meeting-insights-analyzer — Claude Code skill: meeting review

What meeting-insights-analyzer does

Point the skill at a folder of transcripts and it looks for habits that repeat across meetings. How much of each call you spend talking. How often you ask instead of assert. Where your phrasing goes soft. It also counts fillers such as “um”, “like” and “you know”, and flags the moments when they spike.

The skill ships as an instruction file for the agent. It scans the folder for .txt, .md, .vtt, .srt and .docx, checks whether speaker labels and timestamps are present, confirms the date range, and locates your name in the transcripts. If you have not said what you want out of the analysis, it asks first. Then it walks the patterns and backs each one with a timestamped example: what happened, why it matters, what to say instead.

Transcripts have to exist already — the skill does not transcribe audio. Speaker labels and timestamps matter just as much. Without them there is no speaking ratio to compute and no way to tie an example to a moment in the call. A flat wall of text with no speaker turns will stall the analysis at the first step.