verifiers 0.2.2.dev40__py3-none-any.whl → 0.2.2.dev41__py3-none-any.whl

verifiers/v1/harness.py

@@ -9,7 +9,7 @@ from typing import TYPE_CHECKING, ClassVar, Generic, TypeVar
 from verifiers.v1.clients import ModelContext
 from verifiers.v1.configs.harness import HarnessConfig
 from verifiers.v1.decorators import discover_decorated, invoke_all
-from verifiers.v1.errors import HarnessError, boundary
+from verifiers.v1.errors import HarnessError, SandboxError, boundary
 from verifiers.v1.runtimes import ProgramResult, Runtime
 from verifiers.v1.task import TaskData
 from verifiers.v1.types import Messages
@@ -140,6 +140,11 @@ class Harness(ABC, Generic[ConfigT]):
         if result.exit_code != 0:
             # The real cause is at the END of a traceback, so keep the tail.
             detail = (result.stderr or result.stdout).strip()[-2000:] or "<no output>"
+            if not await runtime.alive():
+                raise SandboxError(
+                    f"runtime died under harness {self.config.id!r} "
+                    f"(exit {result.exit_code}): {detail}"
+                )
             raise HarnessError(
                 f"harness {self.config.id!r} exited {result.exit_code}: {detail}"
             )

verifiers/v1/runtimes/base.py

@@ -197,6 +197,16 @@ class Runtime(ABC):
     async def run(self, argv: list[str], env: dict[str, str]) -> ProgramResult:
         pass
 
+    async def alive(self) -> bool:
+        """Whether the box still executes anything. Not every runtime raises when
+        the box is gone — some surface it as `exec`'s own non-zero result,
+        indistinguishable from the command failing. One probe on the failure path
+        tells the two apart before we blame anyone."""
+        try:
+            return (await self.run(["true"], {})).exit_code == 0
+        except Exception:  # noqa: BLE001 - failing to exec at all means the box is gone
+            return False
+
     async def run_program(self, argv: list[str], env: dict[str, str]) -> ProgramResult:
         """Run the harness's MAIN program — the rollout itself (a possibly long-lived, stateful,
         agentic run) — as opposed to the short idempotent infra ops (write / mv / install /

{verifiers-0.2.2.dev40.dist-info → verifiers-0.2.2.dev41.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: verifiers
-Version: 0.2.2.dev40
+Version: 0.2.2.dev41
 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.dev40.dist-info → verifiers-0.2.2.dev41.dist-info}/RECORD

