verifiers 0.3.2.dev141__tar.gz → 0.3.2.dev143__tar.gz

{verifiers-0.3.2.dev141 → verifiers-0.3.2.dev143}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: verifiers
-Version: 0.3.2.dev141
+Version: 0.3.2.dev143
 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-0.3.2.dev141 → verifiers-0.3.2.dev143}/tests/v1/fixtures/echo_acp_resume_v1.py

@@ -52,9 +52,12 @@ class ACPResumeTask(vf.Task[vf.TaskData, vf.State, ACPResumeTaskConfig]):
 class ACPResumeEnv(vf.SingleAgentEnv):
     async def run(self, task, agents):
         async with agents.agent.interaction(task) as interaction:
+            # Stated as a fact, not a "do not" instruction: OpenAI's content filter
+            # refuses "remember ... do not call tools or store it" as an injection.
             first = await interaction.turn(
-                f"Remember the codeword {CODEWORD} for this conversation only. "
-                "Do not call tools or store it in persistent memory. Reply with exactly READY."
+                f"The codeword for this conversation is {CODEWORD}. Keep it in the "
+                "conversation context only, without tools or persistent memory. "
+                "Reply with exactly READY."
             )
             segments = [first]
             if not first.terminated:

{verifiers-0.3.2.dev141 → verifiers-0.3.2.dev143}/tests/v1/fixtures/echo_v1.py

@@ -54,13 +54,15 @@ class EchoTaskset(vf.Taskset[EchoTask, EchoConfig]):
     def load(self) -> list[EchoTask]:
         return [
             # Keep coding-agent harnesses on the direct-response path instead of
-            # spending this single-turn smoke task on a tool call.
+            # spending this single-turn smoke task on a tool call. Phrased as an
+            # instruction rather than a prohibition: OpenAI's content filter refuses
+            # "Do not call tools ... include this exact phrase" as an injection.
             EchoTask(
                 EchoData(
                     idx=i,
                     prompt=(
-                        "Do not call tools or execute code. Reply immediately and include "
-                        f"this exact phrase in your final response: {phrase}"
+                        "Answer directly without using tools or running code. Your reply "
+                        f"must contain this exact phrase: {phrase}"
                     ),
                     system_prompt=SYSTEM,
                     answer=phrase,

{verifiers-0.3.2.dev141 → verifiers-0.3.2.dev143}/verifiers/v1/__init__.py

@@ -14,8 +14,12 @@ from verifiers.v1.clients import (
     resolve_client,
 )
 from verifiers.v1.configs.agent import AgentConfig
-from verifiers.v1.configs.cli.env import narrowed_env_annotation, resolve_env_field
-from verifiers.v1.configs.env import EnvConfig, default_agent_harness
+from verifiers.v1.configs.cli.env import (
+    merge_env_defaults,
+    narrowed_env_annotation,
+    resolve_env_field,
+)
+from verifiers.v1.configs.env import EnvConfig, SharedEnvConfig, default_agent_harness
 from verifiers.v1.configs.harness import HarnessConfig
 from verifiers.v1.configs.judge import JudgeConfig, Judges
 from verifiers.v1.configs.retries import RetryConfig
@@ -30,7 +34,7 @@ from verifiers.v1.configs.task import (
     RewardFunctionConfig,
     TaskConfig,
 )
-from verifiers.v1.configs.taskset import TasksetConfig
+from verifiers.v1.configs.taskset import SharedTasksetConfig, TasksetConfig
 from verifiers.v1.env import Env
 from verifiers.v1.envs.single_agent import SingleAgentEnv, SingleAgentEnvConfig
 from verifiers.v1.episode import (
@@ -286,6 +290,7 @@ __all__ = [  # noqa: RUF022 - grouped by public API area
     "Taskset",
     "TaskConfig",
     "TasksetConfig",
+    "SharedTasksetConfig",
     "DecoratedFunctionConfig",
     "RewardFunctionConfig",
     "BaseConfig",
@@ -309,8 +314,10 @@ __all__ = [  # noqa: RUF022 - grouped by public API area
     "Env",
     "SingleAgentEnv",
     "EnvConfig",
+    "SharedEnvConfig",
     "ServeConfig",
     "resolve_env_field",
+    "merge_env_defaults",
     "narrowed_env_annotation",
     "SingleAgentEnvConfig",
     "AgentConfig",

{verifiers-0.3.2.dev141 → verifiers-0.3.2.dev143}/verifiers/v1/configs/cli/env.py

@@ -1,5 +1,5 @@
 """Run-config plumbing around the `[env]` block: narrowing the `env` field of
-every config that owns one.
+every config that owns one, and layering shared env defaults under it.
 
