inspect-ai 0.3.60__py3-none-any.whl → 0.3.61__py3-none-any.whl

inspect_ai/_cli/eval.py

@@ -592,7 +592,6 @@ def eval_set_command(
     logit_bias: str | None,
     seed: int | None,
     stop_seqs: str | None,
-    suffix: str | None,
     temperature: float | None,
     top_p: float | None,
     top_k: int | None,

inspect_ai/_eval/task/generate.py

@@ -8,6 +8,7 @@ from inspect_ai.model import (
 )
 from inspect_ai.model._cache import epoch
 from inspect_ai.solver import TaskState
+from inspect_ai.solver._limit import SampleLimitExceededError
 from inspect_ai.tool import ToolFunction
 
 
@@ -21,45 +22,50 @@ async def task_generate(
     # track tool_choice (revert to "auto" after first forced call of a tool)
     tool_choice = state.tool_choice
 
-    while True:
-        # If we don't update the epoch here as we go, it's entirely possible
-        # we'd cache the same response for every single epoch, which would
-        # completely defeat the point!
-        epoch.set(state.epoch)
+    try:
+        while True:
+            # If we don't update the epoch here as we go, it's entirely possible
+            # we'd cache the same response for every single epoch, which would
+            # completely defeat the point!
+            epoch.set(state.epoch)
 
-        # call the model
-        state.output = await model.generate(
-            input=state.messages,
-            tools=state.tools,
-            tool_choice=tool_choice,
-            config=config,
-            cache=cache,
-        )
+            # call the model
+            state.output = await model.generate(
+                input=state.messages,
+                tools=state.tools,
+                tool_choice=tool_choice,
+                config=config,
+                cache=cache,
+            )
 
-        # append the assistant message
-        message = state.output.message
-        state.messages.append(message)
+            # append the assistant message
+            message = state.output.message
+            state.messages.append(message)
 
-        # check for completed
-        if state.completed:
-            return state
+            # check for completed
+            if state.completed:
+                return state
 
-        # resolve tool calls if necessary
-        if tool_calls != "none" and message.tool_calls:
-            # call tools and append messages to state
-            state.messages.extend(
-                await call_tools(message, state.tools, config.max_tool_output)
-            )
+            # resolve tool calls if necessary
+            if tool_calls != "none" and message.tool_calls:
+                # call tools and append messages to state
+                state.messages.extend(
+                    await call_tools(message, state.tools, config.max_tool_output)
+                )
 
-            # check for completed or only executing a single tool call
-            if state.completed or tool_calls == "single":
-                return state
+                # check for completed or only executing a single tool call
+                if state.completed or tool_calls == "single":
+                    return state
+
+                # if a tool_call was forced set tool_choice to 'auto'
+                # (otherwise it will get forced over and over again)
+                if isinstance(tool_choice, ToolFunction):
+                    tool_choice = "auto"
 
-            # if a tool_call was forced set tool_choice to 'auto'
-            # (otherwise it will get forced over and over again)
-            if isinstance(tool_choice, ToolFunction):
-                tool_choice = "auto"
+            # no tool calls or not resolving tool calls, we are done!
+            else:
+                return state
 
-        # no tool calls or not resolving tool calls, we are done!
-        else:
-            return state
+    # propagate current state along with sample limit exceeded
+    except SampleLimitExceededError as ex:
+        raise ex.with_state(state)

inspect_ai/_eval/task/run.py

@@ -75,9 +75,9 @@ from inspect_ai.scorer._scorer import unique_scorer_name
 from inspect_ai.solver import Generate, Plan, TaskState
 from inspect_ai.solver._chain import Chain, unroll
 from inspect_ai.solver._fork import set_task_generate
+from inspect_ai.solver._limit import SampleLimitExceededError
 from inspect_ai.solver._solver import Solver
 from inspect_ai.solver._task_state import sample_state, set_sample_state, state_jsonable
-from inspect_ai.util._limit import SampleLimitExceededError
 from inspect_ai.util._sandbox.context import sandbox_connections
 from inspect_ai.util._sandbox.environment import SandboxEnvironmentSpec
 from inspect_ai.util._subtask import init_subtask
@@ -402,7 +402,13 @@ async def task_run(options: TaskRunOptions) -> EvalLog:
         view_notify_eval(logger.location)
 
         try:
-            await send_telemetry("eval_log", eval_log_json_str(eval_log))
+            if (
+                await send_telemetry("eval_log_location", eval_log.location)
+                == "not_handled"
+            ):
+                # Converting the eval log to JSON is expensive. Only do so if
+                # eval_log_location was not handled.
+                await send_telemetry("eval_log", eval_log_json_str(eval_log))
         except Exception as ex:
             py_logger.warning(
                 f"Error occurred sending telemetry: {exception_message(ex)}"
@@ -650,7 +656,7 @@ async def task_run_sample(
                     )
 
                     # capture most recent state for scoring
-                    state = sample_state() or state
+                    state = ex.state or sample_state() or state
                     state.completed = True
 
                 except BaseException as ex:
@@ -734,7 +740,7 @@ async def task_run_sample(
                     error = handle_error(ex)
 
         # handle sandboxenv init errors
-        except BaseException as ex:
+        except Exception as ex:
             error = handle_error(ex)
 
         # complete the sample

inspect_ai/_util/hooks.py

@@ -17,19 +17,29 @@ from .error import PrerequisiteError
 #
 # Telemetry can be optionally enabled by setting an INSPECT_TELEMETRY
 # environment variable that points to a function in a package which
-# conforms to the TelemetrySend signature below.
+# conforms to the TelemetrySend signature below. A return value of True
+# indicates that the telemetry event was handled.
 
-# There are currently two types of telemetry sent:
-#    - model_usage (type ModelUsage)
-#    - eval_log    (type EvalLog)
+# There are currently three types of telemetry sent:
+#    - model_usage       (JSON string of the model usage)
+#    - eval_log_location (file path or URL string of the eval log)
+#    - eval_log          (JSON string of the eval log)
+#                        [only sent if eval_log_location unhandled]
+# The eval_log_location type is preferred over eval_log as it means we can take
+# advantage of the .eval format and avoid loading the whole log into memory.
 
-TelemetrySend = Callable[[str, str], Awaitable[None]]
+TelemetrySend = Callable[[str, str], Awaitable[bool]]
 
 
-async def send_telemetry(type: Literal["model_usage", "eval_log"], json: str) -> None:
+async def send_telemetry(
+    type: Literal["model_usage", "eval_log", "eval_log_location"], json: str
+) -> Literal["handled", "not_handled", "no_subscribers"]:
     global _send_telemetry
     if _send_telemetry:
-        await _send_telemetry(type, json)
+        if await _send_telemetry(type, json):
+            return "handled"
+        return "not_handled"
+    return "no_subscribers"
 
 
 _send_telemetry: TelemetrySend | None = None

inspect_ai/_view/www/package.json

@@ -26,7 +26,7 @@
   },
   "dependencies": {
     "@popperjs/core": "^2.11.8",
-    "asciinema-player": "^3.8.1",
+    "asciinema-player": "^3.8.2",
     "bootstrap": "^5.3.3",
     "bootstrap-icons": "^1.11.3",
     "clipboard": "^2.0.11",

inspect_ai/model/_chat_message.py

@@ -72,8 +72,8 @@ class ChatMessageUser(ChatMessageBase):
     role: Literal["user"] = Field(default="user")
     """Conversation role."""
 
