Ask AI response rendering and anchored plan Q&A

Provenance stamp: This issue was triaged on June 13, 2026 as part of the older-than-two-month GitHub issue triage exercise.

Source issues: #406, #456, #514

## Summary

Ask AI should be useful inside plan and review workflows without making the output hard to read or unreliable. These issues cover three parts of that experience: asking questions about a plan, separating reasoning from final answers, and ensuring provider responses render correctly.

## What the source issues requested

- #406 reported that in-browser AI responses were difficult to read because model thinking and the final response looked the same.
- #456 requested the ability to ask questions about a plan without immediately changing the plan or restarting the feedback loop. Later comments also asked for anchored plan discussion and clearer action after an answer.
- #514 reported that OpenCode AI responses created empty bubbles, suggesting provider events were received but not parsed or rendered correctly.

## Triage determination

Ask AI has evolved since these issues were opened, and some implementation work has shipped. This replacement issue should track the remaining user-facing contract: questions can be anchored to plan or review content, responses render across providers, and the UI clearly separates status or reasoning from the final answer.

## Acceptance criteria

- Users can ask questions about a plan or selected block without sending a denial or approval.
- Anchored Ask AI conversations make it clear what text or block the answer refers to.
- The UI distinguishes final answer text from reasoning, thinking, status, or tool progress.
- OpenCode provider responses render visible text consistently instead of empty bubbles.
- If a previously reported provider rendering bug is already fixed, add a verification note and close the corresponding old source issue.
- The old source issues can be closed once their Ask AI requirements are either implemented or explicitly represented here.