Nice catch and clean fix. The original getLastNVisualLines accumulated every wrapped line for the full text just to slice(-n) at the end, which is wasteful for long text with a small n (per the code, n <= 5). Your pushLine eviction (shift-then-push once lines.length === n) keeps the accumulator bounded and produces the same output.

A couple of things worth double-checking before this lands:

1. Array.prototype.shift() is O(current length) per call. Since n is small in practice this is negligible, but a small ring buffer (fixed-size array + head index) would avoid the shift cost entirely and be a more idiomatic 'bounded queue' if this function is ever called with a larger n. Not blocking, just a note for future-proofing.
2. The trailing-push condition changed from current.length > 0 || lines.length === 0 to current.length > 0 || (!hasMore && lines.length === 0). This looks correct given n > 0 is enforced upstream (lines never drain back to zero once eviction begins), but it's a subtle enough change that it deserves a one-line comment explaining why hasMore is checked here, for the next person reading it.

Test coverage is solid — zero/negative bounds, exact-n, over-n with eviction, and long-line wrapping are all exercised, and they pass. This is a small, self-contained, well-motivated change that's easy to port by hand. Good first PR.