This turns meeting transcripts into structured minutes through iterative refinement, and it's built for messy real-world scenarios. If your transcript has "Speaker 1, Speaker 2" labels, it uses word count and speaking patterns to identify who's who, optionally matching against a team directory. The multi-turn generation approach runs three parallel passes and merges them to avoid losing content, which is smart if you've ever had an LLM drop action items. It auto-generates filenames from content, integrates with transcript cleanup tools, and includes a human review loop with cross-AI comparison support. The speaker identification logic alone makes this worth trying if you're dealing with auto-transcribed meetings.
npx skills add https://github.com/daymade/claude-code-skills --skill meeting-minutes-taker