hud-python 0.2.8__py3-none-any.whl → 0.2.9__py3-none-any.whl

hud/adapters/common/types.py

@@ -217,9 +217,10 @@ class PositionFetch(CLAAction):
 class CustomAction(CLAAction):
     type: Literal["custom"] = "custom"
     action: str
+    args: dict[str, Any] | None = None
 
     def __str__(self) -> str:
-        action_str = f"⚙️ Custom: {self.action}"
+        action_str = f"⚙️ Custom: {self.action} {self.args}"
         action_str += _format_logs_for_display(self.logs, self.reasoning)
         return action_str
 

hud/adapters/operator/adapter.py

@@ -7,6 +7,7 @@ from hud.adapters.common.types import (
     CLAButton,
     CLAKey,
     ClickAction,
+    CustomAction,
     DragAction,
     MoveAction,
     Point,
@@ -27,6 +28,7 @@ class OperatorAdapter(Adapter):
         "arrowleft": "left",
         "arrowright": "right",
         "cmd": "ctrl",
+        "super": "win",
     }
 
     BUTTON_MAP: ClassVar[dict[str, CLAButton]] = {"wheel": "middle"}
@@ -92,6 +94,8 @@ class OperatorAdapter(Adapter):
 
             elif action_type == "response":
                 converted_action = ResponseAction(text=data.get("text", ""))
+            elif action_type == "custom":
+                converted_action = CustomAction(action=data.get("action", ""))
             else:
                 raise ValueError(f"Unsupported action type: {action_type}")
 

hud/agent/operator.py

@@ -92,6 +92,19 @@ class OperatorAgent(Agent[AsyncOpenAI, dict[str, Any]]):
         self.initial_prompt = None
         self.pending_safety_checks = []
 
+        self.base_system_prompt = """
+        You are an autonomous computer-using agent. Follow these guidelines:
+
+        1. Be decisive and complete tasks without asking for confirmation unless absolutely necessary.
+        2. If you need user confirmation for safety-critical actions, use the formal safety check mechanism.
+        3. Do NOT ask questions like "Should I proceed?" or "Would you like me to continue?" - just proceed with the task.
+        4. When you find what you're looking for (e.g., a file to upload), proceed with the action directly.
+        5. Only stop when the task is fully complete or if you encounter an error that prevents completion.
+        6. Trust that the user wants you to complete the entire task they've requested.
+
+        Remember: You wave been given permission to complete the requested task autonomously.
+        """
+
         self.task_run_id = None
 
     async def fetch_response(self, observation: Observation) -> tuple[list[dict[str, Any]], bool]:
@@ -145,6 +158,7 @@ class OperatorAgent(Agent[AsyncOpenAI, dict[str, Any]]):
                 model=self.model,
                 tools=[computer_tool],
                 input=input_param,
+                instructions=self.base_system_prompt,
                 truncation="auto",
                 reasoning={"summary": "auto"},
             )
@@ -225,9 +239,14 @@ class OperatorAgent(Agent[AsyncOpenAI, dict[str, Any]]):
 
             # If we found final text, package it as a 'response' action
             if final_text_response:
-                # No ResponseAgent logic here anymore - just return the response
-                actions = [{"type": "response", "text": final_text_response}]
-                done = True
+                if (
+                    "the task is infeasible" in final_text_response.lower()
+                ):  # Custom action for OSWorld
+                    done = True
+                    actions = [{"type": "custom", "action": "FAIL"}]
+                else:
+                    actions = [{"type": "response", "text": final_text_response}]
+                    done = True
             else:
                 logger.info("No computer calls and no final text message found.")
             # Keep done = True, actions remains empty

hud/job.py

@@ -274,11 +274,11 @@ async def _maybe_resample_action(
             try:
                 decision = await response_agent.determine_response(response_text)
                 if decision == "CONTINUE":
-                    logger.info("ResponseAgent indicated CONTINUE. Retrying...")
-                    obs.text = "Please continue."
+                    logger.info("ResponseAgent indicated CONTINUE. for message: %s", response_text)
+                    obs.text = "Yes, please continue."
                     return obs, False
-                elif decision == "CONTINUE":
-                    logger.warning("Max continue retries reached. Stopping despite CONTINUE.")
+                else:
+                    logger.warning("ResponseAgent indicated STOP for message: %s", response_text)
             except Exception as e:
                 logger.warning("Error using ResponseAgent: %s", e)
     return obs, True

hud/utils/tests/test_version.py

@@ -5,4 +5,4 @@ def test_import():
     """Test that the package can be imported."""
     import hud
 
