I’m using a sequential crew where the last task/agent makes an analysis based on the output of previous ones. The problem is that oftentimes the final task does not take into account the output data of those previous tasks, but makes up its own fake data.
Of course, inside the crew.py, I specifically include the context in the last task, like:
context=[self.task_a(), self.task_b()]
Any help please?
@Ger If you’re using the sequential process, you don’t need to use the context parameter because contexts are taken care of by default. As stated in the docs:
Sequential Process
This method mirrors dynamic team workflows, progressing through tasks in a thoughtful and systematic manner. Task execution follows the predefined order in the task list, with the output of one task serving as context for the next.
If contexts weren’t working with the sequential process, it would indicate a fundamental issue with the CrewAI SDK. While not impossible, this is highly unlikely. Such a problem would likely generate numerous reports on the forum.
Instead, I recommend reviewing and improving your agent and task configurations. See if that solves the issue.
Hi @rokbenko, thank you for the reply! Indeed, in the sequential process we don’t need to use the context parameter. However, in my case, I have this ‘chain’ of task execution:
task_a → task_b → task_c → task_d
The final one (task_d) has to get the outputs of task_a and task_c in order to produce the anticipated results. It does so most of them times, but once in a while it fails and uses fake data.
This is why only for this task I use the context parameter.
In this case, use Flows. Specifically, use the structured state management feature. See the docs.
I’ve used state management in my projects, and it worked perfectly for me every time. For example, take a look at how I handled state management in the main.py file of the Podcast Prepper project I built. I simply saved a variable to the state and passed it as an input when kicking off the next crew. You can check out my YouTube tutorial for a detailed walkthrough of the Podcast Prepper.
Thank you, I haven’t yet used flows but will try it for sure!
@Ger You’re welcome! 
Let me know how it turns out.
Hmm, the frontend is directly related to the current crew setup via the webhook I consume, in order to update after each task is completed. So, for now, I’m not going to flows and break it.
I’m pretty much sure many users implement similar sequential crews where at least one task gets context (output) from more than one previous tasks. Given that, and the problem I am faced with, is there anything additional I could try?