AI Coding Weekly

Pocock writes up the Home Repo pattern with a setup prompt

One repo holds the skills, the steering files and the wiki, and every agent runs there.

Matt Pocock has turned his answer to the multi-repo agent question into a written-up article, and it ships with a prompt you can paste at your own org's repos to set the thing up. He had described the pattern earlier in a post, and says he had been using it for months while assuming it was too boring and obvious to write down.

The pattern is one repo, the home, where all of his skills, steering files, wiki and tooling live. He runs the agent in the home repo. The other repos are reached through pointers in AGENTS.md, written as plain instructions like "For Sandcastle, clone mattpocock/sandcastle to ~/repos/sandcastle" and the same for his skills repo. That is the whole mechanism. No workspace manifest, no monorepo migration, no tool. The agent is told where the code lives and clones it when it needs it.

Why one home

Pocock lists three things the home repo buys you. All the steering files and tooling sit in one place. There is a single spot to run every agent from. And you can make sweeping changes over multiple repos, which is the part that usually falls apart when each repo carries its own half-maintained instructions file and its own copy of the same skill.

He frames the problem as the questions nobody has a clean answer to. Where do you put the skills, the steering files, the tooling. Duplicating them per repo means they drift. Keeping them nowhere means every agent run starts cold.

What this is and is not

This is one working engineer's setup, described by him, not a benchmark or a comparison against other approaches. Pocock says he talked about it in his course and that people there liked it, which is the extent of the evidence on offer. There is no claim here about how it behaves on a large org's repo graph, and the prompt he published is the invitation to find out on your own.

For working engineers, the interesting detail is how low-tech the pointer layer is. The agent reads an instruction in a markdown file telling it to clone a repo to a path, and that is sufficient coordination for an agent that already has a shell. If your current setup involves a wrapper script that stitches context across checkouts, the Home Repo is the version of that with the wrapper deleted.

Matt Pocock
@mattpocockuk
X
Let me propose a solution I've been using for months (that I thought was too boring and obvious to post)
Nov 12, 2026 · View on X

Have the good opinion, not the loud one

One email a day. Under three minutes. What working engineers found out about coding with AI.

Get the next one by email

Coding with AI, read daily so you do not have to. The experiments, the receipts and the arguments from engineers shipping real software. Not a changelog.