Slash commands

Built-in slash commands for chat — session control, model and mode, prompt-shaping helpers, and the built-in skills invoked the same way.

Product stage: Development build
Last reviewed: 2026-08-28
Successful result: The change is visible in the active project.

Who this is for: anyone extending chat with / commands — session control, model and mode switches, or one of the built-in skills.

Type / in the chat input to open the command picker. Filter by name; navigate with the keyboard.

Session and export

Command	What it does
`/help`	Show the list of available commands
`/clear`	Clear the current session
`/fork`	Fork the session into a new one
`/compact`	Condense older turns to free context (alias: `/condense`)
`/export`	Copy the conversation to the clipboard as Markdown
`/generate <task>`	Draft a starting prompt for a new task, pre-filled in the composer

Model and mode

Command	What it does
`/model [provider/]model`	List the active model, or switch to a different one for this session
`/mode [mode]`	List the execution modes, or switch this session to one (`chat`, `plan`, `confirm`, `autoedit`, `fullaccess`)

See permission modes for what each mode allows.

Prompt-shaping commands

These insert a system prompt for the turn and then continue as a normal message — they do not run tools on their own.

Command	What it does
`/plan [task]`	Ask the model for a concrete, numbered plan
`/explain [scope]`	Explain the selected code, file, or architecture
`/summarise [scope]`	Summarise the conversation or a file
`/translate <language>`	Translate the selected text or docs

Other built-ins

Command	What it does
`/web <url>`	Open the URL in the in-app browser panel (requires the browser panel enabled in Settings → Privacy)
`/diagram [type] <source>`	Render a diagram via Kroki
`/roleplay [off	brief]`
`/goal <objective>`	Start a long-horizon goal loop for this session
`/goal status`	Show the active goal loop’s progress
`/goal clear`	Abandon the active goal loop and unfreeze the execution mode
`/fuse <prompt>`	Fan the prompt out to a panel of models at once (via OpenRouter) and combine the responses. Advanced; each model in the panel bills separately.

Built-in skills

Miton also ships a handful of default skills, invoked the same way — type /name or $name:

Command	What it does
`/code-review`	Comprehensive PR-style code review
`/security-scan`	Security-focused scan of the current code
`/documentation`	Draft or update documentation for the current scope
`/debug`	Investigate a failure with a debugging-oriented prompt
`/deploy`	Walk through a deployment for the current project
`/wiki`	Work with the project wiki
`/miton` (or `$miton`)	Ask about Miton itself — capabilities, surfaces, privacy, models — or change an allowlisted setting

Sub-agents are not slash commands

Sub-agents (Explorer, Verifier, Debugger, Test Runner, Orchestrator, and others) are spawned with an @name mention in the chat input — for example @explorer where is the auth code? — not with a slash command. See Sub-agents for the full list and how they run.

Command history

Recent commands appear in the picker. History is per user (not per project) and stored locally.

Limits

Slash commands are single invocations — multi-turn work continues as a normal session
Plugin-provided commands follow the plugin model; they are not user-authored YAML files
Whispered commands such as /kim still run when typed in full; they stay off /help and this table