verifiers 0.2.2.dev92__tar.gz → 0.2.2.dev94__tar.gz

{verifiers-0.2.2.dev92 → verifiers-0.2.2.dev94}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: verifiers
-Version: 0.2.2.dev92
+Version: 0.2.2.dev94
 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.dev92 → verifiers-0.2.2.dev94}/tests/v1/test_e2e.py

@@ -231,11 +231,12 @@ async def test_acp_resume_with_tool(run_v1, harness, harness_runtime, tmp_path):
     assert trace.ok, trace.errors
     assert trace.stop_condition == "user_closed"
     assert trace.rewards["resumed"].score == 1.0
-    assert trace.tools  # ACP-native tools, or Pi's MCP adapter meta-tool
     segments = trace.info["acp_segments"]
     assert len(segments) == 2
     assert segments[0]["terminated"] is False
     assert segments[1]["terminated"] is False
+    # Native MCP tools need not appear in the intercepted model request that
+    # populates trace.tools; the ACP transcript is the source of truth for use.
     assert "tool" in segments[1]["roles"]
     assert segments[1]["tool_outputs"]
     if harness == "rlm":

{verifiers-0.2.2.dev92 → verifiers-0.2.2.dev94}/verifiers/v1/acp/runner.py

@@ -36,6 +36,7 @@ from acp.schema import (
 )
 
 MAX_PACKET_BYTES = 128 * 1024 * 1024
+LATE_UPDATE_GRACE_SECONDS = 1.0
 
 
 class VerifiersACPClient(Client):
@@ -43,6 +44,7 @@ class VerifiersACPClient(Client):
         self.visible_reply = ""
         self.message_id: str | None = None
         self.tool_calls: dict[str, str] = {}
+        self.output_changed = asyncio.Condition()
 
     def reset(self) -> None:
         self.visible_reply = ""
@@ -50,22 +52,23 @@ class VerifiersACPClient(Client):
         self.tool_calls = {}
 
     async def session_update(self, session_id: str, update: Any, **kwargs: Any) -> None:
-        if isinstance(update, ToolCall):
-            self.tool_calls[update.tool_call_id] = update.status or "pending"
-            return
-        if isinstance(update, ToolCallUpdate):
-            if update.status:
-                self.tool_calls[update.tool_call_id] = update.status
-            return
-        if not isinstance(update, AgentMessageChunk) or not isinstance(
-            update.content, TextContentBlock
-        ):
-            return
-        message_id = getattr(update, "message_id", None)
-        if message_id is not None and message_id != self.message_id:
-            self.visible_reply = ""
-            self.message_id = message_id
-        self.visible_reply += update.content.text
+        async with self.output_changed:
+            if isinstance(update, ToolCall):
+                self.tool_calls[update.tool_call_id] = update.status or "pending"
+            elif isinstance(update, ToolCallUpdate):
+                if update.status:
+                    self.tool_calls[update.tool_call_id] = update.status
+            elif isinstance(update, AgentMessageChunk) and isinstance(
+                update.content, TextContentBlock
+            ):
+                message_id = getattr(update, "message_id", None)
+                if message_id is not None and message_id != self.message_id:
+                    self.visible_reply = ""
+                    self.message_id = message_id
+                self.visible_reply += update.content.text
+            else:
+                return
+            self.output_changed.notify_all()
 
     async def request_permission(
         self,
@@ -172,6 +175,25 @@ async def prompt(
     except RequestError as error:
         detail = error.data.get("details") if isinstance(error.data, dict) else None
         raise RuntimeError(detail or str(error)) from error
+
+    # ACP 0.11 dispatches notifications in background tasks but resolves a request
+    # response directly in its receive loop. An agent that sends its final
+    # session/update immediately before session/prompt returns can therefore wake
+    # this coroutine before the update handler has run. Wait specifically for text:
+    # a completed tool update may also arrive first and must not hide a later reply.
+    def has_visible_reply() -> bool:
+        return bool(client.visible_reply.strip())
+
+    if not has_visible_reply():
+        async with client.output_changed:
+            try:
+                await asyncio.wait_for(
+                    client.output_changed.wait_for(has_visible_reply),
+                    timeout=LATE_UPDATE_GRACE_SECONDS,
+                )
+            except asyncio.TimeoutError:  # noqa: UP041 - Python 3.10 compatibility
+                pass
+
     tool_statuses = list(client.tool_calls.values())
     completed_tool_turn = (
         config.get("allow_empty_tool_reply", False)
@@ -179,7 +201,7 @@ async def prompt(
         and bool(tool_statuses)
         and all(status in ("completed", "failed") for status in tool_statuses)
     )
-    if not client.visible_reply.strip() and not completed_tool_turn:
+    if not has_visible_reply() and not completed_tool_turn:
         raise RuntimeError(
             "ACP agent produced no visible reply "
             f"(stop_reason={response.stop_reason}, tool_statuses={tool_statuses})"

{verifiers-0.2.2.dev92 → verifiers-0.2.2.dev94}/verifiers/v1/runtimes/prime.py

@@ -308,14 +308,16 @@ class PrimeRuntime(Runtime):
     async def run_background(
         self, argv: list[str], env: dict[str, str], log: str
     ) -> None:
-        # `&` backgrounds inside the sandbox; the job returns immediately, the process
-        # lives until the sandbox is deleted in stop().
-        inner = f"nohup {shlex.join(argv)} > {shlex.quote(log)} 2>&1 &"
-        result = await self.run(["sh", "-c", inner], env)
-        if result.exit_code != 0:
-            raise SandboxError(
-                f"prime background launch failed: {result.stderr.strip()}"
+        command = f"exec {shlex.join(argv)} > {shlex.quote(log)} 2>&1"
+        try:
+            await self._client.start_background_job(
+                self.info.id,
+                command,
+                working_dir=self.config.workdir,
+                env=env,
             )
+        except Exception as e:
+            raise SandboxError(f"prime background launch failed: {e}") from e
 
     async def _read(self, path: str) -> bytes:
         # Avoid background-job log limits and base64 overhead by downloading binary data directly.