This is a great example of making the system trustworthy: Inbox → scripted processing → git diff review → reset if needed. The diff-as-trust-layer feels like the missing piece for a lot of “AI organizes my notes” ideas.
Also +1 on “limited rules” being more important than fancy models — once the structure is predictable, Claude becomes a genuinely useful retrieval and research assistant.
Out of curiosity, where does the loop close for you today? After “deep research” produces a conclusion, do you write back into a MOC / decision log / project note, or does it mostly live in chat + commits? I’m exploring a similar loop (context → retrieval → suggestion → human review → write-back) with the same constraint: keep it auditable and reversible (diff-friendly). Details are in my HN profile/bio if you want to compare approaches.
Also +1 on “limited rules” being more important than fancy models — once the structure is predictable, Claude becomes a genuinely useful retrieval and research assistant.
Out of curiosity, where does the loop close for you today? After “deep research” produces a conclusion, do you write back into a MOC / decision log / project note, or does it mostly live in chat + commits? I’m exploring a similar loop (context → retrieval → suggestion → human review → write-back) with the same constraint: keep it auditable and reversible (diff-friendly). Details are in my HN profile/bio if you want to compare approaches.