Skip to content

Commit c9edce5

Browse files
committed
add
1 parent ab28244 commit c9edce5

File tree

2 files changed

+8
-15
lines changed

2 files changed

+8
-15
lines changed

eval_protocol/mcp/execution/manager.py

Lines changed: 5 additions & 12 deletions
Original file line numberDiff line numberDiff line change
@@ -169,21 +169,14 @@ async def _execute_with_semaphore(idx):
169169
max_tool_calls=getattr(policy, "max_tools_per_turn", None),
170170
)
171171
if trajectory.terminated:
172-
if trajectory.termination_reason in {
173-
TerminationReason.CONTROL_PLANE_SIGNAL,
174-
TerminationReason.USER_STOP,
175-
}:
176-
evaluation_rows[idx].rollout_status.status = "finished"
177-
elif trajectory.termination_reason in {TerminationReason.MAX_STEPS, TerminationReason.INTERRUPTED}:
178-
evaluation_rows[idx].rollout_status.status = "stopped"
179-
evaluation_rows[idx].rollout_status.error_message = trajectory.control_plane_summary.get(
180-
"termination_reason", trajectory.termination_reason
181-
)
182-
else:
172+
if trajectory.termination_reason == TerminationReason.ERROR:
183173
evaluation_rows[idx].rollout_status.status = "error"
184-
evaluation_rows[idx].rollout_status.error_message = trajectory.control_plane_summary.get(
174+
evaluation_rows[idx].rollout_status.reason= trajectory.control_plane_summary.get(
185175
"error_message", None
186176
)
177+
else:
178+
evaluation_rows[idx].rollout_status.status = "finished"
179+
evaluation_rows[idx].rollout_status.reason = trajectory.termination_reason
187180
else:
188181
evaluation_rows[idx].rollout_status.status = "running"
189182

eval_protocol/models.py

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -229,10 +229,10 @@ class RolloutStatus(BaseModel):
229229
error: Rollout failed.
230230
stopped: Rollout terminated unexpectedly (e.g. max step, control plane signal, user stop).
231231
"""
232-
status: Literal["running", "finished", "error", "stopped"] = Field(
233-
"finished", description="Status of the rollout."
232+
status: Literal["running", "finished", "error"] = Field(
233+
"running", description="Status of the rollout."
234234
)
235-
error_message: Optional[str] = Field(None, description="Error message if the rollout failed.")
235+
reason: Optional[str] = Field("", description="reason of the rollout status, mapped to values in TerminationReason")
236236

237237

238238
class EvaluationRow(BaseModel):

0 commit comments

Comments
 (0)