-    tool_call_id: str | None = Field(default=None)
-    """ID of tool call this message has the content payload for."""
+    tool_call_id: list[str] | None = Field(default=None)
+    """ID(s) of tool call(s) this message has the content payload for."""
 
 
 class ChatMessageAssistant(ChatMessageBase):

inspect_ai/model/_model.py

@@ -33,7 +33,6 @@ from inspect_ai._util.trace import trace_action
 from inspect_ai.tool import Tool, ToolChoice, ToolFunction, ToolInfo
 from inspect_ai.tool._tool_def import ToolDef, tool_defs
 from inspect_ai.util import concurrency
-from inspect_ai.util._limit import SampleLimitExceededError
 
 from ._cache import CacheEntry, CachePolicy, cache_fetch, cache_store
 from ._call_tools import disable_parallel_tools, tool_call_view, tools_info
@@ -764,40 +763,104 @@ def resolve_tool_model_input(
 def tool_result_images_as_user_message(
     messages: list[ChatMessage],
 ) -> list[ChatMessage]:
-    return functools.reduce(tool_result_images_reducer, messages, [])
+    """
+    To conform to models lacking support for images in tool responses, create an alternate message history that moves images into a fabricated user message.
+
+    Tool responses will have images replaced with "Image content is included below.", and the new user message will contain the images.
+    """
+    init_accum: ImagesAccumulator = ([], [], [])
+    chat_messages, user_message_content, tool_call_ids = functools.reduce(
+        tool_result_images_reducer, messages, init_accum
+    )
+    # if the last message was a tool result, we may need to flush the pending stuff here
+    return maybe_adding_user_message(chat_messages, user_message_content, tool_call_ids)
+
+
+ImagesAccumulator = tuple[list[ChatMessage], list[Content], list[str]]
+"""
+ImagesAccumulator is a tuple containing three lists:
+- The first list contains ChatMessages that are the result of processing.
+- The second list contains ContentImages that need to be inserted into a fabricated user message.
+- The third list contains the tool_call_id's associated with the tool responses.
+"""
 
 
 def tool_result_images_reducer(
-    messages: list[ChatMessage],
+    accum: ImagesAccumulator,
     message: ChatMessage,
-) -> list[ChatMessage]:
+) -> ImagesAccumulator:
+    messages, pending_content, tool_call_ids = accum
     # if there are tool result images, pull them out into a ChatUserMessage
-    if isinstance(message, ChatMessageTool) and isinstance(message.content, list):
-        tool_message = ChatMessageTool(
-            content=message.content.copy(),
-            tool_call_id=message.tool_call_id,
-            function=message.function,
+    if (
+        isinstance(message, ChatMessageTool)
+        and isinstance(message.content, list)
+        and any([isinstance(c, ContentImage) for c in message.content])
+    ):
+        init_accum: ImageContentAccumulator = ([], [])
+        new_user_message_content, edited_tool_message_content = functools.reduce(
+            tool_result_image_content_reducer, message.content, init_accum
         )
-        assert isinstance(tool_message.content, list)
-        messages.append(tool_message)
-
-        user_content: list[Content] = []
-        for i in range(0, len(tool_message.content)):
-            if isinstance(tool_message.content[i], ContentImage):
-                user_content.append(message.content[i])
-                tool_message.content[i] = ContentText(
-                    text="Image content is in the message below."
+
+        return (
+            messages
+            + [
+                ChatMessageTool(
+                    content=edited_tool_message_content,
+                    tool_call_id=message.tool_call_id,
+                    function=message.function,
                 )
-        if len(user_content) > 0:
-            messages.append(
-                ChatMessageUser(content=user_content, tool_call_id=message.tool_call_id)
-            )
+            ],
+            pending_content + new_user_message_content,
+            tool_call_ids + ([message.tool_call_id] if message.tool_call_id else []),
+        )
 
     else:
-        messages.append(message)
+        return (
+            maybe_adding_user_message(messages, pending_content, tool_call_ids)
+            + [message],
+            [],
+            [],
+        )
 
-    # return messages
-    return messages
+
+ImageContentAccumulator = tuple[list[Content], list[Content]]
+"""
+ImageContentAccumulator is a tuple containing two lists of Content objects:
+- The first list contains ContentImages that will be included in a fabricated user message.
+- The second list contains modified content for the tool message with images replaced with text.
+"""
+
+
+def tool_result_image_content_reducer(
+    acc: ImageContentAccumulator, content: Content
+) -> ImageContentAccumulator:
+    """
+    Reduces the messages Content into two separate lists: one for a fabricated user message that will contain the images and one for modified tool message with the images replaced with text.
+
+    Returns:
+      ImageContentReducer: A tuple containing two lists of Content objects.
+        - The first list contains the images that will be included in a fabricated user message.
+        - The second list contains modified content for the tool message with images replaced with text.
+    """
+    new_user_message_content, edited_tool_message_content = acc
+    if isinstance(content, ContentImage):
+        return new_user_message_content + [content], edited_tool_message_content + [
+            ContentText(text="Image content is included below.")
+        ]
+
+    else:
+        return new_user_message_content, edited_tool_message_content + [content]
+
+
+def maybe_adding_user_message(
+    messages: list[ChatMessage], content: list[Content], tool_call_ids: list[str]
+) -> list[ChatMessage]:
+    """If content is empty, return messages, otherwise, create a new ChatMessageUser with it and return a new messages list with that message added."""
+    return (
+        messages + [ChatMessageUser(content=content, tool_call_id=tool_call_ids)]
+        if content
+        else messages
+    )
 
 
 # Functions to reduce consecutive user messages to a single user message -> required for some models
@@ -884,6 +947,7 @@ def handle_sample_message_limit(input: str | list[ChatMessage]) -> None:
         active_sample_message_limit,
         set_active_sample_total_messages,
     )
+    from inspect_ai.solver._limit import SampleLimitExceededError
 
     total_messages = 1 if isinstance(input, str) else len(input)
     message_limit = active_sample_message_limit()
@@ -910,6 +974,7 @@ def record_model_usage(model: str, usage: ModelUsage) -> None:
         active_sample_token_limit,
         set_active_sample_total_tokens,
     )
+    from inspect_ai.solver._limit import SampleLimitExceededError
 
     # record usage
     set_model_usage(model, usage, sample_model_usage_context_var.get(None))

inspect_ai/model/_providers/anthropic.py

@@ -291,9 +291,6 @@ class AnthropicAPI(ModelAPI):
         elif "content filtering" in error:
             content = "Sorry, but I am unable to help with that request."
             stop_reason = "content_filter"
-        else:
-            content = error
-            stop_reason = "unknown"
 
         if content and stop_reason:
             return ModelOutput.from_content(

inspect_ai/solver/__init__.py

@@ -6,6 +6,7 @@ from ._chain import chain
 from ._critique import self_critique
 from ._fork import fork
 from ._human_agent.agent import human_agent
+from ._limit import SampleLimitExceededError
 from ._multiple_choice import MultipleChoiceTemplate, multiple_choice
 from ._plan import Plan, plan
 from ._prompt import chain_of_thought, prompt_template, system_message, user_message
@@ -37,6 +38,7 @@ __all__ = [
     "TaskState",
     "Generate",
     "MultipleChoiceTemplate",
+    "SampleLimitExceededError",
 ]
 
 

inspect_ai/solver/_basic_agent.py

@@ -13,6 +13,7 @@ from inspect_ai.solver._chain import chain
 from inspect_ai.tool._tool import Tool, ToolResult, tool
 from inspect_ai.tool._tool_with import tool_with
 
+from ._limit import SampleLimitExceededError
 from ._prompt import system_message
 from ._solver import Generate, Solver, solver
 from ._task_state import TaskState
@@ -167,61 +168,70 @@ def basic_agent(
             # track attempts
             attempts = 0
 
