verifiers 0.3.2.dev87__tar.gz → 0.3.2.dev88__tar.gz

{verifiers-0.3.2.dev87 → verifiers-0.3.2.dev88}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: verifiers
-Version: 0.3.2.dev87
+Version: 0.3.2.dev88
 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-0.3.2.dev87 → verifiers-0.3.2.dev88}/verifiers/v1/acp/__init__.py

@@ -58,6 +58,7 @@ class ACPConfig:
     mcp_urls: dict[str, str] | None = None
     system_prompt: str | None = None
     session_meta: JsonObject | None = None
+    client_capabilities: JsonObject | None = None
 
 
 class ACPHarness(Harness[ConfigT]):
@@ -265,6 +266,7 @@ class ACPHarnessSession(HarnessSession):
             "mcp_urls": self.mcp_urls,
             "system_prompt": self.config.system_prompt or "",
             "session_meta": self.config.session_meta or {},
+            "client_capabilities": self.config.client_capabilities or {},
         }
         async with self._lock:
             if self._closed:

{verifiers-0.3.2.dev87 → verifiers-0.3.2.dev88}/verifiers/v1/acp/runner.py

@@ -194,7 +194,9 @@ class ACPSession:
             self.connection = agent_process[0]
             initialized = await self.connection.initialize(
                 protocol_version=PROTOCOL_VERSION,
-                client_capabilities=ClientCapabilities(),
+                client_capabilities=ClientCapabilities.model_validate(
+                    config["client_capabilities"]
+                ),
             )
             self.capabilities = initialized.agent_capabilities
             session = await self.connection.new_session(

{verifiers-0.3.2.dev87 → verifiers-0.3.2.dev88}/verifiers/v1/harnesses/codex/harness.py

@@ -6,7 +6,7 @@ import logging
 import re
 from collections import Counter
 
-from verifiers.v1.acp import ACPConfig, ACPHarness
+from verifiers.v1.acp import ACPConfig, ACPHarness, ACPTurn
 from verifiers.v1.clients import ModelContext
 from verifiers.v1.configs.harness import HarnessConfig, PinnedVersion
 from verifiers.v1.harnesses.node import NODE_BIN_DIR, ensure_node
@@ -47,6 +47,16 @@ class CodexHarness(ACPHarness[CodexHarnessConfig]):
     SUPPORTS_MCP = True
     SUPPORTS_SKILLS = True
 
+    def acp_turn_result(self, trace: Trace, result: ACPTurn) -> None:
+        # codex-acp returns terminal failures in metadata with stop_reason=end_turn.
+        failure = (
+            result.response_metadata.get("jetbrains", {})
+            .get("air", {})
+            .get("sessionFailure")
+        )
+        if failure and failure["phase"] == "active":
+            raise RuntimeError(f"Codex {failure['category']}: {failure['safeMessage']}")
+
     async def setup(self, runtime: Runtime) -> None:
         await self.install_skills(runtime, SKILLS_DIR)
         await ensure_node(runtime)
@@ -101,6 +111,13 @@ class CodexHarness(ACPHarness[CodexHarnessConfig]):
             # Codex reads MCP servers from the config written by build_env().
             mcp_urls={},
             system_prompt=system_prompt,
+            client_capabilities={
+                "_meta": {
+                    "jetbrains": {
+                        "air": {"version": 1, "capabilities": ["sessionFailure"]}
+                    }
+                }
+            },
         )
 
     async def cleanup(self, trace: Trace, runtime: Runtime) -> None: