Matt Pocock describes a home repo: one version-controlled starting point for an agent that needs to work across several independent repositories. The original link starts at 5:09; this capture covers the whole 8:53 video.
The problem
- Duplicating skills, agent instructions and tooling across many repositories makes keeping them in sync tedious.
- User-level configuration is accessible everywhere, but Pocock worries it can drift and conflates personal preferences with a team’s shared setup.
- A single agent session can already work in multiple directories; the home repo organizes the common context around that capability.
The home repo
- Start the agent in one shared repository containing skills, steering files, documentation and custom tools.
- Its
AGENTS.mdorCLAUDE.mdpoints to the other repositories by local path; in Pocock’s setup the agent can clone a missing repo and use worktrees for concurrent work. - Code and its separate Git history stay in the individual repos. Shared instructions live at the starting point; repo-specific instructions stay with their project.
- Pocock suggests grouping pointers by area as the list grows, so the agent can discover the relevant repos without loading hundreds of entries at once.
A hub for tools and knowledge
- Pocock’s home repo also houses a private Markdown wiki and tools connecting to services such as Discord, GitHub, Gmail and YouTube.
- His example combines a Discord message, a code fix and a course update in one workflow, reusing the same skills and tools across projects.
- He prefers a monorepo when practical; the home repo is an alternative when organizational or public/private boundaries make one impractical.
His companion article and setup prompt propose inventorying repos and shared material first, then getting approval before applying a migration. The video describes his workflow, not an independently measured improvement.