Nothing to hand? Load the — a manager who never quite says the project is late — the , where the lead talks 60% and one voice never gets in, or the , which should come back strong.
Paste the transcript
Speaker-labeled lines from any source — Zoom, Meet, Granola, Otter, Fireflies — as pasted text or a dropped .vtt / .srt / .txt file, read locally. The free speaker table counts what can be counted before any AI runs: turns, talk share, questions, filler words, hedging phrases and cut-off turns, per speaker. Name yourself so the coaching centers on you, and pick a focus — conflict, balance, listening or facilitation.
The AI reviews it
A strong/mixed/needs-work verdict; quoted patterns — hedging on critical feedback, tension deflected, interruptions, missed listening moments — each with the exact line, why it matters and the better phrasing; a five-area communication health check; speaking statistics; and a twelve-item checklist scored pass, fail or not-observable. Every speaker the free table detected is weighed or explicitly set aside, and the review's numbers are reconciled against the mechanical counts.
Practice and re-measure
Strengths worth keeping, a growth plan, and ordered next steps starting with something usable in your very next meeting — tick them off as you practise. Signed in, your review history is saved to your SkillSafe account, so it follows you to another device; signed out it stays in this browser only. Compare any past review with your latest to see which checklist items flipped, which patterns keep coming back, and how talk share, questions, fillers and interruptions moved. Patterns and the speaker table export as CSV; the report as Markdown or JSON.
Derived from the @composiohq/meeting-insights-analyzer skill.