-            # main loop (state.completed checks message_limit and token_limit)
-            while not state.completed:
-                # generate output and append assistant message
-                state.output = await get_model().generate(
-                    input=state.messages, tools=state.tools, cache=cache
-                )
-                state.messages.append(state.output.message)
-
-                # check for context window overflow
-                if state.output.stop_reason == "model_length":
-                    from inspect_ai.log._transcript import transcript
-
-                    transcript().info("Agent terminated: model context window exceeded")
-                    break
-
-                # resolve tools calls (if any)
-                if state.output.message.tool_calls:
-                    # call tool functions
-                    tool_results = await call_tools(
-                        state.output.message, state.tools, max_output=max_tool_output
+            try:
+                # main loop (state.completed checks message_limit and token_limit)
+                while not state.completed:
+                    # generate output and append assistant message
+                    state.output = await get_model().generate(
+                        input=state.messages, tools=state.tools, cache=cache
                     )
-                    state.messages.extend(tool_results)
-
-                    # was an answer submitted?
-                    answer = submission(tool_results)
-                    if answer:
-                        # set the output to the answer for scoring
-                        state.output.completion = answer
-
-                        # exit if we are at max_attempts
-                        attempts += 1
-                        if attempts >= max_attempts:
-                            state.completed = True
-                            break
-
-                        # exit if the submission is successful
-                        answer_scores = await score(state)
-                        if score_value_fn(answer_scores[0].value) == 1.0:
-                            state.completed = True
-                            break
-
-                        # otherwise notify the model that it was incorrect and continue
-                        else:
-                            response_message = (
-                                incorrect_message(state, answer_scores)
-                                if callable(incorrect_message)
-                                else incorrect_message
-                            )
-                            state.messages.append(
-                                ChatMessageUser(content=response_message)
-                            )
-
-                # no tool calls, urge the model to continue
-                else:
-                    state.messages.append(ChatMessageUser(content=continue_message))
+                    state.messages.append(state.output.message)
+
+                    # check for context window overflow
+                    if state.output.stop_reason == "model_length":
+                        from inspect_ai.log._transcript import transcript
+
+                        transcript().info(
+                            "Agent terminated: model context window exceeded"
+                        )
+                        break
+
+                    # resolve tools calls (if any)
+                    if state.output.message.tool_calls:
+                        # call tool functions
+                        tool_results = await call_tools(
+                            state.output.message,
+                            state.tools,
+                            max_output=max_tool_output,
+                        )
+                        state.messages.extend(tool_results)
+
+                        # was an answer submitted?
+                        answer = submission(tool_results)
+                        if answer:
+                            # set the output to the answer for scoring
+                            state.output.completion = answer
+
+                            # exit if we are at max_attempts
+                            attempts += 1
+                            if attempts >= max_attempts:
+                                state.completed = True
+                                break
+
+                            # exit if the submission is successful
+                            answer_scores = await score(state)
+                            if score_value_fn(answer_scores[0].value) == 1.0:
+                                state.completed = True
+                                break
+
+                            # otherwise notify the model that it was incorrect and continue
+                            else:
+                                response_message = (
+                                    incorrect_message(state, answer_scores)
+                                    if callable(incorrect_message)
+                                    else incorrect_message
+                                )
+                                state.messages.append(
+                                    ChatMessageUser(content=response_message)
+                                )
+
+                    # no tool calls, urge the model to continue
+                    else:
+                        state.messages.append(ChatMessageUser(content=continue_message))
+
+            # propagate current state along with sample limit exceeded
+            except SampleLimitExceededError as ex:
+                raise ex.with_state(state)
 
             return state
 

inspect_ai/{util → solver}/_limit.py

@@ -1,5 +1,7 @@
 from typing import Literal
 
+from ._task_state import TaskState
+
 
 class SampleLimitExceededError(Exception):
     """Exception raised when a sample limit is exceeded.
@@ -18,9 +20,20 @@ class SampleLimitExceededError(Exception):
         value: int,
         limit: int,
         message: str | None = None,
+        state: TaskState | None = None,
     ) -> None:
         self.type = type
         self.value = value
         self.limit = limit
         self.message = f"Exceeded {type} limit: {limit:,}"
+        self.state = state
         super().__init__(message)
+
+    def with_state(self, state: TaskState) -> "SampleLimitExceededError":
+        return SampleLimitExceededError(
+            self.type,
+            value=self.value,
+            limit=self.limit,
+            message=self.message,
+            state=state,
+        )

inspect_ai/solver/_task_state.py

@@ -22,7 +22,6 @@ from inspect_ai.scorer._metric import Score
 from inspect_ai.scorer._target import Target
 from inspect_ai.tool import Tool, ToolChoice
 from inspect_ai.tool._tool_def import ToolDef
-from inspect_ai.util._limit import SampleLimitExceededError
 from inspect_ai.util._store import Store, store_jsonable
 from inspect_ai.util._store_model import SMT
 
@@ -173,7 +172,7 @@ class TaskState:
         self.metadata = metadata
         """Metadata from the `Sample` for this `TaskState`"""
 
-        self._messages: list[ChatMessage] = ChatMessageList(messages)
+        self._messages: list[ChatMessage] = ChatMessageList(messages, self)
         """
         Chat conversation history for sample.
 
@@ -272,7 +271,7 @@ class TaskState:
     @messages.setter
     def messages(self, messages: list[ChatMessage]) -> None:
         """Set messages in chat history."""
-        self._messages = ChatMessageList(messages)
+        self._messages = ChatMessageList(messages, self)
 
     @property
     def max_messages(self) -> int | None:
@@ -319,8 +318,32 @@ class TaskState:
 
     @property
     def completed(self) -> bool:
-        """Is the task completed."""
-        return self._completed
+        """Is the task completed.
+
+        Additionally, checks message and token limits and raises if they are exceeded.
+        """
+        from inspect_ai.log._samples import set_active_sample_total_messages
+
+        from ._limit import SampleLimitExceededError
+
+        # update messages
+        set_active_sample_total_messages(len(self.messages))
+
+        if self._completed:
+            return True
+        elif self.message_limit and len(self.messages) >= self.message_limit:
+            raise SampleLimitExceededError(
+                "message",
+                value=len(self.messages),
+                limit=self.message_limit,
+                state=self,
+            )
+        elif self.token_limit and self.token_usage >= self.token_limit:
+            raise SampleLimitExceededError(
+                "token", value=self.token_usage, limit=self.token_limit, state=self
+            )
+        else:
+            return self._completed
 
     @completed.setter
     def completed(self, completed: bool) -> None:
@@ -403,7 +426,8 @@ def sample_jsonable(sample: Sample) -> dict[str, Any]:
 
 
 class ChatMessageList(list[ChatMessage]):
-    def __init__(self, iterable: Iterable[ChatMessage]):
+    def __init__(self, iterable: Iterable[ChatMessage], parent_state: TaskState):
+        self.parent_state = parent_state
         items, length = self._iterable_length(iterable)
         self._check_size(length)
         super().__init__(items)
@@ -411,12 +435,18 @@ class ChatMessageList(list[ChatMessage]):
     def _check_size(self, additional_items: int = 1) -> None:
         from inspect_ai.log._samples import active_sample_message_limit
 
+        from ._limit import SampleLimitExceededError
+
         messages_limit = active_sample_message_limit()
         if messages_limit is not None:
             messages = len(self) + additional_items
             if messages > messages_limit:
                 raise SampleLimitExceededError(
-                    "message", value=messages, limit=messages_limit
+                    "message",
+                    value=messages,
+                    limit=messages_limit,
+                    message=None,
+                    state=self.parent_state,
                 )
 
     def append(self, item: ChatMessage) -> None:

inspect_ai/tool/_tools/_web_browser/_web_browser.py

@@ -345,7 +345,9 @@ async def web_browser_cmd(cmd: str, *args: str) -> str:
     if sandbox_env:
         store = store_as(WebBrowserStore)
         if not store.session_id:
-            result = await sandbox_env.exec(["python3", WEB_CLIENT_NEW_SESSION])
+            result = await sandbox_env.exec(
+                ["python3", WEB_CLIENT_NEW_SESSION], timeout=180
+            )
 
             if not result.success:
                 raise RuntimeError(

inspect_ai/tool/beta/_computer/_resources/Dockerfile

@@ -33,8 +33,6 @@ RUN apt-get update && \
 
 # Userland apt-get'able apps
 RUN apt-get install -y --no-install-recommends \
-      # A simple image viewer.
-      xpaint \
       # A calculator application.
       galculator && \
     apt-get clean
@@ -78,7 +76,7 @@ RUN useradd -m -s /bin/bash -d $HOME $USERNAME
 RUN echo "${USERNAME} ALL=(ALL) NOPASSWD: ALL" >> /etc/sudoers
 USER ${USERNAME}
 WORKDIR $HOME
-COPY --chown=$USERNAME:$USERNAME image_home_dir/ $HOME
+ADD --chown=$USERNAME:$USERNAME image_home_dir/ $HOME
 
