What problem does it solve?
Raw meeting transcripts are long, noisy, and hard to act on; this Skill converts them into structured, evidence-based meeting minutes with decisions, action items, and speaker attribution while minimizing content loss.
Core Features & Use Cases
- Multi-pass generation with UNION merge: Spawns parallel subagents that each produce complete minutes, then merges all versions so no decision, action item, or diagram is lost.
- Speaker identification: Resolves anonymous labels like "Speaker 1/2" via source-platform relabeling first, with feature-analysis fallback mapped against a team directory in context.md.
- Retrieval self-test: A fresh-context subagent extracts future-query claims from the transcript and hit-tests them against the minutes, promoting missed directives with verbatim quotes.
- Use Case: Given a Feishu Minutes export of a two-hour API design review, produce minutes with key decisions (each backed by quotes), an action item table with owners and deadlines, Mermaid sequence diagrams, and a parking lot of deferred items.
Quick Start
Ask the AI to generate meeting minutes from your transcript file, optionally providing a context.md team directory for speaker identification.