breakthrough-edu
UserBootstrap your own personal AI operating system -- a 4-tool stack (Claude Code + Lark Base + Obsidian + Lark Tasks) plus a character system that remembers you across sessions. Install: npx skills add breakthrough-edu/personal-ai-os-bootstrap
Categories
Indexed Skills (3)
personal-ai-os-bootstrap
Walks the builder through bootstrapping a personal AI operating system based on a 4-tool stack (Claude Code, Lark Base, Obsidian, Lark Tasks) and a character system (profile + companion-soul skill + two-layer memory). MUST trigger when the builder says "set up my personal AI OS", "bootstrap my AI assistant system", "build a JARVIS-like setup", "configure my Claude workflow", "I want a Claude that remembers me across sessions", "set up Lark Base + Obsidian + Claude Code", "personal AI OS", or similar integrated-bootstrap intent. Do NOT trigger for single-tool questions, casual AI use, or anything that does not involve building a multi-tool integrated system.
meeting-scribe
Turn meeting audio into three artifacts -- a corrected transcript, an HTML visual canvas, and a Markdown summary with AI insights -- transcribing locally and cross-platform with Whisper (no cloud, no external transcription app). Runs on Apple Silicon mac via whisperkit-cli, on Intel mac / Windows / Linux via faster-whisper, or via whisper.cpp. MUST trigger when the user mentions a meeting transcript or recording, the audio drop-zone folder, processing / correcting / cleaning / fixing a transcript, transcribing an audio file, Whisper output, or generating meeting notes from a recording. Also trigger on Chinese phrasings -- 处理 transcript, 整理会议纪要, 修正 transcript, 清理 transcript, 修订 transcript, 转录纠错, 会议转写, 帮我做个会议总结, 把会议 transcript 处理一下, 处理一下昨天会议, 整理今天会议, 会议纪要, 转写文本, 把录音转成文字, 处理这个录音, 处理这个音频. Also trigger when the user drops an audio file (.wav / .mp3 / .m4a / .flac / .ogg ...) or a .txt transcript into the configured drop-zone folder. Behavior is driven entirely by a per-user config at ~/.config/meeting-transcripts/
youtube-transcript-ingestion
Batch-grab transcripts from any YouTube channel and save them as readable markdown files with clean frontmatter. Guided 5-step flow -- channel, keyword + date filter, output folder, optional context-delta comparison, rate-limit strategy -- using yt-dlp for metadata and markitdown for transcripts. Optionally compares results against your existing notes to surface only what is genuinely new. Trigger phrases -- grab YouTube transcripts, ingest YouTube channel, yt-dlp transcripts, markitdown YouTube, research a creator's videos, build a corpus from a channel, filter content overload from a channel.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.