The system will process the inbox and suggest actions for each item. If an item is related to a project, it will recognize it and propose suitable next steps, such as creating a GitHub issue.
Implementation detaiils
Whenever a new to-do in the inbox gets created, we kick off an async job:
The job creates a set of function calls to abilities/tools, which do not get called.
They are prepared and saved in todo post meta.
The TODO UI can present these suggested action and the user can choose to perform them.
This should probably be achieved by using gpt-5 with thinking on the structured output mode. The structured output should be an array of types to the abilities/tools calls.
Alternatively, we can fake function calls wit the same architecture as abilites.
We DO NOT WANT to actually triggere these tools / abilities as part of this, we onnly want ot present suggestions
The system will process the inbox and suggest actions for each item. If an item is related to a project, it will recognize it and propose suitable next steps, such as creating a GitHub issue.
Implementation detaiils
Whenever a new to-do in the inbox gets created, we kick off an async job:
The job creates a set of function calls to abilities/tools, which do not get called.
They are prepared and saved in todo post meta.
The TODO UI can present these suggested action and the user can choose to perform them.
This should probably be achieved by using gpt-5 with thinking on the structured output mode. The structured output should be an array of types to the abilities/tools calls.
Alternatively, we can fake function calls wit the same architecture as abilites.
We DO NOT WANT to actually triggere these tools / abilities as part of this, we onnly want ot present suggestions