verifiers 0.3.2.dev88__tar.gz → 0.3.2.dev90__tar.gz

{verifiers-0.3.2.dev88 → verifiers-0.3.2.dev90}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: verifiers
-Version: 0.3.2.dev88
+Version: 0.3.2.dev90
 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.dev88 → verifiers-0.3.2.dev90}/pyproject.toml

@@ -239,7 +239,6 @@ markers = [
     "rlm: v1 e2e cases on the rlm harness",
     "kimi_code: v1 e2e cases on the kimi-code harness",
     "pi: v1 e2e cases on the pi harness",
-    "pool: v1 e2e cases on the pool harness",
     "codex: v1 e2e cases on the codex harness",
     "asyncio: marks tests as async tests",
 ]

{verifiers-0.3.2.dev88 → verifiers-0.3.2.dev90}/tests/v1/conftest.py

@@ -24,7 +24,7 @@ Every combination carries its axes' pytest marks, so subsets select with `-m`:
     uv run pytest tests/v1 -n auto -m modal                       # only modal (needs local setup)
 
 Marks: runtimes `subprocess` / `docker` / `prime` / `modal`, placement `colocated`,
-harnesses `null` / `bash` / `rlm` / `kimi_code` / `pi` / `pool` / `openclaw` / `codex` /
+harnesses `null` / `bash` / `rlm` / `kimi_code` / `pi` / `openclaw` / `codex` /
 `claude_code` / `hermes_agent`.
 A mark is applied per axis, so it selects every case touching that value on ANY axis; for one exact
 combination use `-k` on the test id (e.g. `-k "harness-in-docker-with-tool-in-subprocess"`).

{verifiers-0.3.2.dev88 → verifiers-0.3.2.dev90}/tests/v1/test_e2e.py

@@ -90,9 +90,7 @@ ACP_RESUME_PLACEMENTS = [
         marks=[mark.pi, mark.docker],
         id="pi-responses-acp-in-docker",
     ),
-    pair("pool", "docker", "pool-acp-in-docker"),
     pair("openclaw", "docker", "openclaw-acp-in-docker"),
-    pair("pool", "prime", "pool-acp-in-prime"),
     pair("rlm", "prime", "rlm-acp-in-prime-vm"),
     pytest.param(
         "prime-agent",

{verifiers-0.3.2.dev88 → verifiers-0.3.2.dev90}/verifiers/v1/harness.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import asyncio
 import logging
 import os
 from abc import ABC, abstractmethod
@@ -99,23 +100,32 @@ class Harness(ABC, Generic[ConfigT]):
         """Upload each `config.skills` folder into `runtime` at `dest/<folder name>` —
         the program's fixed skill discovery location, which a supporting harness's
         `setup` passes."""
+        uploads = []
+        executables = []
         for skill in self.config.skills:
             # Resolve so `.`/`..` entries get their real folder name (and can't
             # place files outside `dest`).
             skill = skill.resolve()
             if not skill.is_dir():
                 raise ValueError(f"skill {str(skill)!r} is not a folder")
-            executables = []
-            for file in sorted(skill.rglob("*")):
+            for file in skill.rglob("*"):
                 if not file.is_file():
                     continue
                 target = f"{dest}/{skill.name}/{file.relative_to(skill).as_posix()}"
-                await runtime.write(target, file.read_bytes())
+                uploads.append((target, file.read_bytes()))
                 if os.access(file, os.X_OK):
                     executables.append(target)
-            if executables:
-                # `write` moves bytes, not modes; restore the execute bits scripts need.
-                await runtime.run(["chmod", "+x", *executables], {})
+        # Settle every upload before returning or surfacing an error.
+        results = await asyncio.gather(
+            *(runtime.write(target, data) for target, data in uploads),
+            return_exceptions=True,
+        )
+        for result in results:
+            if isinstance(result, BaseException):
+                raise result
+        if executables:
+            # `write` moves bytes, not modes; restore the execute bits scripts need.
+            await runtime.run(["chmod", "+x", *executables], {})
 
     async def _check_result(
         self, trace: Trace, runtime: Runtime, result: ProgramResult

{verifiers-0.3.2.dev88 → verifiers-0.3.2.dev90}/verifiers/v1/harnesses/__init__.py

