verifiers 0.2.1.dev60__tar.gz → 0.2.1.dev62__tar.gz

{verifiers-0.2.1.dev60 → verifiers-0.2.1.dev62}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: verifiers
-Version: 0.2.1.dev60
+Version: 0.2.1.dev62
 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-0.2.1.dev60 → verifiers-0.2.1.dev62}/pyproject.toml

@@ -252,7 +252,7 @@ markers = [
     "modal: v1 e2e cases touching the modal runtime (local-only setup; excluded in CI)",
     "colocated: v1 e2e cases with a user/tool server colocated in the harness runtime",
     "null: v1 e2e cases on the null harness",
-    "default: v1 e2e cases on the default harness",
+    "bash: v1 e2e cases on the bash harness",
     "rlm: v1 e2e cases on the rlm harness",
     "kimi_code: v1 e2e cases on the kimi-code harness",
     "codex: v1 e2e cases on the codex harness",

{verifiers-0.2.1.dev60 → verifiers-0.2.1.dev62}/tests/v1/conftest.py

@@ -15,12 +15,12 @@ Every matrix value carries a pytest mark, so subsets select with `-m`:
     uv run pytest tests/v1 -n auto                                # the whole matrix (needs modal setup)
     uv run pytest tests/v1 -n auto -m "not prime and not modal"  # the CI matrix (host + docker only)
     uv run pytest tests/v1 -n auto -m docker                      # any case touching the docker runtime
-    uv run pytest tests/v1 -n auto -m default                     # only the default harness
+    uv run pytest tests/v1 -n auto -m bash                        # only the bash harness
     uv run pytest tests/v1 -n auto -m prime                       # only prime (real sandboxes; local)
     uv run pytest tests/v1 -n auto -m modal                       # only modal (needs local setup)
 
 Marks: runtimes `subprocess` / `docker` / `prime` / `modal`, placement `colocated`,
-harnesses `null` / `default` / `rlm` / `kimi_code` / `codex` / `claude_code`. A mark is applied per
+harnesses `null` / `bash` / `rlm` / `kimi_code` / `codex` / `claude_code`. 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"`). prime/modal provision real
 remote sandboxes (slow, infra-flaky, need setup), so they're local-only — CI runs
@@ -104,7 +104,7 @@ def tool_runtime(request) -> dict:
     return {"runtime": {"type": request.param}}
 
 
-# Harnesses, composed with the runtime fixtures, each carrying its harness mark (`-m default`, ...).
+# Harnesses, composed with the runtime fixtures, each carrying its harness mark (`-m bash`, ...).
 # Built-ins are bundled in the `harnesses` package; the agent CLIs (`rlm` / `kimi-code` / `codex` /
 # `claude-code`) install their dependencies at rollout. `compact` (an example harness) and
 # `terminus-2` (drives the host tmux) are excluded. `test_agentic` skips `null` (a chat loop with no
@@ -112,7 +112,7 @@ def tool_runtime(request) -> dict:
 @pytest.fixture(
     params=[
         pytest.param("null", marks=pytest.mark.null, id="null"),
-        pytest.param("default", marks=pytest.mark.default, id="default"),
+        pytest.param("bash", marks=pytest.mark.bash, id="bash"),
         pytest.param("rlm", marks=pytest.mark.rlm, id="rlm"),
         pytest.param("kimi-code", marks=pytest.mark.kimi_code, id="kimi-code"),
         pytest.param("codex", marks=pytest.mark.codex, id="codex"),

{verifiers-0.2.1.dev60 → verifiers-0.2.1.dev62}/verifiers/v1/cli/init.py

@@ -235,7 +235,7 @@ A v1 verifiers environment, scaffolded with `init`.
 
 ```bash
 uv pip install -e .        # install this package (or register it in your project)
-uv run eval {dash} -n 3    # evaluate a few tasks with the default harness
+uv run eval {dash} -n 3    # evaluate a few tasks with the bash harness
 ```
 
 ## Layout

{verifiers-0.2.1.dev60 → verifiers-0.2.1.dev62}/verifiers/v1/dialects/chat.py

@@ -125,7 +125,7 @@ def parse_tools(raw: list[dict] | None) -> list[Tool] | None:
 
 
 # --- vf -> chat wire ----------------------------------------------------------
-# `message_to_wire` (chat-only): used by `extend` (user-sim turn injection), the default harness
+# `message_to_wire` (chat-only): used by `extend` (user-sim turn injection), the bash harness
 # (a Messages prompt), and the train client (its generate request). The proxy preserves its parsed
 # native JSON independently and does not use this serializer.
 

