repo-user

name	repo-user
description	Operate the `repo` personal-git-hosting orchestrator — init / stage / push / pull / sync repository identities across account peers. Trigger when the user wants to create a personal git repo, set up peer sync between machines, work with one of the four repo types (bare / stage / child / clone), or resolve cross-peer conflicts.

repo-user skill

1. Design principles

• Educational. Reading bin/repo end-to-end teaches the four-typed repo identity model.
• Functional. Each verb wraps a small set of git(1) calls.
• Decentralized. Each account holds its own bare; sync is symmetric pull/push between peers.
• Simple. repo calls only account and config at runtime.

2. The model

A repository identity is a slug <name> with four forms across the local + peer topology:

Form	Where	Purpose
bare	$XDG_DATA_HOME/repo/<name>.git	canonical receive-pack target
stage	~/src/<name> (working tree)	local user pushes here
child	~/src/<name> pointing at peer bare	seeds peer repos
clone	~/src/<name> (read-only clone)	vanilla clone

Sync is git pull/push against peer bares; SSH endpoints come from account remote-url <peer> <purpose> (FEAT-044).

3. Workflow recipes

1. Init a new repo identity.

   repo init notes
   repo bare notes
   repo stage notes
   

2. Inspect.

   repo source-home   # ~/src
   repo data-home     # ~/.local/share/repo
   repo identity      # the slug
   

3. Sync to a peer.

   repo sync notes [e-mail]
   repo push notes [e-mail]
   repo pull notes [e-mail]
   

4. Status across all repos.

   repo status
   

5. Conflict resolution.

   repo conflicts
   repo resolve notes
   

6. Rename across all forms.

   repo rename old new
   

7. Seed a peer's bare from a local stage.

   repo child notes [e-mail]
   cd ~/src/notes && git push [e-mail] master
   

4. Guardrails

1. Never force-push to a bare without explicit confirmation. A bare is the canonical receive-pack target for one or more stages; rewriting its history breaks every clone / child / peer downstream. If git push --force feels necessary, confirm with the human owner first.
2. Don't manually git remote add after repo stage. repo manages remotes via the account remote-url resolver; manual edits drift out of sync.
3. repo clean <name> runs git clean -f — removes untracked files. Untracked directories and gitignored files survive; run git clean -fdx in the worktree if you need the deeper sweep.
4. Conflict resolution defaults to --theirs. Override by editing the conflicted files manually before repo resolve.
5. Sync depends on account online <peer> — if the peer is unreachable, the verb warn-and-skips silently. Verify with account status first.
6. repo data-home is the security boundary. Any account with SSH access there can push to your bares. Treat it like ~/.ssh.

5. Where to read more

• man repo
• man account's remote-url (FEAT-044)
• This package's CLAUDE.md