Slash commands
Typing / opens the command palette. It lists every command available in the current project: Claude Code's built‑ins, plus whatever your project, skills and plugins add.
Using the palette
↑ ↓ move, Tab completes, Enter completes — and sends straight away if the command takes no arguments. Esc closes it and leaves your text alone.
Matching considers the command's name, its aliases and its description, best match first. So /revi finds code-review through its review alias.
Commands are grouped by where they come from: Claude Code's own, a plugin, or an MCP server.
Local commands
Some commands are handled by Claude Code itself and never reach the model — /context, /usage, /clear. Their answers are shown as a panel, visibly not Claude speaking, and their turn footer says what ran rather than reporting a cost of $0.0000.
Two are worth knowing in Vitrino specifically:
/clearends the conversation. Claude Code starts a new session id, and nothing above that point reaches Claude again. Vitrino draws an unmissable marker and follows the new session, so reopening the tab later resumes the conversation after the clear rather than the discarded one./compactsummarises the conversation so far to free context. The context meter offers the same thing as Compact now.
Commands you will not see
Names beginning __ are Claude Code's internal plumbing — wired for the harness to call, not written to be typed. Vitrino leaves them out of the palette.
They are hidden, not blocked. Typing one still sends it, if you know what you are doing.
MCP prompts are named mcp__server__prompt and are listed — the __ is in the middle, not at the start.
Commands that change mid-session
Claude Code pushes a fresh list whenever it changes, so installing a plugin or adding a project command makes it appear in the palette without restarting.