• Search the codebase for all comments with the "@todo" tag.
• Create an index of all todos and which file, function & line number they appear in.
• Categorize them
• Group the ones that appear to refer to the same todo
• Add an estimate for the size / effort of each todo

This seems like a good job for an LLM.

• Try using repo2txt to compile & summarize the backend codebase in one file
• Use Claude.ai (or Gemini 1.5 Pro if z longer context window is needed) to query the codebase and extract and summarize the todos as described above
• Or write a script to extract the todos & some of their context, then use the LLMs just for the summarization steps