Explorer
Home Blog Contact Claude SDD

Local Life Management (LLM? Get It?)

Jan 21, 2026

Earlier this month I wrote about setting up a Meta-Repo for managing my side projects, and last week I wrote about how I was experimenting with using it to manage my life, not just my side projects. Since a handful of people reached out asking to see the code, I’ve cleaned it up a little and put it on GitHub: Local Life Manager

This is very much an experiment at the moment, and it’s changing rapidly, so treat the README as the source of truth for its current workflows and instructions, but here I can go a little more into what this is and how I am using it.

The Big Idea: Files as Context

To rehash what I mentioned in the last blog post about this, we talk a lot about “context windows” in AI. But even with 200k tokens, you can’t shove your entire life into a prompt every time you open a terminal. So I created a meta-repo, a repo that sits on top of my other repos, run Claude from there, and let it use all of that as the context it needs.

This is roughly broke up into three parts:

  1. The Brain: A strictly formatted Obsidian vault. Everything about my personal life, daily journal, tasks, notes on things I’ve studied, notes on YouTube videos I’ve watched or books I’ve read.
  2. The Hands: A collection of Claude Code Skills that know exactly how to read and write to those files.
  3. The Workspace: A separation between “Ideas” (planning/issue tracking/random ideas) and “Spaces” (actual code).

When I run /good-morning, Claude doesn’t “know” my schedule. It reads my-vault/02 Calendar/2026-01-21.md. It reads my todo.md. It checks my GitHub notifications. It builds the context dynamically, does the job, and then forgets it all when the session ends.

The persistence is the filesystem.

What’s actually in the box?

If you clone the repo, you’re getting the skeleton of my daily workflow:

  1. 49 Custom Skills: Everything from /brief (for scoping new ideas) to /critique (for having an AI roast your bad ideas before you build them) to /daily-review (for logging off without guilt).
  2. Agent Personas: Pre-prompted definitions for “Product Managers,” “Code Architects,” and “Critics” that I spin up for specific tasks.
  3. The “Meta-Repo” Structure: A folder hierarchy that forces you to separate thinking about code from writing code.

Currently it’s all very opinionated to my workflow, but maybe that will change over time.

You can also kind of think of it as two pillars.

Project Management

This is a continuation of the work I’ve been writing about for a while around developing my own spec-driven-development workflow for LLMs. You can read those posts for more information, but the basics are there are two directories: ideas/ and spaces/.

ideas/

These are the planning docs. I keep my project briefs, analysis docs, random notes around ideas I have, all of that in here. There are several Skills and custom Subagents dedicated to helping with this ideation and planning process.

spaces/

This is where all of the code lives. It might be more intuitive to name this projects/ but I am trying to stick to the proper PMI definition of a project being a scoped piece of work intended to be finished, not just an ongoing bucket. So I chose to use the spaces terminology, for now at least.

Crucially this directory is in .gitignore. Everything in it is its own repo for obvious reasons.

Life Management

This is where my Obsidian vault comes in. I have this inside of the repo, but it is also in the gitignore as I commit that separately, and use the Obsidian sync functionality. Again, this is where I keep all of my notes and tasks and everything about my personal life. I am building out some Skills and Subagents to help me out with that kind of thing, everything from planning my day to setting my personal yearly goals. My take on the “Second Brain” movement (if you don’t know just search for it on YouTube…).

The thing I am working on it now is workflows around auto fetching and summarizing content from certain YouTube channels and RSS feeds. Like all of you I’m sure, I find myself jsut drowning in all of the things I feel like I need to stay on top of. This shouldn’t be a replacement from watching/reading the really improtant things, but (hopefully) by presummarzing them I can better filter out the ones that are actually worth my time. It’s a little clunky, the state is just saved in a JSON file, which won’t scale at all, so improvements need to be made there. Still, it works, and it has been helpful.

I’m also playing around more with the idea of the guided tutor sessions. Asking Claude to teach me about something, having it ask me to explain things back in my own words (which studies show is the best way to learn something), keeping track of these learning sessions so we can actually have a structured, long term “curriculum” (for lack of a better word). Still early, but promising.

Rough Edges and Warnings

Look, I’m using this every day, but it’s an ongoing experiment.

  • It’s brittle: If you move a folder it expects to be there, it will hallucinate or crash.
  • It’s personal: The Skills are tuned to my workflow. You’ll likely need to rewrite half of them to make them feel right for you.
  • It’s a template, not a library: You’re meant to fork this and butcher it.

The README in the repository is the only source of truth. I’m updating it as I break things and fix them. If you’re curious, go there. Read the code. Read the Skills. Maybe clone it and try tweaking it to your preferences.

If you do have a chance to play with it please drop me a line and let me know what you think!