verifiers 0.3.2.dev141__tar.gz → 0.3.2.dev142__tar.gz

{verifiers-0.3.2.dev141 → verifiers-0.3.2.dev142}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: verifiers
-Version: 0.3.2.dev141
+Version: 0.3.2.dev142
 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.dev141 → verifiers-0.3.2.dev142}/tests/v1/fixtures/echo_acp_resume_v1.py

@@ -52,9 +52,12 @@ class ACPResumeTask(vf.Task[vf.TaskData, vf.State, ACPResumeTaskConfig]):
 class ACPResumeEnv(vf.SingleAgentEnv):
     async def run(self, task, agents):
         async with agents.agent.interaction(task) as interaction:
+            # Stated as a fact, not a "do not" instruction: OpenAI's content filter
+            # refuses "remember ... do not call tools or store it" as an injection.
             first = await interaction.turn(
-                f"Remember the codeword {CODEWORD} for this conversation only. "
-                "Do not call tools or store it in persistent memory. Reply with exactly READY."
+                f"The codeword for this conversation is {CODEWORD}. Keep it in the "
+                "conversation context only, without tools or persistent memory. "
+                "Reply with exactly READY."
             )
             segments = [first]
             if not first.terminated:

{verifiers-0.3.2.dev141 → verifiers-0.3.2.dev142}/tests/v1/fixtures/echo_v1.py

@@ -54,13 +54,15 @@ class EchoTaskset(vf.Taskset[EchoTask, EchoConfig]):
     def load(self) -> list[EchoTask]:
         return [
             # Keep coding-agent harnesses on the direct-response path instead of
-            # spending this single-turn smoke task on a tool call.
+            # spending this single-turn smoke task on a tool call. Phrased as an
+            # instruction rather than a prohibition: OpenAI's content filter refuses
+            # "Do not call tools ... include this exact phrase" as an injection.
             EchoTask(
                 EchoData(
                     idx=i,
                     prompt=(
-                        "Do not call tools or execute code. Reply immediately and include "
-                        f"this exact phrase in your final response: {phrase}"
+                        "Answer directly without using tools or running code. Your reply "
+                        f"must contain this exact phrase: {phrase}"
                     ),
                     system_prompt=SYSTEM,
                     answer=phrase,

{verifiers-0.3.2.dev141 → verifiers-0.3.2.dev142}/verifiers/v1/harnesses/codex/harness.py

@@ -61,14 +61,15 @@ class CodexHarness(ACPHarness[CodexHarnessConfig]):
     SUPPORTS_TOOL_INTERCEPTION = True
 
     def acp_turn_result(self, trace: Trace, result: ACPTurn) -> None:
-        # codex-acp returns terminal failures in metadata with stop_reason=end_turn.
+        # codex-acp returns terminal failures in metadata with stop_reason=end_turn; a
+        # `warning` is a retry it already recovered from.
         failure = (
             result.response_metadata.get("jetbrains", {})
             .get("air", {})
             .get("sessionFailure")
         )
-        if failure and failure["phase"] == "active":
-            raise RuntimeError(f"Codex {failure['category']}: {failure['safeMessage']}")
+        if failure and failure["severity"] == "error":
+            raise RuntimeError(f"Codex {failure['category']}: {failure['title']}")
 
     async def setup(self, runtime: Runtime) -> None:
         await ensure_node(runtime)