What problem does it solve? Meeting recordings and transcripts contain decisions, commitments, and sentiment that get lost without durable records, yet they are often the most confidential artifacts in a project. This Skill produces structured meeting minutes (ATAs) and qualitative summaries while keeping sensitive audio and transcripts inside the project's data perimeter. ## Core Features & Use Cases - Confidentiality-aware pipeline: Classifies material sensitivity, consults data governance (ADRs), and processes confidential content on-premise with local models. - Local transcription: Uses faster-whisper for local ASR of audio/video, with glossary prompts for technical terms and cross-checking against automatic transcription speaker labels. - Hybrid workload split: Routes qualitative synthesis (summary, sentiment) to a local LLM (e.g., Ollama) and structured fields to a more capable model only with explicit authorization. - Anti-hallucination and provenance: Records only what was said, marks missing data as "não registrado", and logs what was processed on-premise versus externally. - Use Case: Given a confidential client interview recording (.mp4), transcribe it locally with Whisper, produce an ATA with decisions and action items plus a sentiment/observations section, and store it in the restricted project area without committing it. ## Quick Start Use the meeting-minutes skill to transcribe this meeting recording and produce an ATA with decisions, action items, and a sentiment summary.
