Skills and slash commands

A skill is a reusable procedure written as one SKILL.md document. Save it on a channel, keep it for yourself, or give it to an agent, then run it by typing its name as a /command.

Skills are how you stop re-explaining the same work. Write down your launch checklist, your review steps or your brand voice once, and everyone who runs the command gets the same steps, the same checks, and the same idea of what finished looks like.

Three places a skill can live

Where you save a skill decides who can run it and where it works.

Run a skill

Type / at the start of a message to open the list of commands you can run there, then pick one. A skill is in that list once it's a command (see Choose which skills are /commands). In a channel, the list draws on that channel's skills, the agent you're talking to, your own, the workspace skills you wrote, and the skills of agents you own that are on this channel's team. A channel's skills show up only in that channel's chat and in tasks started from it.

A coding agent can bring its own skills too. Once the coding agent you're talking to, or one of yours on the channel's team, has run at least once, the skills its runtime reports (Claude Code's, for example) show in the / list, and picking one runs it in that agent's own runtime.

A /command doesn't have to lead the message. Type one mid-sentence, such as "can you /launch-brief this spec", and its steps apply to that message.

Desktop and mobile show a badge on each skill row: Workspace, Channel, Job or Personal. An agent skill shows the agent's name when available, otherwise Agent. On your phone, use the same / list above the message box to see a skill's badge before choosing it.

Every /command comes from a skill. Nothing that ships with Nebula is a command, so a name that no skill claims isn't one you can run.

The list is what's yours, not everything anyone has ever written - a big workspace would otherwise bury your own commands under teammates' ones. A teammate's skill, or a skill that isn't a command, still runs if you know it: type its name at the start of a message, and as long as nothing in your own list claims that name, it's found.

When you run one in a channel, what you typed posts as your message. The work runs as a task in a thread under it, and the result reports back there.

In the message, the command shows as a violet pill with the skill's name. On desktop the pill draws for everyone in the conversation, whichever teammate ran it: click it to see what the command does - its description, how to run it, who wrote it, where it's shared from, the agent it runs as and its SKILL.md. A command that isn't yours yet offers Add to my commands, which copies it into your own commands so you can run it anywhere you work, and that's how a teammate's personal command gets shared. When a command can't be added the button is greyed with the reason beside it, and once it's yours it reads In my commands. On your phone a pill draws for a command your own list offers, and tapping it opens the same explanation read-only.

Choose which skills are /commands

Skills that agents write for themselves while they work don't crowd the / list: a skill shows there only once someone makes it a command. Taking one out of the list keeps the skill, and agents can still read and use it.

See every skill

On desktop, Settings → Workspace → Skills lists every skill in the workspace, at every scope. Filter by who wrote it or by agent, open one to read or edit its SKILL.md, or use the trash icon to delete it. Channel settings → Skills shows the same list narrowed to what a /command can reach in that channel. A skill that a job writes while it runs can't be changed there. A workspace or channel skill can be changed or deleted by whoever saved it last, or by someone who manages that workspace or channel. On an agent, anyone who can edit the agent can change its skills.

On your phone, Settings → Skills (under Workspace Brain) lists the same skills. You can read and delete them there, but you create and edit them on desktop.

Create a skill

A finished channel skill might look like this:

---
name: launch-brief
description: Turn a feature spec into a one-page launch brief for q4-launch.
command: true
---

1. Read the linked spec and pull out the problem, the audience and the ship date.
2. Draft a headline and three benefit bullets in our brand voice.
3. List open questions for Maya before anything goes to Launch Copywriter.

Done means one page, no jargon, and every claim traceable to the spec.

Editing a skill later bumps its version, which shows next to its name in the list.

Which skill runs when names clash

The same name can mean different things in different places. When more than one skill shares a name, the most specific one wins.

A shared workspace skill only ever runs as that fallback, when no channel, agent or personal skill uses the name.

Say Maya keeps a personal /launch-brief, and q4-launch has its own. In q4-launch, everyone - Maya included - gets the channel's version. In every other channel, Maya gets hers.

Install a skill from a link

Found a skill someone published? Ask Nebula to install it from a direct link to its SKILL.md. Nebula fetches it, runs a safety scan and installs it on the agent you name, right in the chat. If the scan flags a problem, nothing is installed and Nebula tells you why.

The conversation below is illustrative. Replace the example URL with a direct HTTPS link to a SKILL.md you have reviewed, and name an agent you can edit.

Y

You

Nebula, install the launch checklist skill from https://example.com/launch-checklist/SKILL.md onto Launch Copywriter.

Nebula

Installed. The SKILL.md passed its safety scan, and launch-checklist now shows under Skills in Launch Copywriter's Brain.

FAQ

Related