cleaning

Author: DidierRLopes

40-vanilla-agent-dashboard-widgets/README.md

@@ -1,13 +1,16 @@
-# 36 - Vanilla Agent Dashboard Widgets
+# 40 - Vanilla Agent Dashboard Widgets
 
-This example demonstrates a simple agent that receives the full list of widgets present on the current dashboard and passes that context directly to the LLM. The model decides which widget(s) to use and issues a function call accordingly.
+This example demonstrates a simple agent that lists all widgets available on the current dashboard, showing their metadata and parameters in a structured format.
 
 Key behaviors:
 
 - Exposes `agents.json` with `widget-dashboard-search` enabled so the Workspace sends dashboard widget metadata (as `widgets.secondary`, etc.).
-- If the user has selected primary widgets, the agent immediately issues a function call to fetch data for them.
-- Otherwise, the agent does not select widgets heuristically. It appends the full dashboard widget list to the prompt and instructs the LLM to respond with a `get_widget_data` JSON function call when needed.
-- Falls back to a plain LLM reply if no data is needed.
+- Lists all widgets from both explicit context (primary) and dashboard context (secondary).
+- Shows detailed widget information including:
+  - Widget metadata (name, description, ID, category, UUID)
+  - Parameters table with type, default, current values, options, and descriptions
+- Organizes dashboard widgets by tabs for better visibility.
+- Does NOT automatically fetch widget data - focuses purely on widget discovery and listing.
 
 ## Run locally
 
@@ -18,4 +21,4 @@ Key behaviors:
 
 ## Test
 
-- From this directory: `poetry run pytest tests`
+- From this directory: `poetry run pytest tests`

40-vanilla-agent-dashboard-widgets/tests/test_agent.py

@@ -32,7 +32,7 @@ def test_agents_json_has_dashboard_search_feature_enabled():
     assert agent["features"]["widget-dashboard-search"] is True
 
 
-def test_query_recognizes_dashboard_widgets_from_secondary():
+def test_query_lists_dashboard_widgets_from_secondary():
     test_payload_path = (
         Path(__file__).parent.parent.parent
         / "testing"
@@ -43,17 +43,19 @@ def test_query_recognizes_dashboard_widgets_from_secondary():
 
     # Simulate no explicit primary selection
     payload["widgets"]["primary"] = []
-    # Ask to list widgets instead of retrieving; ensures non-LLM path is exercised
+    # Ask to list widgets
     payload["messages"][0]["content"] = "What widgets are available in the dashboard?"
 
     response = test_client.post("/v1/query", json=payload)
     assert response.status_code == 200
 
-    # We expect a regular message listing dashboard widgets (secondary only)
+    # We expect a message listing dashboard widgets (secondary only)
     CopilotResponse(response.text).has_any("copilotMessage", "Company News")
+    # Should show dashboard context header
+    CopilotResponse(response.text).has_any("copilotMessage", "Dashboard Context")
 
 
-def test_query_respects_primary_selection_and_calls_get_widget_data():
+def test_query_lists_primary_widgets_when_selected():
     # Use same payload but keep primary set
     test_payload_path = (
         Path(__file__).parent.parent.parent
@@ -66,25 +68,50 @@ def test_query_respects_primary_selection_and_calls_get_widget_data():
     response = test_client.post("/v1/query", json=payload)
     assert response.status_code == 200
 
-    # We expect a function call – let the UI handle the actual retrieval
-    CopilotResponse(response.text).has_any(
-        "copilotFunctionCall", {"function": "get_widget_data"}
-    )
+    # We expect a message listing widgets including primary context
+    CopilotResponse(response.text).has_any("copilotMessage", "Explicit Context")
+    # Should also show dashboard context if available
+    CopilotResponse(response.text).has_any("copilotMessage", "Dashboard Context")
 
 
-def test_query_lists_dashboard_widgets():
-    # Ask to list widgets and expect a direct message with names
+def test_query_shows_widget_metadata_and_parameters():
     test_payload_path = (
         Path(__file__).parent.parent.parent
         / "testing"
         / "test_payloads"
         / "retrieve_widget_from_dashboard.json"
     )
     payload = json.load(open(test_payload_path))
-    payload["widgets"]["primary"] = []
-    payload["messages"][0]["content"] = "What widgets are available in the dashboard?"
+    payload["messages"][0]["content"] = "Show me widget details"
 
     response = test_client.post("/v1/query", json=payload)
     assert response.status_code == 200
 
-    CopilotResponse(response.text).has_any("copilotMessage", "Company News")
+    # Should show widget metadata fields
+    CopilotResponse(response.text).has_any("copilotMessage", "Description")
+    CopilotResponse(response.text).has_any("copilotMessage", "UUID")
+    # Should show parameters table
+    CopilotResponse(response.text).has_any("copilotMessage", "Parameters")
+
+
+def test_query_does_not_fetch_widget_data():
+    """Test that the agent only lists widgets and does not fetch data"""
+    test_payload_path = (
+        Path(__file__).parent.parent.parent
+        / "testing"
+        / "test_payloads"
+        / "retrieve_widget_from_dashboard.json"
+    )
+    payload = json.load(open(test_payload_path))
+    payload["messages"][0]["content"] = "Hello"
+
+    response = test_client.post("/v1/query", json=payload)
+    assert response.status_code == 200
+    
+    # Should NOT have any function calls for get_widget_data
+    response_text = response.text
+    assert "get_widget_data" not in response_text
+    assert "copilotFunctionCall" not in response_text
+    
+    # Should only have messages listing widgets
+    CopilotResponse(response.text).has_any("copilotMessage")