Skillz: Use Claude skills with any agent (Codex, Gemini, Copilot, etc...) (pypi.org)

Skillz is an experimental Python package and CLI (published on PyPI as v0.1.4) that exposes Anthropic‑style SKILL.md directories as MCP tools so any FastMCP client—agents like Codex, Gemini, Copilot integrations, or custom MCP clients—can discover and invoke them. The server recursively registers one MCP tool per skill slug, returns the authored Markdown instructions plus rich metadata (name, description, license, allowed tools), and enumerates resources with deterministic resource://skillz/{slug}/ URIs and runnable flags so clients can stream files via ctx.read_resource or fetch them directly. It supports stdio, HTTP, and SSE transports, defaults to a ~/.skillz skills root, and is launched via the uvx skillz CLI (requires Python 3.12+ and uv). Technically, Skillz can optionally execute helper scripts inside a temporary workspace: script invocations accept args, env, files, stdin, and workdir, enforce timeouts, and return exit code, stdout/stderr, and duration. Script execution is deliberately minimal (temporary copy + pared-down env), so the project warns it’s experimental and potentially unsafe—use only with trusted content or hardened sandboxes. For the AI/ML community, Skillz offers a lightweight interoperability layer to share reusable, documented agent tools across different agent runtimes, speeding prototype integrations and tool discovery—but it also highlights the need for stronger sandboxing and security controls before production use.