Skills
A folder with a SKILL.md, in the same format Claude Code, Codex and skills.sh use.
A skill is a folder with a SKILL.md: a name and a description up top, instructions below, and
any files it refers to next to it. It is the same format Claude Code, Codex and the
skills.sh catalog use, so skills from Vercel, Anthropic or any public
GitHub repository work as they are.
The host reads skills from .agents/skills, .claude/skills and .plural/skills in the
project and in the home folder; a project skill wins over a global one with the same name. The
agent sees each skill's name and description and loads the full instructions with the skill
tool when a task matches.
Commands
| Command | What it does |
|---|---|
/skills | List the installed skills |
/skills search <query> | Search skills.sh, most installed first |
/skills add <owner/repo@skill> | Install into the project's .agents/skills (host) |
/skills add <owner/repo@skill> --global | Install into ~/.agents/skills for every session (host) |
/skills rm <name> | Remove a skill Plural installed (host) |
/skill:<name> [request] | Run a skill by hand: the agent gets its instructions and your request |
/skill:grill-me my plan works for anyone who can prompt: the room sees my plan marked
via /skill:grill-me, and the agent gets the skill's instructions followed by the request.
Every installed skill also costs a line of the agent's prompt on every request. A skill you only
ever run by hand can stay out of it: add disable-model-invocation: true to its frontmatter,
and the agent no longer sees it, while /skill:<name> still runs it. /skills marks such
skills.
---
name: grill-me
description: Pressure-tests a plan before you commit to it.
disable-model-invocation: true
---The agent can search the catalog on its own and suggest a skill. It can install one too, but every install waits for the host's approval like an edit does, and plan mode refuses it.
A skill is someone else's instructions
The agent will follow an installed skill's instructions with the host's permissions. Install skills only from sources you trust — see Security → a third-party skill steering the agent.