feat: update default planner model to gpt-5.4 (#170)

* fix: skip close_all and notification cleanup by default

The close_all PowerShell command (Get-Process | CloseMainWindow) and
notification cleanup (taskkill OneDrive) crash the WAA Flask server,
making it unresponsive for the rest of the run. This happened on every
test run.

Now these only run when clean_desktop=True is explicitly set. Default
behavior skips them entirely — the task runs against whatever state
the desktop is in, which is more reliable than crashing the server
trying to clean up.

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

* feat: update default planner model to gpt-5.4

Update the OpenAI default model in ApiAgent from gpt-5.1 to gpt-5.4
for higher-quality planning and reasoning in both the main agent and
the WAA deploy copy.

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

---------

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

Author: abrichr

openadapt_evals/adapters/waa/live.py

@@ -664,34 +664,30 @@ def reset(self, task: BenchmarkTask) -> BenchmarkObservation:
 
         import requests
 
-        # Try to close all windows for clean state via /execute_windows
-        try:
-            close_cmd = (
-                "import subprocess; "
-                "subprocess.run("
-                "['powershell', '-Command', "
-                "'Get-Process | Where-Object {$_.MainWindowTitle -ne \\\"\\\"} | "
-                "ForEach-Object { $_.CloseMainWindow() }'], "
-                "timeout=15)"
-            )
-            resp = requests.post(
-                f"{self.config.server_url}/execute_windows",
-                json={"command": close_cmd},
-                timeout=10.0,
-            )
-            if resp.status_code == 200:
-                logger.info("Closed all windows for clean state")
-            else:
-                logger.warning("close_all failed: HTTP %s", resp.status_code)
-        except Exception as e:
-            logger.warning(f"Failed to close windows (non-fatal): {e}")
-
-        # Optionally apply deterministic desktop policy before task setup.
-        if self.config.clean_desktop or self.config.force_tray_icons:
+        # Close windows and dismiss notifications ONLY if explicitly requested.
+        # The close_all and notification cleanup commands can crash the WAA Flask
+        # server (PowerShell Get-Process hangs, taskkill blocks), leaving the
+        # server unresponsive for the rest of the run.
+        if self.config.clean_desktop:
+            try:
+                close_cmd = (
+                    "import subprocess; "
+                    "subprocess.run("
+                    "['powershell', '-Command', "
+                    "'Get-Process | Where-Object {$_.MainWindowTitle -ne \\\"\\\"} | "
+                    "ForEach-Object { $_.CloseMainWindow() }'], "
+                    "timeout=10)"
+                )
+                resp = requests.post(
+                    f"{self.config.server_url}/execute_windows",
+                    json={"command": close_cmd},
+                    timeout=10.0,
+                )
+                if resp.status_code == 200:
+                    logger.info("Closed all windows for clean state")
+            except Exception as e:
+                logger.warning("close_all failed (non-fatal): %s", e)
             self._apply_clean_desktop_policy(requests)
-        else:
-            # Best-effort cleanup even when full clean policy is disabled.
-            self._dismiss_notifications(requests)
 
         # LibreOffice can surface a modal "Document Recovery" dialog after
         # dirty shutdowns. Pre-clean recovery state before setup to avoid

openadapt_evals/agents/api_agent.py

@@ -293,7 +293,7 @@ class ApiAgent(BenchmarkAgent):
     # Default models for each provider
     DEFAULT_MODELS = {
         "anthropic": "claude-sonnet-4-5-20250929",
-        "openai": "gpt-5.1",
+        "openai": "gpt-5.4",
     }
 
     def __init__(

openadapt_evals/waa_deploy/api_agent.py

@@ -177,7 +177,7 @@ class ApiAgent:
     # Default models for each provider
     DEFAULT_MODELS = {
         "anthropic": "claude-sonnet-4-5-20250929",
-        "openai": "gpt-5.1",
+        "openai": "gpt-5.4",
     }
 
     def __init__(