verifiers 0.2.2.dev33__tar.gz → 0.2.2.dev35__tar.gz

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: verifiers
-Version: 0.2.2.dev33
+Version: 0.2.2.dev35
 Summary: Verifiers: Environments for LLM Reinforcement Learning
 Project-URL: Homepage, https://github.com/primeintellect-ai/verifiers
 Project-URL: Documentation, https://github.com/primeintellect-ai/verifiers

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/pyproject.toml

@@ -57,9 +57,9 @@ dependencies = [
 
 [dependency-groups]
 dev = [
-    "ruff>=0.9.0",
+    "ruff>=0.16.0",
     "pre-commit>=2.19.0",
-    "ty>=0.0.1a29,<0.0.22",
+    "ty>=0.0.63",
     "pytest>=7.0.0,<9.1.0",
     "pytest-asyncio>=0.21.0",
     "pytest-cov>=4.0.0",
@@ -135,6 +135,9 @@ scratchpad-v1 = { path = "environments/scratchpad_v1", editable = true }
 kuhn-poker-v1 = { path = "environments/kuhn_poker_v1", editable = true }
 
 [tool.uv.exclude-newer-package]
+# Bounded cutoffs for the explicitly requested tool upgrades.
+ruff = "2026-07-27"
+ty = "2026-07-27"
 # PrimeIntellect-published on PyPI (trusted publisher)
 prime-tunnel = false
 prime-sandboxes = false

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/conftest.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 """Pytest configuration and fixtures for verifiers tests."""
 
 import logging

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/test_browser_env.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 """Tests for BrowserEnv integration (browser_env, dom_mode, cua_mode).
 
 These tests verify the BrowserEnv class and its mode implementations

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/test_cli_agent_env.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 """Tests for CliAgentEnv and HarborEnv."""
 
 import asyncio

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/test_client_auth_errors.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 import httpx
 import pytest
 from anthropic import AuthenticationError as AnthropicAuthenticationError

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/test_composable_env.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 """Tests for the composable architecture: Task, TaskSet, SandboxTaskSet, SandboxSpec."""
 
 import importlib

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/test_context_token_metrics.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 """Tests for per-turn context token metrics.
 
 Tests the trajectory-based context token computation

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/test_decorator_ranks.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 """Tests for decorator priority ordering functionality."""
 
 import pytest

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/test_env_server.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 """Tests for the env server and client.
 
 Covers:

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/test_environment_extra.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 """Additional tests for verifiers.envs.environment.Environment.
 
 Covers:

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/test_gepa_cli.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 import argparse
 from pathlib import Path
 

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/test_imports.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 import verifiers
 
 

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/test_nemorl_client.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 from unittest.mock import AsyncMock, patch
 
 import pytest

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/test_openai_chat_completions_token_client.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 from typing import Any, cast
 
 import httpx

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/test_opencode_rlm_env.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 """Tests for the OpenCodeRLMEnv class."""
 
 import asyncio

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/test_openenv_client.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 from typing import Any
 
 import verifiers as vf

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/test_renderer_client.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 import asyncio
 from functools import lru_cache
 from unittest.mock import patch

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/test_renderer_e2e.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 """Tier-1 end-to-end tests for the renderer → RendererClient → Environment path.
 
 Exercises the full TITO wire protocol with a scripted stand-in for vLLM:

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/test_rlm_composable_env.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 """Tests for RLM harness integration with ComposableEnv.
 
 Validates that rlm_harness() produces a Harness with the correct metrics

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/test_rubric.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 """Tests for the Rubric class."""
 
 from typing import cast

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/test_sandbox_mixin.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 import asyncio
 from unittest.mock import AsyncMock, MagicMock, patch
 

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/test_save_utils.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 """Tests for verifiers.utils.save_utils serialization behavior.
 