-    assert hud.__version__ == "0.2.8"
+    assert hud.__version__ == "0.2.9"

hud/version.py

@@ -4,4 +4,4 @@ Version information for the HUD SDK.
 
 from __future__ import annotations
 
-__version__ = "0.2.8"
+__version__ = "0.2.9"

{hud_python-0.2.8.dist-info → hud_python-0.2.9.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hud-python
-Version: 0.2.8
+Version: 0.2.9
 Summary: SDK for the HUD evaluation platform.
 Project-URL: Homepage, https://github.com/hud-evals/hud-sdk
 Project-URL: Bug Tracker, https://github.com/hud-evals/hud-sdk/issues

{hud_python-0.2.8.dist-info → hud_python-0.2.9.dist-info}/RECORD

@@ -1,13 +1,13 @@
 hud/__init__.py,sha256=kjjq-l2msg9HcfYQ4sL8c0-StQIlsl2qLwh8Tx0nKro,1210
 hud/exceptions.py,sha256=pifKvSqxj9_g4NfARVyH5a-lTThhi9XW06tIXaBakQw,5526
 hud/gym.py,sha256=JNWlO2GXev0xIjoTI9HMEbcQgGpzc6fku7-RYoYAxHI,4996
-hud/job.py,sha256=_OKcdeWdoT7f3wiR7fZFjTVJs0OZCfMrxEG_cXYR6v4,26965
+hud/job.py,sha256=0vWbr3E5bYstVRzXS_6l-57JGUFcrZpmFrNkOSQ8Aa0,26969
 hud/settings.py,sha256=rx2zc3abJmf9ztwMHRYf9rGqgGprdRPCRhvJstsgyzc,1674
 hud/task.py,sha256=vDcjKUo8la0AUTP7mwMc2nYwe0tkbnrWwM9-Kvf3Ugg,8773
 hud/taskset.py,sha256=9IRwHeAdsk_IEibayM-hElE3gTp0mgmi-huN67h9-tc,7019
 hud/trajectory.py,sha256=ctAwrGIkdULr4xI6G-1Dp2fhDol4o_PmnPcqTzAEIUc,3797
 hud/types.py,sha256=xqrBb4rPKVkoLVwnyGk4PUrVKayCjKcUD_--n4OrxIM,2954
-hud/version.py,sha256=orhJ7MoupCr28yXngRez5EQfHbk4n6aHC0VjxM1WoGw,104
+hud/version.py,sha256=b76M28Br1tS7hHak1DIKBa5REyYNMWNZAmwUbRwGyuw,104
 hud/adapters/__init__.py,sha256=zz24KdC_e9TJPgWo6y57_8SzevEE5ak4Cm6tXzMxwRk,266
 hud/adapters/claude/__init__.py,sha256=i7QEF-29FLb9qxp1eYtXs-adIk_tG54tL-9g6d3xodk,100
 hud/adapters/claude/adapter.py,sha256=vCpotJ5gzQs4PP2iCXVavIcyG8c_4m1P6fuXStwUxSo,6675
@@ -15,11 +15,11 @@ hud/adapters/claude/tests/__init__.py,sha256=9GZj0rz4tTkiPnLfxTmyBPr-s8UZc3gph6W
 hud/adapters/claude/tests/test_adapter.py,sha256=cAdHEoqLngLiV7QwlWJ0KuNgb1vNv9WZTPQMnxhMDKI,18319
 hud/adapters/common/__init__.py,sha256=BjdZWJVs_AKtpFrt-tNsdQRjnz7D97DFEQirJ-r0mp8,118
 hud/adapters/common/adapter.py,sha256=GETzlsEl-uYkL-U4cQHBnfLAvm1dbXec4fKC2ypR1L0,5821
-hud/adapters/common/types.py,sha256=DpBu30o32tFEcTdMF8j-IKLN9cNRA9Luko8FKIB8K20,9904
+hud/adapters/common/types.py,sha256=6frue7_gZlSYtOHhF2tFHqzjltzzHsTVs6-H-jQwZ4Y,9955
 hud/adapters/common/tests/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 hud/adapters/common/tests/test_adapter.py,sha256=rTD36LjvytHqMIyOLDyrn0RLIkd20s6f6dwoBEarJaw,8744
