fix: avoid asking questions at the end of thinking

Author: SchrodingersCattt

agents/matmaster_agent/flow_agents/plan_make_agent/prompt.py

@@ -41,7 +41,7 @@ def get_static_plan_system_block(available_tools_with_info: str) -> str:
 - The tool name in text MUST exactly match the "tool_name" field value (or "llm_tool" when tool_name is null).
 
 ### RE-PLANNING LOGIC:
-If the input contains errors from previous steps, analyze the failure and adjust the current plan (e.g., fix parameters or change tools) to resolve the issue. Mention the fix in the "step_description" while still following the required format.
+If the input contains errors from previous steps, analyze the failure and adjust the current plan (e.g., fix parameters or change tools) to resolve the issue. Mention the fix in the "step_description" while still following the required format. Do not ask the user whether to fix—output the adjusted plan directly. Do not end intro/overall with a question.
 
 ### MULTI-PLAN GENERATION (NEW):
 Generate MULTIPLE alternative plans (at least 3, unless impossible) that all satisfy the user request.

agents/matmaster_agent/flow_agents/thinking_agent/prompt.py

@@ -32,7 +32,7 @@ def get_static_system_block(available_tools_with_info: str) -> str:
         Deconstruct user intent into Input Data (Source) and Desired Output (Target).
         Identify key entities (SMILES, chemical formulas, files).
         Determine if the user wants *Learning* (docs/tutorials) or *Execution* (simulations).
-        If <Session Memory> is provided: use it. It contains prior insights, parameters, or findings from this session; align your plan with these when relevant (e.g. preferred methods, previously chosen parameters, or constraints the user implied earlier).
+        If <Session Memory> is provided: use it only to align tool choice and parameters (e.g. preferred methods, previously chosen parameters). Do not reason about "whether to fix" past errors—only use memory to constrain the next tool chain.
     </phase_1_analysis>
 
     <phase_2_drafting>
@@ -72,7 +72,7 @@ def get_static_system_block(available_tools_with_info: str) -> str:
 
 ### FINAL TRIGGER
 - This is a planning phase. You verify the plan's validity.
-- You MUST end your response with exactly one line: `Revision needed` (to trigger the validation loop). Do not output `READY` here.
+- You MUST end your response with exactly one line: `Revision needed` (to trigger the validation loop). Do not output `READY` here. Do not end with a question.
 
 ### STATIC CONTEXT (Tools)
 <Available Tools With Info>
@@ -206,7 +206,7 @@ def get_static_revision_system_block(available_tools_with_info: str) -> str:
 [Your reasoning for the correction]
 The system will automatically run another validation round on your corrected plan. Continue until you can output "Verification passed." when all checks pass.
 
-End with "Verification passed." ONLY when all prerequisites definitively pass. Do not use hedging; say PASS or FAIL clearly.
+End with "Verification passed." ONLY when all prerequisites definitively pass. Do not use hedging; say PASS or FAIL clearly. Do not end with a question.
 """