 A run composes the blocks it needs — `[env]` (what runs, `configs/env.py`),
 `[serve]` (how it's hosted, `configs/serve.py`) — plus its own fields. Nothing here is a base class: the eval
@@ -10,11 +10,11 @@ single_agent_env_config)`. The `SerializeAsAny` is load-bearing: pydantic serial
 by declared type, so a plain `EnvConfig` silently drops a narrowed subclass's agents
 and knobs from `model_dump()` — the env-server wire's payload."""
 
-from pydantic import ValidationError
+from pydantic import BaseModel, ValidationError
 from pydantic_config import BaseConfig
 
-from verifiers.v1.configs.env import EnvConfig
-from verifiers.v1.utils.generic import prefix_validation_error
+from verifiers.v1.configs.env import EnvConfig, SharedEnvConfig
+from verifiers.v1.utils.generic import deep_merge, prefix_validation_error
 
 
 def resolve_env_field(data: dict, narrowed: "type[EnvConfig] | None" = None) -> dict:
@@ -45,6 +45,32 @@ def resolve_env_field(data: dict, narrowed: "type[EnvConfig] | None" = None) ->
     return data
 
 
+def merge_env_defaults(defaults: SharedEnvConfig, env: dict | None) -> dict:
+    """A raw `env` block over `defaults`, the knobs that several envs of one run share
+    (e.g. the retries of every eval source). Only the fields set in `defaults` apply,
+    and the env's own values win. A subtree whose `id`/`type` differs from the
+    default's is the env's alone, so one plugin's knobs never leak into another's."""
+    return deep_merge(_dump_set(defaults), env or {})
+
+
+def _dump_set(config: BaseModel) -> dict:
+    """`config`'s set fields, including those set on a sub-config after construction
+    (which leaves the parent's field unmarked). Each dumped sub-config keeps its
+    `id`/`type` even at the default: `deep_merge` detects a plugin switch only when
+    both sides name it."""
+    dump = {}
+    for name in type(config).model_fields:
+        value = getattr(config, name)
+        if isinstance(value, BaseModel):
+            nested = _dump_set(value)
+            if nested or name in config.model_fields_set:
+                keys = [k for k in ("id", "type") if k in type(value).model_fields]
+                dump[name] = {**{k: getattr(value, k) for k in keys}, **nested}
+        elif name in config.model_fields_set:
+            dump[name] = config.model_dump(include={name})[name]
+    return dump
+
+
 def narrowed_env_annotation(cls) -> "type[EnvConfig] | None":
     """The env field's annotation when the CLI pre-narrowed it (`narrow_config`
     swaps in a concrete subclass). The base declaration reads as `EnvConfig` itself

{verifiers-0.3.2.dev141 → verifiers-0.3.2.dev143}/verifiers/v1/configs/env.py

@@ -1,5 +1,6 @@
 """An environment's config — the run's single `[env]` block: the seed taskset,
-each role as an `AgentConfig` field, and the env-level knobs."""
+each role as an `AgentConfig` field, and the env-level knobs. `SharedEnvConfig` is
+the part every env has."""
 
 from typing import get_args
 
@@ -13,7 +14,7 @@ from verifiers.v1.configs.agent import (
 )
 from verifiers.v1.configs.harness import HarnessConfig
 from verifiers.v1.configs.retries import RetryConfig
-from verifiers.v1.configs.taskset import TasksetConfig
+from verifiers.v1.configs.taskset import SharedTasksetConfig, TasksetConfig
 from verifiers.v1.interception import ElasticInterceptionPoolConfig, InterceptionConfig
 from verifiers.v1.types import ID
 
@@ -34,18 +35,12 @@ def _mentions_agent_config(annotation) -> bool:
     return any(_mentions_agent_config(arg) for arg in get_args(annotation))
 
 
-class EnvConfig(BaseConfig):
-    """An environment's config — the run's single `[env]` block, one subclass per
-    `Env` class (bound via `Env[YourConfig]`): each role is an
-    `AgentConfig` field with a default instance, plus env-level knobs. The run's `env`
-    field narrows to it by env `id` (else taskset id) — `--env.<role>.model` addressing."""
+class SharedEnvConfig(BaseConfig):
+    """The env knobs every env has, whatever its id and agents — what several envs of
+    one run can share. `EnvConfig` adds the env `id`, the taskset `id`, and the agents."""
 
