meeting-mom

Meeting Minutes (MOM) Generator

You transform messy meeting transcripts into minutes that people actually read.

Why this matters

Most meeting transcripts are walls of text nobody revisits. The goal here is to extract the signal — decisions made, questions still open, and who owes what by when — and present it so someone who skipped the meeting can catch up in 90 seconds.

Input

The user provides:

1. A transcript — could be a .txt file, .docx file, or pasted text. May have speaker labels (e.g., "John: I think we should...") or may be raw speech-to-text without them.
2. Meeting context (optional but helpful) — meeting title, date, attendees, purpose. If the user doesn't provide this, infer what you can from the transcript itself.

Reading the transcript

• For .txt or .md files: read directly
• For .docx files: use pandoc to extract text:

  pandoc input.docx -t plain -o /tmp/transcript.txt
  

  If pandoc isn't available, use the docx skill's approach (unpack ZIP, read XML).
• For pasted text: work with it directly in context

Output

You produce two files, saved to the user's workspace:

1. [meeting-name]-mom.html — A self-contained, single-file HTML page using the template in assets/mom-template.html
2. [meeting-name]-mom.md — The same content as clean Markdown

Use a sensible filename derived from the meeting topic or date (e.g., q1-planning-mom.html, 2026-02-26-standup-mom.md). Lowercase, hyphens, no spaces.

How to process the transcript

Step 1: Read the full transcript

Read the entire transcript before extracting anything. You need the full picture to understand context — something said in minute 5 might reframe what was decided in minute 45.

Step 2: Identify the structure

As you read, identify these elements:

• Participants: Who spoke? What were their roles (if inferable)?
• Topics discussed: Group the conversation into coherent topics/agenda items. Meetings rarely follow a clean agenda, so you'll need to cluster related discussion even if it was scattered across the timeline.
• Decisions: Anything where the group agreed on a direction. Look for phrases like "let's go with", "we decided", "agreed", "the plan is", consensus moments.
• Action items: Anything someone committed to doing. Look for "I'll", "can you", "let's make sure", "by Friday", "owner:", assignment language. Infer owners and deadlines from context — if someone says "I'll send that over tomorrow", the owner is that person and the deadline is tomorrow relative to the meeting date. Flag uncertain inferences with a ⚠️ marker so the user can verify.
• Open questions: Anything raised but not resolved. Parking lot items. "We should think about...", "TBD", "let's revisit", trailing discussions without conclusions.
• Key context/background: Important information shared that isn't a decision or action but matters for understanding (e.g., "Q3 revenue was down 12%", "the API migration is blocked on vendor response").