verifiers 0.3.2.dev107__py3-none-any.whl → 0.3.2.dev109__py3-none-any.whl

verifiers/v1/agent.py

@@ -540,6 +540,13 @@ class Agent:
     ) -> dict:
         """Resolve one run's runtime config, pairing checks, timeouts,
         interception — shared by `run` and `interaction`."""
+        harness = self.harness
+        skills = [*task.data.skills, *harness.config.skills]
+        if skills:
+            # Skill installations can hold run-specific state, such as RLM's package environment.
+            harness = type(harness)(
+                harness.config.model_copy(update={"skills": skills})
+            )
         if runtime is not None:
             _check_borrowed_placement(task, runtime, self.runtime_config)
             runtime_config = runtime.config
@@ -550,7 +557,7 @@ class Agent:
             )
             run_is_local = runtime_is_local(runtime_config)
         validate_pairing(
-            self.harness,
+            harness,
             type(task),
             runtime_config,
             tools=[*task.toolsets(task.config), *shared_tools.values()],
@@ -558,7 +565,7 @@ class Agent:
         timeouts = resolve_rollout_timeouts(self.timeout, task)
         return {
             "agent_config": self.config,
-            "harness": self.harness,
+            "harness": harness,
             "ctx": self.ctx,
             "runtime_config": runtime_config,
             "timeouts": replace(

verifiers/v1/configs/agent.py

@@ -14,7 +14,7 @@ class TimeoutConfig(BaseConfig):
     """Timeout (in seconds) for different phases of an agent's run."""
 
     setup: float | None = None  # one shared budget: task setup + provisioning
-    """Timeout (in seconds) for the task + harness setup hooks."""
+    """Timeout (in seconds) for task/harness setup through session preparation."""
     rollout: float | None = None
     """Timeout (in seconds) for the agent's solve attempt. Unset: the task's own
     timeout, else 4 hours. `0` disables the timeout."""

verifiers/v1/configs/harness.py

@@ -4,7 +4,7 @@ from __future__ import annotations
 
 import os
 from pathlib import Path
-from typing import Annotated
+from typing import Annotated, Literal
 
 from pydantic import ConfigDict, Field, FiniteFloat
 from pydantic_config import BaseConfig
@@ -15,6 +15,14 @@ PinnedVersion = Annotated[str, Field(pattern=r"^[A-Za-z0-9._+-]+$")]
 """A release/tag a harness pins its program install to."""
 
 
+SkillSource = Path | Annotated[dict[Literal["runtime"], str], Field(min_length=1)]
+
+
+def skill_destination(skill: SkillSource, dest: str) -> str:
+    """Runtime roots merge into `dest`; host folders keep their resolved name."""
+    return dest if isinstance(skill, dict) else f"{dest}/{skill.resolve().name}"
+
+
 class HarnessConfig(BaseConfig):
     id: ID = "bash"
     """Installed harness package, set through the seat's
@@ -44,10 +52,10 @@ class HarnessConfig(BaseConfig):
     tool_timeout: FiniteFloat = Field(600.0, gt=0)
     """Seconds a single MCP tool call may take; raise it for tools that boot a VM."""
     disabled_tools: list[str] | None = None
-    skills: list[Path] = Field(default_factory=list)
-    """Skill folders to upload into the program's skill discovery directory — each
-    lands at `<skills dir>/<folder name>`. Only harnesses whose program discovers
-    skills natively (`SUPPORTS_SKILLS`) accept them."""
+    skills: list[SkillSource] = Field(default_factory=list)
+    """Host skill folders or `{runtime: path}` roots of skills inside the runtime.
+    Sources are installed in order; later files override matching earlier files.
+    Only harnesses with native skill support (`SUPPORTS_SKILLS`) accept them."""
 
     @property
     def name(self) -> str:

verifiers/v1/harness.py

@@ -8,7 +8,7 @@ from collections.abc import Mapping
 from typing import TYPE_CHECKING, ClassVar, Generic, TypeVar
 
 from verifiers.v1.clients import ModelContext
-from verifiers.v1.configs.harness import HarnessConfig
+from verifiers.v1.configs.harness import HarnessConfig, skill_destination
 from verifiers.v1.errors import HarnessError, SandboxError, boundary
 from verifiers.v1.runtimes import ProgramResult, Runtime
 from verifiers.v1.task import TaskData
@@ -45,8 +45,8 @@ class Harness(ABC, Generic[ConfigT]):
     where model-directed execution changes the rules: the subprocess-on-host
     warning, the judge env's sandbox requirement."""
     SUPPORTS_SKILLS: ClassVar[bool] = False
-    """Whether the program discovers SKILL.md skills — its `setup` calls
-    `install_skills` with the program's fixed discovery location; configuring
+    """Whether the program discovers SKILL.md skills — it calls
+    `install_skills` with a run-scoped discovery location; configuring
     `skills` on a harness without support is rejected up front."""
     NEEDS_CONTAINER: ClassVar[bool] = True
     """Whether the program must run in a container runtime: True for every harness
@@ -97,35 +97,64 @@ class Harness(ABC, Generic[ConfigT]):
         """Provision this harness in `runtime` before its execution timeout starts."""
 
     async def install_skills(self, runtime: Runtime, dest: str) -> None:
-        """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 = []
+        """Install `config.skills` in the program's discovery directory.
+        Runtime roots are copied in place; host skill folders are uploaded."""
         for skill in self.config.skills:
+            target_dir = skill_destination(skill, dest)
+            if isinstance(skill, dict):
+                result = await runtime.run(
+                    [
+                        "bash",
+                        "-e",
+                        "-o",
+                        "pipefail",
+                        "-c",
+                        r"""
+[ -d "$1" ]
+source=$(CDPATH= cd -- "$1/." && pwd -P)
+mkdir -p -- "$2"
+target=$(CDPATH= cd -- "$2/." && pwd -P)
+[ "$source" != "$target" ] || exit 0
+# A source such as "." must not copy the destination back into itself.
+exclude=$(printf '%s' "./${target#"${source%/}/"}" | sed 's/[][\\*?]/\\&/g')
+tar -C "$source" --exclude="$exclude" -cf - . | tar -xpf - -C "$target"
+""",
+                        "vf-skills",
+                        skill["runtime"],
+                        target_dir,
+                    ],
+                    {},
+                )
+                if result.exit_code:
+                    raise RuntimeError(
+                        f"installing runtime skills from {skill['runtime']!r} failed: {result.stderr}"
+                    )
+                continue
             # 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")
+            uploads = []
+            executables = []
             for file in skill.rglob("*"):
                 if not file.is_file():
                     continue
-                target = f"{dest}/{skill.name}/{file.relative_to(skill).as_posix()}"
+                target = f"{target_dir}/{file.relative_to(skill).as_posix()}"
                 uploads.append((target, file.read_bytes()))
                 if os.access(file, os.X_OK):
                     executables.append(target)
-        # 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], {})
+            # Finish this source before the next one overwrites matching files.
+            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/v1/harnesses/claude_code/harness.py

