verifiers 0.3.1.dev34__tar.gz → 0.3.1.dev36__tar.gz

{verifiers-0.3.1.dev34 → verifiers-0.3.1.dev36}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: verifiers
-Version: 0.3.1.dev34
+Version: 0.3.1.dev36
 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
@@ -51,7 +51,7 @@ Provides-Extra: browser
 Requires-Dist: python-dotenv>=1.0.0; extra == 'browser'
 Requires-Dist: stagehand>=3.4.0; extra == 'browser'
 Provides-Extra: harbor
-Requires-Dist: harbor==0.20.0; (python_version >= '3.12') and extra == 'harbor'
+Requires-Dist: harbor==0.21.0; (python_full_version >= '3.12') and extra == 'harbor'
 Provides-Extra: modal
 Requires-Dist: modal>=1.4.0; extra == 'modal'
 Provides-Extra: nemo-gym

{verifiers-0.3.1.dev34 → verifiers-0.3.1.dev36}/pyproject.toml

@@ -109,7 +109,7 @@ browser = [
 ]
 harbor = [
     # Harbor currently publishes Python 3.12+ packages; keep core verifiers installable on 3.11.
-    "harbor==0.20.0; python_version >= '3.12'",
+    "harbor==0.21.0 ; python_full_version >= '3.12'",
 ]
 modal = [
     "modal>=1.4.0",
@@ -186,6 +186,7 @@ web-search-interception = { path = "environments/web_search_interception", edita
 # Bounded cutoffs for the explicitly requested tool upgrades.
 # Full UTC timestamps: bare dates resolve to local-tz midnight and churn the
 # lockfile across timezones.
+harbor = "2026-08-11T00:00:00Z"
 ruff = "2026-07-28T00:00:00Z"
 ty = "2026-07-28T00:00:00Z"
 # PrimeIntellect-published on PyPI (trusted publisher)

{verifiers-0.3.1.dev34 → verifiers-0.3.1.dev36}/verifiers/v1/cli/eval/main.py

@@ -73,8 +73,12 @@ def main(argv: list[str] | None = None) -> None:
     # config a resume typically re-runs — would overwrite the previous run.
     run_path = output_path(config)
     if config.clean and not config.resume and run_path.exists():
-        if not run_path.resolve().is_relative_to(config.output_dir.resolve()):
-            raise SystemExit("--clean requires run.dir to remain under output_dir")
+        output_dir = config.output_dir.resolve()
+        resolved_run_path = run_path.resolve()
+        if resolved_run_path == output_dir or not resolved_run_path.is_relative_to(
+            output_dir
+        ):
+            raise SystemExit("--clean requires run.dir to name a child of output_dir")
         shutil.rmtree(run_path)
     traces_file = run_path / TRACES_FILE
     if not config.resume and traces_file.exists() and traces_file.stat().st_size > 0:

{verifiers-0.3.1.dev34 → verifiers-0.3.1.dev36}/verifiers/v1/cli/validate.py

@@ -485,8 +485,10 @@ def main(argv: list[str] | None = None) -> None:
     # A named run directory is re-entered only by `--resume` or wiped by `--clean`: any
     # other write into it would overwrite the previous run's results.
     if config.clean and not config.resume and out.exists():
-        if not out.resolve().is_relative_to(config.output_dir.resolve()):
-            raise SystemExit("--clean requires run.dir to remain under output_dir")
+        output_dir = config.output_dir.resolve()
+        resolved_out = out.resolve()
+        if resolved_out == output_dir or not resolved_out.is_relative_to(output_dir):
+            raise SystemExit("--clean requires run.dir to name a child of output_dir")
         shutil.rmtree(out)
     results_file = out / RESULTS_FILE
     if not config.resume and results_file.exists() and results_file.stat().st_size > 0:

{verifiers-0.3.1.dev34 → verifiers-0.3.1.dev36}/verifiers/v1/harnesses/claude_code/harness.py

