compabob
A customizable Claude Code setup for knowledge workers: agents, safety hooks, skills, memory, and an Obsidian knowledge base. Clone, run setup, make it yours.
A starter configuration kit for Claude Code that gives knowledge workers persistent memory, an organized note vault, and custom slash commands across sessions, no coding required, just run the setup script.
Compabob is a starter configuration package for Claude Code, the AI coding assistant made by Anthropic. It is aimed at knowledge workers, meaning people who spend their days writing, planning, researching, or coordinating rather than primarily writing software. The goal is to turn a fresh Claude Code installation into a personalized work assistant that remembers who you are, stores notes in an organized knowledge base, and follows consistent safety rules across every session.
Out of the box, Claude Code starts each session with no memory of previous conversations. Compabob wraps it with a structure that persists across sessions: a constitution file that defines how the assistant should behave, a memory folder where it saves things you tell it, a personal knowledge base organized as an Obsidian vault, and a set of named slash commands for recurring tasks like preparing for a meeting or generating a morning summary. A setup script asks a few questions about who you are and what you do, then scaffolds all of this in about a minute.
The package also includes several specialized sub-agents for different kinds of work, such as an analyst agent, a communications agent, and a strategy agent. These are separate AI personas the assistant can hand work off to depending on what the task requires. Safety hooks run alongside normal operation and act as guardrails to prevent the assistant from sending anything externally without your approval.
Installation requires the Claude Code CLI, Python 3.10 or later, git, and a paid Claude plan. The README includes a step-by-step walkthrough for people who have never used a terminal, covering installation on macOS, Linux, and Windows via WSL. An update script pulls new versions of the kit without touching your personal vault, memory, or configuration files, so customizations survive updates.
The project is maintained by its author as a single-user template rather than a product. There is no signup, no telemetry, and no cost beyond the Claude subscription you already have. Everything runs locally on your own machine. The README describes the intended first-hour workflow as getting one full loop working: take notes from a real meeting, ask the assistant to retrieve and summarize them, then use that context to draft something.
Where it fits
- Take notes from a meeting and ask the assistant to retrieve and summarize them later in the same or a future session.
- Use built-in slash commands to generate a morning summary or prepare talking points for an upcoming meeting.
- Store a personal knowledge base in Obsidian that the AI can read and update automatically across sessions.
- Hand off specialized work to sub-agents for analysis, communications drafting, or strategy planning.