@@ -15,7 +15,6 @@ ACP_VERSION = "0.67.0"
 CLAUDE_BIN = f"{PACKAGES_DIR}/node_modules/.bin/claude"
 ACP_BIN = f"{PACKAGES_DIR}/node_modules/.bin/claude-agent-acp"
 CLAUDE_CONFIG_ROOT = ".vf-claude"
-SKILLS_DIR = ".claude/skills"
 ACP_INSTALL = r"""
 set -e
 export PATH="/var/tmp/vf-node/bin:$PATH"
@@ -39,7 +38,6 @@ class ClaudeCodeHarness(ACPHarness[ClaudeCodeHarnessConfig]):
     SUPPORTS_SKILLS = True
 
     async def setup(self, runtime: Runtime) -> None:
-        await self.install_skills(runtime, SKILLS_DIR)
         await ensure_node(runtime)
         versions = {"version": self.config.version, "acp_version": ACP_VERSION}
         directory = CLAUDE_ACP_DIR.format(**versions)
@@ -74,6 +72,7 @@ class ClaudeCodeHarness(ACPHarness[ClaudeCodeHarnessConfig]):
     ) -> ACPConfig:
         system_prompt, prompt = self.resolve_prompt(data)
         config_dir = self.config_dir(trace)
+        await self.install_skills(runtime, f"{config_dir}/skills")
         versions = {"version": self.config.version, "acp_version": ACP_VERSION}
         session_meta = {
             "claudeCode": {

verifiers/v1/harnesses/codex/harness.py

@@ -22,7 +22,6 @@ PACKAGES_DIR = f"{CODEX_DIR}/acp"
 ACP_VERSION = "1.2.0"
 CODEX_BIN = f"{PACKAGES_DIR}/node_modules/.bin/codex"
 ACP_BIN = f"{PACKAGES_DIR}/node_modules/.bin/codex-acp"
-SKILLS_DIR = ".agents/skills"
 INSTALL = r"""
 set -e
 export PATH="/var/tmp/vf-node/bin:$PATH"
@@ -58,7 +57,6 @@ class CodexHarness(ACPHarness[CodexHarnessConfig]):
             raise RuntimeError(f"Codex {failure['category']}: {failure['safeMessage']}")
 
     async def setup(self, runtime: Runtime) -> None:
-        await self.install_skills(runtime, SKILLS_DIR)
         await ensure_node(runtime)
         logger.info(
             "codex: ensuring Codex %s and codex-acp %s are installed",
@@ -137,6 +135,7 @@ class CodexHarness(ACPHarness[CodexHarnessConfig]):
         mcp_urls: dict[str, str],
     ) -> dict[str, str]:
         home = self.trace_home(trace)
+        await self.install_skills(runtime, f"{home}/skills")
         mcp_config = "features={mcp_2026_07_28=true}\n" + (
             "mcp_servers={"
             + ",".join(

verifiers/v1/harnesses/kimi_code/harness.py

@@ -8,7 +8,7 @@ import tomli_w
 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.harnesses.utils.install import ensure_installed, remove_dir
 from verifiers.v1.runtimes import Runtime
 from verifiers.v1.task import TaskData
 from verifiers.v1.trace import Trace
@@ -18,7 +18,6 @@ logger = logging.getLogger(__name__)
 BINARY = "/tmp/vf-kimi-code/bin/kimi"
 KIMI_HOME = ".vf-kimi-code"
 ACP_COMMAND = [BINARY, "acp"]
-SKILLS_DIR = f"{KIMI_HOME}/skills"
 
 INSTALL = r"""
 set -e
@@ -52,7 +51,6 @@ class KimiCodeHarness(ACPHarness[KimiCodeHarnessConfig]):
     SUPPORTS_SKILLS = True
 
     async def setup(self, runtime: Runtime) -> None:
-        await self.install_skills(runtime, SKILLS_DIR)
         logger.info(
             "kimi-code: ensuring Kimi Code %s is installed", self.config.version
         )
@@ -77,6 +75,8 @@ class KimiCodeHarness(ACPHarness[KimiCodeHarnessConfig]):
         data: TaskData,
     ) -> ACPConfig:
         kimi_home = f"{KIMI_HOME}/{trace.id}"
+        skills_dir = f"{kimi_home}/skills"
+        await self.install_skills(runtime, skills_dir)
         provider_type = {
             "chat_completions": "openai",
             "responses": "openai_responses",
@@ -88,7 +88,7 @@ class KimiCodeHarness(ACPHarness[KimiCodeHarnessConfig]):
             else endpoint
         )
         config = {
-            "extra_skill_dirs": [SKILLS_DIR] if self.config.skills else [],
+            "extra_skill_dirs": [skills_dir] if self.config.skills else [],
             **(
                 {
                     "permission": {
@@ -125,3 +125,6 @@ class KimiCodeHarness(ACPHarness[KimiCodeHarnessConfig]):
             prompt=prompt,
             system_prompt=system_prompt,
         )
+
+    async def cleanup(self, trace: Trace, runtime: Runtime) -> None:
+        await remove_dir(runtime, f"{KIMI_HOME}/{trace.id}", "Kimi Code state")

verifiers/v1/harnesses/openclaw/harness.py

@@ -1,9 +1,7 @@
 """Run OpenClaw's Gateway-backed ACP agent against interception."""
 
-import asyncio
 import json
 import logging
-import secrets
 
 from verifiers.v1.acp import ACPConfig, ACPHarness
 from verifiers.v1.clients import ModelContext
@@ -130,22 +128,6 @@ class OpenClawHarness(ACPHarness[OpenClawHarnessConfig]):
     SUPPORTS_SKILLS = True
 
