    return result

tools/search.py

@@ -0,0 +1,40 @@
+import os
+import json
+from tools._session import _session
+
+
+def web_search(query: str) -> str:
+    """Search Google via Serper API for current information.
+    Use this for factual questions, recent events, or anything requiring web search.
+    """
+    api_key = os.getenv("SERPER_API_KEY")
+    if not api_key:
+        return "SERPER_API_KEY not set."
+    try:
+        response = _session.post(
+            "https://google.serper.dev/search",
+            headers={"X-API-KEY": api_key, "Content-Type": "application/json"},
+            data=json.dumps({"q": query, "num": 5}),
+            timeout=15,
+        )
+        response.raise_for_status()
+        data = response.json()
+
+        parts = []
+        # Answer box (direct answer)
+        if "answerBox" in data:
+            ab = data["answerBox"]
+            answer = ab.get("answer") or ab.get("snippet") or ""
+            if answer:
+                parts.append(f"Direct answer: {answer}")
+
+        # Organic results
+        for r in data.get("organic", [])[:5]:
+            title = r.get("title", "")
+            link = r.get("link", "")
+            snippet = r.get("snippet", "")
+            parts.append(f"Title: {title}\nURL: {link}\nSnippet: {snippet}")
+
+        return "\n---\n".join(parts) if parts else "No results found."
+    except Exception as e:
+        return f"Search error: {e}"

tools/wikipedia.py

@@ -0,0 +1,51 @@
+import os
+import requests
+
+# Wikipedia needs the local proxy to be accessible from China
+_PROXY = os.getenv("LOCAL_PROXY", "http://127.0.0.1:7890")
+_wiki_session = requests.Session()
+_wiki_session.proxies = {"http": _PROXY, "https": _PROXY}
+_wiki_session.trust_env = False
+_wiki_session.headers.update({
+    "User-Agent": "Mozilla/5.0 (compatible; ResearchAgent/1.0; educational use)"
+})
+
+
+def wikipedia_search(query: str) -> str:
+    """Search Wikipedia for encyclopedic information."""
+    # Use Wikipedia REST API directly
+    search_url = "https://en.wikipedia.org/w/api.php"
+    try:
+        # Step 1: search for page title
+        search_resp = _wiki_session.get(search_url, timeout=15, params={
+            "action": "query",
+            "list": "search",
+            "srsearch": query,
+            "srlimit": 3,
+            "format": "json",
+        })
+        search_resp.raise_for_status()
+        results = search_resp.json().get("query", {}).get("search", [])
+        if not results:
+            return "No Wikipedia articles found."
+
+        # Step 2: fetch content of top result
+        title = results[0]["title"]
+        content_resp = _wiki_session.get(search_url, timeout=15, params={
+            "action": "query",
+            "titles": title,
+            "prop": "extracts",
+            "exintro": False,
+            "explaintext": True,
+            "format": "json",
+        })
+        content_resp.raise_for_status()
+        pages = content_resp.json().get("query", {}).get("pages", {})
+        page = next(iter(pages.values()))
+        text = page.get("extract", "")
+        if not text:
+            return f"Wikipedia article '{title}' has no extractable content."
+        url = f"https://en.wikipedia.org/wiki/{title.replace(' ', '_')}"
+        return f"Wikipedia: {title}\nURL: {url}\n\n{text[:3000]}"
+    except Exception as e:
+        return f"Wikipedia search failed: {e}"