Symbol

Prompts and skills are two ways to make a capsule invocable in your AI client, either a prompt you call on demand or a skill it can load on its own.

A capsule is knowledge your AI assistant can already read and use. Prompts and skills go a step further: they let you turn a capsule into something your assistant can invoke directly, in one of two ways:

A prompt: a reusable, fill-in-the-blanks message you drop into a conversation on demand.
A skill: a packaged capability your assistant can discover and follow on its own.

They look similar in your assistant's slash-command menu, but they do different jobs.

At a glance

	Prompt	Skill
What it is	A reusable message template	An installed capability
How you use it	You type it (in Claude Code, `/mcp__symbol__senior-code-reviewer`)	Your assistant picks it automatically from its description, or you type it
What happens	The composed text is dropped into the chat as your message	Your assistant loads the instructions and acts on them
Variables	Supports fill-in variables at use time	None; the content is fixed when you publish
Availability	Live through your Symbol connection	Installed once, then available across sessions

Rule of thumb: a prompt is text you send; a skill is a job your assistant runs.

Prompts

Turn on the prompt toggle for a capsule and it appears in any connected AI client as a slash command (in Claude Code, /mcp__symbol__senior-code-reviewer). Selecting it inserts the capsule's content as your next message, a fast way to reuse a persona, a checklist, or a set of instructions without copy-pasting.

In Claude Code

Symbol prompts surface as MCP prompts, which Claude Code exposes as slash commands. See Slash commands: MCP prompts and Use MCP prompts as commands.

Variables

A prompt can contain {{variables}}. There are two kinds:

Field values: filled in automatically from the capsule's Type fields. If a capsule has a profession field set to "software engineer", then {{profession}} is already filled when you invoke the prompt.
Caller variables: left open for whoever uses the prompt to fill in. When you invoke the prompt, your assistant asks for (or infers) these values.

You can review and change how each {{variable}} behaves from the Variables view on the capsule page, a click-to-edit view of the prompt source where each token shows whether it's a filled-in field value or an open blank the caller supplies.

Try it

Open a capsule that contains {{variables}}, turn on Prompt, then in your AI client invoke it (in Claude Code, /mcp__symbol__<name>). Field values arrive pre-filled; open variables become questions your assistant fills in.

Skills

Publishing a capsule as a skill packages it as a reusable capability your AI client can load. Unlike a prompt, which you invoke by name each time, a skill can be selected automatically by your assistant when its description matches what you're doing.

A published skill's name comes from the capsule's Type and its alias (for example agents-senior-code-reviewer), or its numeric reference when it has no alias. Its description is drawn from the capsule's title and summary, so write those to read like "use this when…"; that's what helps your assistant know when to reach for it.

A few things to know:

Encrypted capsules can't be published as skills. Their content can't be rendered for distribution.
Skills have a size limit. A very large capsule may be rejected or warned about at publish time; keep skills focused.
Skills are shareable. Published skills are distributed through your workspace, so teammates who install it get the same capability.

Getting the latest version

Both prompts and skills are cached by your AI client, so a fresh edit doesn't always appear instantly.

Prompts: if you edit a prompt and don't see the change, reconnect Symbol in your client. Starting a new chat alone may not refresh the list, because the client keeps its connection (and its cached prompts) between chats.
Skills: published updates are picked up the next time your client starts a session. Republishing with no changes is a no-op, so you never get spurious "update available" churn.

Editing a capsule in Symbol updates it immediately on the server. If the change isn't showing in your assistant, it's the client's cache: reconnect (prompts) or start a fresh session (skills).

Troubleshooting

A prompt isn't showing up:

Confirm the Prompt toggle is on for the capsule.
Make sure the capsule isn't encrypted or archived; neither can be a prompt.
Check you're connected with the account and workspace that owns the capsule.
Reconnect Symbol to refresh the list.

A skill isn't showing up or is out of date:

Confirm the capsule is actually published as a skill (this is separate from the prompt toggle).
Make sure it isn't encrypted, and that it's within the skill size limit.
Confirm you've installed the workspace's skills where you're using them — see the Plugin Marketplace page for the install command and how to recover a lost install URL.
Start a new session to pick up the latest published version.

FAQ

Do I have to choose one? No. A capsule can be a prompt, a skill, or both. Pick by intent: send text → prompt; have your assistant do work, discoverable on its own → skill.

Can I pass values into a skill like I do a prompt? No. Variables are a prompt feature, resolved when you invoke the prompt. A skill is fixed content published ahead of time, so any {{...}} in it is just literal text.

I edited the capsule but the output didn't change. You're seeing a cached copy: reconnect Symbol (prompt) or start a new session (skill).

Can an encrypted capsule be a prompt or a skill? Neither. Encrypted content can't be rendered on the server, so it's excluded from both.

What if I need a whole subagent, not just a capability? That's an agent — its own system prompt, tools, and model, packaged from a capsule and delivered as project files, ephemeral --agents JSON, or a plugin.

Related: Capsules · Types · Agents

Prompts & Skills