meeting-minutes-taker_skill

Overview

This skill transforms raw meeting transcripts into high-fidelity, structured meeting minutes with iterative human-in-the-loop review. It generates evidence-backed minutes (decisions, action items, quotes), performs speaker identification when labels are generic, and supports intelligent file naming and multi-pass merging to avoid content loss. The workflow is designed for production use, preserving intermediate files for audit and iteration.

How this skill works

The skill reads a transcript and optional project context file, performs feature-based speaker identification, and runs multiple independent complete-minute generations in parallel. It writes each pass to transcript-specific intermediate files, UNION-merges all versions (including diagrams and quotes), and runs a final completeness review against the original transcript. The draft is then presented to the user for confirmation and iterative edits until approval.

When to use it

• You have a raw meeting transcript and need formal, evidence-based minutes or summaries.
• Multiple versions of minutes must be merged without losing content or diagrams.
• Existing minutes need auditing against the original transcript to find missing items.
• Transcript uses anonymous labels like "Speaker 1/2/3" and you want accurate speaker mapping.
• You need an output file with an intelligent filename and preserved intermediate artifacts for review.

Best practices

• Pre-process low-quality transcripts with a transcript-cleaner before generating minutes.
• Provide a context.md team directory to improve speaker identification accuracy.
• Confirm suggested intelligent filename and any speaker mappings before finalizing.
• Run three parallel complete-pass generations and UNION-merge to minimize omissions.
• Keep intermediate files in a transcript-specific intermediate/ folder for auditing.

Example use cases

• Turn an hour-long product design transcript into a structured minutes document with decisions, action items, and supporting quotes.
• Merge minutes from two different AI tools into a single authoritative record without losing content.
• Identify and map anonymous speakers to a team directory to assign action owners correctly.
• Generate Mermaid diagrams from architecture discussion segments and include them in minutes.
• Iteratively refine minutes with stakeholders until every item is validated against the transcript.