diff --git a/src/conductor/engine/workflow.py b/src/conductor/engine/workflow.py
index 484192b..89f812a 100644
--- a/src/conductor/engine/workflow.py
+++ b/src/conductor/engine/workflow.py
@@ -1448,6 +1448,23 @@ async def _execute_loop(self, current_agent_name: str) -> dict[str, Any]:
                                 "stderr": script_output.stderr,
                                 "exit_code": script_output.exit_code,
                             }
+                            # Auto-parse JSON stdout: if stdout is valid JSON
+                            # object, merge its fields into output so they're
+                            # accessible as output.field_name in templates and
+                            # route conditions (like LLM structured outputs).
+                            try:
+                                parsed = json.loads(script_output.stdout)
+                                if isinstance(parsed, dict):
+                                    shadowed = set(parsed.keys()) & set(output_content.keys())
+                                    if shadowed:
+                                        logger.debug(
+                                            "Script '%s' JSON output shadows built-in fields: %s",
+                                            agent.name,
+                                            ", ".join(sorted(shadowed)),
+                                        )
+                                    output_content.update(parsed)
+                            except (json.JSONDecodeError, ValueError):
+                                pass
                             self.context.store(agent.name, output_content)
                             self.limits.record_execution(agent.name)
                             self.limits.check_timeout()
diff --git a/tests/test_engine/test_workflow.py b/tests/test_engine/test_workflow.py
index 14843a4..589e1b6 100644
--- a/tests/test_engine/test_workflow.py
+++ b/tests/test_engine/test_workflow.py
@@ -298,6 +298,46 @@ def mock_handler(agent, prompt, context):
         # Workflow.input.goal should not be in agent2's context since it's not in input list
         assert "other" not in agent2_context.get("workflow", {}).get("input", {})
 
+    @pytest.mark.asyncio
+    async def test_script_json_stdout_parsed_into_output(self) -> None:
+        """Test that script stdout containing JSON is auto-parsed into output fields."""
+        config = WorkflowConfig(
+            workflow=WorkflowDef(name="json-script", entry_point="detector"),
+            agents=[
+                AgentDef(
+                    name="detector",
+                    type="script",
+                    command="pwsh",
+                    args=[
+                        "-Command",
+                        'Write-Output \'{"plan_exists": true, "route": "planning"}\'; exit 0',
+                    ],
+                    routes=[
+                        RouteDef(to="planner", when="route == 'planning'"),
+                        RouteDef(to="$end"),
+                    ],
+                ),
+                AgentDef(
+                    name="planner",
+                    type="script",
+                    command="pwsh",
+                    args=["-Command", "Write-Output 'done'; exit 0"],
+                    routes=[RouteDef(to="$end")],
+                ),
+            ],
+        )
+
+        provider = CopilotProvider(mock_handler=lambda a, p, c: {})
+        engine = WorkflowEngine(config, provider)
+        await engine.run({})
+
+        det = engine.context.agent_outputs["detector"]
+        assert det["plan_exists"] is True
+        assert det["route"] == "planning"
+        assert "stdout" in det
+        assert det["exit_code"] == 0
+        assert "planner" in engine.context.agent_outputs
+
 
 class TestWorkflowEngineRouting:
     """Tests for workflow routing."""