 # configure Firefox to skip all 'first run' UI
 RUN mkdir -p $HOME/.mozilla/firefox-esr/profile.default && \

inspect_ai/tool/beta/_computer/_resources/entrypoint/x11vnc_startup.sh

@@ -5,7 +5,7 @@ echo "starting vnc"
     -forever \
     -shared \
     -wait 50 \
-    -cursor most \
+    -multiptr \
     -cursor arrow \
     -rfbport 5900 \
     -nopw \

inspect_ai/tool/beta/_computer/_resources/image_home_dir/.config/xfce4/xfconf/xfce-perchannel-xml/xfce4-screensaver.xml

@@ -0,0 +1,10 @@
+<?xml version="1.0" encoding="UTF-8"?>
+
+<channel name="xfce4-screensaver" version="1.0">
+  <property name="saver" type="empty">
+    <property name="mode" type="int" value="0" />
+  </property>
+  <property name="lock" type="empty">
+    <property name="enabled" type="bool" value="false" />
+  </property>
+</channel>

inspect_ai/util/__init__.py

@@ -3,7 +3,6 @@ from inspect_ai._util.trace import trace_action, trace_message
 from ._concurrency import concurrency
 from ._console import input_screen
 from ._display import DisplayType, display_type
-from ._limit import SampleLimitExceededError
 from ._panel import InputPanel, input_panel
 from ._resource import resource
 from ._sandbox import (
@@ -37,7 +36,6 @@ __all__ = [
     "input_panel",
     "input_screen",
     "OutputLimitExceededError",
-    "SampleLimitExceededError",
     "resource",
     "subprocess",
     "SandboxEnvironment",

inspect_ai/util/_sandbox/self_check.py

@@ -32,6 +32,7 @@ async def self_check(sandbox_env: SandboxEnvironment) -> dict[str, bool | str]:
     for fn in [
         test_read_and_write_file_text,
         test_read_and_write_file_binary,
+        test_write_file_text_utf,
         test_read_and_write_file_including_directory_absolute,
         test_read_and_write_file_including_directory_relative,
         test_read_file_zero_length,
@@ -64,33 +65,39 @@ async def self_check(sandbox_env: SandboxEnvironment) -> dict[str, bool | str]:
 
 
 async def _cleanup_file(sandbox_env: SandboxEnvironment, filename: str) -> None:
-    res = await sandbox_env.exec(["rm", filename])
+    res = await sandbox_env.exec(["rm", "-f", "--", filename])
     assert res.success
 
 
 async def test_read_and_write_file_text(sandbox_env: SandboxEnvironment) -> None:
-    await sandbox_env.write_file(
-        "test_read_and_write_file_text.file", "great #content\nincluding newlines"
-    )
-    written_file_string = await sandbox_env.read_file(
-        "test_read_and_write_file_text.file", text=True
-    )
+    file_name = "test_read_and_write_file_text.file"
+    await sandbox_env.write_file(file_name, "great #content\nincluding newlines")
+    written_file_string = await sandbox_env.read_file(file_name, text=True)
     assert "great #content\nincluding newlines" == written_file_string, (
         f"unexpected content: [{written_file_string}]"
     )
-    await _cleanup_file(sandbox_env, "test_read_and_write_file_text.file")
+    await _cleanup_file(sandbox_env, file_name)
+
+
+async def test_write_file_text_utf(sandbox_env: SandboxEnvironment) -> None:
+    utf_content = "✨☽︎✨🌞︎︎✨🚀✨"
+    file_name = "test_write_file_text_utf.file"
+    await sandbox_env.write_file(file_name, utf_content)
+    file_with_utf_content = await sandbox_env.read_file(file_name, text=True)
+    assert isinstance(file_with_utf_content, str)
+    assert file_with_utf_content == utf_content
+    await _cleanup_file(sandbox_env, file_name)
 
 
 async def test_read_and_write_file_binary(sandbox_env: SandboxEnvironment) -> None:
+    file_name = "test_read_and_write_file_binary.file"
     await sandbox_env.write_file(
-        "test_read_and_write_file_binary.file", b"\xc3\x28"
+        file_name, b"\xc3\x28"
     )  # invalid UTF-8 from https://stackoverflow.com/a/17199164/116509
 
-    written_file_bytes = await sandbox_env.read_file(
-        "test_read_and_write_file_binary.file", text=False
-    )
+    written_file_bytes = await sandbox_env.read_file(file_name, text=False)
     assert b"\xc3\x28" == written_file_bytes
-    await _cleanup_file(sandbox_env, "test_read_and_write_file_binary.file")
+    await _cleanup_file(sandbox_env, file_name)
 
 
 async def test_read_and_write_file_including_directory_absolute(
@@ -101,6 +108,7 @@ async def test_read_and_write_file_including_directory_absolute(
     written_file_string = await sandbox_env.read_file(file_name, text=True)
     assert "absolutely enjoying being in a directory" == written_file_string
     await _cleanup_file(sandbox_env, file_name)
+    await sandbox_env.exec(["rmdir", "/tmp/test_rw_including_directory_absolute"])
 
 
 async def test_read_and_write_file_including_directory_relative(
@@ -111,20 +119,23 @@ async def test_read_and_write_file_including_directory_relative(
     written_file_string = await sandbox_env.read_file(file_name, text=True)
     assert "relatively enjoying being in a directory" == written_file_string
     await _cleanup_file(sandbox_env, file_name)
+    await sandbox_env.exec(["rmdir", "test_rw_including_directory_relative"])
 
 
 async def test_read_file_zero_length(sandbox_env: SandboxEnvironment) -> None:
-    await sandbox_env.exec(["touch", "zero_length_file.file"])
-    zero_length = await sandbox_env.read_file("zero_length_file.file", text=True)
+    file_name = "zero_length_file.file"
+    await sandbox_env.exec(["touch", file_name])
+    zero_length = await sandbox_env.read_file(file_name, text=True)
     assert isinstance(zero_length, str)
     assert zero_length == ""
+    await _cleanup_file(sandbox_env, file_name)
 
 
 async def test_read_file_not_found(sandbox_env: SandboxEnvironment) -> None:
-    file = "nonexistent"
+    file_name = "nonexistent"
     with Raises(FileNotFoundError) as e_info:
-        await sandbox_env.read_file(file, text=True)
-    assert file in str(e_info.value)
+        await sandbox_env.read_file(file_name, text=True)
+    assert file_name in str(e_info.value)
 
 
 async def test_read_file_not_allowed(sandbox_env: SandboxEnvironment) -> None:
@@ -134,22 +145,23 @@ async def test_read_file_not_allowed(sandbox_env: SandboxEnvironment) -> None:
     with Raises(PermissionError) as e_info:
         await sandbox_env.read_file(file_name, text=True)
     assert file_name in str(e_info.value)
+    await sandbox_env.exec(["chmod", "+r", file_name])
     await _cleanup_file(sandbox_env, file_name)
 
 
 async def test_read_file_is_directory(sandbox_env: SandboxEnvironment) -> None:
-    file = "/etc"
+    file_name = "/etc"
     with Raises(IsADirectoryError) as e_info:
-        await sandbox_env.read_file(file, text=True)
+        await sandbox_env.read_file(file_name, text=True)
     assert "directory" in str(e_info.value)
 
 
 async def test_read_file_nonsense_name(
     sandbox_env: SandboxEnvironment,
 ) -> None:
-    file = "https:/en.wikipedia.org/wiki/Bart%C5%82omiej_Kasprzykowski"
+    file_name = "https:/en.wikipedia.org/wiki/Bart%C5%82omiej_Kasprzykowski"
     with Raises(FileNotFoundError) as e_info:
-        await sandbox_env.read_file(file, text=True)
+        await sandbox_env.read_file(file_name, text=True)
     assert "wikipedia" in str(e_info.value)
 
