The Chvor Registry: A Community Marketplace for AI Skills and Tools

Chvor started as a self-hosted AI assistant you could run on your own hardware, talk to on your own terms, and trust with your own data. But a standalone assistant, no matter how capable, only gets you so far. The real power comes when a platform develops an ecosystem — when the community starts building, sharing, and composing behaviors that none of us would have created alone.

That is the idea behind the Chvor Registry: a community marketplace where developers can discover, install, and publish AI skills, tools, and templates for the Chvor platform. Today we are opening it up and inviting you to be part of it.

What the Registry Contains

The registry organizes contributions into three kinds of entries:

Skills are YAML behavior definitions that shape how Chvor responds in specific contexts. A skill might teach Chvor to perform systematic code reviews, draft release notes from commit history, or triage incoming support tickets. Skills are declarative — they describe what Chvor should do, not how the underlying model works — which makes them easy to read, modify, and share.

Tools are MCP (Model Context Protocol) server integrations that give Chvor access to external systems. A tool might connect Chvor to a web search API, a database, a Slack workspace, or a CI pipeline. Tools extend what Chvor can actually do in the world beyond generating text.

Templates are curated bundles of skills and tools that work well together. Think of them as starter kits: a “full-stack developer” template might combine a code-reviewer skill, a git-diff tool, a documentation-generator skill, and a deployment-status tool into a single installable package. Templates give new users a fast on-ramp and give experienced users a way to share opinionated setups.

Where to Find It

You can browse the full registry at https://registry.chvor.ai. Every entry includes a description, version history, dependency list, download count, and category tags so you can find what you need quickly.

If you want to contribute your own skills or tools, head to the Creator Portal at https://registry-creators.chvor.ai. We will walk through the publishing flow in detail below.

Installing Skills and Tools

There are two ways to install from the registry, and both are designed to feel effortless.

From the CLI, use the chvor install command followed by the entry kind and its identifier:

chvor install skill code-reviewer
chvor install tool web-search
chvor install template fullstack-dev

From chat, just ask Chvor directly. Saying something like “install the code-reviewer skill” will trigger the same resolution and installation process behind the scenes.

Either way, dependencies auto-resolve. If a skill declares that it depends on a specific tool, that tool gets installed automatically. You do not need to trace the dependency graph yourself.

The Index Format

Every entry in the registry is described by a consistent set of fields:

• id — a unique slug like code-reviewer or web-search
• kind — one of skill, tool, or template
• name — a human-readable display name
• description — a short summary of what the entry does
• version — a semver string (e.g., 1.0.0, 2.3.1)
• author — the GitHub username or team that published it
• category — one of ai, communication, data, developer, file, productivity, or web
• tags — freeform labels for search and filtering
• downloads — total install count
• sha256 — a hash of the entry contents for integrity verification
• requires — any credentials or environment prerequisites
• dependencies — other registry entries this one depends on

This format is intentionally simple. It is meant to be easy to parse, easy to validate, and easy for tooling to consume.

Lockfile Tracking and Auto-Updates

When you install a registry entry, Chvor records its version and SHA-256 hash in a lockfile at ~/.chvor/data/registry-lock.json. This file serves as the source of truth for what is installed and at which exact version.

Chvor checks for updates automatically every six hours. If a newer version of an installed skill or tool is available in the registry, Chvor will pull it down and update the lockfile.

There is one important exception: if you have modified a skill locally, the SHA-256 hash of your local copy will differ from the hash recorded at install time. When Chvor detects this mismatch, it skips the auto-update for that entry. Your local changes are respected. You will never lose a customization because of a background update.

This design gives you the best of both worlds — automatic freshness for entries you have not touched, and full control over entries you have tailored to your workflow.

Publishing to the Registry

Contributing to the registry is straightforward. Here is the full flow:

1. Sign in with GitHub. Head to the Creator Portal at https://registry-creators.chvor.ai and authenticate with your GitHub account. This links your submissions to your identity and lets us show proper attribution.

2. Submit your skill or tool for review. Upload your YAML file and fill in the required metadata. You can preview how it will appear in the registry before submitting.

3. Automated validation runs first. Your submission passes through a series of automated checks:

• Valid YAML frontmatter and structure
• Semver-compliant version string
• No embedded secrets or API keys
• File size under 50KB
• Slug validation (lowercase, alphanumeric, hyphens only)

For tools specifically, additional checks apply:

• MCP command must be on the allowlist of supported runtimes
• No hardcoded secrets in the server configuration
• Transport type must be stdio or http

4. Admin review. After automated validation passes, a human reviewer evaluates the submission for quality, safety, and completeness. We check that the description is accurate, the behavior is well-defined, and the entry does what it claims to do.

5. Published to the live registry. Once approved, your entry appears in the registry and becomes installable by anyone running Chvor.

The turnaround for review is typically within 48 hours. We are working on scaling this as the community grows.

Anatomy of a Skill

To make this concrete, here is what a skill YAML file looks like:

name: code-reviewer
description: Systematic code review with actionable feedback
version: 1.0.0
author: chvor-team
type: workflow
category: developer
tags: [code-review, quality]
requires:
  credentials: [github-token]
dependencies: [json-formatter]

The name and description fields are what users see when browsing the registry. The type field indicates the behavioral pattern — in this case, a multi-step workflow. The category and tags help with discovery. The requires block declares that this skill needs a GitHub token to function. And the dependencies array tells Chvor to auto-install the json-formatter tool if it is not already present.

This declarative format means you do not need to write code to create a skill. You describe the behavior, declare the dependencies, and let the platform handle the rest.

Building the Ecosystem Together

The registry is not just a distribution mechanism — it is a bet on how AI development should work. We believe the most useful AI behaviors will not come from a single team behind closed doors. They will come from practitioners who encounter real problems, build solutions, and share them with others facing the same challenges.

A DevOps engineer who builds a skill for incident triage can share it with every on-call team running Chvor. A data scientist who wires up a tool for querying Snowflake can save the next person hours of configuration. A team lead who assembles a template for onboarding new developers can make that institutional knowledge portable.

Over time, we see the registry becoming a shared library of best practices for working with AI — not abstract theory, but concrete, installable, composable building blocks. Skills that depend on tools, templates that bundle skills, and a community that keeps raising the bar on what self-hosted AI can do out of the box.

We are just getting started. Browse the registry at https://registry.chvor.ai, try installing a few entries, and when you build something worth sharing, submit it to the Creator Portal. The platform is only as good as what the community puts into it.