diff --git a/agent/component/agent_with_tools.py b/agent/component/agent_with_tools.py
index 91b467cc5..f931f2f18 100644
--- a/agent/component/agent_with_tools.py
+++ b/agent/component/agent_with_tools.py
@@ -277,19 +277,15 @@ class Agent(LLM, ToolBase):
         else:
             user_request = history[-1]["content"]
 
-        def build_task_desc(prompt: str, user_request: str, tool_metas: list[dict], user_defined_prompt: dict | None = None) -> str:
+        def build_task_desc(prompt: str, user_request: str, user_defined_prompt: dict | None = None) -> str:
             """Build a minimal task_desc by concatenating prompt, query, and tool schemas."""
             user_defined_prompt = user_defined_prompt or {}
 
-            tools_json = json.dumps(tool_metas, ensure_ascii=False, indent=2)
-
             task_desc = (
                 "### Agent Prompt\n"
                 f"{prompt}\n\n"
                 "### User Request\n"
                 f"{user_request}\n\n"
-                "### Tools (schemas)\n"
-                f"{tools_json}\n"
             )
 
             if user_defined_prompt:
@@ -368,7 +364,7 @@ class Agent(LLM, ToolBase):
                 hist.append({"role": "user", "content": content})
 
         st = timer()
-        task_desc = build_task_desc(prompt, user_request, tool_metas, user_defined_prompt)
+        task_desc = build_task_desc(prompt, user_request, user_defined_prompt)
         self.callback("analyze_task", {}, task_desc, elapsed_time=timer()-st)
         for _ in range(self._param.max_rounds + 1):
             if self.check_if_canceled("Agent streaming"):
diff --git a/rag/prompts/next_step.md b/rag/prompts/next_step.md
index e84aa8f4e..3c45a0d08 100644
--- a/rag/prompts/next_step.md
+++ b/rag/prompts/next_step.md
@@ -94,7 +94,7 @@ This content will NOT be shown to the user.
 ## Step 2: Structured Reflection (MANDATORY before `complete_task`)
 
 ### Context
-- Goal: {{ task_analysis }}
+- Goal: Reflect on the current task based on the full conversation context
 - Executed tool calls so far (if any): reflect from conversation history
 
 ### Task Complexity Assessment