fix(eval): surfaced subagent abort reason

- Used `||` so empty stderr no longer masks the real abort reason.

packages/coding-agent/src/eval/__tests__/agent-bridge.test.ts

@@ -231,6 +231,27 @@ describe("runEvalAgent", () => {
 		});
 		await expect(runEvalAgent({ prompt: "fail" }, { session: makeSession() })).rejects.toThrow("boom");
 	});
+
+	it("surfaces the abort reason when an aborted subagent has empty stderr", async () => {
+		// An aborted subagent returns exitCode 1 with stderr "" and error
+		// undefined; the real reason lives in abortReason. The bridge must not
+		// collapse the failure message to "" (which the Python prelude renders as
+		// the info-free "bridge call '__agent__' failed").
+		mockAgents();
+		vi.spyOn(taskExecutor, "runSubprocess").mockImplementation(async options =>
+			singleResult(options, {
+				exitCode: 1,
+				output: "",
+				stderr: "",
+				aborted: true,
+				abortReason: "Subagent runtime limit exceeded (task.maxRuntimeMs=1000)",
+			}),
+		);
+
+		await expect(runEvalAgent({ prompt: "hello" }, { session: makeSession() })).rejects.toThrow(
+			"Subagent runtime limit exceeded (task.maxRuntimeMs=1000)",
+		);
+	});
 });
 
 describe("agent() through eval runtimes", () => {

packages/coding-agent/src/eval/agent-bridge.ts

@@ -280,7 +280,7 @@ export async function runEvalAgent(args: unknown, options: EvalAgentBridgeOption
 
 	if (result.exitCode !== 0 || result.error) {
 		const failureMessage =
-			result.error ?? result.stderr ?? result.abortReason ?? `agent() subagent '${agentName}' failed.`;
+			result.error || result.stderr || result.abortReason || `agent() subagent '${agentName}' failed.`;
 		throw new ToolError(failureMessage);
 	}