verifiers 0.2.2.dev38__tar.gz → 0.2.2.dev40__tar.gz

{verifiers-0.2.2.dev38 → verifiers-0.2.2.dev40}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: verifiers
-Version: 0.2.2.dev38
+Version: 0.2.2.dev40
 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-0.2.2.dev38 → verifiers-0.2.2.dev40}/tests/v1/conftest.py

@@ -17,7 +17,8 @@ one without `indirect=True` fails loudly.
 Every combination carries its axes' pytest marks, so subsets select with `-m`:
 
     uv run pytest tests/v1 -n auto                                # everything (needs modal setup)
-    uv run pytest tests/v1 -n auto -m "not prime and not modal"  # the CI set (host + docker only)
+    uv run pytest tests/v1 -n auto -m "not e2e"                   # deterministic CI matrix
+    uv run pytest tests/v1 -n auto -m "e2e and not prime and not modal"  # live CI job
     uv run pytest tests/v1 -n auto -m docker                      # any case touching the docker runtime
     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)
@@ -27,8 +28,8 @@ Marks: runtimes `subprocess` / `docker` / `prime` / `modal`, placement `colocate
 harnesses `null` / `bash` / `rlm` / `kimi_code` / `pi` / `pool` / `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 `-m "not prime and not modal"`.
+prime/modal provision real remote sandboxes (slow, infra-flaky, need setup), so they're local-only.
+CI runs deterministic tests across the Python matrix and the remaining live E2Es once.
 """
 
 import os

{verifiers-0.2.2.dev38 → verifiers-0.2.2.dev40}/verifiers/v1/harnesses/bash/program.py

@@ -1,6 +1,6 @@
 # /// script
 # requires-python = ">=3.10"
-# dependencies = ["openai", "mcp", "httpx", "tenacity"]
+# dependencies = ["openai", "mcp>=1.24.0,<2", "httpx", "tenacity"]
 # ///
 """Secrets arrive through argv so local tool subprocesses do not inherit them."""
 

{verifiers-0.2.2.dev38 → verifiers-0.2.2.dev40}/verifiers/v1/harnesses/null/program.py

@@ -1,6 +1,6 @@
 # /// script
 # requires-python = ">=3.11"
-# dependencies = ["openai", "mcp", "httpx", "tenacity"]
+# dependencies = ["openai", "mcp>=1.24.0,<2", "httpx", "tenacity"]
 # ///
 """The interception endpoint and secret arrive through argv rather than the environment."""
 

{verifiers-0.2.2.dev38 → verifiers-0.2.2.dev40}/verifiers/v1/harnesses/rlm/harness.py

@@ -94,6 +94,8 @@ class RLMHarness(Harness[RLMHarnessConfig]):
         ensure = shlex.quote(f"[ -x {RLM_BIN} ] || ({install})")
         guarded = f"mkdir -p {RLM_DIR} && flock {RLM_DIR}/install.lock sh -c {ensure}"
         env = {**self.config.resolved_env, "RLM_HOME": RLM_HOME}
+        extra_uv_args = env.get("RLM_EXTRA_UV_ARGS", "")
+        env["RLM_EXTRA_UV_ARGS"] = f"{extra_uv_args} --with mcp~=1.28".strip()
         result = await runtime.run(["sh", "-c", guarded], env)
         if result.exit_code != 0:
             raise RuntimeError(f"rlm install failed: {result.stderr.strip()[-500:]}")