     async def setup(self, runtime: Runtime) -> None:
-        if not hasattr(self, "_staged_skills_dir"):
-            self._staged_skills_dir = (
-                f".vf-openclaw/staged-skills-{secrets.token_hex(8)}"
-            )
-            self._skills_setup_lock = asyncio.Lock()
-        if self.config.skills:
-            async with self._skills_setup_lock:
-                # A complete tree is immutable, so concurrent setups can safely reuse it.
-                ready_path = f"{self._staged_skills_dir}/.ready"
-                ready = await runtime.run(["test", "-f", ready_path], {})
-                if ready.exit_code != 0:
-                    await remove_dir(
-                        runtime, self._staged_skills_dir, "OpenClaw skills"
-                    )
-                    await self.install_skills(runtime, self._staged_skills_dir)
-                    await runtime.write(ready_path, b"")
         directory = OPENCLAW_DIR.format(version=self.config.version)
         binary = OPENCLAW_BIN.format(version=self.config.version)
         logger.info("openclaw: ensuring OpenClaw %s is installed", self.config.version)
@@ -233,14 +215,7 @@ class OpenClawHarness(ACPHarness[OpenClawHarnessConfig]):
             # OpenClaw treats an empty allowlist as all; a no-match key disables the catalog.
             config.setdefault("skills", {})["allowBundled"] = ["__none__"]
         await runtime.write(config_path, json.dumps(config).encode())
-        if self.config.skills:
-            copied = await runtime.run(
-                ["cp", "-R", self._staged_skills_dir, skills_dir], {}
-            )
-            if copied.exit_code != 0:
-                raise RuntimeError(
-                    f"failed to isolate OpenClaw skills: {copied.stderr.strip()[-500:]}"
-                )
+        await self.install_skills(runtime, skills_dir)
 
         env = {
             **self.config.resolved_env,

verifiers/v1/harnesses/pi/harness.py

@@ -7,9 +7,9 @@ from typing import Literal
 
 from verifiers.v1.acp import ACPConfig, ACPHarness
 from verifiers.v1.clients import ModelContext
-from verifiers.v1.configs.harness import HarnessConfig, PinnedVersion
+from verifiers.v1.configs.harness import HarnessConfig, PinnedVersion, skill_destination
 from verifiers.v1.harnesses.node import NODE_BIN_DIR, ensure_node
-from verifiers.v1.harnesses.utils.install import ensure_installed
+from verifiers.v1.harnesses.utils.install import ensure_installed, remove_dir
 from verifiers.v1.runtimes import Runtime
 from verifiers.v1.task import TaskData
 from verifiers.v1.trace import Trace
@@ -21,7 +21,6 @@ KEY_VAR = "PI_INTERCEPT_KEY"
 PI_DIR = "/var/tmp/vf-pi"
 PACKAGES_DIR = f"{PI_DIR}/mcp"
 PI_BIN = f"{PACKAGES_DIR}/node_modules/.bin/pi"
-SKILLS_DIR = ".agents/skills"
 MCP_VERSION = "2.28.0"
 ACP_VERSION = "0.0.33"
 MCP_ADAPTER = f"{PACKAGES_DIR}/node_modules/pi-mcp-adapter/index.ts"
@@ -61,7 +60,6 @@ class PiHarness(ACPHarness[PiHarnessConfig]):
     SUPPORTS_SKILLS = True
 
     async def setup(self, runtime: Runtime) -> None:
-        await self.install_skills(runtime, SKILLS_DIR)
         await ensure_node(runtime)
         logger.info(
             "pi: ensuring Pi %s and pi-acp %s are installed",
@@ -93,6 +91,8 @@ class PiHarness(ACPHarness[PiHarnessConfig]):
     ) -> ACPConfig:
         system_prompt, prompt = self.resolve_prompt(data)
         agent_dir = f".vf-pi-agent-{trace.id}"
+        skills_dir = f"{agent_dir}/skills"
+        await self.install_skills(runtime, skills_dir)
         reasoning = ctx.sampling.reasoning_effort not in (
             None,
             "none",
@@ -159,8 +159,7 @@ class PiHarness(ACPHarness[PiHarnessConfig]):
         skill_args = [
             arg
             for skill in self.config.skills
-            # Resolve like `install_skills` so the path matches what it wrote.
-            for arg in ("--skill", f"{SKILLS_DIR}/{skill.resolve().name}")
+            for arg in ("--skill", skill_destination(skill, skills_dir))
         ]
         pi_args = [
             PI_BIN,
@@ -190,3 +189,6 @@ class PiHarness(ACPHarness[PiHarnessConfig]):
             # Pi's extension owns the task-scoped MCP configuration.
             mcp_urls={},
         )
+
+    async def cleanup(self, trace: Trace, runtime: Runtime) -> None:
+        await remove_dir(runtime, f".vf-pi-agent-{trace.id}", "Pi state")

verifiers/v1/harnesses/prime_agent/harness.py

@@ -7,7 +7,7 @@ from typing import Literal
 
 from verifiers.v1.acp import ACPConfig, ACPHarness, ACPTurn
 from verifiers.v1.clients import ModelContext
-from verifiers.v1.configs.harness import HarnessConfig
+from verifiers.v1.configs.harness import HarnessConfig, skill_destination
 from verifiers.v1.harnesses.node import NODE_BIN_DIR, ensure_node
 from verifiers.v1.harnesses.utils.install import ensure_installed, remove_dir
 from verifiers.v1.runtimes import Runtime
@@ -25,7 +25,6 @@ PRIME_AGENT_COMMIT: Literal["81ae3cb34d27d38ee37f9e205a1e73694993b344"] = (
 PRIME_AGENT_VERSION = "0.9.1"
 PRIME_AGENT_DIR = "/var/tmp/vf-prime-agent"
 STATE_ROOT = "/tmp/vf-prime-agent-runs"
-SKILLS_DIR = ".agents/skills"
 PROVIDER = "intercept"
 LIFECYCLE_META_NAMESPACE = "ai.primeintellect.prime-agent"
 KEY_VAR = "PRIME_AGENT_INTERCEPT_KEY"
@@ -155,7 +154,6 @@ class PrimeAgentHarness(ACPHarness[PrimeAgentHarnessConfig]):
         statuses.append(status)
 
     async def setup(self, runtime: Runtime) -> None:
-        await self.install_skills(runtime, SKILLS_DIR)
         await ensure_node(runtime)
         logger.info("prime-agent: ensuring commit %s is installed", self.config.commit)
         await ensure_installed(
@@ -191,6 +189,7 @@ class PrimeAgentHarness(ACPHarness[PrimeAgentHarnessConfig]):
 
         root = self._root(trace)
         agent_dir = f"{root}/agent"
+        skills_dir = f"{agent_dir}/skills"
         created = await runtime.run(
             [
                 "mkdir",
@@ -207,6 +206,7 @@ class PrimeAgentHarness(ACPHarness[PrimeAgentHarnessConfig]):
             raise RuntimeError(
                 f"prime-agent state directory failed: {created.stderr.strip()[-500:]}"
             )
+        await self.install_skills(runtime, skills_dir)
         reasoning = ctx.sampling.reasoning_effort not in (
             None,
             "none",
@@ -251,7 +251,7 @@ class PrimeAgentHarness(ACPHarness[PrimeAgentHarnessConfig]):
         if self.config.autonomous:
             args.append("--autonomous")
         for skill in self.config.skills:
-            args += ["--skill", f"{SKILLS_DIR}/{skill.resolve().name}"]
+            args += ["--skill", skill_destination(skill, skills_dir)]
         if system_prompt:
             args += ["--append-system-prompt", system_prompt]
 

verifiers/v1/harnesses/rlm/harness.py

@@ -2,6 +2,7 @@
 
 import hashlib
 import logging
+import secrets
 import shlex
 from typing import Any, Literal
 
@@ -18,7 +19,7 @@ from pydantic_config import BaseConfig
 from verifiers.v1.acp import ACPConfig, ACPHarness, ACPTurn, JsonObject
 from verifiers.v1.clients import ModelContext
 from verifiers.v1.configs.harness import HarnessConfig
-from verifiers.v1.harnesses.utils.install import ensure_installed
+from verifiers.v1.harnesses.utils.install import ensure_installed, remove_dir
 from verifiers.v1.runtimes import Runtime
 from verifiers.v1.task import TaskData
 from verifiers.v1.trace import Trace
@@ -140,11 +141,15 @@ class RLMHarness(ACPHarness[RLMHarnessConfig]):
     APPENDS_SYSTEM_PROMPT = True
     SUPPORTS_MCP = True
     SUPPORTS_SKILLS = True
+    _skills_install_dir: str | None = None
 
     async def setup(self, runtime: Runtime) -> None:
-        # Before the installer: install.sh packages the skills it finds.
-        await self.install_skills(runtime, SKILLS_DIR)
+        if self.config.skills:
+            # Editable skill packages and uv's tool environment must belong to this run.
+            self._skills_install_dir = f"{RLM_CACHE_DIR}-skills-{secrets.token_hex(16)}"
         directory = self._install_dir()
+        skills_dir = f"{directory}/skills" if self.config.skills else SKILLS_DIR
+        await self.install_skills(runtime, skills_dir)
         binary = f"{directory}/bin/rlm"
         checkout = f"{directory}/checkout"
         ready = f"{directory}/.ready"
@@ -155,7 +160,9 @@ class RLMHarness(ACPHarness[RLMHarnessConfig]):
             "{ apt-get update -qq && apt-get install -y -qq git; } && "
             f"rm -rf {checkout} && git clone https://{RLM_REPO} {checkout} && "
             f"git -C {checkout} checkout {shlex.quote(self.config.version)} && "
+            f"sed -i 's|/task/rlm-skills|{skills_dir}|g' {checkout}/install.sh && "
             f"UV_INSTALL_DIR={directory}/bin UV_TOOL_BIN_DIR={directory}/bin "
+            f"UV_TOOL_DIR={directory}/tools "
             f"RLM_CHECKOUT_PATH={checkout} bash {checkout}/install.sh && "
             f"touch {ready})"
         )
@@ -279,6 +286,10 @@ class RLMHarness(ACPHarness[RLMHarnessConfig]):
 
     async def cleanup(self, trace: Trace, runtime: Runtime) -> None:
         await runtime.run(["rm", "-rf", f"{RLM_STATE_DIR}/{trace.id}"], {})
+        if self._skills_install_dir is not None:
+            await remove_dir(
+                runtime, self._skills_install_dir, "RLM skill installation"
+            )
 
     @staticmethod
     def _home(trace: Trace) -> str:
@@ -286,4 +297,4 @@ class RLMHarness(ACPHarness[RLMHarnessConfig]):
 
     def _install_dir(self) -> str:
         cache_key = hashlib.sha256(self.config.version.encode()).hexdigest()
-        return f"{RLM_CACHE_DIR}-{cache_key}"
+        return self._skills_install_dir or f"{RLM_CACHE_DIR}-{cache_key}"

verifiers/v1/rollout.py

@@ -43,7 +43,7 @@ class RolloutTimeouts:
     """Per-stage rollout timeouts, each bounding one rollout stage."""
 
     setup: float | None = None
-    """Timeout (in seconds) for the task + harness setup hooks."""
+    """Timeout (in seconds) for task/harness setup through session preparation."""
     agent: float | None = None
     """Timeout (in seconds) for the agent's solve attempt."""
     finalize: float | None = None
@@ -221,7 +221,7 @@ class Rollout:
             self.trace.timing.boot.end = now
             self.trace.timing.setup.start = now
             self.trace.notify()
-            # Task setup and harness provisioning share one setup-stage deadline.
+            # Setup hooks and harness session preparation share one deadline.
             setup_deadline = (
                 None
                 if self._timeouts.setup is None
@@ -271,7 +271,10 @@ class Rollout:
             # Setup and service provisioning are complete. Apply the runtime's
             # execution policy while preserving the framework routes the agent uses.
             await runtime.prepare_execution([self._endpoint, *self._urls.values()])
-            async with boundary(HarnessError, "opening harness session"):
+            async with (
+                boundary(HarnessError, "opening harness session"),
+                asyncio.timeout_at(setup_deadline),
+            ):
                 harness_data = self.trace.task.data
                 if (
                     self._session.request_interceptors

verifiers/v1/runtimes/prime.py

@@ -1,10 +1,4 @@
-"""Remote Prime sandbox runtime.
-
-`expose` (sandbox port -> public URL) uses the SDK's native exposure (`client.expose`), so a
-host-side harness/framework can reach a tool server hosted in the sandbox. The reverse
-direction (a program in the sandbox reaching a host service) is the shared host-side
-`Tunnel` (interception.tunnel), not the runtime's concern.
-"""
+"""Remote Prime VM sandbox runtime."""
 
 import asyncio
 import base64
@@ -25,7 +19,6 @@ from pydantic import Field, model_validator
 from verifiers.v1.configs.runtime import NetworkPolicyConfig
 from verifiers.v1.errors import SandboxError
 from verifiers.v1.runtimes.base import (
-    SERVICE_PORT,
     BaseRuntimeInfo,
     ProgramResult,
     Runtime,
@@ -73,15 +66,11 @@ class PrimeConfig(NetworkPolicyConfig):
     type: Literal["prime"] = "prime"
     image: str = "python:3.11-slim"
     """Docker image to run. Any pullable ref works: on the first use of an image, the
-    platform auto-builds what the sandbox needs from it (a VM image for `vm` sandboxes,
-    ~10 minutes) and caches the result, so later sandboxes on the same ref start in
+    platform auto-builds the VM image that the sandbox needs from it (~10 minutes)
+    and caches the result, so later sandboxes on the same ref start in
     seconds."""
     workdir: str | None = None
     """Working directory override; None uses the task's workdir, or /app."""
-    vm: bool = True
-    """Run as a micro-VM rather than a container (kernel features / stronger isolation)."""
-    guaranteed: bool = False
-    """Request guaranteed (vs best-effort) capacity."""
     region: str | None = None
     """Region to provision in (None = provider-chosen)."""
     labels: list[str] = Field(default_factory=list)
@@ -106,10 +95,6 @@ class PrimeConfig(NetworkPolicyConfig):
     def _validate_egress(self) -> "PrimeConfig":
         if not self.network_restricted:
             return self
-        if not self.vm:
-            raise ValueError(
-                "Prime allow/block egress lists require a VM sandbox (vm=true)"
-            )
         if not self.allow:
             return self
         validate_egress_lists(
@@ -156,11 +141,7 @@ class PrimeRuntime(Runtime):
 
