fix(benchmarks): Harden Claude UI process cleanup

Detach timed Claude commands before process-group termination and fix RocketSim preflight launch detection for direct app/path commands.

Co-Authored-By: OpenAI Codex <noreply@openai.com>

Author: cameroncooke

src/benchmarks/claude-ui/__tests__/preflight-commands.test.ts

@@ -22,6 +22,20 @@ describe('Claude UI benchmark preflight commands', () => {
     expect(preflightCommandsWithFocusResign({ commands })).toBe(commands);
   });
 
+  it('detects simple and path-based RocketSim launch commands', () => {
+    expect(
+      preflightCommandsWithFocusResign({
+        commands: ['open RocketSim', 'open /Applications/RocketSim.app'],
+        simulatorId: 'SIM-123',
+      }),
+    ).toEqual([
+      'open RocketSim',
+      "open -a Simulator --args -CurrentDeviceUDID 'SIM-123'",
+      'open /Applications/RocketSim.app',
+      "open -a Simulator --args -CurrentDeviceUDID 'SIM-123'",
+    ]);
+  });
+
   it('shell-quotes simulator IDs used by the focus command', () => {
     expect(
       preflightCommandsWithFocusResign({

src/benchmarks/claude-ui/harness.ts

@@ -228,7 +228,7 @@ function runCommand(opts: {
       cwd: opts.cwd,
       env: opts.env ?? process.env,
       stdio: ['pipe', 'pipe', 'pipe'],
-      detached: opts.terminalJsonResultGraceMs !== undefined,
+      detached: opts.terminalJsonResultGraceMs !== undefined || opts.timeoutMs !== undefined,
     });
 
     const clearTerminalResultTimer = (): void => {

src/benchmarks/claude-ui/preflight-commands.ts

@@ -17,7 +17,7 @@ function shellSingleQuote(value: string): string {
 }
 
 function isRocketSimAppLaunchCommand(command: string): boolean {
-  return /^\s*open\s+.*(?:^|\s)RocketSim(?:\.app)?\s*$/.test(command);
+  return /^\s*open\s+(?:.*(?:\s|\/))?RocketSim(?:\.app)?\s*$/.test(command);
 }
 
 export function preflightCommandsWithFocusResign(opts: {