Skillz: Anthropic‑Style Skills for Any MCP Client (elite-ai-assisted-coding.dev)

Skillz is a tiny, experimental MCP (Model Client Protocol) server that exposes Anthropic-style "skills"—folders containing SKILL.md—as tools usable by any MCP client, not just Claude. It scans a root directory for SKILL.md files, registers one tool per skill (tool name = skill slug, e.g., algorithmic-art), uses the skill’s front-matter description for the tool description, and when invoked returns the skill’s instructions plus absolute file paths to bundled resources. Optionally the server can execute a helper script packaged with the skill via script + script_payload. You can run it with a single command (uv run skillz.py /path/to/skills) and point your MCP client at the server over stdio, HTTP, or SSE. This matters because skills are a compact way to load task-specific context and instructions dynamically—avoiding context bloat—yet today they’re effectively limited to Anthropic’s Claude toolchain. Skillz is a proof-of-concept shim that could let other agent frameworks adopt skills, accelerating cross-platform interoperability and nudging skills toward becoming a general, standardizable mechanism for agent tooling. Important caveats: it’s experimental and potentially unsafe—scripts can run arbitrary code—so treat skills as untrusted code, sandbox execution (containers/VMs), and don’t run unvetted skills on sensitive machines.