     @property
     def supports_live_processes(self) -> bool:
-        return self.config.vm
-
-    @property
-    def published_port(self) -> int | None:
-        return SERVICE_PORT
+        return True
 
     async def start(self) -> None:
         from prime_sandboxes import AsyncSandboxClient, CreateSandboxRequest
@@ -186,14 +167,7 @@ class PrimeRuntime(Runtime):
             "memory_gb": self.config.memory,
             "disk_size_gb": self.config.disk,
             "gpu_count": gpu_count,
-            # -1 is prime's convention for no lifetime limit; containers with an
-            # idle timeout must carry a finite lifetime as a safety fallback (which
-            # must exceed the idle timeout)
-            "timeout_minutes": (
-                -1
-                if self.config.vm or idle_minutes is None
-                else max(EFFECTIVELY_UNBOUNDED_SECONDS // 60, idle_minutes + 1)
-            ),
+            "timeout_minutes": -1,
             "idle_timeout_minutes": idle_minutes,
             "gpu_type": gpu_type,
             "region": self.config.region,
@@ -216,8 +190,6 @@ class PrimeRuntime(Runtime):
                                 dict.fromkeys([*BASE_LABELS, *self.config.labels])
                             ),
                             docker_image=self.config.image,
-                            vm=self.config.vm,
-                            guaranteed=self.config.guaranteed,
                             environment_vars=self.env,
                             **{k: v for k, v in options.items() if v is not None},
                         )
@@ -323,11 +295,6 @@ class PrimeRuntime(Runtime):
     async def open_process(
         self, argv: list[str], env: dict[str, str]
     ) -> RuntimeProcess:
-        if not self.config.vm:
-            raise SandboxError(
-                "persistent harness sessions on Prime require a VM sandbox; "
-                "set runtime.prime.vm=true"
-            )
         try:
             process = await self._client.open_process(
                 self.info.id,
@@ -339,22 +306,11 @@ class PrimeRuntime(Runtime):
             raise SandboxError(f"prime live process failed to start: {e}") from e
         return PrimeProcess(process)
 
-    async def expose(self, port: int) -> str | None:
-        # Publish a server hosted IN the sandbox via the SDK's native port exposure → a public
-        # HTTPS URL. Removed when the sandbox is deleted in stop(), so a tool in its own prime
-        # sandbox needs no host tunnel. Port exposure is region-gated: many regions (incl. the
-        # backend default, which lands in us-central) 400 it; `us` supports it. TODO: re-enable the
-        # prime cases in the e2e `skip_if_unexposable` guard once prime exposes ports in any region.
-        try:
-            exposed = await self._client.expose(self.info.id, port)
-        except Exception as e:  # surface prime's exposure constraints actionably
-            raise SandboxError(
-                "prime port exposure failed — port exposure isn't supported in this sandbox's "
-                "region; pin `tools.runtime.region` to a region that supports it (e.g. `us`), or "
-                f"use a colocated / docker / modal tools.runtime instead. ({e})"
-            ) from e
-        logger.info("prime: exposed sandbox port %d at %s", port, exposed.url)
-        return exposed.url.rstrip("/")
+    async def expose(self, port: int) -> str:
+        raise SandboxError(
+            "Prime VM sandboxes do not support port exposure; colocate the service "
+            "with its consumer, or use a docker or modal service runtime"
+        )
 
     async def run_background(
         self, argv: list[str], env: dict[str, str], log: str
@@ -371,7 +327,7 @@ class PrimeRuntime(Runtime):
             raise SandboxError(f"prime background launch failed: {e}") from e
 
     async def _read(self, path: str, max_bytes: int | None = None) -> bytes:
-        if max_bytes is not None and self.config.vm:
+        if max_bytes is not None:
             try:
                 # VM execute_command uses bash and returns the complete output stream.
                 result = await self._client.execute_command(
@@ -386,8 +342,6 @@ class PrimeRuntime(Runtime):
             if result.exit_code:
                 raise SandboxError(f"read {path!r}: {result.stderr.strip()[-500:]}")
             return base64.b64decode(result.stdout)
-        if max_bytes is not None:
-            return await super()._read(path, max_bytes)
         # Avoid background-job log limits and base64 overhead by downloading binary data directly.
         # The temporary file is removed on every exit, and its byte read stays off the event loop.
         target = (

verifiers/v1/task.py

@@ -20,6 +20,7 @@ from typing import TYPE_CHECKING, Any, ClassVar, Generic, Self
 from pydantic import BaseModel, ConfigDict, Field
 from typing_extensions import TypeVar
 
+from verifiers.v1.configs.harness import SkillSource
 from verifiers.v1.configs.task import TaskConfig
 from verifiers.v1.errors import TaskError, boundary
 from verifiers.v1.state import StateT
@@ -97,11 +98,13 @@ class TaskData(BaseModel):
     workdir: str | None = None
     """Optional working directory to use for the task. Only relevant for tasks that run in a container."""
 
+    skills: list[SkillSource] = Field(default_factory=list)
+    """Skill sources installed before the harness's configured skills for this task."""
+
     network_allow: list[str] = Field(default_factory=lambda: ["*"])
     """Execution-time destinations requested by this task. `*` leaves the runtime
     allowlist unchanged; a concrete list replaces a wildcard or retains entries also
-    present in an existing allowlist. Prime runtimes accept host-level entries and
-    require `vm=true`."""
+    present in an existing allowlist. Prime runtimes accept host-level entries."""
     network_block: list[str] = Field(default_factory=list)
     """Execution-time destinations denied by this task and combined with runtime
     blocks. Non-empty concrete allowlists cannot be combined with blocklists. Docker

verifiers/v1/tasksets/harbor/env.py

@@ -37,7 +37,7 @@ class HarborEnv(IsolatedVerifierEnv, vf.Env[HarborEnvConfig]):
             await agents.agent.run(task)
             return
         # Resolve the verifier's box before the solve, so an impossible pairing
-        # (e.g. a restricted Prime verifier without vm=true) costs nothing
+        # (e.g. an invalid network policy) costs nothing
         # rather than a full agent run.
         self.verifier_config(task)
         await agents.agent.run(task.defer_scoring(), collect_artifacts=True)

