iflytek/skillhub

SkillHub is a self-hosted registry and governance layer for agent skills, not a collection of skills itself. Teams publish SKILL.md-based skill packages into scoped namespaces, version them, review them, and let colleagues search or install them through a CLI. It fits organizations that already produce skills and need a private place to distribute them under access control, rather than a public catalogue.

What changes in practice

Instead of passing skill folders around or cloning a public repo, an author publishes a package with a semantic version, tags such as beta or stable, and automatic latest tracking. Namespaces carry their own members and Owner/Admin/Member roles, so team admins review within their scope and platform admins gate promotion to the global scope; these governance actions are audit-logged. Visibility rules mean users only see skills they are authorized to view. Because the format is the standard SKILL.md layout, existing skill folders can be published into the registry directly. The project explicitly positions itself as infrastructure complementary to open content collections such as anthropics/skills, and reminds republishers to check each skill's own license — some document skills there are source-available rather than open source.

Installation paths

Consumers install skills with the first-party CLI (@astron-team/skillhub), which logs in with an API token and supports --agent/--dir targets. Directories can be pointed at specific agent layouts, covering OpenClaw, Hermes Agent, DeepSeek Harness, HarnessClaw and the in-house astron-agent. A ClawHub-compatible layer covers search, inspection and installation, but the README states its publish protocol is not compatible, so publishing uses the SkillHub CLI; native CLI APIs are described as the primary supported path while protocol compatibility continues to expand.

Deployment and prerequisites

Self-hosting is the documented default: release images from GHCR via a runtime script or Docker Compose, or basic Kubernetes manifests and a Helm chart under deploy/k8s/ and charts/skillhub/; the chart guide covers required secrets, TLS, external data services and upgrade constraints. Docker and Docker Compose are needed for container deployments. Source development needs Java 21, Node.js 22.13+/24.x and the declared pnpm version; CLI development needs Bun. Published server and web images target amd64, arm64 and riscv64, while the scanner image is only amd64/arm64. Production notes recommend setting the public URL, keeping PostgreSQL/Redis bound locally, using external S3/OSS storage, and changing the default bootstrap admin password — the validation script rejects the default. Operators should also note the monitoring example is a starting point, not a working scrape setup, and that upload validation uses a configurable file-extension allowlist.

Limits

SkillHub does not supply skill content itself, and the documented scope covers skill management and distribution rather than evaluating skill quality. Storage is pluggable (local filesystem or S3/MinIO) via configuration. The README's integrations are described by the project; no independent outcome data or benchmark is supplied.

HotRepo's take

Recorded trending-board samples show 5,282 stars on 2026-10-24 and 5,285 on 2026-10-25. That two-day sample alone does not indicate momentum; use the deployment and governance docs rather than the star count when judging fit.

Project source and documentation