| /** |
| * The `/skill:<name>` invocation grammar (issue #1148) — one string, because |
| * every client that reads or writes a draft has to agree on where a token |
| * starts and ends: the TUI's highlighter and autocomplete, the Desktop |
| * composer's chips, and Runtime's submit-time parser and stripper. |
| * |
| * It lives in core rather than in Runtime because the Desktop composer needs |
| * it to render a draft, long before anything is sent, and the renderer has no |
| * business depending on Runtime for that. |
| * |
| * A token is valid at the start of the text or after whitespace, so paths and |
| * URLs (`a/skill:b`, `https://x/skill:y`) never produce false positives. |
| * `<name>` uses the skill id charset; resolution downstream matches by id |
| * first, then by display name. |
| * |
| * Always construct a fresh `RegExp` from this source at the point of use: a |
| * shared instance with the `g` flag carries `lastIndex` between calls. |
| */ |
| export const SKILL_INVOCATION_TOKEN_SOURCE = String.raw`(?:^|(?<=\s))\/skill:([A-Za-z0-9._-]+)`; |