 Covers:

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/test_tool_utils.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 """Tests for the tool_utils module."""
 
 from typing import Annotated, Optional

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/v1/fixtures/duet_v1.py

@@ -9,10 +9,10 @@ should land one record carrying two role-stamped traces.
 
 import asyncio
 
-import verifiers.v1 as vf
-
 from echo_v1 import EchoTaskset, lenient_match
 
+import verifiers.v1 as vf
+
 
 class DuetEnvConfig(vf.EnvConfig):
     # Both seats pin the lean `null` chat loop.
@@ -42,4 +42,4 @@ class DuetTaskset(EchoTaskset):
     pass
 
 
-__all__ = ["DuetTaskset", "DuetEnv"]
+__all__ = ["DuetEnv", "DuetTaskset"]

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/v1/fixtures/echo_acp_resume_v1.py

@@ -93,7 +93,7 @@ class ACPResumeTaskset(vf.Taskset[ACPResumeTask, ACPResumeConfig]):
         ]
 
 
-__all__ = ["ACPResumeTaskset", "ACPResumeEnv"]
+__all__ = ["ACPResumeEnv", "ACPResumeTaskset"]
 
 
 if __name__ == "__main__":

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/v1/fixtures/echo_user_sim_v1.py

@@ -62,4 +62,4 @@ class EchoUserSimTaskset(vf.Taskset[EchoUserSimTask, EchoUserSimConfig]):
         ]
 
 
-__all__ = ["EchoUserSimTaskset", "EchoUserSimEnv"]
+__all__ = ["EchoUserSimEnv", "EchoUserSimTaskset"]

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/v1/fixtures/echo_v1.py

@@ -6,6 +6,8 @@ the model's reply. It's a fixture taskset for the v1 e2e suite (in tests/v1/fixt
 by id `echo-v1` via pytest's `pythonpath`).
 """
 
+from pydantic import Field
+
 import verifiers.v1 as vf
 
 SYSTEM = "Repeat the user's message back to them exactly, with no extra words."
@@ -39,7 +41,9 @@ class EchoTask(vf.Task[EchoData]):
 
 
 class EchoConfig(vf.TasksetConfig):
-    phrases: list[str] = ["hello world", "ping", "verifiers"]
+    phrases: list[str] = Field(
+        default_factory=lambda: ["hello world", "ping", "verifiers"]
+    )
 
 
 class EchoTaskset(vf.Taskset[EchoTask, EchoConfig]):

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/v1/test_envs.py

@@ -72,7 +72,9 @@ def test_eval(taskset: str):
         "--sampling.max-tokens", "512", "--rich", "false",
     ]  # fmt: skip
     try:
-        proc = subprocess.run(cmd, capture_output=True, text=True, timeout=EVAL_TIMEOUT)
+        proc = subprocess.run(
+            cmd, capture_output=True, text=True, timeout=EVAL_TIMEOUT, check=False
+        )
     except subprocess.TimeoutExpired:
         pytest.fail(f"Timed out after {EVAL_TIMEOUT}s evaluating {taskset}")
     assert proc.returncode == 0, (

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/v1/test_judges.py

@@ -6,6 +6,7 @@ import json
 import re
 
 import pytest
+from pydantic import Field
 
 import verifiers.v1 as vf
 from verifiers.v1.graph import MessageNode
@@ -72,7 +73,9 @@ def fake_judge_model(monkeypatch):
                     "reason": "cites Paris" if "Paris" in text else "no Paris",
                     "verdict": "yes" if "Paris" in text else "no",
                 }
-                for name, text in re.findall(r"^- ([^:]+): (.+)$", messages, re.M)
+                for name, text in re.findall(
+                    r"^- ([^:]+): (.+)$", messages, re.MULTILINE
+                )
             ]
             response = JudgeResponse(
                 text=json.dumps({"verdicts": verdicts}),
@@ -167,7 +170,12 @@ def test_judges_reject_shared_reward_keys():
 
     # class-level DEFAULTS are held to the same rule (they bypass the before-hook)
     class TwoDefaults(vf.TaskConfig):
-        judges: vf.Judges = [vf.ReferenceJudgeConfig(), vf.ReferenceJudgeConfig()]
+        judges: vf.Judges = Field(
+            default_factory=lambda: [
+                vf.ReferenceJudgeConfig(),
+                vf.ReferenceJudgeConfig(),
+            ]
+        )
 
     with pytest.raises(ValueError, match="share a reward key"):
         TwoDefaults()
@@ -227,7 +235,8 @@ async def test_reference_score(fake_judge_model):
 
 async def test_reference_score_messages_prompt(fake_judge_model):
     # A Messages-form prompt still reaches the judge as text (via TaskData.prompt_text).
-    from verifiers.v1.types import TextContentPart, UserMessage as UM
+    from verifiers.v1.types import TextContentPart
+    from verifiers.v1.types import UserMessage as UM
 
