Feature Request: Extend Summarise using LLM task to print the results in the out dir

[acurat]

One of the use cases I have to is look at legacy COBOL code and generate summaries and use that output to create requirements documents for greenfield app rewrite.

I have been having a fair amount of success with RAG & LLMs but I really like the bottom up summarization approach rekt takes. I would like to see if we get better results with this tool.

It would be great if the task can:

1. Print/output all the individual summaries in a json(?) format.
2. Also, output the summary of the entire program in the text(?) format bringing all the AST and data structure summaries together.

Is this doable?

[avishek-sen-gupta]

1. How about a JSON which reflects the syntactic structure of the code, but has an additional summary field for every node?
2. Are you looking for all the summaries appended together in a flattened form? Or, a single LLM summary of all of these summaries?

[acurat]

1. That works 👍
2. I can append it from the JSON above or by fishing it out of Neo4j. So really looking for the latter - Single LLM summary of the entire program.

[avishek-sen-gupta]

The first piece of functionality should be working in commit a28e1ecf345cf9153a16afac9e93fcc2ce419868. The only commands you need to run are "BUILD_BASE_ANALYSIS WRITE_LLM_SUMMARY".
Note that this does not populate the graph DB with the summaries, but simply outputs it as a JSON file.
For now, populating the graph DB is a separate task.
Let me know if this fits your use case.

[acurat]

This works great, thanks!