This skill lets an agent read, write and maintain an Obsidian vault of Markdown files. Notes are written for a future agent to read, with a short preamble, rich frontmatter, sources and recency markers. New sources rewrite existing pages, contradictions are reconciled and every write is propagated to the daily note, boards and indexes.
It also bootstraps new vaults from presets, keeps an operation log and ships a research toolkit that pulls X posts, web research, YouTube and podcast transcripts into the vault.
What You Get
- A bootstrap script and presets for executive, builder, creator and researcher vaults, plus an assistant mode for running a vault for someone else
- 49 slash commands across operations, thinking tools, context and research
- Scheduled agents for a morning brief, nightly consolidation, a weekly review and vault health checks
- An optional MCP server that exposes the vault to other clients such as Claude Desktop and Cursor
Common Use Cases
- Save the decisions, people and tasks from a meeting into the right notes in one command
- Ingest a PDF, audio file, image or YouTube link so existing pages are updated
- Run a vault-first deep research pass that fills only the gaps in what you already know
- Challenge a decision against your past failures and reversed choices
What Makes It Unique
Where a typical notes skill appends, this one rewrites existing pages, resolves contradictions and writes synthesis pages for patterns across sources. It adds a shared logbook for several coding agents, scheduled maintenance agents and an AI-first note format designed for retrieval by later sessions.
Under The Hood
The agent reads and writes Markdown files across your vault and updates index and log files. It runs Python scripts from the skill folder through the shell, for example to bootstrap a vault, migrate logs or refresh stats, and a setup script can add a session-start hook to your Claude settings file. Research commands call outside services such as Grok, Perplexity, Gemini, the YouTube API and podcast feeds with keys you put in an env file, and ingest can use local Whisper, ffmpeg and LibreOffice. The install script and some of the docs download a script from the internet and run it directly, the installer also uses git, and several adapters touch credentials, so read them before running. Scheduled agents run on their own without you present, and the skill itself tells the agent not to pipe remote install scripts into a shell for you. Read the scripts and hooks before enabling them.