feat: redesign the advisor tool row - #28069

Summary

Redesigns the advisor tool row in the Coder Agents chat to match the density and conventions of the other tool cards, adds optional intent labelling, and returns an explicit budget-exhausted message to the agent.

Problem

The advisor row stood out from adjacent tool calls. The header stacked a label plus multiple pill chips (model name, "N uses left") and wrapped a long question across lines, and the body was a separate surfaced panel with its own "Advice" chip. The bare N left counter also assumed the reader already knew the advisor is quota-limited per turn.

What changed

• The header is a single status line whose label is the model-supplied model_intent when present (rendered like the exec tool's intent), falling back to a status verb ("Consulting / Consulted / Failed to consult the advisor"). Header text color matches the other tool rows.
• The full question moved into the expandable card, above the guidance, where it stays readable and selectable without crowding the header. The card keeps the shared bordered, max-h-64 scroll treatment used by the subagent and exec tools.
• Dropped the live "N left" quota counter and the model-name chip from the header.
• The advisor tool now accepts an optional model_intent arg (same pattern as the exec tool's) so the agent can label the call.
• When the per-turn advisor budget is exhausted, the tool returns an explicit agent-facing message telling the parent to stop calling the advisor, instead of only a limit_reached type tag.

Every state (collapsed, expanded, running, streaming, error, limit, long question, with intent) is covered by the existing AdvisorTool Storybook stories.

Generated by Coder Agents (DanielleMaywood).