Inference Gateway Skills

Curated catalog of Agent Skills for the Inference Gateway ecosystem.

This repository serves two purposes:

1. Catalog - catalog.json is the generated index served at https://registry.inference-gateway.com/skills/. It is not hand-edited - scripts/build-catalog.mjs rebuilds it from a single source-of-truth input: skills.yaml, which lists every skill (local or external) as one entry. Local entries are read from skills/<name>/SKILL.md in this repo; external entries are fetched from upstream at the pinned ref.
2. Skill bodies - folders under skills/ contain skill content that the Inference Gateway maintainers have authored, vendored, or adapted. Each folder retains its own LICENSE.

Installing a skill

Besides infer skills install <name>, every skill in the catalog can be installed with the skills CLI - the open standard for the agent skills ecosystem (Claude Code, Codex, Cursor, OpenCode and more):

# every skill authored in this repo
npx skills add inference-gateway/skills

# a single skill, by its catalog name
npx skills add inference-gateway/skills --skill go

# or by the catalog entry's `source` URL (works for external skills too)
npx skills add https://github.com/inference-gateway/skills/tree/main/skills/go

# list what's available without installing
npx skills add inference-gateway/skills --list

Add -g to install globally instead of into the current project, and -a claude-code to target a specific agent.

Adding a skill

Open a pull request that:

• Adds one entry to skills.yaml. See the comment block at the top of that file for the entry schema.
  • Skill body in this repo: set url: https://github.com/inference-gateway/skills and path: skills/<name>/SKILL.md. Also add skills/<name>/SKILL.md with valid Agent Skills frontmatter (name matching the folder, description 1-1024 chars; license: recommended).
  • Skill body in another repo: set url to the upstream repo and pin ref: to a release tag. The build job fetches the upstream SKILL.md, validates the frontmatter, and merges the entry into catalog.json.
• For skills derived from a third party and vendored into this repo, preserves the upstream LICENSE inside the skill folder and adds a NOTICE file at the repo root recording the attribution.

You do not edit catalog.json by hand - the build script regenerates it. Run bun install && bun run build and commit the regenerated catalog.json in the same pull request; CI fails if it is stale. This keeps the catalog consistent with every release tag, which is what @latest on the CDN serves.

The catalog is versioned as a whole via the repo's git tag (see Releases), so per-entry refs aren't needed - consumers pin to a catalog version.

The catalog is consumed by:

• registry.inference-gateway.com/skills/
  • human-browsable listing.
• registry.inference-gateway.com/skills/index.json
  • machine-readable index used by infer skills search / infer skills install <name> in the inference-gateway CLI.

Security scanning

Every catalog skill is security-scanned with NVIDIA SkillSpector - local skills and external ones at their pinned ref. Run bun run scan locally (warn-only) or see docs/security-scanning.md for the threshold policy and the CI workflow.

Licensing

• Repository-level files (README.md, catalog.json, LICENSE) and IG-authored skills under skills/ are Apache-2.0.
• Skills derived from a third-party project carry their own license inside the skill folder; that license takes precedence for that skill's contents.