@@ -14,7 +14,7 @@ from verifiers.v1.trace import Trace
 
 CLAUDE_ACP_DIR = "/var/tmp/vf-claude-agent-acp-{version}-{acp_version}"
 PACKAGES_DIR = f"{CLAUDE_ACP_DIR}/packages"
-ACP_VERSION = "0.65.0"
+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"
@@ -32,7 +32,7 @@ touch {ready}
 
 
 class ClaudeCodeHarnessConfig(HarnessConfig):
-    version: str = Field(default="2.1.223", pattern=r"^[A-Za-z0-9._+-]+$")
+    version: str = Field(default="2.1.232", pattern=r"^[A-Za-z0-9._+-]+$")
     """Claude Code release to install, pinned for reproducibility."""
 
 

{verifiers-0.3.1.dev34 → verifiers-0.3.1.dev36}/verifiers/v1/harnesses/codex/harness.py

@@ -21,7 +21,7 @@ logger = logging.getLogger(__name__)
 
 CODEX_DIR = "/var/tmp/vf-codex-{version}-{acp_version}"
 PACKAGES_DIR = f"{CODEX_DIR}/acp"
-ACP_VERSION = "1.1.10"
+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"
@@ -38,7 +38,7 @@ touch {ready}
 
 
 class CodexHarnessConfig(HarnessConfig):
-    version: str = Field(default="0.146.1", pattern=r"^[A-Za-z0-9._+-]+$")
+    version: str = Field(default="0.147.0", pattern=r"^[A-Za-z0-9._+-]+$")
     """Codex release to install, pinned for reproducibility."""
     multi_agent: bool = False
     """Enable Codex's native multi-agent v2 tools."""

{verifiers-0.3.1.dev34 → verifiers-0.3.1.dev36}/verifiers/v1/harnesses/kimi_code/harness.py

@@ -39,7 +39,7 @@ env \
 
 
 class KimiCodeHarnessConfig(HarnessConfig):
-    version: str = Field(default="0.34.0", pattern=r"^[A-Za-z0-9._+-]+$")
+    version: str = Field(default="0.36.0", pattern=r"^[A-Za-z0-9._+-]+$")
     """Kimi Code release to install, pinned for reproducibility."""
     transport: Literal["chat_completions", "responses", "anthropic_messages"] = (
         "chat_completions"

{verifiers-0.3.1.dev34 → verifiers-0.3.1.dev36}/verifiers/v1/harnesses/pi/harness.py

@@ -23,7 +23,7 @@ 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 = "2.20.1"
+MCP_VERSION = "2.25.0"
 ACP_VERSION = "0.0.33"
 MCP_ADAPTER = f"{PACKAGES_DIR}/node_modules/pi-mcp-adapter/index.ts"
 ACP_BIN = f"{PACKAGES_DIR}/node_modules/.bin/pi-acp"
@@ -46,7 +46,7 @@ fi
 
 
 class PiHarnessConfig(HarnessConfig):
-    version: str = Field(default="0.84.0", pattern=r"^[A-Za-z0-9._+-]+$")
+    version: str = Field(default="0.84.1", pattern=r"^[A-Za-z0-9._+-]+$")
     """Pi release to install, pinned for reproducibility."""
     transport: Literal["chat_completions", "responses", "anthropic_messages"] = (
         "chat_completions"

{verifiers-0.3.1.dev34 → verifiers-0.3.1.dev36}/verifiers/v1/harnesses/terminus_2/harness.py

@@ -15,7 +15,7 @@ logger = logging.getLogger(__name__)
 
 
 class Terminus2HarnessConfig(HarnessConfig):
-    version: str = Field(default="0.20.0", pattern=r"^[A-Za-z0-9._+-]+$")
+    version: str = Field(default="0.21.0", pattern=r"^[A-Za-z0-9._+-]+$")
     """Harbor release to install, pinned for reproducibility."""