Maybe a method to overcome model context length limitations

[cexer]

Even Claude only supports max 200K contexts, which is not enough for larger projects or for more iterations.

There is an optimization method that I think of, don't know if it is feasible:
1.When task is completed and archived as a file, make a summary, which can include a reference to related file path (rather than the full file content).
2.In subsequent tasks, the request contains only the summary text of the previous tasks, rather than the full conversation contents.
3.If more specific information from the previous Task is needed, retrieve the archive file of the previous Task to read.

There is a project use this method : https://github.com/sztimhdd/Looping-Claude