I record voice-journal entries during my evening walks using my favourite app, Monologue, on my phone.
The raw transcript is obviously quite messy. It captures wind blowing, ums, likes, basically, bird chirping, and whatever else is happening around me. Monologue even labels the background sounds, which is honestly pretty cool.
So, to get these voice entries back into Obsidian, I created a skill that cleans up filler words, fixes obvious grammar, adds wikilinks to entities like people, apps, and companies, and splits everything into short paragraphs.
The interesting part was not the cleanup itself. It was how much reinforcement agents need before they stop “improving” your writing. For journal entries, I don't want a summary. I want my exact voice and tone, just with less noise.
Haiku still fails to oblige and acts smart. Sonnet or GPT-mini gets the job done well.
It's a great way to just brain-dump while walking, but the raw transcripts are an absolute mess. Full of rambling sentences, filler words, and whatever weird ways my brain jumps between topics. This skill makes them usable again without making them sound less like me.
The stack
- Agent: Claude
- Tool:
skill-creatorskill