{verifiers-0.2.1.dev60 → verifiers-0.2.1.dev62}/verifiers/v1/env.py

@@ -87,11 +87,11 @@ def pool_serve_kwargs(pool: StaticPoolConfig | ElasticPoolConfig) -> dict:
 class EnvConfig(BaseConfig):
     """The taskset that loads tasks and the harness that runs them."""
 
-    # SerializeAsAny: these hold resolved subclasses (e.g. MathConfig, DefaultHarnessConfig);
+    # SerializeAsAny: these hold resolved subclasses (e.g. MathConfig, BashHarnessConfig);
     # without it model_dump() narrows to the base type and drops the subclass fields, so the
     # env-server subconfig the orchestrator writes would lose taskset/harness-specific knobs.
     taskset: SerializeAsAny[TasksetConfig] = TasksetConfig()
-    harness: SerializeAsAny[HarnessConfig] = HarnessConfig(id="default")
+    harness: SerializeAsAny[HarnessConfig] = HarnessConfig(id="bash")
     timeout: TimeoutConfig = TimeoutConfig()
     retries: RetryConfig = RetryConfig()
     max_turns: int | None = None

{verifiers-0.2.1.dev60 → verifiers-0.2.1.dev62}/verifiers/v1/harness.py

@@ -31,7 +31,7 @@ logger = logging.getLogger(__name__)
 
 
 class HarnessConfig(BaseConfig):
-    id: ID = "default"
+    id: ID = "bash"
     """Local package or Hub `org/name[@version]`, set through `--harness.id`."""
     runtime: RuntimeConfig = SubprocessConfig()
     """Runtime for the harness program; tool servers choose their placement separately."""

{verifiers-0.2.1.dev60 → verifiers-0.2.1.dev62}/verifiers/v1/harnesses/__init__.py

@@ -1,9 +1,9 @@
+from verifiers.v1.harnesses.bash import BashHarness, BashHarnessConfig
 from verifiers.v1.harnesses.claude_code import (
     ClaudeCodeHarness,
     ClaudeCodeHarnessConfig,
 )
 from verifiers.v1.harnesses.codex import CodexHarness, CodexHarnessConfig
-from verifiers.v1.harnesses.default import DefaultHarness, DefaultHarnessConfig
 from verifiers.v1.harnesses.kimi_code import KimiCodeHarness, KimiCodeHarnessConfig
 from verifiers.v1.harnesses.mini_swe_agent import (
     MiniSWEAgentHarness,
@@ -15,12 +15,12 @@ from verifiers.v1.harnesses.rlm import RLMHarness, RLMHarnessConfig
 from verifiers.v1.harnesses.terminus_2 import Terminus2Harness, Terminus2HarnessConfig
 
 __all__ = [
+    "BashHarness",
+    "BashHarnessConfig",
     "ClaudeCodeHarness",
     "ClaudeCodeHarnessConfig",
     "CodexHarness",
     "CodexHarnessConfig",
-    "DefaultHarness",
-    "DefaultHarnessConfig",
     "KimiCodeHarness",
     "KimiCodeHarnessConfig",
     "NullHarness",

verifiers-0.2.1.dev62/verifiers/v1/harnesses/bash/__init__.py

@@ -0,0 +1,6 @@
+from verifiers.v1.harnesses.bash.harness import (
+    BashHarness,
+    BashHarnessConfig,
+)
+
+__all__ = ["BashHarness", "BashHarnessConfig"]

{verifiers-0.2.1.dev60/verifiers/v1/harnesses/default → verifiers-0.2.1.dev62/verifiers/v1/harnesses/bash}/harness.py

@@ -25,7 +25,7 @@ SEARCH_PROMPT = (
 )
 
 
-class DefaultHarnessConfig(HarnessConfig):
+class BashHarnessConfig(HarnessConfig):
     edit: bool = True
     """Offer the local `edit` tool (single-occurrence string replacement in a file) alongside
     `bash`. On by default; set `--harness.edit false` for a bash-only agent."""
@@ -36,7 +36,7 @@ class DefaultHarnessConfig(HarnessConfig):
     the agent's `bash` subprocesses don't inherit it."""
 