verifiers/v1/tasksets/harbor/taskset.py

@@ -393,6 +393,7 @@ def verifier_box_data(data: HarborData) -> HarborData:
             "upload_environment": data.upload_environment if fresh else False,
             "env": dict(verifier.env),
             "healthcheck": verifier.healthcheck,
+            "skills": [],
             "mcp_servers": [],
             "network_allow": list(verifier.network_allow),
             "network_block": [],
@@ -612,6 +613,7 @@ def parse_task(task_dir: Path, idx: int, harbor_config: HarborConfig) -> HarborD
         tags=meta.get("tags", []),
         task_dir=str(task_dir),
         upload_environment=upload_environment,
+        skills=[{"runtime": environment.skills_dir}] if environment.skills_dir else [],
         **environment.model_dump(
             include={"env", "healthcheck", "mcp_servers"}, mode="json"
         ),
@@ -729,9 +731,7 @@ def parse_verifier_environment(
             "the task never declared",
             task_dir.name,
         )
-    unsupported = [
-        field for field in ("skills_dir", "tpu") if getattr(environment, field, None)
-    ]
+    unsupported = [field for field in ("tpu",) if getattr(environment, field, None)]
     if environment.os != TaskOS.LINUX or unsupported:
         raise ValueError(
             f"{task_dir.name}: verifier environment declares "

{verifiers-0.3.2.dev107.dist-info → verifiers-0.3.2.dev109.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: verifiers
-Version: 0.3.2.dev107
+Version: 0.3.2.dev109
 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
@@ -36,7 +36,7 @@ Requires-Dist: numpy>=2.1.0
 Requires-Dist: openai<3.0.0,>=2.54.0
 Requires-Dist: prime-pydantic-config[toml]>=0.4.3
 Requires-Dist: prime-runs>=0.1.3
-Requires-Dist: prime-sandboxes>=0.2.39
+Requires-Dist: prime-sandboxes>=0.3.0
 Requires-Dist: prime-tunnel>=0.1.8
 Requires-Dist: pydantic>=2.12.3
 Requires-Dist: pyzmq>=27.1.0

