verifiers 0.2.1.dev53__tar.gz → 0.2.1.dev55__tar.gz

{verifiers-0.2.1.dev53 → verifiers-0.2.1.dev55}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: verifiers
-Version: 0.2.1.dev53
+Version: 0.2.1.dev55
 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.dev53 → verifiers-0.2.1.dev55}/pyproject.toml

@@ -73,14 +73,14 @@ dev = [
     "nltk>=3.9.2",
     "textarena==0.7.4",
 ]
-# v1 example environments vendored under environments/, resolved by id via [tool.uv.sources].
+# V1 examples and support packages, resolved from their local sources below.
 # Installed by default (tool.uv.default-groups) so the v1 CLIs (eval <id>, serve) run straight
 # from this repo. The built-in tasksets/harnesses ship inside verifiers (verifiers/v1/...).
 examples = [
     "compact",
     "gsm8k-v1", "glossary-v1", "deepwiki-v1", "wiki-search-v1", "code-golf-v1",
     "reverse-text-v1", "color-codeword-v1",
-    "wordle-v1", "alphabet-sort-v1", "scratchpad-v1",
+    "wordle-v1", "openenv-v1", "openenv-wordle-v1", "alphabet-sort-v1", "scratchpad-v1",
 ]
 
 [project.optional-dependencies]
@@ -144,6 +144,8 @@ gsm8k-v1 = { path = "environments/gsm8k_v1", editable = true }
 code-golf-v1 = { path = "environments/code_golf_v1", editable = true }
 reverse-text-v1 = { path = "environments/reverse_text_v1", editable = true }
 wordle-v1 = { path = "environments/wordle_v1", editable = true }
+openenv-v1 = { path = "packages/openenv_v1", editable = true }
+openenv-wordle-v1 = { path = "environments/openenv_wordle_v1", editable = true }
 color-codeword-v1 = { path = "environments/color_codeword_v1", editable = true }
 alphabet-sort-v1 = { path = "environments/alphabet_sort_v1", editable = true }
 scratchpad-v1 = { path = "environments/scratchpad_v1", editable = true }

{verifiers-0.2.1.dev53 → verifiers-0.2.1.dev55}/verifiers/v1/mcp/launch.py

@@ -182,6 +182,10 @@ async def serve_in_runtime(
     the current rollout task from the adjacent `/task` endpoint rather than a launch argument.
     """
     env = {"VF_CONFIG": server.config.model_dump_json()}
+    if runtime.type == "subprocess":
+        # Keep provider temp files in the runtime workdir so cleanup removes them.
+        assert runtime.info.id is not None
+        env["TMPDIR"] = runtime.info.id
     if state_url:
         env["VF_STATE_URL"] = state_url
         env["VF_STATE_SECRET"] = state_secret
@@ -194,12 +198,19 @@ async def serve_in_runtime(
     else:
         port_file = f"/tmp/vf-port-{uuid.uuid4().hex}"
         env["MCP_PORT_FILE"] = port_file
-    if runtime.type == "subprocess":
-        python = sys.executable
-    else:
+    python = sys.executable
+    if runtime.type != "subprocess":
         python = await _install_in_sandbox(server, runtime)
+    command = [python, "-m", type(server).__module__]
+    if runtime.type != "subprocess":
+        # Providers may invoke uv after the install shell exits, so preserve its PATH.
+        command = [
+            "sh",
+            "-c",
+            f'export PATH="$HOME/.local/bin:$PATH"; exec {shlex.join(command)}',
+        ]
     log = f"vf_tool_{server.server_name}.log"
-    await runtime.run_background([python, "-m", type(server).__module__], env, log)
+    await runtime.run_background(command, env, log)
     if fixed is not None:
         port = fixed
     else:

{verifiers-0.2.1.dev53 → verifiers-0.2.1.dev55}/verifiers/v1/mcp/server.py

@@ -127,6 +127,7 @@ class ServerBase(Generic[ConfigT, StateT]):
         self._state_adapter = TypeAdapter(self._state_cls)
         self._inert_state: StateT = self._state_cls()  # type: ignore[assignment]
         self._state_client: AsyncClient | None = None
+        self._exit_stack = contextlib.AsyncExitStack()
 
     @property
     def state(self) -> StateT:
@@ -250,43 +251,36 @@ class ServerBase(Generic[ConfigT, StateT]):
         if port_file:
             Path(port_file).write_text(str(sock.getsockname()[1]))
 
-        async def _setup() -> None:
-            await self.setup()
-            # Shared taskset-scoped servers have no task channel and skip this hook.
-            await self._setup_task_from_channel(*self._state_channel())
-
-        asyncio.run(_setup())
-        # These servers are reached by the harness through localhost or a tunnel, never a browser.
-        from mcp.server.transport_security import TransportSecuritySettings
-
-        security = TransportSecuritySettings(enable_dns_rebinding_protection=False)
-        mcp = FastMCP(
-            self.server_name,
-            json_response=True,
-            stateless_http=True,
-            transport_security=security,
-        )
-        self._register(mcp)
-        app = mcp.streamable_http_app()
-        mcp_lifespan = app.router.lifespan_context
-
-        @contextlib.asynccontextmanager
-        async def serving_lifespan(starlette):
+        async def serve() -> None:
             import httpx
+            from mcp.server.transport_security import TransportSecuritySettings
 
             try:
                 async with (
+                    self._exit_stack,
                     httpx.AsyncClient(timeout=STATE_TIMEOUT) as client,
-                    mcp_lifespan(starlette),
                 ):
                     self._state_client = client
-                    yield
+                    await self.setup()
+                    await self._setup_task_from_channel(*self._state_channel())
+                    # These servers are reached through localhost or a tunnel, never a browser.
+                    security = TransportSecuritySettings(
+                        enable_dns_rebinding_protection=False
+                    )
+                    mcp = FastMCP(
+                        self.server_name,
+                        json_response=True,
+                        stateless_http=True,
+                        transport_security=security,
+                    )
+                    self._register(mcp)
+                    app = mcp.streamable_http_app()
+                    server = uvicorn.Server(uvicorn.Config(app, log_level="critical"))
+                    await server.serve(sockets=[sock])
             finally:
                 self._state_client = None
 
-        app.router.lifespan_context = serving_lifespan
-        server = uvicorn.Server(uvicorn.Config(app, log_level="critical"))
-        asyncio.run(server.serve(sockets=[sock]))
+        asyncio.run(serve())
 
     @classmethod
     def _config_cls(cls) -> type[BaseConfig]: