Scale computer-use 2.0 with open-source drivers, cross-OS sandboxes, and benchmarks for training, evaluation, and data generation.
Rust agentai-agentapplecomputer-usecomputer-use-agentcontainerizationcuadesktop-automationhacktoberfestlumemacosmanusoperatorswiftvirtualizationvirtualization-frameworkwindowswindows-sandbox
fix(cua-driver-bench): count the Windows cua-driver process as a tool overlay, not a raise (#6623)
Francesco Bonacci42d2892804cb
diff --git a/libs/cua-driver-bench/automated-eval/macos_bench/swift/summarize_sentinel.py b/libs/cua-driver-bench/automated-eval/macos_bench/swift/summarize_sentinel.pyindex 445e7d754a..7cdb32ba0b 100644--- a/libs/cua-driver-bench/automated-eval/macos_bench/swift/summarize_sentinel.py+++ b/libs/cua-driver-bench/automated-eval/macos_bench/swift/summarize_sentinel.py@@ -21,9 +21,9 @@ SENTINEL_BID = "ai.cua.benchsentinel"DEVIATION_PX = 10.0HID_KEYS = ("move", "down", "key", "scroll")# Amendment 15: overlay and indicator windows of the computer-use tools themselves (the agent cursor, the-# "ChatGPT Computer Use" banner) are not raises. They are logged as metadata (`overlay_owners`) only.+# "ChatGPT Computer Use" banner, the `cua-driver` process on Windows) are not raises. They are logged as metadata (`overlay_owners`) only.OVERLAY_OWNER_RE = re.compile(- r"(?i)cua ?driver|chatgpt computer use|computer use|codex|app-cu-helper|claude"+ r"(?i)cua[ _-]?driver|chatgpt computer use|computer use|codex|app-cu-helper|claude")FRONT_EVENTS = ("s", "front", "armed", "disarmed")# Idle values only ever grow between samples unless a real event resets them.diff --git a/libs/cua-driver-bench/automated-eval/macos_bench/swift/tests/test_summarize_sentinel.py b/libs/cua-driver-bench/automated-eval/macos_bench/swift/tests/test_summarize_sentinel.pyindex 15f141e69d..213dec6c06 100644--- a/libs/cua-driver-bench/automated-eval/macos_bench/swift/tests/test_summarize_sentinel.py+++ b/libs/cua-driver-bench/automated-eval/macos_bench/swift/tests/test_summarize_sentinel.py@@ -308,6 +308,15 @@ class WindowsRaisedTests(unittest.TestCase):result["overlay_owners"], ["ChatGPT Computer Use", "Cua Driver Bench Script"])+ def test_windows_driver_process_name_is_overlay(self) -> None:+ # On Windows the owner is the executable stem, `cua-driver`, with a hyphen.+ seq = [["cua-driver"]] * 4 + [["BenchLab"]] * 2+ lines = [self.line("armed", 0, [])] + [self.line("s", i + 1, a) for i, a in enumerate(seq)]+ result = self.summarize(lines)+ self.assertEqual(result["windows_raised"], 1)+ self.assertEqual(result["raised_by"], ["BenchLab"])+ self.assertEqual(result["overlay_owners"], ["cua-driver"])+def test_user_activity_counts(self) -> None:lines = [self.line("armed", 0, [])] + [self.line("s", i, []) for i in range(1, 12)]lines += [