Build meeting protocols from transcripts in tmp/: validate inputs (Voice Recognition transcript required), generate artifacts under notes/meetings/artifacts/, create a meeting note under notes/meetings/ (YYYY-MM-DD-slug.md) with a copy-ready Russian protocol block, and update notes/work/typos.md via kb.py typos. Triggered by /meeting-protocol, "протокол встречи", "разобрать транскрипт".
Use this skill when the user asks to generate a meeting protocol/report from transcripts in tmp/ and move artifacts into the knowledge base.
scripts/speech2text.py
(base URL via SPEECH2TEXT_BASE_URL, token via SPEECH2TEXT_TOKEN or ~/.config/speech2text/token).date '+%Y-%m-%d %H:%M (%A) %Z'
ls -la tmp
tmp/TEXT| TIME| SPEAKER (do not rely only on filename).SPEAKER_XX and mention the limitation.term:proposed_fix + 1 exact quote (1–2 lines) + 1 short note “о чём речь”.notes/meetings/artifacts/YYYY-MM-DD-<slug>/kontur-tolk-transcription.md (raw)voice-rec-transcription-raw.md (raw)voice-rec-transcription.md (sorted by start time; derived from raw)tmp/ so tmp/ stays temporary.notes/meetings/YYYY-MM-DD-<slug>.mdТема.
Коротко о встрече. (2–4 предложения без буллетов)
Обсуждаемые темы. (1–5 абзацев по 2–4 предложения; буллеты допустимы, но в основном текст)
Итоги (1–2 абзаца по 2–4 предложения; буллеты допустимы)
Следующие шаги/Договоренности
.venv/bin/python scripts/kb.py typos --query "TYPO" | tail -10
.venv/bin/python scripts/kb.py typos --add "TYPO" "FIX"
kb-memory policy (safety gate + “don’t duplicate KB”).Note:<path> entity exists (entityType Note): mcp__server-memory__search_nodes, then mcp__server-memory__create_entities if missing.pointer: observation via mcp__server-memory__add_observations: 1–2 lines of what was decided + the note path.Project:<...> / Person:<...> via mcp__server-memory__create_relations if participants/projects are clear.notes/meetings/YYYY-MM-DD-<slug>.md.