@@ -20,7 +20,6 @@ from verifiers.v1.harnesses.mini_swe_agent import (
 from verifiers.v1.harnesses.null import NullHarness, NullHarnessConfig
 from verifiers.v1.harnesses.openclaw import OpenClawHarness, OpenClawHarnessConfig
 from verifiers.v1.harnesses.pi import PiHarness, PiHarnessConfig
-from verifiers.v1.harnesses.pool import PoolHarness, PoolHarnessConfig
 from verifiers.v1.harnesses.prime_agent import (
     PrimeAgentHarness,
     PrimeAgentHarnessConfig,
@@ -49,8 +48,6 @@ __all__ = [
     "OpenClawHarnessConfig",
     "PiHarness",
     "PiHarnessConfig",
-    "PoolHarness",
-    "PoolHarnessConfig",
     "PrimeAgentHarness",
     "PrimeAgentHarnessConfig",
     "RLMHarness",

verifiers-0.3.2.dev88/verifiers/v1/harnesses/pool/__init__.py

@@ -1,3 +0,0 @@
-from verifiers.v1.harnesses.pool.harness import PoolHarness, PoolHarnessConfig
-
-__all__ = ["PoolHarness", "PoolHarnessConfig"]

verifiers-0.3.2.dev88/verifiers/v1/harnesses/pool/harness.py

@@ -1,96 +0,0 @@
-"""Run Poolside's native ACP server against interception."""
-
-import json
-import shlex
-
-from verifiers.v1.acp import ACPConfig, ACPHarness
-from verifiers.v1.clients import ModelContext
-from verifiers.v1.configs.harness import HarnessConfig, PinnedVersion
-from verifiers.v1.harnesses.utils.install import ensure_installed
-from verifiers.v1.runtimes import Runtime
-from verifiers.v1.task import TaskData
-from verifiers.v1.trace import Trace
-
-POOL_DIR = "/tmp/vf-pool-{version}"
-SKILLS_DIR = ".poolside/skills"
-INSTALL = r"""
-set -e
-command -v curl >/dev/null || (apt-get update -qq && apt-get install -y -qq curl ca-certificates >/dev/null)
-command -v tar >/dev/null || (apt-get update -qq && apt-get install -y -qq tar >/dev/null)
-case "$(uname -s)" in Linux) os=linux ;; Darwin) os=darwin ;; *) echo "unsupported os: $(uname -s)" >&2; exit 1 ;; esac
-case "$(uname -m)" in aarch64|arm64) arch=arm64 ;; x86_64|amd64) arch=amd64 ;; *) echo "unsupported arch: $(uname -m)" >&2; exit 1 ;; esac
-mkdir -p {dir}
-curl -fsSL "https://github.com/poolsideai/pool/releases/download/v{version}/pool-$os-$arch.tar.gz" | tar -xz -C {dir}
-mv "{dir}/pool-$os-$arch" "{dir}/pool"
-chmod +x "{dir}/pool"
-"""
-
-
-class PoolHarnessConfig(HarnessConfig):
-    version: PinnedVersion = "1.0.15"
-    """Pool release to install, pinned for reproducibility."""
-
-
-class PoolHarness(ACPHarness[PoolHarnessConfig]):
-    APPENDS_SYSTEM_PROMPT = True
-    SUPPORTS_MCP = True
-    SUPPORTS_SKILLS = True
-
-    async def setup(self, runtime: Runtime) -> None:
-        await self.install_skills(runtime, SKILLS_DIR)
-        directory = POOL_DIR.format(version=self.config.version)
-        binary = f"{directory}/pool"
-        script = INSTALL.replace("{version}", self.config.version).replace(
-            "{dir}", directory
-        )
-        await ensure_installed(
-            runtime,
-            directory=directory,
-            ready=f"[ -x {binary} ]",
-            install=script,
-            env=self.config.resolved_env,
-            label="Pool",
-            shell=("bash", "-o", "pipefail", "-c"),
-        )
-        await super().setup(runtime)
-
-    async def prepare_acp(
-        self,
-        ctx: ModelContext,
-        trace: Trace,
-        runtime: Runtime,
-        endpoint: str,
-        secret: str,
-        mcp_urls: dict[str, str],
-        data: TaskData,
-    ) -> ACPConfig:
-        system_prompt, prompt = self.resolve_prompt(data)
-        env = {
-            **self.config.resolved_env,
-            # Standalone provider mode sends this bearer and model to interception.
-            "POOLSIDE_API_KEY": secret,
-            "POOLSIDE_STANDALONE_BASE_URL": endpoint,
-            "POOLSIDE_STANDALONE_MODEL": ctx.model,
-        }
-        pool_home = f".vf-pool/{trace.id}"
-        # Values are Pool tool names such as `shell`, `read`, or `edit`.
-        tools = {name: {"disabled": True} for name in self.config.disabled_tools or []}
-        settings = shlex.quote(json.dumps({"tools": tools}))
-        command = [
-            "sh",
-            "-c",
-            (
-                f'export HOME="$PWD/{pool_home}/home" '
-                f'XDG_CONFIG_HOME="$PWD/{pool_home}/config" '
-                f'XDG_STATE_HOME="$PWD/{pool_home}/state"; '
-                f"exec {POOL_DIR.format(version=self.config.version)}/pool acp "
-                f"--sandbox disabled --settings {settings}"
-            ),
-        ]
-        return ACPConfig(
-            env=env,
-            command=command,
-            prompt=prompt,
-            mcp_urls=mcp_urls,
-            system_prompt=system_prompt,
-        )