{verifiers-0.3.2.dev107.dist-info → verifiers-0.3.2.dev109.dist-info}/RECORD

@@ -1,18 +1,18 @@
 verifiers/__init__.py,sha256=yDYsckVQbUkwoh9mU_hBeCLGvS_aWgAxuoNm5LxQpDI,517
 verifiers/cli/commands/init.py,sha256=DaT1NGrK2lnQWvHlrLkZ1D2Te49ll2Eh-DyOAhmCzfU,147
 verifiers/v1/__init__.py,sha256=zzXyzUkCJCXfPRajYNNQlgIAz6_-uu9oYNb-XsbIjeU,8434
-verifiers/v1/agent.py,sha256=rFjoAn9vpRXoyIZTzMgIuGb6HczTSuOenaB-U-x09to,32489
+verifiers/v1/agent.py,sha256=Lc4ySCWBCZRqED-M2LwpPcewAH9k4AuS_ubq-E6-wDQ,32808
 verifiers/v1/env.py,sha256=Hon7gcepciMYgki_jN4-l4XCcKRZRYuYm7c17nfeiFE,18520
 verifiers/v1/episode.py,sha256=Fft0R2W6INAVeJhoE9b8h7L_KQSz7XuxP4FT_XhcsUY,5741
 verifiers/v1/errors.py,sha256=uQySmCUrjDYFkBIii1vBFPiEcPFPQZ9EGpas_6HvcQI,5279
 verifiers/v1/graph.py,sha256=b3Z3lCnsALnAvY82pDnjhODFYakkeLV_hXgGOvbryrQ,40737
-verifiers/v1/harness.py,sha256=sXEjVdrS4A2SO35C_uEebBVeUt8VL7dui7IneAp2MRs,14124
+verifiers/v1/harness.py,sha256=kCJ4lvpZ1LZSCpZZPNaNlPU1kQ1U1YCX9TZ6DVKoTQI,15260
 verifiers/v1/judge.py,sha256=uqn-djkK2_7Q3kOjkDo6ipOvV_HOvuLa9y6AoDjRux8,9618
-verifiers/v1/rollout.py,sha256=0zkVYNUDBag3Bms4uz6TNbim3prx0AwFZiuwkpr2lKI,24308
+verifiers/v1/rollout.py,sha256=oka3sNdv1EbE6LatFFRMJH-5ti6FcqFuD8xa7oVUyDA,24405
 verifiers/v1/semantic.py,sha256=VQ_buw6Ya6QtRoh_fiYI5Avs5RmIQCfWH0H0bQc4aoc,4704
 verifiers/v1/session.py,sha256=7iarEDRH2epID_O8ai-1HwveXze_o8shnLw9pFMFX5o,18900
 verifiers/v1/state.py,sha256=EckF2bWp-vV4b1jYJ9sLI5xrfGuI5spIgYYwW926toI,595
-verifiers/v1/task.py,sha256=p3DsvVIzMRPcoEkHJhtqLueJKQ0RCvnn-Tp7ISw6Ox8,12251
+verifiers/v1/task.py,sha256=7iJLXQo7JKX5jG87atIwT_HZlSto_3a7yB3w__uQQQw,12427
 verifiers/v1/taskset.py,sha256=fp2E0IEhL_Ybj9cegZwljfTmW27p_30zTWHFKw_kXE4,4374
 verifiers/v1/trace.py,sha256=uOlrvP1-C4TYtcK3S6zcCNK6UlS78hUoL-KGvJL0IsQ,32011
 verifiers/v1/types.py,sha256=PH1SxRWXOjPCgQD9aZgM18_bECvFSNQ25gkCbulmTKU,9857
@@ -43,10 +43,10 @@ verifiers/v1/clients/client.py,sha256=zqC_AkiD9pl0kxxIbupNiehbS-YnVdp3EMZ0LOHTd8
 verifiers/v1/clients/eval.py,sha256=cJucyUFsbF1JlG8nF0Tlw2EBMSoAF0YMklgYiOSoNpg,8322
 verifiers/v1/clients/train.py,sha256=w6bUY6GCBYlscOxz-TBWjbDbFoNSyEfQgDynbB7Ef80,18698
 verifiers/v1/configs/__init__.py,sha256=X7u6X7B3ieD1HZl79Tv1kg-yPHbT0QLvAebaGqxVtvA,317
-verifiers/v1/configs/agent.py,sha256=Aczd-QlNYin6yGjierybBu9wDocjSPZXtKyC0knE5QU,3326
+verifiers/v1/configs/agent.py,sha256=PiWMNueum67vnyLLpIgxXZDb3FTQEUFqo2Z1siOOJqs,3342
 verifiers/v1/configs/client.py,sha256=fPWtBJv5lPkqttwk0AF2ubfLdzxX2XCO0Zr-A9V4mtk,4440
 verifiers/v1/configs/env.py,sha256=-Bcdd4MEPfekVc_fnpiwsb9UDwCNdNCKrAHGoQuuXu8,7784
-verifiers/v1/configs/harness.py,sha256=OetAXz5ErJPvNxUd2_6iiP0JItCnJl5eS9aH6VoNywE,2569
+verifiers/v1/configs/harness.py,sha256=ZrR-Egzfft95oMoqQ78bAq9TqP882Q68M_47Ajw9lE8,2918
 verifiers/v1/configs/judge.py,sha256=CBmGgyHXVpcBctvrFD5GlW75wpeFXVFajq_xiMaAVDk,2142
 verifiers/v1/configs/retries.py,sha256=-nPlgmz7J_NZ4GOYG6oCc3thXo4La2SufBmvQGvvPGo,821
 verifiers/v1/configs/runtime.py,sha256=IBX3FxiJ2RJaiX2aKghQ15clFHvJpPUXy8GWQOpNhYM,2967
@@ -91,27 +91,27 @@ verifiers/v1/harnesses/browser_use/__init__.py,sha256=yxm2IG5TDLadxZEkX8SL26XWbc
 verifiers/v1/harnesses/browser_use/harness.py,sha256=sIPRxF-dxiB5xXddCCCOMP6BF1kO01Azyr9AR16x4aw,5582
 verifiers/v1/harnesses/browser_use/program.py,sha256=5g5AnRS-brYSMmHcteLf8-XDLaTDNIOjG5dym91blOc,10051
 verifiers/v1/harnesses/claude_code/__init__.py,sha256=JkZQylMSA3o1fvBx_NM41y_umuahmOSqFjTrJ2hryTw,171
-verifiers/v1/harnesses/claude_code/harness.py,sha256=Racxbs1EAH3gRwSeUptAIa3kB-EUm_nreUnHMwRUHuE,4119
+verifiers/v1/harnesses/claude_code/harness.py,sha256=pGfyhoxOwjZTQXwM3YwazAxsFVO9uQ3M8eDCPdS9WDk,4101
 verifiers/v1/harnesses/codex/__init__.py,sha256=ocyBvlpSO8c3pT6D1ebrQohoXYESQcCyPT5kN3j1-N4,132
-verifiers/v1/harnesses/codex/harness.py,sha256=S6WLiEpOmeGHbwCYHhQyNkXSVYFbwBbCvPgk2tAuDis,7516
+verifiers/v1/harnesses/codex/harness.py,sha256=UKWQQkV-1Ek5OpjXO8ROPpHj7PZDLxiHqwbOccI-7NE,7492
 verifiers/v1/harnesses/hermes_agent/__init__.py,sha256=k84Y8UTTf1KOw5LGqPb2dk-hYFagPevDJBOF7qsR_vo,176
 verifiers/v1/harnesses/hermes_agent/harness.py,sha256=dSsE2VZ8jaM-U14QXO3J7P-U6NMpSuOz3DUMtDWpWk0,3622
 verifiers/v1/harnesses/hermes_agent/program.py,sha256=RaIYm6NFD8KLXERGXp6rOE4QrQjlcTnm1Fu33xYsiHw,570
 verifiers/v1/harnesses/kimi_code/__init__.py,sha256=1BEotSfuzmVcE7UrxHeAREM5d7J20cGkonHuowlrzu4,161
-verifiers/v1/harnesses/kimi_code/harness.py,sha256=NCkNIdOFDNiUrEmLYb-r-SjyunR5Cmu1g4F1lwuhGK0,4163
+verifiers/v1/harnesses/kimi_code/harness.py,sha256=QIprdVQRSmImUdmGnuCOaIbUtiqtaTHHHj5AH56jltA,4333
 verifiers/v1/harnesses/mini_swe_agent/__init__.py,sha256=1OBu2Y-XqL72-dGhFdgB6sr5uX_peEEXT9vEjPy1ovU,182
 verifiers/v1/harnesses/mini_swe_agent/harness.py,sha256=HKGtlDyhXDJqNtDFeb3D4JYUerIgXK6HAmyqOm1mkRo,2473
 verifiers/v1/harnesses/mini_swe_agent/program.py,sha256=Beej-JLIgSqnqIqB746oKU6kJYcynvppIbmwGa3nJJ0,167