     task = QAData(
         idx=0,
@@ -400,7 +409,7 @@ async def test_reference_empty_response_short_circuits(fake_judge_model):
 async def test_reference_list_answer(fake_judge_model):
     # A list-valued answer field is judged as multiple acceptable answers, one per line.
     class MultiTask(vf.TaskData):
-        aliases: list[str] = []
+        aliases: list[str] = Field(default_factory=list)
 
     task = MultiTask(idx=0, prompt="q?", aliases=["Paris", "Lutetia"])
     trace = make_trace()
@@ -569,11 +578,11 @@ async def test_rubric_off_menu_answer_raises(tmp_path, monkeypatch):
 
 def test_rubric_choices_validation(tmp_path):
     with pytest.raises(ValueError, match="at least two"):
-        rubric_judge(
+        _ = rubric_judge(
             tmp_path, body='[[criteria]]\nname = "x"\ntext = "t"\nchoices = ["only"]\n'
         ).criteria
     with pytest.raises(ValueError, match="duplicate options"):
-        rubric_judge(
+        _ = rubric_judge(
             tmp_path,
             body='[[criteria]]\nname = "x"\ntext = "t"\nchoices = ["a", "a"]\n',
         ).criteria

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/tests/v1/test_scoring.py

@@ -10,13 +10,11 @@ def test_compare_stdout_results_keeps_numeric_tolerance() -> None:
 
 
 def test_parse_pytest_outcomes_strips_xfail_xpass_reasons() -> None:
-    output = "\n".join(
-        [
-            "XFAIL tests/test_mod.py::test_xfail - known bug - still tracked",
-            "XPASS tests/test_mod.py::test_xpass - always xfail - unexpectedly passed",
-            "FAILED tests/test_mod.py::test_param[a - b] - assert left - right",
-            "PASSED tests/test_mod.py::test_ok",
-        ]
+    output = (
+        "XFAIL tests/test_mod.py::test_xfail - known bug - still tracked\n"
+        "XPASS tests/test_mod.py::test_xpass - always xfail - unexpectedly passed\n"
+        "FAILED tests/test_mod.py::test_param[a - b] - assert left - right\n"
+        "PASSED tests/test_mod.py::test_ok"
     )
 
     assert vf.parse_pytest_outcomes(output) == {

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/verifiers/__init__.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 from importlib.metadata import PackageNotFoundError
 from importlib.metadata import version as _version
 

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/verifiers/cli/commands/build.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 """Environment build command module for external hosts."""
 
 from verifiers.scripts.build import main

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/verifiers/cli/commands/eval.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 """Evaluation command module for external hosts."""
 
 from verifiers.scripts.eval import (

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/verifiers/cli/commands/gepa.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 """GEPA command module for external hosts."""
 
 from verifiers.scripts.gepa import main

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/verifiers/cli/commands/init.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 """Environment initialization command module for external hosts."""
 
 from verifiers.scripts.init import main

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/verifiers/cli/commands/install.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 """Environment install command module for external hosts."""
 
 from verifiers.scripts.install import main

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/verifiers/cli/plugins/prime.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 """Prime-hosted command plugin contract."""
 
 from dataclasses import dataclass

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/verifiers/clients/__init__.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 from verifiers.clients.anthropic_messages_client import AnthropicMessagesClient
 from verifiers.clients.client import Client
 from verifiers.clients.nemorl_chat_completions_client import (

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/verifiers/clients/anthropic_messages_client.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 import functools
 import json
 import time

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/verifiers/clients/client.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 import logging
 from abc import ABC, abstractmethod
 from typing import TYPE_CHECKING, Generic, TypeVar

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/verifiers/clients/nemorl_chat_completions_client.py

@@ -108,8 +108,8 @@ def _attach_trajectory_tokens_to_prompt(
             continue
         msg = prompt[i]
         if (prompt_ids := tokens.get("prompt_ids")) is not None:
-            msg.prompt_token_ids = list(prompt_ids)
+            msg.prompt_token_ids = list(prompt_ids)  # ty:ignore[invalid-assignment]
         if (completion_ids := tokens.get("completion_ids")) is not None:
-            msg.generation_token_ids = list(completion_ids)
+            msg.generation_token_ids = list(completion_ids)  # ty:ignore[invalid-assignment]
         if (completion_logprobs := tokens.get("completion_logprobs")) is not None:
-            msg.generation_log_probs = list(completion_logprobs)
+            msg.generation_log_probs = list(completion_logprobs)  # ty:ignore[invalid-assignment]

