humanize-notes

What This Skill Does

Takes raw, messy brain-dumps or voice-to-text notes and rewrites them into clean, well-structured, natural-sounding paragraphs. The result should read like a thoughtful human wrote it, not like a machine formatted it. Produces the cleaned text in chat.

Before rewriting, read and apply the shared writing principles in ../_shared/human-voice.md (sound human, preserve voice, don't invent, stay proportional, plain punctuation).

Steps

1. Gather the raw notes.
   • If notes were passed inline as arguments, use $ARGUMENTS as the content.
   • If no arguments were given, ask the user to paste their raw notes or transcript.
2. Clean up the mechanics.
   • Fix grammar, spelling, and punctuation.
   • Remove filler words and verbal tics ("um", "uh", "like", "you know", "sort of").
   • Strip transcription artifacts (false starts, repeated words, stray timestamps, "[inaudible]" markers).
   • Collapse repetition: if the same point is made twice, keep the clearest version.
3. Structure the content.
   • Reorganize scattered thoughts into a logical order.
   • Default to flowing paragraphs.
   • Add short, plain headings ONLY when the notes clearly cover multiple distinct topics. Don't force headings onto single-topic notes.
4. Keep it human. Apply the shared human-voice principles. For notes specifically, keep the writer's first-person perspective (a brain-dump is usually the writer thinking out loud, so "I", "we", and "my" should survive the rewrite).
5. Present the result in a clearly delimited block so it's easy to copy. Then offer to adjust (more concise, more formal, add headings, break into sections, etc.).

Notes

The shared human-voice principles above already cover not inventing content, preserving the writer's voice, staying proportional, and plain punctuation. These are specific to cleaning up notes:

• Preserve meaning. Reorganizing scattered thoughts into a logical order is the goal; changing what the person actually meant is not.
• Mechanics only when in doubt. If you can't tell whether something is a real point or a transcription artifact, keep it and flag it rather than deleting it.