Ability to click elements by text, not by ref

[radiantspace]

What's really lacking is ability to refer elements by their text, as we have in playwright proper. So if we have that information in the snapshot, why don't allow that with playwright-cli click command? Ofc agents can read snapshots for every step, but thats sub optimal.

The use case is a custom skill I make with playwright and it has something like that:

playwright-cli click f3e519 # click "Add filter" button
playwright-cli click f3e529 # click "Date" filter option
playwright-cli click f3e572 # click "Newest created" option

What I want to have is human readable and something that can be immediately reasoned about (and those are well documented playwright APIs honestly, so LLMs should also be able to apply those):

playwright-cli click "button[aria-label='Add filter']"
playwright-cli click "button[role='menuitem']:has-text('Newest created')"
playwright-cli click "button[has-text='Alt tag is empty']"

[pavelfeldman]

When my agent drives the page, it knows refs. Mapping desired element to ref is the LLM's business. But we do print the resolved element for you:

playwright-cli check e21
### Ran Playwright code
```js
await page.getByRole('listitem').filter({ hasText: 'Buy groceries' }).getByLabel('Toggle Todo').check();
`` `