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/verifiers/clients/openai_chat_completions_client.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 import functools
 from collections.abc import Iterable, Mapping
 from typing import Any, TypeAlias, cast
@@ -231,7 +233,7 @@ class OpenAIChatCompletionsClient(
                     role="assistant",
                     content=cast(Any, normalize_content(message.content)),
                     tool_calls=cast(Any, oai_tool_calls),
-                    reasoning_content=message.reasoning_content,  # type: ignore[arg-type]
+                    reasoning_content=message.reasoning_content,  # type: ignore[arg-type]  # ty:ignore[invalid-key]
                 )
             elif isinstance(message, ToolMessage):
                 return ChatCompletionToolMessageParam(

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/verifiers/clients/openai_chat_completions_token_client.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 from collections.abc import Mapping
 from typing import Any, Optional, cast
 
@@ -288,7 +290,7 @@ class OpenAIChatCompletionsTokenClient(OpenAIChatCompletionsClient):
         if tool_call_ids:
             dummy_assistant: OpenAIChatMessage = ChatCompletionAssistantMessageParam(
                 role="assistant",
-                reasoning_content="",  # type: ignore[typeddict-unknown-key]
+                reasoning_content="",  # type: ignore[typeddict-unknown-key]  # ty:ignore[invalid-key]
                 tool_calls=[
                     ChatCompletionMessageFunctionToolCallParam(
                         id=tc_id,
@@ -301,7 +303,7 @@ class OpenAIChatCompletionsTokenClient(OpenAIChatCompletionsClient):
         else:
             dummy_assistant: OpenAIChatMessage = ChatCompletionAssistantMessageParam(
                 role="assistant",
-                reasoning_content="",  # type: ignore[typeddict-unknown-key]
+                reasoning_content="",  # type: ignore[typeddict-unknown-key]  # ty:ignore[invalid-key]
                 content="x",
             )
 

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/verifiers/clients/openai_completions_client.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 from openai import (
     AsyncOpenAI,
 )

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/verifiers/clients/openai_responses_client.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 import time
 from collections.abc import Iterable, Mapping
 from typing import Any, TypeAlias

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/verifiers/clients/renderer_client.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 """Renderer-based client.
 
 All tokenization happens client-side via a Renderer from the renderers package.

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/verifiers/decorators.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 import inspect
 from typing import Any, Callable, Literal, TypeVar, overload
 

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/verifiers/envs/env_group.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 import json
 from collections.abc import Mapping
 from typing import TYPE_CHECKING, Any, cast, final
@@ -292,7 +294,7 @@ class EnvGroup(vf.Environment):
         return dataset
 
     @final
-    async def run_rollout(  # type: ignore[override]
+    async def run_rollout(  # type: ignore[override]  # ty:ignore[override-of-final-method]
         self,
         input: RolloutInput,
         client: Client | ClientConfig,
@@ -328,10 +330,10 @@ class EnvGroup(vf.Environment):
             state_columns,
             env.env_client,
         )
-        return _set_info_route(output, route)  # type: ignore[return-value]
+        return _set_info_route(output, route)  # type: ignore[return-value]  # ty:ignore[invalid-return-type]
 
     @final
-    async def run_group(  # type: ignore[override]
+    async def run_group(  # type: ignore[override]  # ty:ignore[override-of-final-method]
         self,
         group_inputs: list[RolloutInput],
         client: Client | ClientConfig,
@@ -340,7 +342,7 @@ class EnvGroup(vf.Environment):
         max_retries: int = 0,
         state_columns: list[str] | None = None,
         env_client: EnvClient | None = None,
-    ) -> list[vf.RolloutOutput]:
+    ) -> list[vf.RolloutOutput]:  # ty:ignore[invalid-method-override]
         target_env_client = env_client or self.env_client
         if target_env_client is not None:
             if not isinstance(client, ClientConfig):
@@ -381,7 +383,7 @@ class EnvGroup(vf.Environment):
             state_columns,
             env.env_client,
         )
-        return [_set_info_route(output, route) for output in outputs]  # type: ignore[return-value]
+        return [_set_info_route(output, route) for output in outputs]  # type: ignore[return-value]  # ty:ignore[invalid-return-type]
 
