A collection of skills I use with Claude Code to improve my life in various ways. Designed to pair with an AI-managed Obsidian vault — the skills read from and write to the vault.

• summarize/ — drop in a YouTube video, article, PDF, EPUB, or podcast and it writes a summary note into your vault with wikilinks to every person and concept mentioned. Asks detailed vs minimal mode on each run — detailed creates reference notes for every wikilink, minimal leaves them dangling and saves ~85% tokens.
• summarize-call/ — drop in a call recording (video or audio) and it transcribes with speaker labels, summarizes, and writes a call note + transcript + person notes for the participants. Same detailed / minimal mode as summarize.

Two prompt-callout templates land in your vault under 01 Updates/ after install:

• 📌 Daily Brief.md — your day's calendar, email, messages, tasks, weather. Sections + sources are configurable in the prompt callout.
• 📰 Daily News.md — one-page news digest, top story per topic. Topics + languages configurable in the prompt callout.

These aren't skills — they're files with > [!prompt] callouts that contain the agent instructions. The agent reads the callout and follows it. Customize by editing the callout in Obsidian.

To run an update, paste either of these into Claude Code:

update my daily brief — read the prompt callout in "01 Updates/📌 Daily Brief.md" and follow it

update my daily news — read the prompt callout in "01 Updates/📰 Daily News.md" and follow it

For an automatic daily run, set up a recurring task:

• /loop — runs locally on your machine (machine has to be on at the scheduled time)
• /schedule — runs in the cloud (works while your machine is off on their servers)

Two Obsidian plugins that pair well with the skills — see obsidian-plugins/ for details:

• unread-dot (vendored in this repo) — blue dot next to notes with unread: true in frontmatter. The summarize skills set this flag on every new note, so this plugin gives you a visual "you have new stuff" indicator in the file explorer.
• flashcards-obsidian (vendored in this repo) — turn ==highlighted== text, Question::Answer syntax, or #card tags into Anki cards. Fork of reuseman/flashcards-obsidian. Requires Anki + AnkiConnect.

Open Claude Code in any directory and paste this:

Install the ai-life-skills pack from https://github.com/reysu/ai-life-skills. Clone the repo to ~/src/ai-life-skills, ask me where I want the new Obsidian vault to live, create the vault folder with the full folder structure the skills expect, symlink every skill in the repo into ~/.claude/skills/, copy vault/CLAUDE.md from the repo to the new vault's root, copy the daily-brief and daily-news prompt-callout templates from vault/01 Updates/ into the vault's 01 Updates/ folder, and ask me whether to also install the bundled unread-dot Obsidian plugin into the vault's .obsidian/plugins/ folder.

Claude will:

1. Clone the repo
2. Ask where to put the vault (default: ~/ai-vault)
3. Create the vault folder with the expected structure (see below)
4. Symlink summarize and summarize-call into ~/.claude/skills/
5. Copy the person-note template into your vault's _Templates/ folder
6. Copy vault/CLAUDE.md to your vault root (so AI agents know your vault's conventions)
7. Copy 📌 Daily Brief.md and 📰 Daily News.md into your vault's 01 Updates/ folder
8. Optionally install the unread-dot Obsidian plugin into your vault's .obsidian/plugins/

Restart Claude Code so it picks up the new skills. Then run /summarize or /summarize-call.

Recommended: use a new, dedicated Obsidian vault for these skills rather than your existing personal vault. The skills create and modify many notes/folders automatically, and keeping it separate avoids polluting notes you've written yourself.

You can point the skills at an existing vault if you want — tell Claude the path instead of creating a new one and it'll only create any missing folders. Just note the recommendation above about a dedicated vault.

If you just want one skill and already have a vault:

mkdir -p ~/src
git clone https://github.com/reysu/ai-life-skills ~/src/ai-life-skills
ln -s ~/src/ai-life-skills/summarize ~/.claude/skills/summarize
# or:
ln -s ~/src/ai-life-skills/summarize-call ~/.claude/skills/summarize-call

The skills share a templates/ folder at the repo root — leave it where it is, both skills reference it with a relative path.

/summarize https://youtube.com/watch?v=...

Summary length is proportional to source length — a 10-minute video gets a short summary, a 3-hour podcast gets a long one, a 600-page book gets chapter-by-chapter treatment. Quotes and a link to the transcript go in the frontmatter.

For a book or PDF, drop the file into your vault first (I use _Attachments/) then call the skill with the filename:

/summarize The Singularity Is Near.epub

Drop the recording anywhere and call the skill with the path:

/summarize-call ~/Downloads/call-with-alex.mp4

It'll ask whether to transcribe locally (free, private, slower — walks you through installing whisper and pyannote if you don't have them) or with ElevenLabs Scribe (paid, faster, one API call). Then it writes a call note, the transcript, and person notes for the participants.

Both skills ask "detailed or minimal?" on each run. To skip the prompt — useful for scheduled tasks, /loop, or just typing faster — pass the mode in the invocation:

/summarize https://youtube.com/... minimal
/summarize https://youtube.com/... detailed
/summarize-call ~/call.mp4 --minimal
/summarize-call ~/call.mp4 -d

Accepted tokens:

• Minimal: minimal, fast, quick, --minimal, -m
• Detailed: detailed, deep, full, --detailed, -d

If neither is present, the skill prompts interactively. There's no default — you either pass it or pick when asked.

What the modes do:

• Detailed: creates reference notes for every wikilink in the output, person notes for every person mentioned (researches public figures), uses the highest-quality model available
• Minimal: writes the summary/call note only, leaves wikilinks dangling, creates person notes for creators/guests (or call participants) only, uses Sonnet — saves ~85% on tokens for typical content

your vault/
├── 01 Updates/
├── 02 Daily/YYYY/MM/   # daily notes, named MM-DD-YY ddd.md
├── 03 Meetings/        # call notes + transcripts
├── 04 People/
├── 05 Projects/
├── 06 Research/
├── 07 References/
├── 08 Summaries/
├── _Templates/
├── _Attachments/       # drop ebooks/PDFs here before telling claude to summarize them
└── _Bases/             # optional, only if you use Obsidian Bases

Easy-mode install creates this structure for you. If you're using an existing vault, the skills prompt before creating any missing folders on first run. You can also rename any of them in the Configuration block at the top of each SKILL.md.

If you run Claude Code from outside your vault, set VAULT_ROOT:

export VAULT_ROOT="/path/to/vault"

Otherwise the skills walk up from your current directory looking for .obsidian/.

• summarize uses yt-dlp, defuddle, pdftotext, and pandoc
• summarize-call uses ffmpeg (always), plus either mlx_whisper + pyannote.audio (local path) or an ELEVENLABS_API_KEY (cloud path)

Each skill checks what's missing on first run and asks before installing anything.

• macOS 15 (Darwin 25) on Apple Silicon, Python 3.11+, Claude Code CLI
• Local transcription path assumes a Mac with MPS; the skill auto-detects CUDA / MPS / CPU and falls back to CPU on unsupported devices (slower but functional)
• ElevenLabs Scribe path works on any OS with Python + requests

MIT.