 
@@ -159,24 +171,28 @@ async def test_read_file_limit(sandbox_env: SandboxEnvironment) -> None:
     # Patch limit down to 1KiB for the test to save us from writing a 100 MiB file.
     with mock.patch.object(SandboxEnvironmentLimits, "MAX_READ_FILE_SIZE", 1024):
         with Raises(OutputLimitExceededError) as e_info:
-            await sandbox_env.read_file("large.file", text=True)
+            await sandbox_env.read_file(file_name, text=True)
         assert "limit of 100 MiB was exceeded" in str(e_info.value)
     await _cleanup_file(sandbox_env, file_name)
 
 
 async def test_write_file_zero_length(sandbox_env: SandboxEnvironment) -> None:
-    await sandbox_env.write_file("zero_length_file.file", "")
-    zero_length = await sandbox_env.read_file("zero_length_file.file", text=True)
+    file_name = "zero_length_file.file"
+    await sandbox_env.write_file(file_name, "")
+    zero_length = await sandbox_env.read_file(file_name, text=True)
     assert isinstance(zero_length, str)
     assert zero_length == ""
+    await _cleanup_file(sandbox_env, file_name)
 
 
 async def test_write_file_space(sandbox_env: SandboxEnvironment) -> None:
-    space = "✨☽︎✨🌞︎︎✨🚀✨"
-    await sandbox_env.write_file("file with space.file", space)
-    file_with_space = await sandbox_env.read_file("file with space.file", text=True)
+    space = "to the moon"
+    file_name = "file with space.file"
+    await sandbox_env.write_file(file_name, space)
+    file_with_space = await sandbox_env.read_file(file_name, text=True)
     assert isinstance(file_with_space, str)
     assert file_with_space == space
+    await _cleanup_file(sandbox_env, file_name)
 
 
 async def test_write_file_is_directory(
@@ -192,6 +208,9 @@ async def test_write_file_is_directory(
             "content cannot go in a directory, dummy",
         )
     assert "directory" in str(e_info.value)
+    await sandbox_env.exec(
+        ["rm", "-rf", "/tmp/inspect_ai_test_write_file_is_directory"]
+    )
 
 
 async def test_write_file_without_permissions(
@@ -203,6 +222,8 @@ async def test_write_file_without_permissions(
     with Raises(PermissionError) as e_info:
         await sandbox_env.write_file(file_name, "this won't stick")
     assert file_name in str(e_info.value)
+    await sandbox_env.exec(["chmod", "+w", file_name])
+    await _cleanup_file(sandbox_env, file_name)
 
 
 async def test_write_file_exists(
@@ -213,6 +234,7 @@ async def test_write_file_exists(
     await sandbox_env.write_file(file_name, "altered content")
     altered_content = await sandbox_env.read_file(file_name, text=True)
     assert altered_content == "altered content"
+    await _cleanup_file(sandbox_env, file_name)
 
 
 async def test_exec_output(sandbox_env: SandboxEnvironment) -> None:
@@ -305,6 +327,7 @@ async def test_cwd_absolute(sandbox_env: SandboxEnvironment) -> None:
     current_dir_contents = (await sandbox_env.exec(["ls"], cwd=cwd_directory)).stdout
     assert "test_cwd_absolute.file" in current_dir_contents
     await _cleanup_file(sandbox_env, file_name)
+    await sandbox_env.exec(["rmdir", cwd_directory])
 
 
 async def test_exec_stdout_is_limited(sandbox_env: SandboxEnvironment) -> None:

{inspect_ai-0.3.60.dist-info → inspect_ai-0.3.61.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: inspect_ai
-Version: 0.3.60
+Version: 0.3.61
 Summary: Framework for large language model evaluations
 Author: UK AI Safety Institute
 License: MIT License
@@ -68,7 +68,7 @@ Requires-Dist: pytest-asyncio; extra == "dev"
 Requires-Dist: pytest-cov; extra == "dev"
 Requires-Dist: pytest-dotenv; extra == "dev"
 Requires-Dist: pytest-xdist; extra == "dev"
-Requires-Dist: ruff==0.9.3; extra == "dev"
+Requires-Dist: ruff==0.9.4; extra == "dev"
 Requires-Dist: textual-dev>=0.86.2; extra == "dev"
 Requires-Dist: types-PyYAML; extra == "dev"
 Requires-Dist: types-beautifulsoup4; extra == "dev"

