rnsk-toolkits

Monorepo for @rnsk/toolkits — an open-source registry of AI agent connectors consumed by Runstack.

What this repo is

A place to build, test, and publish toolkit definitions — manifests, tool implementations, OAuth specs, and icons. Runstack imports the published npm package and registers tools into its own agent runtime.

Not @rnsk/tools — that is the MCP client SDK.

Repository layout

packages/toolkits/   Published npm package (@rnsk/toolkits)
sandbox/             Local UI + server to test toolkits through @rnsk/bot (not published)
.github/             CI, issue/PR templates, CODEOWNERS
CONTRIBUTING.md      Contributor guide

Package documentation: packages/toolkits/README.md

Quick start

git clone https://github.com/deepraj21/rnsk-toolkits.git
cd rnsk-toolkits
npm install
npm run sandbox        # → http://localhost:5173

Sandbox

The sandbox lets toolkit authors test local, unbuilt toolkit changes in two ways:

• Chat: the real @rnsk/bot widget, running the same searchTool → executeTool flow as Runstack, with a free Gemini or OpenRouter model.
• Tool runner: runs one tool directly, with no LLM.

You can scope both to a single toolkit from the UI. npm run sandbox:npm runs the published package instead, for comparison.

See sandbox/README.md for setup (all optional), credentials, and the HTTP API.

Typical contributor flow

1. Add a toolkit under packages/toolkits/src/toolkits/<id>/
2. Register the manifest in packages/toolkits/src/index.ts
3. npm run sandbox, then pick your toolkit with only in the sidebar and test it in the Tool runner and Chat
4. npm run validate && npm run build in packages/toolkits
5. Open a PR

Versioning

Pre-1.0 (0.0.x): patch releases add toolkits and tools. Pin an exact version in production (e.g. "0.0.4").

Contributing

See CONTRIBUTING.md.

Security

SECURITY.md

License

MIT