Hey HN,

We built Bluebag, a runtime that lets you import Agent Skills from skills.sh and use them in your production AI agents.

Demo: https://www.bluebag.ai/playground Blog: https://www.bluebag.ai/blog/import-skills-sh-into-bluebag Docs: https://bluebag.ai/docs

The problem: skills.sh has 100+ open-source Agent Skills (PDF processing, code review, data extraction, etc.). They work great in Claude and Cursor. But if you want to use them in your own agent with Vercel AI SDK or LangChain, you need to build sandboxing, dependency management, file storage, etc.

What we built: Import a skill, get tools back. We handle the infrastructure.

    import { Bluebag } from "@bluebag/ai-sdk";
    
    const bluebag = new Bluebag({ apiKey: process.env.BLUEBAG_API_KEY });
    const config = await bluebag.enhance({ model, messages });
    const result = streamText(config);

To import a skill, just swap the URL:

    skills.sh/owner/skill → bluebag.ai/owner/skill

That's it. The skill runs in an isolated sandbox with deps pre-installed.

Stack: Isolated VMs per tenant, skills loaded at /skills/{name}, tools for bash/code execution/file access injected automatically.

Works with: Vercel AI SDK, LangChain, any model (Claude, GPT-4, Gemini, Llama).

Would love feedback, especially on what skills you'd want to run in production.