Related work¶
Skills Gateway is not the only product occupying the "central place for AI agent skills" space. This page positions it against the closest-named alternative, so the difference in trust model is explicit rather than implied by the name.
LiteLLM Skills Gateway¶
LiteLLM's Skills Gateway is a
centralized registry for Claude Code skills: teams register a GitHub URL
via API or UI, the system auto-detects the skill name (including nested
subdirectories), an admin flips an enable endpoint, and the skill appears on
an unauthenticated hub page and in a generated marketplace manifest that
Claude Code clients add with /plugin marketplace add. Skills carry semver,
descriptions, keywords, and a domain/namespace hierarchy.
It solves the discovery and distribution half of the problem. It is a registry of pointers: the content authority remains the upstream repository, and "publishing" gates the visibility of a listing, not the bytes a client receives.
The difference in one table¶
| Concern | LiteLLM Skills Gateway | Skills Gateway |
|---|---|---|
| What is registered | A pointer to an upstream GitHub URL | The upstream URL, ingested into a quarantine repository |
| What clients receive | Content fetched from upstream at install time | Bytes from a SHA-pinned, human-approved snapshot served by the read-only facade |
| Publish gate | Listing visibility (enable/disable) | Human approval of a specific snapshot; held updates never displace the approved one |
| Rug-pull protection | None — upstream refs stay mutable and authoritative | The core of the design: snapshots and held updates |
| Revocation | Disable the listing; already-installed content unaffected | Revocation unpublishes the served ref itself |
| Audit | Not described | Append-only ledger of every fetch and administrative action |
| Facade access | Unauthenticated public hub | PAT-only facade — a deliberate trust boundary |
| Single marketplace URL | Generated manifest of external pointers | The virtual catalog, strictly derived from approved-and-served snapshots, self-contained relative sources only |
| Skill metadata for discovery | Semver, keywords, domains/namespaces per skill | Marketplace-granular catalog; skill-level browse is not a current capability |
What this means¶
The two products are complementary ends of the same space. A registry answers "where do I find skills?"; Skills Gateway answers "how do I know the skills my developers install are the ones we reviewed — and stay that way?". A registry of pointers inherits the threat model unmitigated: a compromised or mutated upstream flows straight to every consumer, because nothing sits between the listing and the clone.
The one capability the comparison surfaces as a genuine gap on this side is skill-level discovery metadata (categories, keywords, per-skill search in the portal). That is a product feature, not an architectural difference, and it composes cleanly on top of the existing model if the estate grows large enough to need it.