{inspect_ai-0.3.60.dist-info → inspect_ai-0.3.61.dist-info}/RECORD

@@ -3,7 +3,7 @@ inspect_ai/__main__.py,sha256=oWX4YwDZDg3GS3-IG0yPGoSEOfSzWihELg7QmrUlxjM,67
 inspect_ai/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 inspect_ai/_cli/cache.py,sha256=nOX9ysB3rZ-V8b_ryTpkgpoAynNlq4Op_fIqAIl4EVg,3910
 inspect_ai/_cli/common.py,sha256=5smokbnQGpOe72WXlpDy1QWuWbjxILKnl5UPvgfW0Yk,3678
-inspect_ai/_cli/eval.py,sha256=xxYGk5ewUUmGPIwJ1wp8TcCjxvttp5FxovhgsPfaFL0,31992
+inspect_ai/_cli/eval.py,sha256=LW2KH4iENl5QF10ItTwHzHM7v26RPS7-1XevaQjSa5E,31968
 inspect_ai/_cli/info.py,sha256=d5W7VA5buinGcsdQjWqlsMM6iSNNGRUHQrr4JS2k8nk,1749
 inspect_ai/_cli/list.py,sha256=GreVEhJRpagiCpzVc3FSGhcdpTq9B8Jh--mfgs4ueFQ,2454
 inspect_ai/_cli/log.py,sha256=boSzHZkiabhnYWHLRVsZVENCryG-MDaNHIIbpSp0Mcs,5729
@@ -55,11 +55,11 @@ inspect_ai/_eval/task/__init__.py,sha256=yhBZbjh67QfHy7IdyFGMyQIfBflQLPLkhmz5eEv
 inspect_ai/_eval/task/constants.py,sha256=quAKMw-4-3xKd1T_KwXCZvHYoKRXt1ZGuaHbBcWJwnA,72
 inspect_ai/_eval/task/epochs.py,sha256=Ci7T6CQniSOTChv5Im2dCdSDrP-5hq19rV6iJ2uBcH8,700
 inspect_ai/_eval/task/error.py,sha256=gJnd8X7LHpPz5zcOq_az6ONZICGJ0_VpSz9yhF0yRyY,1233
-inspect_ai/_eval/task/generate.py,sha256=Edm-_6Wp1mkb7XpGkfTAqobWPfjJeWB0sb8-76UjNrc,1999
+inspect_ai/_eval/task/generate.py,sha256=lm066fbZOX7o3NB57rbwwec-ZaIFE745fiuacPCmo20,2352
 inspect_ai/_eval/task/images.py,sha256=Tg3I7d7ThCYP_Lf-H5JA7xH-sH2W-m1c1YfswDwplt4,3949
 inspect_ai/_eval/task/log.py,sha256=TjyLglP-3gMMDkfmxYbsxTvGIWY9FEyVtB09Fyrm_PA,6166
 inspect_ai/_eval/task/results.py,sha256=GkWlgHexm0BPyxrmqLY6YSXz3AqXYMVLXNnKCYfW7ic,15785
-inspect_ai/_eval/task/run.py,sha256=rCV98hW3Pnz9Mu6L9h8QLs79Khy9D0lXiNyrR5qW1V4,34607
+inspect_ai/_eval/task/run.py,sha256=FdTysq-2dMFEXWIs7m8ONbXXDabx9V2saJecBurlQKs,34894
 inspect_ai/_eval/task/rundir.py,sha256=QXetLfqi1lRo-PcIuu7maQpVO57c2ebnsjfZk0lsAFk,2001
 inspect_ai/_eval/task/sandbox.py,sha256=kwG1SQawZFDle3hzqMe4hSdnuvShkKsmMIrcC2gnYHM,6120
 inspect_ai/_eval/task/task.py,sha256=ahlM-0MJc_4wFCWTGQIsnapUn0epka_9jRVK-xdapHY,13800
@@ -83,7 +83,7 @@ inspect_ai/_util/file.py,sha256=49NXD2R_qVDMScBfifiHKhQ6ypB4OyX6cA3ym1k0-1U,1226
 inspect_ai/_util/format.py,sha256=RWmK4JcB7NwRy4rXtUa1JJ52_KhxcvREhMMCFVHvzwQ,1179
 inspect_ai/_util/git.py,sha256=nHCtZMLjMyFjSC_9bksBXeFz4xqxZfY6lfXr_qg2n1E,760
 inspect_ai/_util/hash.py,sha256=N25e4B_Lp9ZFSQIrtMO4x-KrZopJL6gKhs-NO41pxzA,289
-inspect_ai/_util/hooks.py,sha256=eIvQCc3L3_sPUPHY2tsv20IanmvOvoa7cAaDuP_aiYI,3780
+inspect_ai/_util/hooks.py,sha256=8QnHCQQY_2XMYPkiPvixUgFY0E_niZvQhQDMI-eCdhM,4353
 inspect_ai/_util/html.py,sha256=X62FY8gpEJ2ZQoDu2y8aQAbiBUIHKsd7DA9rWCIleo8,168
 inspect_ai/_util/http.py,sha256=c4yvH48ZkETZ7sNDuNzBR0NUS4r-6WzCaolW9my13ns,3628
 inspect_ai/_util/images.py,sha256=W7QJHyzuXhfy3VsLhKTzddSo1g69O9RxnTyhat48Wyo,1312
@@ -127,7 +127,7 @@ inspect_ai/_view/www/favicon.svg,sha256=b9AHYZaO2zBzeKH6G4PwXZMGGW_UxY0omKHam-c9
 inspect_ai/_view/www/index.html,sha256=LDaPH75d5TlepHfIY3wQG0aBcHTpa90Bhx0er_ES45s,910
 inspect_ai/_view/www/jsconfig.json,sha256=vt1gPPYezOFeV9nofA93CmVJAKGb1QeKGuyvEn1CXgk,383
 inspect_ai/_view/www/log-schema.json,sha256=4PRhm3dJvhwPhEZCDUFMwzHcIsPNLGGPuuqXHIwSUxk,95006
-inspect_ai/_view/www/package.json,sha256=zQ4TrahqhBIsJAiiSQ--dW9KhH61-IBNTqtPRzS3tU4,1189
+inspect_ai/_view/www/package.json,sha256=y2cHvK7QKQcVk2v66ldn-syN649xnAjTVHju4QFJY2s,1189
 inspect_ai/_view/www/postcss.config.cjs,sha256=jQ-QIJFuBVUTZXbFpOvUJk4MsLBr_yPOQbRwHD0ZohE,340
 inspect_ai/_view/www/tsconfig.json,sha256=JjaLdpt13sgJYHDWdThRIr0gHzpRzEOKL4E2awt9r9s,277
 inspect_ai/_view/www/vite.config.js,sha256=jmSUrV0YzGCcinfyKcmy2bDRUE10mmPUxMAen0bX8jY,1103
@@ -304,17 +304,17 @@ inspect_ai/log/_recorders/recorder.py,sha256=yvW_D99QxUQmnF5EiGsWIVetBXdssMUcsq5
 inspect_ai/model/__init__.py,sha256=gYBnBjfEEG_GQhu_lhwQ2tW9U4nXLW0GtRJNGfwYy3k,2121
 inspect_ai/model/_cache.py,sha256=IQXhMas58Pjdma894GHGTtHYpmnf_Ojz_eE0kHaQVPs,13567
 inspect_ai/model/_call_tools.py,sha256=Vy3uXVpHY9b0gQrXu3KGmvEOWXntSCxbpJ0elTAQ0So,18128
-inspect_ai/model/_chat_message.py,sha256=21x9MJVyAzKM_XO72X6fG6Ei1Fy8xklSdAgdmDS_RLU,4442
+inspect_ai/model/_chat_message.py,sha256=rJ1_pBn0sQdsr4R_DQUrg2PvRzi4VaYc2N_ttikuFy0,4454
 inspect_ai/model/_conversation.py,sha256=SFumVE67sq-mgSfqaZw2xwE8kow5NxF6FU8VbXsvc8k,1988
 inspect_ai/model/_generate_config.py,sha256=WjlFH6WtfyIpF6TMcSFmIUxyyB0D4quZLIqMd82oEW8,8757
 inspect_ai/model/_image.py,sha256=kpO2Bn_-c-dK80HuPOPH1eSNmcoc39kofwf4yTTiTFE,477
-inspect_ai/model/_model.py,sha256=N8keDFLPXps-3O07GrPC1ZocjdOnaaNI6tbkUsN5clQ,34114
+inspect_ai/model/_model.py,sha256=9rLk3F33By0Gc8d7_LdS9r4sTth11QxC5tGu3m_4suo,37020
 inspect_ai/model/_model_call.py,sha256=r6ObHZwm7jy1qX2qxvAP6iPV3BhdGThL-VH-QEawQhA,2017
 inspect_ai/model/_model_output.py,sha256=X6dEH3L4XPs5H8cWQeVF8tlkDMNRqP3StJlPA_z140E,7029
 inspect_ai/model/_openai.py,sha256=XhYu_Rdc5jLGkrgdIkbniNWlQVBx9iYj2DdDTK1U12o,12871
