How to Give Claude Code a Persistent Brain with Obsidian and 8 Custom Commands

The Problem: Every Session Starts from Scratch

Every new Claude Code session is a blank slate. You waste precious tokens and mental energy re-explaining your project's architecture, decisions, and file structure. This context loss kills momentum and makes multi-session projects inefficient.

The Solution: An Obsidian Vault as a Project Brain

A developer solved this by creating a structured Obsidian vault that acts as a persistent, external memory for Claude Code. The vault is organized like a company with clear departments:

• 01_RnD/ for architecture and technical decisions.
• 02_Product/ for product specs and user stories.
• 03_Marketing/ for content and launch plans.
• 04_Community/, 05_Legal/, etc.

Each department folder contains an index.md file summarizing its state. The core is an execution_plan.md that outlines tasks with dependencies, forming a directed acyclic graph (DAG).

The Workflow: 8 Custom Commands That Change Everything

The power comes from eight custom Claude Code commands that interact with this vault.

Start a Session: /resume
This command reads the execution_plan.md and the latest handoff_note.md. It outputs a precise summary of where you left off and which tasks are unblocked and ready to work on next. No more manual catch-up.

During Work: Automatic Context Injection
When you ask Claude to work on a feature, your prompt can instruct it to first read the relevant files from the vault. For example: "Please review 01_RnD/api_architecture.md and 02_Product/feature_x_spec.md before implementing the new endpoint." Claude has the full project history at its fingertips.

End a Session: /wrap-up
This is the magic. The command:

1. Updates the execution_plan.md with completed tasks.
2. Updates any department index.md files that were affected.
3. Creates a new handoff_note.md summarizing the session's work, decisions, and next immediate steps.
   This note becomes the starting point for the next session's /resume command, creating a perfect cycle.

The Superpower: Parallel Agent Execution

Because the execution plan is a dependency graph, you can identify tasks that are unblocked and independent. The developer used Git worktrees to create separate working directories, then spawned multiple Claude Code sessions (agents) simultaneously.

• Agent 1: Works on the backend API in worktree/backend/.
• Agent 2: Works on the React frontend in worktree/frontend/.
  Both agents pull context from the same central Obsidian vault, ensuring consistency while parallelizing development. This is how the developer built an entire open-source tool (clsh) with a monorepo, packages, demos, and marketing content in one weekend.

How to Implement This System

1. Create Your Vault: Set up an Obsidian vault in your project root (e.g., ./project_brain). Use the department folder structure.
2. Seed the Content: Use an initial Claude Code session to draft your execution_plan.md and key architecture files. Treat this as a foundational investment.
3. Build the Commands: The core commands (/resume, /wrap-up) are shell scripts or Node.js scripts that read and write Markdown files. They can be integrated as custom commands in your Claude Code terminal.
4. Integrate into Your Flow: Start every session with /resume. End every session with /wrap-up. Train yourself to reference vault files in your prompts.

The template and commands are project-agnostic. You can adapt this "company brain" structure for any software project, research paper, or complex planning task with Claude Code.