feat(agentic-rl): normalize int error_code to str in output models

Add a shared before-validator so RolloutOutput/RewardOutput accept int
error codes without hard-failing validation, while downstream always
receives str (the protocol type stays string).

Author: agentscope-ai-java

dashscope/finetune/reinforcement/component/data/base_data_model.py

@@ -26,6 +26,17 @@ def _generate_ro_id(length: int) -> str:
     return "ro-" + hashlib.sha1(uuid.uuid4().bytes).hexdigest()[:length]
 
 
+def _normalize_error_code(v: Any) -> Any:
+    """Coerce ``int`` error codes to ``str``; pass other values through.
+
+    The protocol keeps ``error_code`` as a string, so passing an int no
+    longer hard-fails validation and downstream always receives a str.
+    """
+    if isinstance(v, int) and not isinstance(v, bool):
+        return str(v)
+    return v
+
+
 # ========================================================================== #
 #                              Enum Definitions                              #
 # ========================================================================== #

dashscope/finetune/reinforcement/component/data/reward_output.py

@@ -5,11 +5,12 @@
 Data model definitions for Reward processor output results.
 """
 
-from typing import Dict, Optional
-from pydantic import BaseModel, Field
+from typing import Any, Dict, Optional
+from pydantic import BaseModel, Field, field_validator
 
 from dashscope.finetune.reinforcement.component.data.base_data_model import (
     TaskStatus,
+    _normalize_error_code,
 )
 
 
@@ -60,5 +61,11 @@ class RewardOutput(BaseModel):
         description="Error code if the reward computation failed.",
     )
 
+    @field_validator("error_code", mode="before")
+    @classmethod
+    def normalize_error_code(cls, v: Any) -> Any:
+        """Normalize int error codes to str (protocol stays string)."""
+        return _normalize_error_code(v)
+
     class Config:
         extra = "allow"

dashscope/finetune/reinforcement/component/data/rollout_output.py

@@ -6,12 +6,13 @@
 Corresponds to the RolloutOutput structure in protocol.py.
 """
 
-from typing import Optional
-from pydantic import BaseModel, Field
+from typing import Any, Optional
+from pydantic import BaseModel, Field, field_validator
 
 from dashscope.finetune.reinforcement.component.data.base_data_model import (
     TaskStatus,
     AgentOutput,
+    _normalize_error_code,
 )
 
 
@@ -45,5 +46,11 @@ class RolloutOutput(BaseModel):
         description="Error code (when failed).",
     )
 
+    @field_validator("error_code", mode="before")
+    @classmethod
+    def normalize_error_code(cls, v: Any) -> Any:
+        """Normalize int error codes to str (protocol stays string)."""
+        return _normalize_error_code(v)
+
     class Config:
         extra = "allow"

tests/unit/test_agentic_rl_components.py

@@ -279,6 +279,22 @@ def test_rollout_output_failure(self):
         assert output.error == "Model inference timeout"
         assert output.agent_output is None
 
+    def test_rollout_output_error_code_int_normalized(self):
+        output = RolloutOutput(
+            status=TaskStatus.FAILED,
+            error="Model inference timeout",
+            error_code=3001,
+        )
+        assert output.error_code == "3001"
+        assert isinstance(output.error_code, str)
+
+    def test_rollout_output_error_code_str_passthrough(self):
+        output = RolloutOutput(
+            status=TaskStatus.FAILED,
+            error_code="RATE_LIMIT",
+        )
+        assert output.error_code == "RATE_LIMIT"
+
     def test_rollout_output_defaults(self):
         output = RolloutOutput()
         assert output.status == TaskStatus.SUCCESS
@@ -335,6 +351,15 @@ def test_reward_output_failure(self):
         )
         assert output.status == TaskStatus.FAILED
 
+    def test_reward_output_error_code_int_normalized(self):
+        output = RewardOutput(
+            reward=Reward(reward_score=0.0),
+            status=TaskStatus.FAILED,
+            error_code=3002,
+        )
+        assert output.error_code == "3002"
+        assert isinstance(output.error_code, str)
+
     def test_reward_output_missing_reward(self):
         with pytest.raises(ValidationError):
             RewardOutput()