 hud/adapters/operator/__init__.py,sha256=31vTRs268_TOLd-TeQRKau5bDYy78wxCNpJFhD5_l8U,104
-hud/adapters/operator/adapter.py,sha256=heATqYKrTZy4PIM22CfkhgIOPxugDpuF66wOQjZaaxE,3569
+hud/adapters/operator/adapter.py,sha256=Uz4Sr73T57B7v4RRP0uaibHI17N2hBx6Z9YYjgJCUXA,3732
 hud/adapters/operator/tests/__init__.py,sha256=yTsDVusVXZBQL6DnXpLgKQCBRuOYUAVQ8Blk_k5GETk,41
 hud/adapters/operator/tests/test_adapter.py,sha256=4RAXwyxAtkh-1Mlt1zJayRkcv3LWaPNEhDVTpwOZd4A,12942
 hud/agent/__init__.py,sha256=_OxMG3UW1vXSuixdpo09b1jexfWcUbfK44zto8t6_LE,453
@@ -27,7 +27,7 @@ hud/agent/base.py,sha256=hC3mVUMAWo5HHF2b576ScA9UQzsAzcCfPU9S8mDWthA,4080
 hud/agent/claude.py,sha256=FBSKCxICO6XXYCuIrerVL89bVJ-5JxrZJBDeZgzAdJI,9886
 hud/agent/claude_plays_pokemon.py,sha256=4TPibnTFhTb24ISRKAU3pA4waIcISTfZLOdfBMIMqxE,10085
 hud/agent/langchain.py,sha256=H55JNHcGkdl-LVzZEqOFRkuuFEO0D8MI1jCNz9deoko,9012
-hud/agent/operator.py,sha256=DDU2YOmNu00apt3W-k45Ybcl4lQ4vAw-v8NXv055Ut8,9387
+hud/agent/operator.py,sha256=kntMOsdL5tzaGVSnzbGvFD2PMLzW2DEB2wEqN_LArQw,10500
 hud/agent/misc/__init__.py,sha256=-ftYH1T5r7fXKKra6d8jXYmUz9KOTmYwBrPJU-V3S7g,71
 hud/agent/misc/response_agent.py,sha256=3PPsZqNAyUo2ouSV0ylGQj9fJqojfSB2roq2DadUdG0,3048
 hud/agent/tests/__init__.py,sha256=HbAW7FvSvzzKPU5LpveZceU8XTcDkRe1Bmte3OGi2f0,29
@@ -77,9 +77,9 @@ hud/utils/tests/test_common.py,sha256=KqDSMf7gWf1oYCiQ_BXsnvW1wUmyzbOzAT-HNoF7tx
 hud/utils/tests/test_config.py,sha256=dPlXYWuMrxX-NOYbf0vdJ27TJpfacKG8eiKOSGOcfDU,4079
 hud/utils/tests/test_progress.py,sha256=QunwDgi_heQXhDgmC25zgjr-sFUu5FdJ_1aYigMKeIc,6351
 hud/utils/tests/test_telemetry.py,sha256=t0An1RTBaE0dZVEpF4uwuq5k1R-PXFR5k4u71h60tx8,1224
-hud/utils/tests/test_version.py,sha256=koRqYkFbU5-8GDSXbxX3bghwIBRNoDmI4Mp1e0SFJmI,159
+hud/utils/tests/test_version.py,sha256=DJbzDv9gYYRyWl_73yv8kF7dIXz2xXkCn1QJ6eMO3Yk,159
 hud/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-hud_python-0.2.8.dist-info/METADATA,sha256=w8uLSDQlGIKEb_ILt9V2p0a2JcyBJQGdLCnqCAuKBB0,9785
-hud_python-0.2.8.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-hud_python-0.2.8.dist-info/licenses/LICENSE,sha256=yIzBheVUf86FC1bztAcr7RYWWNxyd3B-UJQ3uddg1HA,1078
-hud_python-0.2.8.dist-info/RECORD,,
+hud_python-0.2.9.dist-info/METADATA,sha256=Sy3WVlfOObkJPzwqmao2RHwfErWgPKn1j1zn92zsclQ,9785
+hud_python-0.2.9.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+hud_python-0.2.9.dist-info/licenses/LICENSE,sha256=yIzBheVUf86FC1bztAcr7RYWWNxyd3B-UJQ3uddg1HA,1078
+hud_python-0.2.9.dist-info/RECORD,,