 inspect_ai/model/_registry.py,sha256=Cr2y32EqLnOqLbSWoXHVK4ivTTzCUhJuACxoTyPt8kY,2032
 inspect_ai/model/_render.py,sha256=rWypNUjgrH4NGp0r-ESAze9gZz7lYNjheEP438vRYZE,922
-inspect_ai/model/_providers/anthropic.py,sha256=OJPjOutTTkgMU54bfEIRVbnweGa4UwXsZgYkacYt20M,24752
+inspect_ai/model/_providers/anthropic.py,sha256=4a07C0PGxWyyHGZuKol6EXULsDC7lphqyJgSdaOJ1gY,24674
 inspect_ai/model/_providers/azureai.py,sha256=moIC4-um_Qs3iXbr4DlP6LUL924aF-s5YyQqF0V5ye4,14037
 inspect_ai/model/_providers/bedrock.py,sha256=BiSEQYlGLKqaadGUJxSQuule3JPLZbAIjfhJ36DYQ8k,23357
 inspect_ai/model/_providers/cloudflare.py,sha256=h6ubjf0kxyMM7Aj2tm68tWa-2R7RAXNGp1O6KMvi0Gw,4143
@@ -359,16 +359,17 @@ inspect_ai/scorer/_reducer/__init__.py,sha256=ntoSXbbBia6gN3Uk3tQFQ8lSt8IBSRvwM5
 inspect_ai/scorer/_reducer/reducer.py,sha256=g8F7sTm_FmPcLdavOGv0YuvqZ5_nz2irmQVq37h2Y5A,11494
 inspect_ai/scorer/_reducer/registry.py,sha256=J2tvuuxf4jBC09_SCBZg99Qb2qQUWG8STEsw7ASWpXQ,5388
 inspect_ai/scorer/_reducer/types.py,sha256=uimvzIBRK7x1Dof77gsHYe9PU3hekB1opm9DTAa4sL4,340
-inspect_ai/solver/__init__.py,sha256=v3lps_q6SU4ZHklFjG-vgy92FcOK3jynk9zPs-nBwa4,3356
-inspect_ai/solver/_basic_agent.py,sha256=uJkjMsBP6SycnJxyXBOitU4AE8dBBCTKEbEZCz0NBuM,9607
+inspect_ai/solver/__init__.py,sha256=I8lmfnohTUYyW3aR7sCQhkOBh9R75a0-QrYkR3hG76E,3433
+inspect_ai/solver/_basic_agent.py,sha256=DaPMu_2SuoBamYwd54soxGaW2lcK21Siuf0qYW9Ps9w,10134
 inspect_ai/solver/_chain.py,sha256=F-2ZHE2KOlDAIgH1_Q23whUMH5TjYGvCHhcOgbRxe7I,2234
 inspect_ai/solver/_critique.py,sha256=ddO8J7VqSEsT6kofawpn3PrcUpLjLUMFmJi0hocDZpI,3504
 inspect_ai/solver/_fork.py,sha256=Ge1PwpCHjeZhm2CLAWKss2uFuQd9BGzVinLOW6UOnfE,2880
+inspect_ai/solver/_limit.py,sha256=zaZseJgjbJaBnGdXQHQ5MpU4tzgUyD8FzLvJMGDk3jA,1122
 inspect_ai/solver/_multiple_choice.py,sha256=tSLrwqAHuvX_eccM6OXiRmlx5bx_3g1LcB8GDWWV9C0,11024
 inspect_ai/solver/_plan.py,sha256=Dp1DDTtGe2iTo8CYWKqCOdfBFfTK_0wi2JzIr6qrikI,7042
 inspect_ai/solver/_prompt.py,sha256=PwGtLQ-dnCzxN_74H1NDT7LAhUuuiw2-c6ZSyXgBFgQ,3953
 inspect_ai/solver/_solver.py,sha256=Q-FrkfD97_TufEzuQxzr_LgziCdQipIuy778NWq7vVM,9008
-inspect_ai/solver/_task_state.py,sha256=D2rpC7lycJH601o6xHNrF3LIWgMGFaXkQ1b_38pF-2U,15169
+inspect_ai/solver/_task_state.py,sha256=Timv9_961yPNjh07BBUL0QeHeLLKx6b-QBsN1ocnEvY,16237
 inspect_ai/solver/_transcript.py,sha256=gkH9CC5gYbz7ZzrFD0TkjtKYjWxQP5EthJOkq8NXDOc,1049
 inspect_ai/solver/_use_tools.py,sha256=W7muO8r9eThXydm1GjFF-f6gip9AhzhgAG2GHSE5EpM,2011
 inspect_ai/solver/_util.py,sha256=pthrf-CzC6FnQYSUFLXTYM4wFEJptZrh5POTmV-Jtow,446
@@ -402,7 +403,7 @@ inspect_ai/tool/_tool_with.py,sha256=YBHjhT9PuM2QyUxD_BzhgqFPFfUPoRrTIpXMBXMXlFY
 inspect_ai/tool/_tools/_execute.py,sha256=DkFlvUTvI595H1zH5IKArhbyBo8YZWqq9tvoUMdvlaw,2823
 inspect_ai/tool/_tools/_web_search.py,sha256=YqZ3E65ssdq1X2NSH9Mqt5INXdPVQOdKa3PbKi7XjAY,7828
 inspect_ai/tool/_tools/_web_browser/__init__.py,sha256=dnnzy96pcvMvxD1OGg4hG-doL7Ru7WH0i25Sb9VIXwE,65
-inspect_ai/tool/_tools/_web_browser/_web_browser.py,sha256=PFBXaN18HNKslJzPlMgs_p2fNgPOClbjhYzD_qa_x9M,15541
+inspect_ai/tool/_tools/_web_browser/_web_browser.py,sha256=Bzb5RtxlQuxGgbkhxyaUTaj6VdNClgePBl5IuX2JJhY,15584
 inspect_ai/tool/_tools/_web_browser/_resources/Dockerfile,sha256=Sh1Ht5oBuxZC_8hLzw877CIvM9me_8Q0MxMemR5E_js,431
 inspect_ai/tool/_tools/_web_browser/_resources/README.md,sha256=RAMe6uFUYepkPSqpdCuag0nqASuFEONDI7jOHagYfuI,2607
 inspect_ai/tool/_tools/_web_browser/_resources/accessibility_node.py,sha256=PuOOeF5rDjN9tz-kRQ_UZUXj7MzrjwuFEdhVIYcCcQw,9628
@@ -423,16 +424,16 @@ inspect_ai/tool/beta/_computer/__init__.py,sha256=fq4BSM4aDhtEtE4279xm47NiO6vyiZ
 inspect_ai/tool/beta/_computer/_common.py,sha256=6XK6MBu2ZiRCao_eMlZdjXEvTmbeKQRN0K-8MtBPsk4,4059
 inspect_ai/tool/beta/_computer/_computer.py,sha256=2R-3GLoSvQn8b0rVPur3jMzaRK4nS6i_sDwzicj5XJ8,7433
 inspect_ai/tool/beta/_computer/_computer_split.py,sha256=H3DVCJqpHp_2ra85W_z9s5r-oHTVWwctuEq5fDdy2T4,5588
-inspect_ai/tool/beta/_computer/_resources/Dockerfile,sha256=CsmxeL8nO58fzKKpjFaAKiMVyMcZxWVbLUQsqXOvKNo,3545
+inspect_ai/tool/beta/_computer/_resources/Dockerfile,sha256=aLdS_UK2-bXiDQcfhGjVytPUQR1lb_WAQN8x4Ssa7_I,3498
 inspect_ai/tool/beta/_computer/_resources/README.md,sha256=5JDNaGJ-Ebq6Io57ANFIqgjPoh11aGDSrrgrhyfiqxU,1657
 inspect_ai/tool/beta/_computer/_resources/entrypoint/entrypoint.sh,sha256=IR8sE-b22YO7lwzdDiyjhLTJWIf0X__wA8WE98dwkwM,394
 inspect_ai/tool/beta/_computer/_resources/entrypoint/novnc_startup.sh,sha256=PAbMgSvprnLvbj8A8a59o_yDfm-jiCXxBxsPb004Bf8,383
-inspect_ai/tool/beta/_computer/_resources/entrypoint/x11vnc_startup.sh,sha256=NRTMEL5TTrEFjKK82ZMEqkHrhYIGIlos80CoElQgWoU,996
+inspect_ai/tool/beta/_computer/_resources/entrypoint/x11vnc_startup.sh,sha256=JFcW46u2ioDpGLptmUOMaqtt2YvuFhCTB42cyWRmo8c,993
 inspect_ai/tool/beta/_computer/_resources/entrypoint/xfce_startup.sh,sha256=w_27I4o7usP8SUMzP3lrXeNuISslyy1aywkgpm_2l4Q,209
