Update app.py

app.py

@@ -60,12 +60,8 @@ def _download_task_file(task_id: str, api_url: str = DEFAULT_API_URL) -> tuple[b
     #     {},
     # ]:
     try:
-        resp = requests.get(url, timeout=30)
-        # resp = requests.get(url, headers=headers, timeout=30)
-        # resp.raise_for_status()
-        # with open(local_path, "wb") as fh:
-        #     fh.write(resp.content)
-        # print(f"Downloaded task file: {local_path} ({len(resp.content)} bytes)")
+        headers = {"Authorization": f"Bearer {HF_TOKEN}"}
+        resp = requests.get(url, headers=headers, timeout=30)
     except requests.exceptions.HTTPError as e:
         status = e.response.status_code if e.response is not None else "?"
         print(f"Download attempt for {task_id} returned {status}")
@@ -227,8 +223,12 @@ def call_tools(state: AgentState) -> AgentState:
             state["context"] = web_search.invoke({"query": question})
     elif label == "research":
         print("[TOOL] web search")
-        search_json = web_search.invoke({"query": question})
-        wiki_text = wikipedia_search.invoke({"query": question})
+        # Generate a focused search query (not the full question text)
+        search_query_prompt = f"Write a short Google search query (max 10 words) to answer this question. Output ONLY the query, nothing else.\n\nQuestion: {question}"
+        focused_query = _llm_router.invoke(search_query_prompt).content.strip().strip('"')
+        print(f"[TOOL] search query: {focused_query}")
+        search_json = web_search.invoke({"query": focused_query})
+        wiki_text = wikipedia_search.invoke({"query": focused_query})
         state["context"] = f"{search_json}\n\n{wiki_text}"
     else:
         print("[TOOL] reasoning only (no search)")
@@ -255,17 +255,24 @@ def synthesize_response(state: AgentState) -> AgentState:
     reasoning = _llm_answer.invoke(reasoning_prompt).content.strip()
     print(f"\n[REASONING]\n{reasoning}\n")
 
-    # Pass 2: extract the final answer cleanly
-    extract_prompt = [
-        SystemMessage(content=get_prompt("extract_system")),
-        HumanMessage(
-            content=get_prompt(
-                prompt_key="extract_user",
-                reasoning=reasoning,
-            )
-        ),
-    ]
-    answer = _llm_answer.invoke(extract_prompt).content.strip()
+    # Try to extract FINAL ANSWER directly from reasoning text (avoids second LLM call hallucinating)
+    fa_match = re.search(r"FINAL ANSWER:\s*(.+)", reasoning, re.IGNORECASE)
+    if fa_match:
+        answer = fa_match.group(1).strip().split('\n')[0].strip()
+    elif reasoning.strip():
+        # Pass 2: ask LLM to extract only if no FINAL ANSWER marker found
+        extract_prompt = [
+            SystemMessage(content=get_prompt("extract_system")),
+            HumanMessage(
+                content=get_prompt(
+                    prompt_key="extract_user",
+                    reasoning=reasoning,
+                )
+            ),
+        ]
+        answer = _llm_answer.invoke(extract_prompt).content.strip()
+    else:
+        answer = "ERROR: no reasoning produced"
     state["answer"] = answer
     return state