Show HN: Skrun — Deploy any agent skill as an API

What is Skrun?

Skrun is an open‑source CLI that turns any Agent Skill (a SKILL.md file) into a callable HTTP API — POST /run — so your agent becomes a service, not just a script. It promises typed inputs, structured outputs, stateful key‑value memory, and tool calling baked into the agent bundle. It has been reported that v0.1 ships with a local runtime and that cloud deployment is on the roadmap (the code exposes a RuntimeAdapter interface to make that easier).

How it works

Developers scaffold an agent with skrun init (or import an existing SKILL.md), edit agent.yaml to pick models and inputs/outputs, run locally with skrun dev, test with skrun test, then skrun deploy to publish. The system is multi‑model — reportedly supporting Anthropic, OpenAI, Google, Mistral and Groq with automatic fallback — and supports two styles of tool calling: local CLI scripts bundled with the agent or external MCP servers (think headless browser via Playwright). Examples shipped include code-review, pdf-processing, seo-audit and a data‑analyst agent that takes CSV and returns structured insights.

Why it matters

Want to expose an internal QA workflow or a content‑drafting agent as an API without knitting together auth, hosting, and glue code? Skrun aims to make that trivial. For teams tired of ad‑hoc wrappers and brittle integrations, the promise of a standard SKILL.md → agent.yaml → POST /run pipeline is appealing: simple, testable, and portable. It’s a neat piece of infrastructure work that feels a bit like “serverless for AI agents” — and because it’s MIT‑licensed, you can fork, tweak, and run it yourself.

Sources: github.com/skrun-dev, Hacker News