(CODUX-123) Refactor InvocationMessageBubble onto shared resolver

[numnx]

🧩 Task: Refactor InvocationMessageBubble onto shared resolver

Automated task PR opened by Code UX for T02.

📋 Summary

Sprint: Sprint 195 · Rich Widgets Chat Threads · Task: T02

Let normal Chat Threads render the same rich runtime widgets (reasoning, tool-call, and planning cards) that the invocation/"3D chat" transcript already renders, by centralizing widget resolution in a shared view-model and consuming it from the threads surface.

🤖 Model & Provider

Provider	Codex CLI
Model	gpt-5.5

⏱️ Timing

Started	2026-07-09 02:12:11 UTC
Finished	2026-07-09 02:16:03 UTC
Duration	3m 52s

📊 Token Usage

Metric	Value
Input tokens	70,424
Cached input tokens	752,640
Output tokens	7,185
Total tokens	830,249
Invocations	1
Tool calls	32
Active time	3m 20s

Included usage estimate (subscription/local login): $0.94

✅ QA Review

⏳ QA review runs after this PR is opened. Check the task card in the dashboard for the outcome once it lands.

📝 Full Task Prompt

## Objective
Refactor `dashboard/src/v2/components/chat/InvocationMessageBubble.tsx` to derive its reasoning/tool/planning widgets from the shared `resolveRichWidget` descriptor instead of duplicating the `metadata.kind` branching, keeping the invocation transcript's visual output identical.

## Scope
- dashboard/src/v2/components/chat/InvocationMessageBubble.tsx
- tests/dashboard/v2/chat-message-bubbles.test.tsx (invocation cases)
- Consumes: `resolveRichWidget` and `RichWidgetDescriptor` from dashboard/src/v2/lib/chat-widget-view-models.ts (T01).

## Implementation Requirements
1. Import `resolveRichWidget` and build the descriptor with `resolveRichWidget({ metadata: message.metadata, content: message.contentMarkdown, toolCallsJson: message.toolCallsJson })`.
2. Replace the current early returns for `kind === "reasoning"` and `kind === "tool_call" || "tool_result"` with switches on the descriptor: `reasoning` → the existing full-width `ReasoningWidget` layout; `tool` → the existing full-width `ToolCallWidget` layout, passing `toolName`/`status`/`args`/`output`/`tokens`/`callId` from the descriptor.
3. Replace the trailing planning `widgetData` slot to read from the shared descriptor's `planning` variant, keeping the same `PlanningRequestWidget` markup and container styles.
4. Preserve all remaining bubble behavior (avatar/role resolution, header chips, provider/model/status/error badges, `sanitizeInvocationOutputText` on the message body, the raw `toolCallsJson` fallback block when `!kind`).

## Constraints
- No visual or DOM-structure changes: the same wrapper classes, `pl-11` widths, and sr-only lines must remain.
- Do not alter `mergeInvocationToolMessages` usage in `ChatPage.tsx`.
- Keep the existing behavior for non-widget assistant/user/tool/system messages unchanged.

## Verification
- Run `pnpm test tests/dashboard/v2/chat-message-bubbles.test.tsx` — invocation reasoning, tool_call/tool_result, error-badge, and planning cases still pass.
- Run `pnpm run typecheck:dashboard` — clean.
- Success looks like: invocation transcript renders reasoning cards, merged tool cards, and planning widgets exactly as before, now sourced from `resolveRichWidget`.

🌿 Branch Info

Base: feature/CODUX-195-rich-widgets-chat-threads
Head: task/feature-codux-195-rich-widgets-chat-threads-t02-codex-mrcvftjb

---

🤖 Generated by Code UX