A model may answer with **bold** or `code` markers in the text. OpenCues strips the markers before anything reaches your text and renders the styling natively instead, so you see the emphasis, never the asterisks.
the model returned
you get
Your text itself stays plain: the markers are a wire format stripped at write time. Terminal hosts render all of the styles below; the browser extension renders bold, italic and strikethrough, and skips code, heading and list styling.
Name a word and a style, and only that word changes.
you type
you get
The instruction and the underscore are consumed; the rest of the line is left exactly as written.
| The model writes | You see |
|---|---|
| **bold** | bold |
| *italic* | italic |
| `code` | code styling |
| ~~strike~~ | strikethrough |
| # heading, - list item | heading and list styling at line starts |
Editing inside a styled stretch of text removes its styling: a hand-edited stretch is yours, not the model's, and the styling of that insertion is dropped as a whole.
Styling detection is suppressed around an active _, so volume _ can never have its underscore read as the start of an italic span. And on surfaces that render markdown themselves at send time, such as Discord through the Windows integration, the markers are left in your text untouched, because there they are the correct content.