skreg

A package registry for AI coding assistant skills, built on cryptographic publisher identity and trust.

What is skreg?

skreg is a registry for skills — reusable instruction sets that extend what your AI coding assistant can do. Think of it like npm or crates.io, but for prompts you share across projects and teams.

Browse and install community skills from skreg.ai. Package your own with a single command and publish them for others to use — or run a private registry your team controls.

Trust & Verification

Every package published to skreg carries a cryptographic publisher signature. Two verification tiers are supported:

| Tier | Badge | How | |------|-------|-----| | Self-signed | ◈ self-signed | Publisher generated their own key. Key consistency is enforced at the namespace level — once a key is used it cannot be swapped without an explicit rotation. | | CA-verified | ✦ verified | Publisher obtained a CA-issued cert from the skreg Publisher CA. Proves organisation identity. |

All packages — regardless of tier — pass content, safety, and structure vetting before appearing in search. The registry does not counter-sign packages; only the publisher's key appears in the signature.

Key material is stored in ~/.skreg/keys/ with chmod 700. Run skreg certify to obtain a CA-verified cert. Run skreg rotate to safely rotate your namespace's signing key.

Install

Pre-built binary (macOS and Linux)

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/dymocaptin/skreg/releases/download/v0.1.2/skreg-cli-installer.sh | sh

Homebrew (macOS and Linux)

brew tap dymocaptin/tap
brew install skreg-cli

npm

npm install -g skreg-cli

cargo (build from source)

Requires Rust (rustup.rs).

git clone https://github.com/dymocaptin/skreg.git
cargo install --path skreg/crates/skreg-cli

After installing with any method, verify with:

skreg --version

Using skills

Finding a skill

Browse and search for skills at skreg.ai.

Coming soon: skreg search <query> — find skills directly from the CLI.

Installing a skill

skreg install <namespace>/<name>

For example:

skreg install dymocaptin/color-analysis

Skills are installed to ~/.skreg/packages/ and are available to your AI coding assistant automatically.

Using an installed skill

Coming soon: Native Claude Code integration — skills installed via skreg will be available directly through the Claude Code plugin marketplace.

Building skills

Anatomy of a skill

A skill is a directory with two files:

my-skill/
├── SKILL.md        # The skill prompt and instructions
└── manifest.json   # Metadata

SKILL.md starts with a YAML frontmatter block followed by the instructions your AI assistant will follow when the skill is active:

---
name: color-analysis
description: Analyzes the dominant colors in any image file, producing a ranked
             color palette with hex codes, RGB/HSL values, human-readable color
             names, and percentage breakdowns.
---

# Color Analysis

You are an expert color analyst. When the user provides an image, analyze it
and produce a detailed color palette report.
...

manifest.json declares the package identity:

{
  "namespace": "dymocaptin",
  "name": "color-analysis",
  "version": "1.0.0",
  "description": "Analyzes the dominant colors in any image file, producing a ranked color palette with hex codes, RGB/HSL values, human-readable color names, and percentage breakdowns."
}

Packaging

From inside your skill directory, run:

skreg pack

This produces a <name>-<version>.skill archive ready to publish.

Publishing

First, log in to your registry:

skreg login <your-namespace>

Then publish:

skreg publish

skreg will pack, upload, and vet your skill. Once it passes review it appears on skreg.ai.

Self-hosting

You can run your own skreg registry. Deploy the infrastructure with Pulumi, then point the CLI at your domain:

skreg login <your-namespace> --registry https://registry.example.com

See infra/README.md for the full deployment guide.

Contributing

skreg is open source. See CONTRIBUTING.md for how to build from source, run tests, and submit changes.

License

Apache 2.0 — see LICENSE.