Keep your agent skills behind the firewall

What it does

SkillHub is a private registry for packaging and distributing AI agent skills inside an organization. Teams publish versioned skill packages to namespaces, search and install them through a web UI or CLI, and promote vetted skills from team scope to a global catalog. It runs entirely on-premise via Docker or Kubernetes and plugs into existing infrastructure with S3-compatible storage and OAuth identity providers.

The interesting bit

The governance model mirrors software package registries but adds enterprise guardrails: team admins review namespace-level submissions, platform admins gate global promotions, and every action leaves an audit trail. It also carries a compatibility layer for ClawHub-style clients, so existing tooling isn’t orphaned while the native CLI becomes the primary path.

Key highlights

• Runs fully self-hosted with data sovereignty; no proprietary skills leave the firewall.
• Namespace-based RBAC with Owner, Admin, and Member roles, plus review queues before global promotion.
• Semantic versioning with custom tags (beta, stable) and automatic latest tracking.
• Pluggable storage backend: local filesystem for dev, S3 or MinIO for production.
• Prometheus and Grafana monitoring stack included; basic Kubernetes manifests provided.

Caveats

• The production release template ships with a default bootstrap admin password (ChangeMe!2026), which the validation script rejects but remains a foot-gun if missed.
• Kubernetes manifests are described as “basic,” so expect to customize for real production topology.
• README warns against running standalone Maven builds on submodules due to sibling dependencies; the build is monorepo-sensitive.

Verdict

Organizations building fleets of internal AI agents and tired of emailing ZIP files around should look here. Solo developers or teams already happy with public registries probably don’t need the overhead.