     @final
     async def rollout(
@@ -415,7 +417,7 @@ class EnvGroup(vf.Environment):
             child_input["info"] = info
         else:
             child_input.pop("info", None)
-        return env_name, child_input, route  # type: ignore[return-value]
+        return env_name, child_input, route  # type: ignore[return-value]  # ty:ignore[invalid-return-type]
 
     def _input_env_route(self, input: RolloutInput) -> tuple[str, ...]:
         info = _info_dict(input.get("info"))

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/verifiers/envs/environment.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 import asyncio
 import atexit
 import json
@@ -160,7 +162,7 @@ class Environment(ABC):
 
         if dataset is not None:
             if callable(dataset):
-                self.dataset_source: DatasetBuilder | None = cast(
+                self.dataset_source: DatasetBuilder | None = cast(  # ty:ignore[unsupported-operator]
                     DatasetBuilder, dataset
                 )
             else:
@@ -171,7 +173,7 @@ class Environment(ABC):
 
         if eval_dataset is not None:
             if callable(eval_dataset):
-                self.eval_dataset_source: DatasetBuilder | None = cast(
+                self.eval_dataset_source: DatasetBuilder | None = cast(  # ty:ignore[unsupported-operator]
                     DatasetBuilder, eval_dataset
                 )
             else:
@@ -292,7 +294,7 @@ class Environment(ABC):
                 if few_shot:
                     messages.extend(few_shot)
                 messages.append({"role": "user", "content": prompt_str})
-                return messages
+                return messages  # ty:ignore[invalid-return-type]
 
             if answer_key == "answer":
                 dataset = dataset.map(

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/verifiers/envs/experimental/__init__.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 from verifiers.envs.experimental.sandbox_mixin import SandboxMixin
 
 __all__ = [

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/verifiers/envs/experimental/cli_agent_env.py

@@ -1,3 +1,5 @@
+# ruff: noqa
+
 import asyncio
 import logging
 import os

{verifiers-0.2.2.dev33 → verifiers-0.2.2.dev35}/verifiers/envs/experimental/composable/README.md

@@ -30,13 +30,13 @@ from verifiers.envs.experimental.composable.harnesses.opencode import opencode_h
 from verifiers.envs.experimental.composable import ComposableEnv
 
 # Create a taskset
-taskset = R2EGymTaskSet()                 # 4578 SWE instances
+taskset = R2EGymTaskSet()  # 4578 SWE instances
 
 # Explore instances
-task = taskset[0]                          # one Task
-task.prompt                                # the problem statement
-task.sandbox_spec                          # SandboxSpec(image=..., cpu=4, ...)
-task.sandbox_spec.image                    # per-instance docker image
+task = taskset[0]  # one Task
+task.prompt  # the problem statement
+task.sandbox_spec  # SandboxSpec(image=..., cpu=4, ...)
+task.sandbox_spec.image  # per-instance docker image
 
 # Slice and filter
 small = taskset.take(100)
@@ -46,9 +46,9 @@ filtered = taskset.filter(lambda ex: ...)
 results = await taskset.validate(
     concurrency=50,
     out_path="outputs/validate.jsonl",
-    max_retries=2,       # retry on vf.InfraError
+    max_retries=2,  # retry on vf.InfraError
     sandbox_client_max_workers=100,  # optional sandbox client worker cap override
-    resume=True,         # skip indices already in out_path
+    resume=True,  # skip indices already in out_path
 )
 
 # Run with an agent
@@ -135,7 +135,9 @@ class MySWERubric(vf.Rubric):
         sandbox_client = state["sandbox_client"]
         sandbox_id = state["sandbox_id"]
         # Run tests in the sandbox
-        test_output = await self.taskset._run_tests(sandbox_client, sandbox_id, state, ...)
+        test_output = await self.taskset._run_tests(
+            sandbox_client, sandbox_id, state, ...
+        )
         return float(self.taskset._calculate_reward(test_output, info))
 
     @vf.cleanup