@@ -174,7 +174,7 @@ verifiers/v1/env.py,sha256=QFwhD1NKbEUWGf0l6G2ttK4JLbhUt2MUCDciWaaJ_1s,17834
 verifiers/v1/episode.py,sha256=KWyM9ovTphEIo2260BYu2HvtnOOTxjLRDPrF_jTCiuA,2497
 verifiers/v1/errors.py,sha256=Pj5Om8x1fP3TDPQQn6nUCoSPoZRsy2JeBz8pXhpPrDY,6883
 verifiers/v1/graph.py,sha256=-j8MNdIv1r5ifajv5ggnGPsfBNB74IO3WKLcrpYh9hg,29102
-verifiers/v1/harness.py,sha256=oL_CyPXw2ScCXfM-E0sk-7_y4PqvjxC8gq2jR2jyWUs,10709
+verifiers/v1/harness.py,sha256=cDiDuhp29aqL4CxiqSZz-IT0QlOcVlye_vTTHZblWo8,10948
 verifiers/v1/judge.py,sha256=VnTi9iSfCJMCccYWc1qCTcT7iD3AcomS2cXfHwh1HVw,9423
 verifiers/v1/legacy.py,sha256=kGfx4CEmk17VEqwX6JfUBbjQRMqJ0tL4kkA2kqGypN8,22305
 verifiers/v1/loaders.py,sha256=TnNKY2msVo3p7m-5wjw376PYH-zTu3TLixXRst0VOiM,9985
@@ -292,7 +292,7 @@ verifiers/v1/mcp/launch.py,sha256=lRtcl98BO7imC44bsxO2yljwRQR_KCvJ6P-bd_hS5L8,18
 verifiers/v1/mcp/server.py,sha256=uW8qddXQbtECxIPCU-lkOlFIv3VKWoFAeZFNhwPoyWo,10985
 verifiers/v1/mcp/toolset.py,sha256=tSPXgTNREZyaeCnoFVnDfXxFAj_pStZMPUIgyPrY_Dk,996
 verifiers/v1/runtimes/__init__.py,sha256=TqNtJVahzQuLS4fgCVeT9xdOzTkp0hRAPOkl7eFVDUg,2011
-verifiers/v1/runtimes/base.py,sha256=hv1McFrIZgymLwMLNYPAlsR4At0uz60haVo5eXrFbNU,14196
+verifiers/v1/runtimes/base.py,sha256=sTCiLo51yAU_T-Voc3BTH-FuzyEM0cGY05VzFE4DmYc,14720
 verifiers/v1/runtimes/limiters.py,sha256=ZjjYUFEt20yH8ggL5uU9w9U500AOB44qSUTQ7bYyh0c,2730
 verifiers/v1/runtimes/modal.py,sha256=FnocoJd9lQVQS3V3CoBon7KpKXm1bKWWeU2FAkq2bKU,8825
 verifiers/v1/runtimes/prime.py,sha256=wldpcZnKZHNRldpoubeMgHdl-ZlKWOyeulFu4nHNKwc,15110
@@ -327,8 +327,8 @@ verifiers/v1/utils/logging.py,sha256=OcMHA6NsYux3oIzjPuI95rDWmFBHNcHDjZeNIhXTX-Y
 verifiers/v1/utils/memory.py,sha256=ZkIvGk6uITAH5sKon65LifKPbvZr8mJ__PVc23FZpOQ,1835
 verifiers/v1/utils/sampling.py,sha256=JczGzBn6s3wsIrSY2Hy3hE8m6NreNgsNzhSjDikQqX0,1037
 verifiers/v1/utils/version.py,sha256=-obEo_-l9-D8FLef4hYxncOe-uJpxrM1g2Hig_37Sgs,1607
-verifiers-0.2.2.dev40.dist-info/METADATA,sha256=7FRJgVM3XdGU4JTZSsGsxxaKndzkW26q4EFyz9NsMhE,4540
-verifiers-0.2.2.dev40.dist-info/WHEEL,sha256=lCkmxWfQsSc9CfIClYeavTdQeEX2toPqufh9gI35EQA,87
-verifiers-0.2.2.dev40.dist-info/entry_points.txt,sha256=dF82JUYEFslR1AOW8LZfuBWeZeQoyX4wCRrduklg8-I,551
-verifiers-0.2.2.dev40.dist-info/licenses/LICENSE,sha256=v0RrUsdV3IDoZhrRce297IXS3xMHNJ-_LdLpFAUWb9k,1072
-verifiers-0.2.2.dev40.dist-info/RECORD,,
+verifiers-0.2.2.dev41.dist-info/METADATA,sha256=DBJbZYlN8WMvZtMYyH2rDHocN0aa-l2mDI0NXAyR1kU,4540
+verifiers-0.2.2.dev41.dist-info/WHEEL,sha256=lCkmxWfQsSc9CfIClYeavTdQeEX2toPqufh9gI35EQA,87
+verifiers-0.2.2.dev41.dist-info/entry_points.txt,sha256=dF82JUYEFslR1AOW8LZfuBWeZeQoyX4wCRrduklg8-I,551
+verifiers-0.2.2.dev41.dist-info/licenses/LICENSE,sha256=v0RrUsdV3IDoZhrRce297IXS3xMHNJ-_LdLpFAUWb9k,1072
+verifiers-0.2.2.dev41.dist-info/RECORD,,