refactor: simplify permission request code after review

- Remove hasResult alias, use hasToolResult directly
- Collapse config/resolvedConfig/activeConfig triple to single lookup
- Remove redundant hasattr/isinstance guards on ToolCallEndEvent
- Use getattr for placeholder ID checks
- Clear _permission_worker in bridge finally block (prevent GC leak)
- Remove dead description initializer
- Add PermissionRequest to use-agent-status.ts waiting_input detection

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: Ambient Code Bot

components/frontend/src/components/session/ask-user-question.tsx

@@ -42,8 +42,6 @@ function parseQuestions(input: Record<string, unknown>): AskUserQuestionItem[] {
   return [];
 }
 
-const hasResult = hasToolResult;
-
 export const AskUserQuestionMessage: React.FC<AskUserQuestionMessageProps> = ({
   toolUseBlock,
   resultBlock,
@@ -52,7 +50,7 @@ export const AskUserQuestionMessage: React.FC<AskUserQuestionMessageProps> = ({
   isNewest = false,
 }) => {
   const questions = parseQuestions(toolUseBlock.input);
-  const alreadyAnswered = hasResult(resultBlock);
+  const alreadyAnswered = hasToolResult(resultBlock);
   const formattedTime = formatTimestamp(timestamp);
   const isMultiQuestion = questions.length > 1;
 

components/frontend/src/components/session/permission-request.tsx

@@ -92,9 +92,7 @@ export const PermissionRequestMessage: React.FC<
     }
   };
 
-  const config = STATUS_CONFIG[disabled && status !== "pending" ? status : "pending"];
-  const resolvedConfig = STATUS_CONFIG[status];
-  const activeConfig = disabled ? resolvedConfig : config;
+  const activeConfig = STATUS_CONFIG[disabled && status !== "pending" ? status : "pending"];
   const Icon = activeConfig.icon;
 
   return (

components/frontend/src/hooks/use-agent-status.ts

@@ -5,9 +5,13 @@ import type {
 } from "@/types/agentic-session";
 import type { PlatformMessage } from "@/types/agui";
 
-function isAskUserQuestionTool(name: string): boolean {
-  const normalized = name.toLowerCase().replace(/[^a-z]/g, "");
-  return normalized === "askuserquestion";
+function normalizeToolName(name: string): string {
+  return name.toLowerCase().replace(/[^a-z]/g, "");
+}
+
+function isHumanInTheLoopTool(name: string): boolean {
+  const normalized = normalizeToolName(name);
+  return normalized === "askuserquestion" || normalized === "permissionrequest";
 }
 
 /**
@@ -38,7 +42,7 @@ export function useAgentStatus(
 
       // Check the last tool call on this message
       const lastTc = msg.toolCalls[msg.toolCalls.length - 1];
-      if (lastTc.function?.name && isAskUserQuestionTool(lastTc.function.name)) {
+      if (lastTc.function?.name && isHumanInTheLoopTool(lastTc.function.name)) {
         const hasResult =
           lastTc.result !== undefined &&
           lastTc.result !== null &&

components/runners/ambient-runner/ag_ui_claude_sdk/adapter.py

@@ -317,7 +317,6 @@ async def _can_use_tool(
         # Build a human-readable description of what Claude wants to do.
         file_path = input_data.get("file_path", "")
         command = input_data.get("command", "")
-        description = ""
         if file_path:
             description = f"{tool_name} on {file_path}"
         elif command:
@@ -964,20 +963,18 @@ def flush_pending_msg():
                 if isinstance(message, BaseEvent):
                     # Rewrite placeholder thread/run IDs injected by
                     # can_use_tool (which doesn't know the real IDs).
-                    if hasattr(message, "thread_id") and message.thread_id == _PERM_PLACEHOLDER_ID:
+                    if getattr(message, "thread_id", None) == _PERM_PLACEHOLDER_ID:
                         message.thread_id = thread_id
-                    if hasattr(message, "run_id") and message.run_id == _PERM_PLACEHOLDER_ID:
+                    if getattr(message, "run_id", None) == _PERM_PLACEHOLDER_ID:
                         message.run_id = run_id
 
                     yield message
 
-                    # Detect PermissionRequest halt: the ToolCallEndEvent
-                    # for a PermissionRequest tool signals that we should
-                    # halt just like a frontend tool.
+                    # PermissionRequest halt: ToolCallEndEvent with a
+                    # perm- prefixed ID triggers the same halt as a
+                    # frontend tool.
                     if (
                         isinstance(message, ToolCallEndEvent)
-                        and hasattr(message, "tool_call_id")
-                        and isinstance(message.tool_call_id, str)
                         and message.tool_call_id.startswith(_PERM_TOOL_ID_PREFIX)
                     ):
                         logger.debug(

components/runners/ambient-runner/ambient_runner/bridges/claude/bridge.py

@@ -233,6 +233,9 @@ async def run(
                     # Clear the halt flag for this thread
                     self._halted_by_thread.pop(thread_id, None)
             finally:
+                # Release worker reference so destroyed workers can be GC'd.
+                self._adapter.set_permission_worker(None)
+
                 # Clear caller token immediately — never persist between turns.
                 if self._context:
                     self._context.caller_token = ""