category / P1
What is an AI Skills registry?
A practical definition of an AI Skills registry and how teams use one to govern portable Skills across agent harnesses.
An AI Skills registry is the shared inventory, review surface, and package lifecycle for reusable agent capabilities that need ownership, versioning, and install state.
Definition
Direct answer
An AI Skills registry is a governed catalog for reusable Skill packages. It tracks what each Skill does, who owns it, which version is approved, how the package was verified, and where that version should be installed across agent harnesses.
Why it is more than a prompt library
A prompt library stores reusable text. An AI Skills registry treats each capability as a package with files, metadata, version history, package checksums, approval state, compatibility notes, and install or update status.
That difference matters once teams use Skills across Codex, Claude Code, Claude Desktop, Cursor, ChatGPT, Gemini CLI, and other harnesses. The registry becomes the source of truth for what is safe to run and what needs review.
The registry flow
A governed registry turns a Skill from a local folder into a reviewable, installable package.
- 1
Skill package
A portable Skill starts with files such as `SKILL.md` plus package metadata in `fygment.yaml`, including name, owner, risk, compatibility, and supporting assets.
- 2
Registry
The registry stores the Skill record, package evidence, source metadata, and immutable version history so teammates can find and inspect it.
- 3
Approval and version
Owners or admins approve a specific package version after validation, checksum capture, and review. Approval applies to that version, not every future edit.
- 4
Install and update across harnesses
Install, check, and update workflows compare local harness state with the approved registry version, so Codex, Claude-family targets, and later harnesses can avoid silent drift.
Fygment-specific examples
In Fygment, a portable Skill package can include `SKILL.md`, `fygment.yaml`, scripts, templates, and assets. Package checksums let the product verify that the installed artifact matches the approved version instead of trusting a folder name.
A workspace can use approvals to decide whether a draft Skill becomes installable. Install, check, and update state then show whether the local copy is current, stale, blocked, or waiting for review.
For controlled pilots, Fygment focuses on the governed Skills registry and package workflow first. Broader marketplace discovery and partner-provider import semantics are roadmap work, not live public marketplace claims.
Prompt library vs MCP directory vs AI Skills registry
These tools can complement each other, but they solve different governance problems.
| Model | Best for | What it usually misses |
|---|---|---|
| Prompt library | Saving reusable instructions, examples, and snippets for people to copy. | Package identity, checksums, approval state, install drift, and harness compatibility. |
| MCP directory | Listing tool servers or connectors that agents can call. | Skill package lifecycle, approved versions, local Skill files, and update state. |
| AI Skills registry | Managing reusable agent capabilities as owned, versioned, reviewed, installable packages. | It still needs clear boundaries: live governance now, alpha setup helpers, and roadmap marketplace automation. |
Fygment availability boundaries
Fygment keeps live product claims separate from alpha and roadmap work.
Live now
Governed Skills registry for controlled pilots
Pilot workspaces can use the registry, package workflow, approvals, package evidence, and install or check/update concepts as the first product surface.
Alpha
Setup and harness workflows
CLI and MCP setup paths are being staged as guided workflows for supported harnesses. They should be described as early setup helpers unless a route says otherwise.
Roadmap
Curated marketplace and broader automation
Public marketplace discovery, partner catalogs, commercial listing flows, enterprise rollout policy, and broad control-plane automation remain planned work.
FAQ
What is an AI Skill?
An AI Skill is a reusable agent capability packaged with instructions, metadata, and optional supporting files such as scripts, templates, or assets.
Why does a Skill need a registry?
A registry makes ownership, approved versions, package checksums, review state, and install drift visible to the team.
Is an AI Skills registry the same as an MCP directory?
No. An MCP directory lists callable tools or servers, while a Skills registry manages Skill packages, approvals, versions, and install state.
Does Fygment run every Skill automatically?
No. Fygment starts with governed packaging, approval, and install/update workflows. Runtime and broader automation stay behind explicit product boundaries.
Is Fygment a public marketplace today?
No. Fygment is a controlled pilot product. Curated marketplace and partner-provider workflows are roadmap work with stricter provenance, licensing, and moderation needs.
Related resources
- Managing portable Skills across AI harnesses
Portable Skills become manageable when a governed registry, the fygment CLI, local inventory, lockfiles, and check/update flows keep supported harness installs reviewable.
- The governance model for AI agent toolchains
Agent-toolchain governance is the operating model that keeps reusable Skills, prompts, MCP setup, and local agent configuration owned, versioned, approved, auditable, and safe to install.