Deploy and Host SkillHub on Railway

SkillHub is iFLYTEK's open-source, self-hosted registry for AI agent skills: the SKILL.md packages used by Claude Code, Codex, OpenClaw and other agents. Your team publishes skills to it, and every upload is checked by a security scanner before it goes live. Skills are organised in namespaces with roles and review, and versions are kept. Agents install skills with one CLI command. It also speaks the ClawHub registry protocol, so existing ClawHub-style clients can point at it.

About Hosting SkillHub

This template runs the official v0.2.22 images as five services:

Service	What it does
SkillHub	The web app (nginx) on your Railway domain. It proxies `/api` to the server.
server	The Spring Boot API. Skill packages are stored on its volume.
scanner	The skill security scanner. Every uploaded package is checked for risky behaviour before release.
Postgres	Users, namespaces, skills and versions
Redis	Sessions and the scan queue

Admin account, no default password. On first boot the server creates the user admin with a generated password, BOOTSTRAP_ADMIN_PASSWORD on the server service. Upstream ships the publicly known ChangeMe!2026.
Sign-up closed by default. SkillHub v0.2.x lets anyone with the URL create an account, and every new account gets its own namespace to publish into. The web service blocks the sign-up endpoint unless you set SKILLHUB_ALLOW_SIGNUP=true. OAuth sign-in works once you configure it.
HTTPS-correct. Session cookies are Secure. nginx trusts Railway's forwarded protocol, and the public URL is wired in, so the CLI install commands the UI shows point at your domain.
Bounded memory. The JVM heap is capped with -Xmx512m. Without the cap, the JVM sizes its heap from the plan's RAM limit, not from what the registry needs.

Common Use Cases

A private skill registry for a company's Claude Code, Codex or Cursor users: publish internal skills once, and install them everywhere with one command
Scan third-party agent skills for risky behaviour before your team installs them
Run a public or community skill hub on your own domain, with namespaces per team and an admin review step

Dependencies for SkillHub Hosting

Postgres and Redis (included)
Optional: OAuth apps (GitHub, GitLab, OIDC, Feishu, DingTalk) for single sign-on, SMTP for password resets, and an LLM API key for the scanner's AI analyzer

Deployment Dependencies

Implementation Details

Sign in. Open the SkillHub service's domain and log in as admin, using BOOTSTRAP_ADMIN_PASSWORD from the server service's Variables tab. Change the password in your profile afterwards. The variable is only read when the account is first created.

Publish and install from the CLI. In the web UI, create an API token under your profile. Then:

npm install -g @astron-team/skillhub
export SKILLHUB_REGISTRY=https://
skillhub login --token sk_...
skillhub publish ./my-skill --namespace global
skillhub search ""
skillhub install my-skill --agent claude-code   # or codex, cursor, ...

A skill is a folder with a SKILL.md file that has name and description in its front matter. Each upload is scanned first. A finding of high severity or worse blocks the release.

Let people in. SkillHub v0.2.x has no screen for an admin to create users. Either:

Set SKILLHUB_ALLOW_SIGNUP=true on the SkillHub service and redeploy while your team signs up, then set it back to false.
Or set up single sign-on on the server service.

Either way, admins can promote or disable users under the admin pages. For GitHub single sign-on, set OAUTH2_GITHUB_CLIENT_ID and OAUTH2_GITHUB_CLIENT_SECRET, with the callback https:///login/oauth2/code/github. GitLab, OIDC, Feishu and DingTalk follow the same pattern; see the upstream .env.release.example.

Smarter scans. The scanner's default analyzers are rule-based and need no key. To add its LLM analyzer, set SKILL_SCANNER_LLM_API_KEY, SKILL_SCANNER_LLM_MODEL and optionally SKILL_SCANNER_LLM_BASE_URL on scanner, then set SKILLHUB_SCANNER_USE_LLM=true and SKILLHUB_SCANNER_LLM_PROVIDER on server.

Upgrades. Change the image tag on server and scanner. On the SkillHub service, change the SKILLHUB_WEB_IMAGE build argument in your fork's Dockerfile. Database migrations run automatically when the server starts.

Resources. Measured on Railway, idle after the first boot:

Service	RAM
server	about 750 MB
scanner	about 150 MB
Postgres	about 120 MB
SkillHub web and Redis	about 60 MB together
Total	about 1.1 GB

The Java server is over the Free plan's 0.5 GB per service and close to the Trial plan's 1 GB, so use the Hobby plan or higher.

Why Deploy SkillHub on Railway?

Railway is a singular platform to deploy your infrastructure stack. Railway will host your infrastructure so you don't have to deal with configuration, while allowing you to vertically and horizontally scale it.

By deploying SkillHub on Railway, you are one step closer to supporting a complete full-stack application with minimal burden. Host your servers, databases, AI agents, and more on Railway.