/help opens with the trust map (posture /policy · activity: the receipt and /trace · record: /trace export and replay) and lists every command by theme. /help --all still works and prints the same thing. /<command> --help details any one. --help works as the first or last argument; in the middle it's ordinary data, so /memory add … can store a fact that mentions it. Removal verbs are interchangeable everywhere: remove / rm / delete / del / forget / drop.
| command | does |
|---|---|
/help [cmd] (/?, /h) | every command by theme, or one in detail |
/memory (/mem) | see, add, edit, remove, and review layered memory; why <n>, stale, setup (the first-run questions) |
/skills | your own procedures: list them; show <name>, create <name>, delete <name> (to the Trash, asks first) |
/policy | your trust settings: risk, allow, shortcut, open, egress, airgap |
/think [fast | auto | deep] | how much Saturn reasons before it answers: bare shows the level and what the last turn's passes did; a level sets it (--session for this session only) |
/trace | inspect runs: why, source, invoke, export, replay, on|off|full |
/quit (/exit, /q) | exit, running /memory review first if candidates are pending (--no-review skips it) |
/clear | start a fresh conversation |
/resume (/continue) | sessions: resume the autosave; save [name], list, <name> |
/copy | copy the last answer to the clipboard (macOS) |
/add-dir [path] · /rm-dir <path> | reach another folder for this session · stop reaching it |
/docs (/documents) | the knowledge base: add <path>, remove <name>, rebuild |
/init [--force] | survey the working folder and draft SATURN.md |
/undo [list] | revert the last turn's file changes |
/models (/model) | the model page: your hardware, the tiers priced against it, every other model you've pulled; use <id>, embedder <id>, list |
/tools | the toolkits: on|off <toolkit> (--session), <toolkit> lists its tools and tiers, --all every tool |
/mcp [list | reload] | MCP server status and the tools they add |
/config | view or edit config.yaml; /config <key> [value] [--session]; /config reload |
/notify | scheduled notifications: list, cancel <id>, test, icon [start|stop] |
/update [--check] | self-update (git pull at the install root); your data is never touched |
::at the prompt
!command
run a shell command yourself, without the agent; the output attaches to your next message
@file · @clipboard
attach a file, or what's on the clipboard
/think <request>
run this one request at deep, whatever the level; a level word on its own sets the level instead
/<skill> [request]
run one of your skills (/weekly-review focus on work); /<skill> --help shows it instead. A built-in command's name always wins over a skill's