change to participant notes

app.py

@@ -362,8 +362,87 @@ def run_agent(system_prompt, email, show_raw_output):
 with gr.Blocks(title="Executive Agent Lab") as demo:
     gr.Markdown("# Executive Agent Lab")
     gr.Markdown(
-        "Build your own email triage agent by writing the system prompt, then test it on sample or custom emails."
-    )
+    "*Built by Tan Hong Ming (NUS) · For learning prompt engineering, agentic workflows, and tool use*"
+)
+    with gr.Accordion("How to Use This Lab", open=False):
+        gr.Markdown("""
+            ## What this lab is about
+
+            This lab helps you practice three ideas:
+
+            1. Prompt engineering  
+            2. Agentic workflows  
+            3. Agent tool use  
+
+            You will improve an email triage agent by editing its system prompt.
+
+            ## Suggested workflow
+
+            ### Step 1
+            Click **Load Basic** and then **Run Agent**.
+
+            Observe what the agent does well and what it does poorly.
+
+            ### Step 2
+            Improve the system prompt so the agent can do the following for each email:
+
+            1. Identify the primary intent  
+            2. Classify priority as Low, Medium, or High  
+            3. Assign an owner from: Sales, Support, Finance, HR, Operations, Legal  
+            4. Determine the next action from: Respond, Escalate, Delegate, Request More Information  
+            5. Briefly explain the reasoning  
+            6. Draft a professional reply  
+
+            ### Step 3
+            Test your prompt on different sample emails.
+
+            Compare how the output changes when your instructions become more specific.
+
+            ### Step 4
+            Get the agent to use the status lookup tool when needed.
+
+            The tool call format is:
+
+            `CALL_TOOL: lookup_status <reference_id>`
+
+            Examples:
+
+            `CALL_TOOL: lookup_status SG-18427`  
+            `CALL_TOOL: lookup_status INC-2041`
+
+            The tool should only be used when the email refers to a specific order, shipment, incident, or ticket and current status information is needed.
+
+            ### Step 5
+            After you get it working, look at the source code to understand what happens under the hood.
+
+            ## What to look for in the code
+
+            Here is the simple flow:
+
+            1. The app sends your system prompt and the email to the model  
+            2. The model either:
+            - answers directly, or
+            - outputs a tool call like `CALL_TOOL: lookup_status SG-18427`
+            3. The app detects that tool call in the model output
+            4. The Python function runs the lookup
+            5. The tool result is sent back to the model
+            6. The model produces the final answer
+
+            So when you read the code, focus on these parts:
+
+            - where the model is called
+            - how `CALL_TOOL:` is detected
+            - how `lookup_status()` works
+            - how the tool result is passed back into the model"""
+        )
+
+        gr.Markdown("### Reference")
+
+        gr.Markdown(
+            "Once you have tried building your own prompt, you can load a working example below."
+        )
+
+        load_reference_btn = gr.Button("Load Reference Prompt")
 
     with gr.Row():
         with gr.Column(scale=1):
@@ -403,22 +482,6 @@ with gr.Blocks(title="Executive Agent Lab") as demo:
                     - when to use the tool and when not to use it"""
                 )
 
-            with gr.Accordion("Instructor Tools", open=False):
-                gr.Markdown("""
-                    ### How to use this app in a session
-
-                    1. Ask participants to start with Basic / Structured prompts  
-                    2. Let them test and struggle with outputs  
-                    3. Introduce tool calling concept  
-                    4. Have them refine prompts to trigger tool correctly  
-                    5. Only then reveal the reference prompt  
-
-                    Key teaching point:
-                    Participants are designing agent behavior, not just prompting."""
-                )
-
-                load_reference_btn = gr.Button("Load Reference Prompt")
-
         with gr.Column(scale=1):
             gr.Markdown("## Test Input")
             gr.Markdown("Load a sample email or paste your own.")