 
-class DefaultHarness(Harness[DefaultHarnessConfig]):
+class BashHarness(Harness[BashHarnessConfig]):
     APPENDS_SYSTEM_PROMPT = True
     SUPPORTS_MCP = True
     SUPPORTS_USER_SIM = True
@@ -86,7 +86,7 @@ class DefaultHarness(Harness[DefaultHarnessConfig]):
                 serper_key = os.environ.get("SERPER_API_KEY")
             if not serper_key:
                 raise ValueError(
-                    "default search=true requires SERPER_API_KEY in the eval environment "
+                    "bash search=true requires SERPER_API_KEY in the eval environment "
                     "(the host env or --harness.env)"
                 )
             args += ["--search", f"--serper-key={serper_key}"]

{verifiers-0.2.1.dev60 → verifiers-0.2.1.dev62}/verifiers/v1/harnesses/claude_code/harness.py

@@ -21,7 +21,7 @@ curl -fsSL https://claude.ai/install.sh | HOME={home} bash -s {version}
 
 
 class ClaudeCodeHarnessConfig(HarnessConfig):
-    version: str = Field(default="2.1.207", pattern=r"^[A-Za-z0-9._+-]+$")
+    version: str = Field(default="2.1.214", pattern=r"^[A-Za-z0-9._+-]+$")
     """Claude Code release to install; pinned for reproducibility."""
 
 

{verifiers-0.2.1.dev60 → verifiers-0.2.1.dev62}/verifiers/v1/harnesses/codex/harness.py

@@ -36,7 +36,7 @@ chmod +x {bin}
 
 
 class CodexHarnessConfig(HarnessConfig):
-    version: str = "0.137.0"
+    version: str = "0.144.5"
     """Codex release to install (the `rust-v<version>` GitHub release); pinned for reproducibility."""
     multi_agent: bool = False
     """Enable Codex's native multi-agent v2 tools."""

{verifiers-0.2.1.dev60 → verifiers-0.2.1.dev62}/verifiers/v1/harnesses/kimi_code/harness.py

@@ -32,7 +32,7 @@ env \
 
 
 class KimiCodeHarnessConfig(HarnessConfig):
-    version: str = "0.14.3"
+    version: str = "0.27.0"
     """Kimi Code release to install, pinned for reproducibility."""
 
 

{verifiers-0.2.1.dev60 → verifiers-0.2.1.dev62}/verifiers/v1/harnesses/mini_swe_agent/harness.py

@@ -9,7 +9,7 @@ PROGRAM_SOURCE = (Path(__file__).resolve().parent / "program.py").read_text()
 
 
 class MiniSWEAgentHarnessConfig(HarnessConfig):
-    version: str = "2.2.8"
+    version: str = "2.4.5"
     """mini-swe-agent release to install, pinned for reproducibility."""
 
 

{verifiers-0.2.1.dev60 → verifiers-0.2.1.dev62}/verifiers/v1/harnesses/mini_swe_agent/program.py

@@ -1,6 +1,6 @@
 # /// script
 # requires-python = ">=3.10"
-# dependencies = ["mini-swe-agent=={version}"]
+# dependencies = ["mini-swe-agent=={version}", "litellm[proxy]"]
 # ///
 
 from minisweagent.run.mini import app

{verifiers-0.2.1.dev60 → verifiers-0.2.1.dev62}/verifiers/v1/harnesses/pi/harness.py

@@ -87,7 +87,7 @@ export default async function isolatedMcp(pi) {{
 
 
 class PiHarnessConfig(HarnessConfig):
-    version: str = "0.80.6"
+    version: str = "0.80.10"
     """Pi release to install, pinned for reproducibility."""
 
 

{verifiers-0.2.1.dev60 → verifiers-0.2.1.dev62}/verifiers/v1/loaders.py

@@ -98,12 +98,12 @@ def judge_class(judge_id: str) -> type[Judge]:
 
 def default_harness_id(taskset_id: str) -> str:
     if not taskset_id:
-        return "default"
+        return "bash"
     try:
         module = import_taskset(taskset_id)
         _plugin_class(module, Harness, "harness")
     except (ModuleNotFoundError, TypeError, AttributeError):
-        return "default"
+        return "bash"
     return taskset_id
 
 

verifiers-0.2.1.dev60/verifiers/v1/harnesses/default/__init__.py

@@ -1,6 +0,0 @@
-from verifiers.v1.harnesses.default.harness import (
-    DefaultHarness,
-    DefaultHarnessConfig,
-)
-
-__all__ = ["DefaultHarness", "DefaultHarnessConfig"]