feat: [gh skill CLI, v2.90.0+] org-level managed skill distribution via remote source config — enterprise angle of cross-surface integration (#50148)

Resolved 💬 4 comments Opened Apr 17, 2026 by TrudosKudos Closed Jun 5, 2026

Part of the cross-surface gh skill integration proposed in #50148. This issue covers the managed settings and enterprise distribution surface specifically.

The surface

Managed settings currently support deploying skill files to users across an organization. This works but requires IT to manage file distribution and re-distribute whenever skills change — there’s no link between the deployed skill files and the repository that produced them.

What's missing today

There is no way for an enterprise admin to point managed settings at a GitHub repository and have Claude surfaces across the org automatically sync from it. Admins must instead distribute SKILL.md files directly, with no version tracking, no release pinning, and no self-service update path.

The ask for this surface

Allow a skills.sources declaration in managed settings (see #50148) pointing at an internal private GitHub repository as the authoritative skill source for all users in the organization. Key properties:

  • Works with private repos scoped by org membership and token
  • Supports version pinning — pin the org to a stable release while testing a new version
  • Skills update automatically when the repo gets a new release — no file redistribution required
  • Provides a git-native audit trail for every skill change
  • Appears as a read-only connected source in the Claude.ai customize panel (#50154) so users can see what’s managed without being able to remove it

How it connects to other surfaces

Managed skills.sources should apply consistently across Claude Code CLI (#50150), Claude.ai desktop (#50154), and Cowork (#50158) — so an org admin configures skill distribution once and it takes effect everywhere, without surface-specific workarounds. The full cross-surface picture is in #50148.

References

View original on GitHub ↗

This issue has 4 comments on GitHub. Read the full discussion on GitHub ↗