 verifiers/v1/harnesses/null/__init__.py,sha256=XDeKOPeoXUQi9US9_Z3_yH9YaZ4EiOMKJmUtNNdtD7s,127
 verifiers/v1/harnesses/null/harness.py,sha256=Bq19j8Bh_d4VvpFx9nwv6BKnG1tKweEk-2BSmaXWLkU,1330
 verifiers/v1/harnesses/openclaw/__init__.py,sha256=kNyfPB1sGwEf4K3mTfKN0VEyxun3SGvNpvArVKvH-x8,160
-verifiers/v1/harnesses/openclaw/harness.py,sha256=tucrQb_Osi7KAEtpyO4i55IXI5SFtLnyItB8eV4kjng,10534
+verifiers/v1/harnesses/openclaw/harness.py,sha256=KUSwzLviNmFc2mIZ7jJ5Abon6EZXsEkHS675f82lG-A,9373
 verifiers/v1/harnesses/pi/__init__.py,sha256=1mSnzOf-RdEchdeg6IHtcdobG5KIpAhuTdUIgsCIn74,117
-verifiers/v1/harnesses/pi/harness.py,sha256=l3iZYXhSm4_dGHuxsfN20phCkTzNWYtTr--aZ0CivK4,6517
+verifiers/v1/harnesses/pi/harness.py,sha256=neTkMp_FKywjUXF7Mi0TJjRHKfjG08vv5yuHRGC3LRQ,6624
 verifiers/v1/harnesses/prime_agent/__init__.py,sha256=xfPC4RrlnyYWaf38tPAOZq4YL6M55lMpj5BoRLmaK8I,171
-verifiers/v1/harnesses/prime_agent/harness.py,sha256=cMvkNdqEutzCoX3lrmCQ5dAi_HSDV3Vhw6K1N-LiBtA,10630
+verifiers/v1/harnesses/prime_agent/harness.py,sha256=sSqVDEVtgLQA5KuqBZ8mrRMyi6EIrPz3r1zXWRiE6-g,10660
 verifiers/v1/harnesses/rlm/__init__.py,sha256=hwx51xTdEWzzYgMj-p4ETNcptoub3IPMDbEdJ4uB8Ns,122
-verifiers/v1/harnesses/rlm/harness.py,sha256=xADdXtDey4w4zs8YR_CP-hR55beHDUXUMkMJD13vSsQ,12609
+verifiers/v1/harnesses/rlm/harness.py,sha256=Fn6lTlKxXno1jV1GrlaE9zhS9CnjC2gLlTvAm5Lrtbo,13223
 verifiers/v1/harnesses/terminus_2/__init__.py,sha256=l1pwfg1HyVDEqT9efdnFWcXqaZwtKwe7TrplkyxZSA0,166
 verifiers/v1/harnesses/terminus_2/harness.py,sha256=DMTY1bzvydyjHqzmvBSBGKJsmUzYEgxqd215a7cG7tg,3052
 verifiers/v1/harnesses/terminus_2/program.py,sha256=X8MZdyItU3gbXGXnGOlJ9LWcUTnqIDaxlhkZvoRiqKY,2703
@@ -145,7 +145,7 @@ verifiers/v1/runtimes/base.py,sha256=IKmrJSJltWX9M1nPUZExKluPJM4N82VsnqQGFz0ECV8
 verifiers/v1/runtimes/container.py,sha256=puQD8X9acP6GefoCNEnq40POi-nXPD-729bUjTVShfM,9554
 verifiers/v1/runtimes/limiters.py,sha256=C6cWDD4pwyFwUY4QFke3oiMXwydSNIfSbJSIglUC2kE,3062
 verifiers/v1/runtimes/modal.py,sha256=rFbxdDBi_VIiDA67XyZg-AOsc_XkV8aQWSKnTUAgCmU,17506
-verifiers/v1/runtimes/prime.py,sha256=VWzskEtMEwND6tKKhRtFM5m4Qzt001xCaI4zKqNon80,19490
+verifiers/v1/runtimes/prime.py,sha256=eeYGu60f8PSGtL6g0nJslYzrQLHo6d64JP73N3q3xTw,16982
 verifiers/v1/runtimes/subprocess.py,sha256=QmpQ235_xrX6gowrr8557VgbydOXpLbfFq0KKGfJThk,8774
 verifiers/v1/runtimes/docker/__init__.py,sha256=2wceP-8V-w0dQBEeLtoJFfCZ135DxP2rcdcUBmSWiNU,18543
 verifiers/v1/runtimes/docker/egress.py,sha256=tGwSvHNgV3-uhjshg8sZ4fVRyq8Fg7DbJl_c-lacPi4,23876
@@ -158,8 +158,8 @@ verifiers/v1/serve/server.py,sha256=i6XrY_PMAapkrP54tYX8-K6iGmOmrR7BbQ-3Qhn3w4s,
 verifiers/v1/serve/types.py,sha256=Qw5IjiJhsZz4giSXUBi9dfInBnQZ5_xiUmgH2qp8jPI,1741
 verifiers/v1/tasksets/__init__.py,sha256=5hTKSPLXSZTgKBbeKQI7FEp06c0ZsaQfuwvrvVFLgb8,712
 verifiers/v1/tasksets/harbor/__init__.py,sha256=JxThzHFYEkuSfQbVf3pEnjordgdfe4MU-QOsSSbyovU,326
-verifiers/v1/tasksets/harbor/env.py,sha256=rZEh-1IaHpfsrwSOn7qTq7uV0K7cZZ28bHB7BEWQdnc,3622
-verifiers/v1/tasksets/harbor/taskset.py,sha256=o7weK05AFjeyxQOFqkSvumx_1iGNe7wViF2zskJrR1I,33693
+verifiers/v1/tasksets/harbor/env.py,sha256=ATMkfwmrE_2jpOOffbjLQ91feG7TOtefBGlpwWbRq8c,3604
+verifiers/v1/tasksets/harbor/taskset.py,sha256=3xCxtVVwQSpaLfoV1SaUoNQlpIGaTmwW6PvPyYh3yv4,33780
 verifiers/v1/tasksets/harbor/toolset.py,sha256=_7HXu-uuvqzhmJ1EX6c3KWLg0g0eRrLjCq1kBsZhIwo,3159
 verifiers/v1/tasksets/lean/__init__.py,sha256=oyv-GbJQBvCyGwo7Ysxm2NjZW-aX4j1uDoHU0f-E7xQ,767
 verifiers/v1/tasksets/lean/scoring.py,sha256=sfzsT6MUK0zdMgAQ57_P_0CBaPHy3M7go71QMr5hDrs,10599
@@ -192,8 +192,8 @@ verifiers/v1/utils/prime.py,sha256=UTYRjp9cbjNb6CVmBHda-1wWZAIIfxNOmTyNuT7_wL4,9
 verifiers/v1/utils/retries.py,sha256=Y2ZgrAjn-qNkRKZP_RVNL_7EK0iaRcZeCa404lpGYi4,5417
 verifiers/v1/utils/score.py,sha256=493yJVMw8teCu9JxapxMFPFzI0hNUqdo0Y2nGW4kckk,6200
 verifiers/v1/utils/version.py,sha256=-obEo_-l9-D8FLef4hYxncOe-uJpxrM1g2Hig_37Sgs,1607
-verifiers-0.3.2.dev107.dist-info/METADATA,sha256=8biJ4Q_vmsTiDiaQ__lQE5aS61TV0duA-4L6put9I80,4238
-verifiers-0.3.2.dev107.dist-info/WHEEL,sha256=zOwg4jB6zX2kU910N-cMawjivD6tO8NEWvE12je1bVk,87
-verifiers-0.3.2.dev107.dist-info/entry_points.txt,sha256=iugElcdWPKbQM7uFF0lZ8iUpHsNr17-BwEAAjJWxV3U,259
-verifiers-0.3.2.dev107.dist-info/licenses/LICENSE,sha256=v0RrUsdV3IDoZhrRce297IXS3xMHNJ-_LdLpFAUWb9k,1072
-verifiers-0.3.2.dev107.dist-info/RECORD,,
+verifiers-0.3.2.dev109.dist-info/METADATA,sha256=Mu8pwqCnU-tLCfuuzeDDJbnG_QIR3ctEOrwuU4lVP8o,4237
+verifiers-0.3.2.dev109.dist-info/WHEEL,sha256=zOwg4jB6zX2kU910N-cMawjivD6tO8NEWvE12je1bVk,87
+verifiers-0.3.2.dev109.dist-info/entry_points.txt,sha256=iugElcdWPKbQM7uFF0lZ8iUpHsNr17-BwEAAjJWxV3U,259
+verifiers-0.3.2.dev109.dist-info/licenses/LICENSE,sha256=v0RrUsdV3IDoZhrRce297IXS3xMHNJ-_LdLpFAUWb9k,1072
+verifiers-0.3.2.dev109.dist-info/RECORD,,