Matt Pocock runs every agent from one "home repo"

Matt Pocock says the question he gets most is how to work with agents across multiple repos at once. His answer is the Home Repo, a single repository where all of his skills, steering files and wiki live, and where he launches every agent.

Matt Pocock

@mattpocockuk

X

I run my agent in the home repo, and then inside AGENTS.md I have pointers to the other repos

The other repos are reached by instruction rather than by directory structure. Inside the home repo's AGENTS.md he keeps plain-language pointers like "For skills, clone mattpocock/skills to ~/repos/skills" and "For Sandcastle, clone mattpocock/sandcastle to ~/repos/sandcastle". The agent reads those lines and fetches what it needs.

Pocock lists three things the setup buys him. All steering files and tooling sit in one place. There is a single spot to run all of his agents. And he can make sweeping changes across multiple repos. He describes it as "extremely simple and incredibly effective", and says he talked about it in his course the day before and that people are responding well to it.

Why this shape

The interesting part is where the shared root lives. In a monorepo, the filesystem is the thing that makes many projects feel like one project, and every tool you point at the root sees everything. In Pocock's setup, the agent's context does that job instead. The repos stay separate on disk and in version control, but the agent starts from a directory that already contains the instructions, the skills and the notes, plus a map of where everything else can be cloned.

That also means the pointers are prose, not configuration. There is no manifest format or workspace file here, just lines in AGENTS.md that an agent is expected to read and act on. It works because the agent can run git, so "clone this repo to this path" is an executable instruction rather than a declaration someone else has to implement.

What this is and is not

This is one engineer's working setup, not a measured result. Pocock gives no numbers, no before and after, and no description of what happens when two repos need coordinated edits in the same session. The claim on the table is that it is simple and that it works for him, and that his course audience liked it.

Still, it is cheap to try and easy to undo. If you already keep a scratch directory with your prompts and house rules in it, the only change is turning that directory into a real repo, writing down where your other projects live, and starting the agent there instead of inside whichever project you happened to open last.