In the message box, / runs a command or a skill, and @ points Oda at a file, a folder, an earlier chat or a helper. This page covers how both work; the commands themselves are described in Chat with Oda.
Open the / menu
Type / in the message box: the menu opens above it, listing the commands and the skills of the open chat. It never hides the text you are typing.
The menu opens when the slash is at the start of your message or follows a space or punctuation. Inside a web address, inside a Windows-style path such as C:/, after another slash, or straight after a word, the slash stays ordinary text.
The + control, Add files or run commands, opens the same menu.
Rows are grouped: the built-in entries under Add (File, Goal, Plan, Feedback) and Commands (Compact, Permission, Model, Export), your skills under Skills, and any subagents of this chat under Subagents.
Each row carries an icon, its name, and a short description of what it does, for example Enter or leave plan mode for Plan and Download this Session log as a ZIP archive for Export.
While the rows load, placeholder rows stand in; rows you have already seen stay while the list refines.
A group that cannot be prepared drops out quietly, and the rest of the menu stays usable; if nothing remains, the menu closes itself.
Run a command or skill
Type / at the start of the message box, or use the + control.
Keep typing to narrow the list; the filter matches names. Move the highlight with Up and Down; it wraps at both ends. You can also type a command's name in full.
Press Enter, or click a row, to pick the highlighted entry. Tab settles the highlighted row where a row offers that. Escape or Shift+Tab leaves the menu without running anything and keeps what you typed.
Send the message. The command runs on its own; it does not start a model turn.
The highlight follows the keyboard and the pointer, whichever you used last.
A click outside closes the menu; a click inside the message box does not, and a closed menu stays closed for the same text.
A key pressed while the list is still loading does not run a row from the old list, and an input method's Enter that picks a candidate never runs anything.
Running a skill works the same way: pick it under Skills. A skill is a set of instructions: /<name> sends your message with the skill loaded, and Oda follows those instructions while it works. You create and edit skills in Settings > Skills; see Using skills.
Commands differ in what they do: /plan turns plan mode on or off, /permission changes the access mode, and /export downloads the chat log.
A command starts at the very beginning of your message: a slash and a lowercase name made of letters, digits, underscores and dashes. If the shape does not match, nothing runs:
goal, without the slash, is a word like any other, and /goal, with a space before the slash, stays ordinary text.
Capital letters do not count: /Goal is not a command.
The name ends at a space or at the end of the message, so /goal/path is not a command.
A lone / runs nothing, and a name Oda does not offer is not a command.
Lines like these go to Oda as normal messages, and Oda answers them like any other prompt. Everything you type after a command's name is handed to the command as it is.
While a command holds your message
When your message starts with a command, the command token is highlighted while it holds, and you can keep typing freely around it.
Deleting the token is the only way to release it: put the caret after it and press Backspace. Escape closes the menu but not the command, and deleting only the text after the name keeps it too.
Renaming the command turns it back into ordinary text.
Press Enter to send: the command runs with the rest of the message as its input.
While the command holds, a quiet hint says what to type: describe your task to generate plan for Plan and describe the objective for a long-running task for Goal. When a goal is already active, the hint offers to edit, pause, resume or clear it. Goals are covered in Chat with Oda.
Files and images with commands
Attachments are checked before a command runs. Only a command that accepts attachments takes them: /plan and /goal do; many others, such as /compact, do not.
If you send an attachment to a command that does not take one, nothing runs: the message box shows /{command} does not accept attachments; remove them first, for example /compact does not accept attachments; remove them first, and your text and the attachments stay put, so you can remove them and send again. The same happens when an attachment is not ready or an image batch is over the limit.
Options and confirmations
Some entries open a short list of choices. While it loads it reads Loading options…; with nothing to offer it reads No options; while your choice is being applied it reads Applying….
The search field takes focus when the list opens, so you can type to filter the choices.
If the list cannot be loaded it stays open with the reason and a retry control.
A choice that carries a warning opens a card first: acknowledge the warning, then confirm. Escape cancels without applying anything.
Every command answers in the chat with its own row: it shows Running… while it works, then Completed or Command failed. A running row carries a Running badge and a failed row a Failed badge.
The row expands when the answer runs over one line.
A command's own answer is not sent to the model, and running one uses no model tokens.
If a command cannot go through, one error notice appears and your text stays in the box so you can correct it.
Point Oda at a file, folder or chat
Type @ in the message box. The menu opens, grouped into Files & folders, Sessions and Subagents.
Keep typing to narrow the list, then pick an entry. Your pick is added to the message as a chip.
For a folder, press Enter to add the folder itself, or Tab (or the row's chevron) to look inside first. A breadcrumb shows where you are, and each step of it takes you back out.
To remove a chip, put the caret right after it and press Backspace, or point at the chip and click the ✕.
A reference tells Oda what you mean. Oda opens the file, folder or chat it needs while it works, and the read shows in the chat as a step. See Reading files and images.
The Sessions group lists your earlier chats. Choosing one points Oda at that conversation; see Reference other chats.
Helpers the chat has started appear under Subagents. See Subagents.
Questions, answered
How do I see all the commands?
Type / in the message box: the menu lists the commands and the skills of the open chat, and typing narrows the list. Chat with Oda describes what each command does.
Can I open the menu without typing /?
Yes. The + control, Add files or run commands, opens the same menu, and you pick a row the same way as when you type /.
Why did no menu appear when I typed /?
The slash opens the menu only when it is at the start of the message or follows a space or punctuation. Inside a web address, a Windows-style path such as C:/, after another slash, or straight after a word, it stays ordinary text.
Why did my message go to Oda instead of running a command?
A command only runs when the message starts with a slash and a lowercase name of letters, digits, underscores and dashes. goal without the slash, /Goal with a capital, /goal/path, and a slash typed after a space stay ordinary text: the message goes to Oda as a normal prompt.
Do commands use model tokens?
No. A command runs on its own rather than as a model turn: its answer stays in the chat and is not sent to the model, so running it spends no model tokens.
What happens after I pick an entry?
The command runs and answers in the chat with its own row: Running… while it works, then Completed, or Command failed with what went wrong. Nothing else changes when a command fails; fix the cause and run it again.
Can I add files or images to a command?
Only if the command accepts attachments: /plan and /goal do, while many others, such as /compact, do not. A command that does not take them refuses with /{command} does not accept attachments; remove them first, and your text and attachments stay in the message box.
What happens when a command asks for a choice?
The choices open in a list in the same spot: Loading options… while they load, No options when there is nothing to choose, and Applying… while your choice is applied. If the list fails to load, it stays open with the reason and a retry control. A choice that carries a warning opens a confirmation card first.
How do I cancel a command I started typing?
Delete the command token: put the caret after it and press Backspace. Escape closes the menu but does not release the command, and deleting only the text after the name keeps it. Renaming the command turns it back into ordinary text.
What is the difference between a command and a skill?
Both appear in the / menu; skills are listed by name under Skills. A command runs directly, for example /compact. A skill is a set of instructions: /<name> sends your message with the skill loaded, and Oda follows those instructions while it works. See Using skills.
Can I remove a reference I added?
Yes. Put the caret right after the chip and press Backspace, or point at the chip and click the ✕. Removing a chip changes nothing else in your message.
How do I point Oda at a folder, not only a file?
Choose the folder from the Files & folders group to add the folder itself, or press Tab (or the row's chevron) to look inside first. A breadcrumb shows the way back out.