You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Three things stacked up so the command appeared to do nothing.
A row's cells live under `data`, and column inference skips object-valued
fields — so the table came back listing an id and two timestamps per row and
none of the content the query was run for. `expand` names the wrapper whose
keys become columns, unioned across the page like the top-level ones. A cell
key that shadows a top-level field is shown by its full path, so two different
values never share a header.
A cell containing a newline pushed the rest of its row onto the next line and
every column after it lost alignment; in text mode a tab invented a field that
`cut -f` reads as real. Display cells are now flattened to one line. `sanitize`
still keeps \t and \n — json and yaml must round-trip them, and this is applied
only to finished cells.
A single cell holding an LLM response set the column width for the whole table
and pushed everything after it off-screen, so table cells clamp at 60 columns.
text/json/yaml are untouched: those exist for the whole value.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0148a65fkPhP4N8tgGPYtRTU
0 commit comments