fix: address all code review comments — guard formatToolArgs/formatToolOutput, validate agent selection, remove unenforced requiresConfirmation

Agent-Logs-Url: https://github.com/objectstack-ai/framework/sessions/f75e2ec2-9b8a-4e71-916e-8193b834ad01

Co-authored-by: hotlong <50353452+hotlong@users.noreply.github.com>

Author: Copilot

apps/studio/src/components/AiChatPanel.tsx

@@ -62,7 +62,12 @@ function formatToolArgs(input: unknown): string {
   return entries
     .slice(0, 4)
     .map(([k, v]) => {
-      const val = typeof v === 'string' ? v : JSON.stringify(v);
+      let val: string;
+      try {
+        val = typeof v === 'string' ? v : (JSON.stringify(v) ?? String(v));
+      } catch {
+        val = String(v);
+      }
       const display = val.length > 30 ? val.slice(0, 30) + '…' : val;
       return `${k}: ${display}`;
     })
@@ -80,7 +85,12 @@ function isToolPart(part: UIMessage['parts'][number]): part is Extract<UIMessage
  * Format tool output for display, truncating to a max length.
  */
 function formatToolOutput(output: unknown, maxLen = 80): string {
-  const raw = typeof output === 'string' ? output : JSON.stringify(output);
+  let raw: string;
+  try {
+    raw = typeof output === 'string' ? output : (JSON.stringify(output) ?? '');
+  } catch {
+    raw = String(output ?? '');
+  }
   return raw.length > maxLen ? raw.slice(0, maxLen) + '…' : raw;
 }
 
@@ -284,6 +294,18 @@ export function AiChatPanel() {
   const baseUrl = getApiBaseUrl();
   const { agents, loading: agentsLoading } = useAgentList(baseUrl);
 
+  // Validate persisted agent against fetched list — fall back to general
+  // chat if the previously selected agent is no longer available.
+  useEffect(() => {
+    if (agentsLoading) return;
+    if (selectedAgent === GENERAL_CHAT_VALUE) return;
+    const isValid = agents.some((a) => a.name === selectedAgent);
+    if (!isValid) {
+      setSelectedAgent(GENERAL_CHAT_VALUE);
+      saveSelectedAgent(GENERAL_CHAT_VALUE);
+    }
+  }, [agents, agentsLoading, selectedAgent]);
+
   const initialMessages = useMemo(() => loadMessages() as UIMessage[], []);
 
   const transport = useMemo(

packages/services/service-ai/src/__tests__/metadata-tools.test.ts

@@ -123,12 +123,12 @@ describe('Individual Tool Metadata (.tool.ts)', () => {
     });
   }
 
-  it('should mark create_object as requiresConfirmation', () => {
-    expect(createObjectTool.requiresConfirmation).toBe(true);
+  it('should not set requiresConfirmation on create_object (server-side enforcement not yet implemented)', () => {
+    expect(createObjectTool.requiresConfirmation).toBe(false);
   });
 
-  it('should mark delete_field as requiresConfirmation', () => {
-    expect(deleteFieldTool.requiresConfirmation).toBe(true);
+  it('should not set requiresConfirmation on delete_field (server-side enforcement not yet implemented)', () => {
+    expect(deleteFieldTool.requiresConfirmation).toBe(false);
   });
 
   it('should not mark read-only tools as requiresConfirmation', () => {

packages/services/service-ai/src/tools/create-object.tool.ts

@@ -17,7 +17,11 @@ export const createObjectTool = defineTool({
     'Use this when the user wants to create a new entity, table, or data model.',
   category: 'data',
   builtIn: true,
-  requiresConfirmation: true,
+  // NOTE: requiresConfirmation is intentionally false (default) because the
+  // server-side tool-call loop in AIService.chatWithTools/streamChatWithTools
+  // executes tool calls immediately without checking this flag.  The flag
+  // should only be set once server-side approval gating is implemented to
+  // avoid giving users a false sense of safety.
   parameters: {
     type: 'object',
     properties: {

packages/services/service-ai/src/tools/delete-field.tool.ts

@@ -16,7 +16,10 @@ export const deleteFieldTool = defineTool({
     'Use this when the user explicitly wants to remove an attribute or column from a table.',
   category: 'data',
   builtIn: true,
-  requiresConfirmation: true,
+  // NOTE: requiresConfirmation is intentionally false (default) because the
+  // server-side tool-call loop in AIService.chatWithTools/streamChatWithTools
+  // executes tool calls immediately without checking this flag.  The flag
+  // should only be set once server-side approval gating is implemented.
   parameters: {
     type: 'object',
     properties: {