fix(terminal): stop URL wrap detection swallowing the next line

[matej21]

Problem

Phase 2 of URL detection (links.rs) extends a URL onto the following row when a TUI wraps the line itself, i.e. without a WRAPLINE flag. Its "URL reaches the end of the visible content" guard compared the URL's end against the URL row's own trimmed length — trivially satisfied by every row that ends with a URL. From there the only thing standing between the URL and the next line was the weak-extension guard, which lets through any token containing a /.

Claude Code's PR output hit exactly that shape, so the branch name landed inside the link:

  https://github.com/org/repo/pull/4
  (docs/data-flow-findings-and-system-audit → main).

Detected text before the fix:

https://github.com/org/repo/pull/4(docs/data-flow-findings-and-system-audit

The whole branch name was underlined and clicking the link opened the wrong URL.

Fix

A mid-token wrap fills the row to the layout edge, so a continuation row can never be wider than the row it continues. In the case above the URL row ends at column 46 while the next row runs to 52 — proof that the break was a word break, not a wrap.

• Reject the extension when the next row is wider than the URL's end column, using the same +3 tolerance the other guards use.
• Track url_end_col across iterations so multi-row extensions compare against the row they actually continue.

Phase 1 (real WRAPLINE wraps) is untouched.

Tests

New regression test detect_url_not_extended_when_next_line_is_longer, built from the screenshot above.

cargo test -p okena-terminal → 173 passed, 0 failed. All existing wrap tests still pass, including detect_url_wrapped_tui_narrow_layout and detect_url_wrapped_with_trailing_text, which cover genuine TUI-wrapped URLs.

🤖 Generated with Claude Code

https://claude.ai/code/session_01JZXzxKAXvZc2RrjYWRAbMX