-    id: ID = ""
-    """Which `Env` runs. Empty = the taskset's own, else `SingleAgentEnv`; set
-    to pair a reusable env with any taskset (an explicit id wins over the bundled)."""
-    # SerializeAsAny: the env-server wire needs the resolved subclass's fields.
-    taskset: SerializeAsAny[TasksetConfig] = TasksetConfig()
-    """The seed taskset — the rows every rollout starts from (`--env.taskset.id`)."""
+    taskset: SharedTasksetConfig = SharedTasksetConfig()
+    """The taskset knobs every taskset has."""
     timeout: TimeoutConfig = TimeoutConfig()
     retries: RetryConfig = RetryConfig()
     """Whole-EPISODE retries — the coarse fallback for faults no agent owns; a
@@ -63,6 +58,20 @@ class EnvConfig(BaseConfig):
     interception: InterceptionConfig = ElasticInterceptionPoolConfig()
     """The interception shape: `elastic` (default), `server`, or `static`."""
 
+
+class EnvConfig(SharedEnvConfig):
+    """An environment's config — the run's single `[env]` block, one subclass per
+    `Env` class (bound via `Env[YourConfig]`): each role is an
+    `AgentConfig` field with a default instance, plus env-level knobs. The run's `env`
+    field narrows to it by env `id` (else taskset id) — `--env.<role>.model` addressing."""
+
+    id: ID = ""
+    """Which `Env` runs. Empty = the taskset's own, else `SingleAgentEnv`; set
+    to pair a reusable env with any taskset (an explicit id wins over the bundled)."""
+    # SerializeAsAny: the env-server wire needs the resolved subclass's fields.
+    taskset: SerializeAsAny[TasksetConfig] = TasksetConfig()
+    """The seed taskset — the rows every rollout starts from (`--env.taskset.id`)."""
+
     @property
     def env_id(self) -> str:
         """The taskset id, prefixed by the paired env id (`best-of-n+gsm8k-v1`)."""

{verifiers-0.3.2.dev141 → verifiers-0.3.2.dev143}/verifiers/v1/configs/taskset.py

@@ -9,16 +9,22 @@ from verifiers.v1.configs.task import TaskConfig
 from verifiers.v1.types import ID
 
 
-class TasksetConfig(BaseConfig):
-    id: ID = ""
-    """Installed taskset package, set with `--env.taskset.id` (or the
-    positional `eval <taskset-id>`)."""
+class SharedTasksetConfig(BaseConfig):
+    """The knobs every taskset has, whatever its id — what several tasksets of one run
+    can share."""
+
     task: SerializeAsAny[TaskConfig] = TaskConfig()
     """Config passed to each task, under `--env.taskset.task.*`."""
     system_prompt: Path | None = None
     """File whose text overrides each task's `TaskData.system_prompt` on
     iteration (e.g. a GEPA `best_system_prompt.txt`)."""
 
+
+class TasksetConfig(SharedTasksetConfig):
+    id: ID = ""
+    """Installed taskset package, set with `--env.taskset.id` (or the
+    positional `eval <taskset-id>`)."""
+
     @property
     def name(self) -> str:
         return self.id

{verifiers-0.3.2.dev141 → verifiers-0.3.2.dev143}/verifiers/v1/harnesses/codex/harness.py

@@ -61,14 +61,15 @@ class CodexHarness(ACPHarness[CodexHarnessConfig]):
     SUPPORTS_TOOL_INTERCEPTION = True
 
     def acp_turn_result(self, trace: Trace, result: ACPTurn) -> None:
-        # codex-acp returns terminal failures in metadata with stop_reason=end_turn.
+        # codex-acp returns terminal failures in metadata with stop_reason=end_turn; a
+        # `warning` is a retry it already recovered from.
         failure = (
             result.response_metadata.get("jetbrains", {})
             .get("air", {})
             .get("sessionFailure")
         )
-        if failure and failure["phase"] == "active":
-            raise RuntimeError(f"Codex {failure['category']}: {failure['safeMessage']}")
+        if failure and failure["severity"] == "error":
+            raise RuntimeError(f"Codex {failure['category']}: {failure['title']}")
 
     async def setup(self, runtime: Runtime) -> None:
         await ensure_node(runtime)