Skip to main content
Good candidates include a Wwise audit, a render check or a UCS naming pass.

Create a Command

  1. Open Settings > Commands and click New command.
  2. In Name, type a short name such as ucs-check. You type this after /.
  3. In Instructions, write the request as you’d type it in chat.
  4. Click Save (Cmd+S / Ctrl+S).
Save before you leave the form. Only Cancel and Esc ask “Discard changes?”. Any other way out, such as clicking another Settings tab or starting a new chat, loses unsaved changes.
Names use only lowercase a to z, digits and single hyphens between words (none at the start or end). A name can be up to 64 characters and must be unique. Reagent tidies the name as you type. Avoid new: a command with that name can’t be opened for editing. Write a command like a clear chat message: what to look at, what to do, and when to stop and ask you. A command named ucs-check could contain:

Run a Command

Type / at the start of your message or after a space or line break. A list of your saved commands opens in alphabetical order. Nothing appears until you’ve saved one. There are no built-in commands, and skills don’t appear here. Keep typing to narrow the list. Part of a name or a small typo still matches. Move the highlight with the arrow keys. Press Enter or Tab to insert the highlighted command (the top match, unless you moved the highlight), or click a command. Esc closes the list. While the list is open, Enter inserts instead of sending, so press it again to send. Names that match a saved command turn bold. You can mix commands with your own text and use several in one message:
Reagent replaces each recognized command with its saved instructions and sends everything as one normal message:
  • Your chat shows the short version (/ucs-check). Reagent keeps the full text with the message, and that’s what the model sees.
  • Edits apply from your next message. Reagent reads the saved instructions each time you send. Messages you already sent keep their original text.
  • Commands are text, not settings. A command doesn’t change your permission mode, the model or the tools Reagent can use.
If a command asks for something your plan doesn’t include, the message still sends but Reagent can’t do that part (see Feature Availability).

Manage Commands

On Settings > Commands, click a command or Edit to change it. After a rename, only the new name works. To delete one, click the trash icon (or Delete in the editor) and confirm. This deletes the file and can’t be undone. Each command is a .md file named after it that holds only the instructions. The files live in a commands folder in Reagent’s app data folder. Open in Finder opens the app data folder with commands selected. The label is the same on Windows, where it opens File Explorer. Copying the folder makes a backup. If you add or edit these files yourself:
  • Name each file after its command, following the naming rules above, and end it in .md. Files with other names are ignored.
  • Everything in the file is sent as is, including headings or notes to yourself.
  • Reagent picks up files you add, rename or delete next time you open Settings > Commands or restart Reagent.
Commands belong to this computer and aren’t synced. To copy them, put the .md files in the other computer’s commands folder (click Open in Finder there).

Troubleshooting

Read the message under Name:
  • “Use lowercase letters, digits, and hyphens only”: the name starts or ends with a hyphen. This message also shows briefly while you type a name ending in a hyphen, such as ucs-.
  • “A command with this name already exists”: choose a different name.
Reagent sends /name exactly as you typed it when:
  • It isn’t a separate word, as in SFX/ucs-check, /ucs-check. or /ucs-check,. The name isn’t bold. Put a space after it or end the message with it.
  • The file was deleted outside Reagent. The command can stay in the / list until Reagent picks up the change. Sending it shows “Command /name no longer exists — sent as text.”

Chat

Where you type and send commands.

Skills

Built-in workflows Reagent loads when a task needs them.