Skip to main content
The input is where you give the model instructions. The placeholder says:
Input and bottom pickers

​
Send and newline

Settings → General / Behavior has two send styles: While the model generates, the send button becomes Stop.

​
Completion triggers

Longer triggers win first: @# / @@ match before @. Agents are usually picked from the input toolbar / picker, not from a standalone # completion trigger. That differs from the # agent entry in the VS Code docs.

​
@ file references

Type @ to search files in the current workspace. A selected file becomes a file chip. You can also write:
Dragging a file from the file tree is similar: a regular file becomes a path reference, and an image becomes an attachment. Type @ to attach a project file

​
/ and $

  • /: slash commands for the current runtime plus local commands
  • $: Skills / skill aliases

​
! prompts

Type ! to insert a prompt library template.

​
@@ and @#

  • @@: pick project memory by hand and inject it into the current turn
  • @#: attach a local note card

​
Attachments, paste, and drag-and-drop

The input accepts:
  • File path references
  • Images (paste or drag in)
  • The quick-action toolbar items: attach file / memory / skill / agent / prompt / enhance

​
Bottom pickers

Usually:
  1. Provider / engine
  2. Model
  3. Mode
  4. Tools / attachments / stop / send
Claude has plan / agent / auto permission modes. Codex has collaboration mode and sandbox-related options. Other engines differ by runtime.

​
Message queue

You can type the next message while the previous turn is still generating. Queued messages send automatically after the current turn finishes.

​
Enhance prompt

The input toolbar can rewrite the current draft before you send. Configure it on the enhance / prompt pages in Settings. See Enhance prompt.