Just ask.
Your agent
has the tool.

An agent skill is a folder containing a SKILL.md file: YAML frontmatter with a name and description, followed by instructions in Markdown, plus optional scripts/, references/ and assets/ directories. The agent loads only the name and description at startup, and reads the full instructions only when a task matches. That progressive loading is why an agent can keep many skills available at almost no context cost. The format is defined by the Agent Skills specification, an open standard stewarded by the Agentic AI Foundation and implemented by 40+ tools.

No. A skill written to the specification is a single folder that every conformant agent reads unchanged. The two rules that keep it portable are to reference bundled files by paths relative to the skill root — scripts/watch.sh, never an absolute path — and to keep vendor-specific syntax out of the instructions, since the body is read verbatim by every agent. Every skill in this toolkit is checked against both rules automatically before it can be committed.

Most read ~/.agents/skills/ at the user level and .agents/skills/ inside a project — a vendor-neutral path that several agents give precedence over their own directories. Of the 8 agents whose documentation we verified on 2026-08-11, 7 read it. Claude Code is the exception: it reads ~/.claude/skills/, .claude/skills/ and installed plugins only, which is why it installs from a plugin marketplace here instead.

A skill is procedural knowledge — instructions and scripts, loaded on demand, with no process running. An MCP server is a live tool surface: it runs, holds credentials, and exposes typed tools over a protocol. Reach for MCP when the capability needs a persistent connection or a typed contract; reach for a skill when a Markdown file and a shell script would do the job. This toolkit ships both, and keeps them in separate tiers because they travel differently.

Yes. Everything here is MIT licensed and free to use, modify and redistribute, including in commercial work. There is no account, no telemetry, and no API key required by the toolkit itself — individual skills declare their own dependencies, such as ffmpeg, in a compatibility field you can read before installing.