Meeting Assistant

Problem

Meeting transcripts from voice-to-text tools often contain filler words, homophone errors, fragmented discussion, and small talk. Using them as minutes is hard to read, while decisions and action items are easy to miss. This skill works from already textualized transcripts and solves two tasks: cleaning the transcript into readable prose and condensing scattered discussion into structured minutes.

How It Works

The workflow is serial and cannot be skipped or merged. Task 1 polishes the transcript: it fixes grammar, removes redundant fillers such as “um” and “uh”, merges repeated points, and reorganizes content by topic or theme rather than preserving a raw timeline. After Task 1, it asks whether to continue. Task 2 extracts structured minutes: it groups discussion by agenda item, preserves hard details such as decisions, dates, names, numbers, standards, and ownership, and marks key information with bold, tables, and block quotes. If a speaker name cannot be confirmed, it may mark it as [to confirm: possibly XX] instead of inventing one.

Boundaries

This skill expects the user to provide a full meeting transcript. It is suitable for internal meetings, customer interviews, and project retrospectives. It does not dispatch tasks, manage calendars, query external systems, or infer facts beyond the supplied context. If key speaker identity is ambiguous, users should supplement context. If an action-item table is needed, users should state the desired output structure in the prompt.