[BUG] Workflow task cannot access results from dependent tasks

[foxxqin]

Checks

• I have updated to the lastest minor and patch version of Strands
• I have checked the documentation and this is not expected behavior
• I have searched ./issues and there are no duplicates of my issue

Strands Version

0.2.3

Tools Package Version

1.3.0

Tools used

1. Workflow

Python Version

3.13.5

Operating System

macOS 15.6

Installation Method

pip

Steps to Reproduce

1. Create an Agent using workflow tool.
2. Define the task as following:
   tasks=[ { "task_id": "story_generation", "description": "Generate a simple story (few paragraph).", "system_prompt": "Generate a simple story in English.", "model_provider": "bedrock", "model_settings": {"model_id": "us.anthropic.claude-3-7-sonnet-20250219-v1:0"}, "priority": 5 }, { "task_id": "chinses_translate", "description": "Translate the generated story in Chinese.", "dependencies": ["story_generation"], "system_prompt": "Translate the generated story in Chinese.", "model_provider": "bedrock", "model_settings": {"model_id": "us.anthropic.claude-3-7-sonnet-20250219-v1:0"}, "priority": 3 }, { "task_id": "japanese_translate", "description": "Translate the generated story in Japanese.", "dependencies": ["story_generation"], "system_prompt": "Translate the generated story in Japanese.", "model_provider": "bedrock", "model_settings": {"model_id": "us.anthropic.claude-3-7-sonnet-20250219-v1:0"}, "priority": 3 }, { "task_id": "final_story", "description": "Generate a pretty HTML of the story", "dependencies": ["story_generation", "chinses_translate", "japanese_translate"], "model_provider": "bedrock", "model_settings": {"model_id": "us.anthropic.claude-3-7-sonnet-20250219-v1:0"}, "system_prompt": "Generate a pretty HTML of the story, English version first, followed by Chinese version then Japanese version of the story.", "priority": 2 } ]
3. Start the workflow

Expected Behavior

As documented in Strands Agent website, the task should have context of result of dependent task from previous steps.

Actual Behavior

I got the following error from "chinese_translate" and "japanese_translate" task: I don't see the story you need to translate. And I checked the model invocation log in Bedrock, the result of story generation is not in the context for "chinese_translate" and "japanese_translate" tasks.

Additional Context

No response

Possible Solution

No response

Related Issues

No response

[steven-sg]

+1

We're currently doing something as follows to workaround the issue for the time being.

class PatchedWorkflowManager(StrandsWorkflowManager):
    """
    Patched workflow manager to work around speghetti in Strands that expects the content attribute to exist.
    """

    def _create_task_agent(self, *args, **kwargs) -> StrandsAgent:
        # Establish the original behavior
        create_task_agent_result = super()._create_task_agent(*args, **kwargs)
        original_call = create_task_agent_result.__call__

        @wraps(original_call)
        def patched_call(self, *args, **kwargs):
            patched_result = original_call(*args, **kwargs)

            # Add the content attribute if it doesn't already exist
            if not hasattr(patched_result, "content"):
                setattr(patched_result, "content", [{"text": str(patched_result)}])
            return patched_result

        # Swap out the class at runtime with a patched __call__
        create_task_agent_result.__class__ = type(
            f"{StrandsAgent.__name__}Patched", (StrandsAgent,), {"__call__": patched_call}
        )
        return create_task_agent_result


@tool
def workflow(
    workflow_id: str, workflow_tasks: list[Any], agent: StrandsAgent | None = None
) -> dict[str, Any]:
    """
    Execute a workflow with the specified ID and tasks.

    Args:
        workflow_id: The ID of the workflow to execute (from JSON registry)
        workflow_tasks: The tasks corresponding to a workflow.
        agent: Parent agent
    """
    workflow_manager = PatchedWorkflowManager(parent_agent=agent)
    _ = workflow_manager.create_workflow(workflow_id, workflow_tasks)
    return workflow_manager.start_workflow(workflow_id)

[mehtarac]

Hi, thank you for raising this issue. We will reproduce it and also retry with the graph primitive/tool because we can achieve workflow using a graph as well.