> ## Documentation Index > Fetch the complete documentation index at: https://docs.vistazo.ai/llms.txt > Use this file to discover all available pages before exploring further. # Skills > Written procedures the agent follows, so you stop re-explaining them. A skill is a markdown document your workspace writes for the agent. How your team names things, what a "quarter" means here, the recipe for the report you rebuild every month, which source to trust when two disagree. It is standing instruction rather than a one-off prompt. Write it once and the agent follows it whenever it applies. ## Writing one Skills live under **Skills** in the nav. A skill has a name, a description, and a body. The **description** is what the agent reads when deciding whether a skill is relevant, so it is load-bearing. "How we count active customers" earns its way into a turn; "customer stuff" does not. Vistazo writes a first draft of the description for you, and you can edit it. The editor saves as you type. If a colleague is already editing, the page says so by name and stays read-only until their lock goes stale, at which point you can **take over**. Links in a skill body render as links. It is a real document. Keep it the length you would want to read, since a skill that loads before the work costs tokens every time it applies. ## How the agent uses one Two paths, and both matter. * **You invoke it.** Type `/` in any composer and pick one. Its body is prepended to that turn. * **The agent picks it.** The names and descriptions of the skills you can read sit in the agent's context, so it can reach for the right one on its own when your question matches. The second path is why descriptions deserve care, and why a skill does not need to be invoked to be useful. The catalog holds up to 100 skills and covers only the ones you have access to. ## Versions Skills are versioned like dashboards, under **Settings → History** on the skill. Save a version deliberately, or let one be recorded when your edits settle. Restoring is undoable, because the current state is snapshotted first. ## Who can read and edit By default every full member of the workspace can read and edit every skill. You can narrow that per skill under **Settings → Sharing**, naming individual people or changing the workspace baseline. Two exceptions worth knowing. * **Guests**, meaning people invited by email to specific objects, need an explicit grant and are always read-only. * **Viewer seats** can browse and read skills in the app, but cannot edit them, and their agent gets no skills at all. Neither the catalog nor an invoked body reaches a viewer-seat chat, because skills are instructions for building. See [Roles and seats](/workspace/members#roles-seats-and-guests). A skill can also be shared outside the workspace with a link, like a dashboard. The link renders the document. See [Sharing](/sharing/overview). ## Agents can write them too An agent connected over MCP can author skills with `save_skill`, and the write happens immediately. A skill an agent created or last edited is held back until a person confirms it. It appears in the agent catalog **only to its author** until somebody opens it in the editor and saves. So an agent cannot quietly install standing instructions that steer everyone else's conversations. Vistazo AI, the in-app chat, asks before saving a skill, because a skill is workspace-wide standing instruction rather than an edit to one object. Your own MCP client does not ask, which is what the hold-back above exists to cover.