 inspect_ai/tool/beta/_computer/_resources/entrypoint/xvfb_startup.sh,sha256=hd2naWFFpm3S0cWZ6Lhlpm6KD3L6-g8Zw2dgxchXMUg,1118
+inspect_ai/tool/beta/_computer/_resources/image_home_dir/.config/xfce4/xfconf/xfce-perchannel-xml/xfce4-screensaver.xml,sha256=jNgaNqBCngQlykTlLhmmdc_LLOrH2AMk7pUpLkbCjMY,312
 inspect_ai/tool/beta/_computer/_resources/image_home_dir/Desktop/Firefox Web Browser.desktop,sha256=Odm77RSEiTlMx7cY8odUCO2a8fvIUwHcpEUanpHzbL0,181
 inspect_ai/tool/beta/_computer/_resources/image_home_dir/Desktop/Visual Studio Code.desktop,sha256=jYYu8pcdIhFCC_3cEgO-0z0A6eQO2WQkIVViebSBbpA,184
-inspect_ai/tool/beta/_computer/_resources/image_home_dir/Desktop/XPaint.desktop,sha256=T093gZ3B2aXNd0yo6J31rJ0HeE3ROXPCbgAWxZqtjDA,158
 inspect_ai/tool/beta/_computer/_resources/tool/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 inspect_ai/tool/beta/_computer/_resources/tool/_logger.py,sha256=owkNYe9lyShTLXoMqhK4Qtzacnt5FvSxN8Wqf2MO5XM,652
 inspect_ai/tool/beta/_computer/_resources/tool/_run.py,sha256=xhXdnBK1di9muaO44CEirL9hpGy3NmKbjfMpyeVmn8Y,1595
@@ -440,12 +441,11 @@ inspect_ai/tool/beta/_computer/_resources/tool/_tool_result.py,sha256=cd6JNFhwyI
 inspect_ai/tool/beta/_computer/_resources/tool/_x11_client.py,sha256=rLduqd6JduoM9nMQk2-q7lpK4TCasz2F6_6mexquInI,9566
 inspect_ai/tool/beta/_computer/_resources/tool/computer_tool.py,sha256=0ehJuuUO6AlWUZKt3TyUbWQuwyBmkpsBbHxizZI_0GQ,2574
 inspect_ai/tool/beta/_computer/_resources/tool/requirements.txt,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-inspect_ai/util/__init__.py,sha256=4I9qA1900NIJNYjnGtHFHVoXkWv89pO6h10ePQ2Ohe8,1465
+inspect_ai/util/__init__.py,sha256=Nbr5h9EDqnUFqj1SSm5hJccHp_sz2YB1SCZgFS0NYDk,1388
 inspect_ai/util/_concurrency.py,sha256=Olzk259NPeSXIy5LzID_WEVGnaW6p5CBCBrmlZUYufM,2227
 inspect_ai/util/_console.py,sha256=V1XkIoKcNZo0SgRUOv15zJAWz6-zV6267hC4Oldj8oY,1237
 inspect_ai/util/_conversation.py,sha256=KzqvKfj1tB14cgARZjYyIVG2EpuE-EZKqLGAPIXv1Xs,784
 inspect_ai/util/_display.py,sha256=IUVyzS0PtFo9LeRW_EWvDv7tkGy1rsZGBjqg63uOPOs,1591
-inspect_ai/util/_limit.py,sha256=HMgembPprMvJFeFQy82Gw_BkPX4mqYBP1mGu-aA0n5g,751
 inspect_ai/util/_panel.py,sha256=bSXXV1LDVMt8DeGWEYTfEm3iMz9I02sX5xpBSVUVRdQ,3072
 inspect_ai/util/_resource.py,sha256=MMmtTKtt78pDIp9Uc_OxJom_q8mcKozVqt8kosKRJt0,3420
 inspect_ai/util/_store.py,sha256=QemJe2M-RK6zSFNcd07_92XFjvNtWKgHzBr5eT3KF1I,3786
@@ -459,7 +459,7 @@ inspect_ai/util/_sandbox/environment.py,sha256=Qo7ne28L6fn3igo2Gd0H1lz4vP60IdJGS
 inspect_ai/util/_sandbox/limits.py,sha256=K-GjKfSugOq8KP0wW_oF6qFrXsOnMV0C88QUWkjPJ9o,2164
 inspect_ai/util/_sandbox/local.py,sha256=NkHnR_e7s7RFsBdwfaSR7Yzp6lSUc7Em0Pc9_CFuN4c,3534
 inspect_ai/util/_sandbox/registry.py,sha256=mQwWwqzaCXF1FZ2fcVujpp3WMA35GWnh1w43SoIJAVM,2145
-inspect_ai/util/_sandbox/self_check.py,sha256=o-5skGZzzT1HCh9R6gf_D9J-PmCl0VRbriX4rqUjZ60,14101
+inspect_ai/util/_sandbox/self_check.py,sha256=ZRb2wMRnyiUBJPJqTfLlUO2_ctxBoJ-4soyERfn583A,15222
 inspect_ai/util/_sandbox/service.py,sha256=2os7W8NYBDcaBoaHVfZ1YrI9hvldksmiwqkUYrCRCPo,11258
 inspect_ai/util/_sandbox/docker/cleanup.py,sha256=MK6UlADcWtTDotppeVJga2ibf9Ud-e4V-5ReoNbmhqg,4793
 inspect_ai/util/_sandbox/docker/compose.py,sha256=4aIWWTaTUY9ZWrfSynkRqrUbKlYWrRYoSDX9WrjdHFQ,11473
@@ -468,9 +468,9 @@ inspect_ai/util/_sandbox/docker/docker.py,sha256=sx4PNv_4PDuKlkeYV6ASaZbo0XT-I_V
 inspect_ai/util/_sandbox/docker/internal.py,sha256=fATyk2pdtjSl-D0VPT4dmkXV-gOc5HrPH0EQDW4IAJY,1446
 inspect_ai/util/_sandbox/docker/prereqs.py,sha256=0j6_OauBBnVlpBleADcZavIAAQZy4WewVjbRn9c0stg,3355
 inspect_ai/util/_sandbox/docker/util.py,sha256=pSPsRGymrTmTnEUHiHoQSNqeurPP1mL5kB-105O6EWo,2794
-inspect_ai-0.3.60.dist-info/LICENSE,sha256=aYPffOl9TwBXDQ8g33Jh6AsBhobb3A76qNm7r2HZsps,1079
-inspect_ai-0.3.60.dist-info/METADATA,sha256=xzklRJs-p-vYq21ksrYIVA2XCUvMVksqx5Gvtdpz6T8,4528
-inspect_ai-0.3.60.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
-inspect_ai-0.3.60.dist-info/entry_points.txt,sha256=WGGLmzTzDWLzYfiyovSY6oEKuf-gqzSDNOb5V-hk3fM,54
-inspect_ai-0.3.60.dist-info/top_level.txt,sha256=Tp3za30CHXJEKLk8xLe9qGsW4pBzJpEIOMHOHNCXiVo,11
-inspect_ai-0.3.60.dist-info/RECORD,,
+inspect_ai-0.3.61.dist-info/LICENSE,sha256=aYPffOl9TwBXDQ8g33Jh6AsBhobb3A76qNm7r2HZsps,1079
+inspect_ai-0.3.61.dist-info/METADATA,sha256=8fUvCJ8CAPh0ANPl4OPuHGBmbhbewMoH6F1AA94SRTA,4528
+inspect_ai-0.3.61.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
+inspect_ai-0.3.61.dist-info/entry_points.txt,sha256=WGGLmzTzDWLzYfiyovSY6oEKuf-gqzSDNOb5V-hk3fM,54
+inspect_ai-0.3.61.dist-info/top_level.txt,sha256=Tp3za30CHXJEKLk8xLe9qGsW4pBzJpEIOMHOHNCXiVo,11
+inspect_ai-0.3.61.dist-info/RECORD,,

inspect_ai/tool/beta/_computer/_resources/image_home_dir/Desktop/XPaint.desktop

@@ -1,10 +0,0 @@
-[Desktop Entry]
-Version=1.0
-Type=Application
-Name=XPaint
-Comment=Xpaint painting application
-Exec=xpaint
-Icon=xpaint
-Path=
-Terminal=false
-StartupNotify=false