fix(chat): reuse one row for live activity

[maria-rcks]

Uses one stable live-activity row for the running turn. It starts as Thinking with Lucide’s brain icon. An actually in-progress tool replaces that row and renders its normal Running animation. If parallel tools interleave, the latest still-running tool keeps the live row; once none are running, the newest settled tool stays in the same row with its normal completed, failed, or declined presentation. The next assistant text moves that tool group into history and starts a fresh Thinking row.

Verified with 88 focused timeline tests, scoped lint, and the web typecheck. UI capture was blocked because the isolated tailnet preview rendered blank.

Created by gpt-5.6-sol through the Codex harness.

---

Note

Low Risk
Scoped to chat timeline row derivation and presentation; behavior is covered by focused logic and component tests with no auth or data-path changes.

Overview
The chat timeline now uses a single stable row (live-activity-row) for everything that represents “what the agent is doing right now” on an in-progress turn, instead of separate thinking-indicator-row and work-live:* rows.

Row derivation (deriveMessagesTimelineRows): active-turn tools still collapse into one work-live row, but it always uses LIVE_ACTIVITY_ROW_ID. The displayed entry prefers a still-running tool when one exists; otherwise it shows the latest visible tool with active: false after completion. A separate Thinking row is appended only when the turn is working and no activity row was added yet—so Thinking no longer stacks under a settled tool on the same turn.

UI: initial thinking uses the shared LiveActivityRow with a brain icon (replacing bot for thinking tone). Spacing treats thinking like other activity rows.

Tests lock in the new ids, the thinking→running tool→completed tool handoff on one row, and stable row reuse when assistant text grows.

^{Reviewed by Cursor Bugbot for commit eb18efa. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Reuse single live-activity-row for thinking and tool activity in MessagesTimeline

• Replaces the separate thinking-indicator-row and per-group work-live:{identity} ids with one shared LIVE_ACTIVITY_ROW_ID constant in MessagesTimeline.logic.ts
• deriveMessagesTimelineRows now emits a single activity row that shows initial thinking (kind: thinking) when no tool is active, or the latest tool activity (kind: work-live) when one exists; the thinking row is only appended when no activity row was created
• Renders a brain icon for the thinking state and aligns bottom padding for thinking rows with work rows in MessagesTimeline.tsx
• Risk: any out-of-tree code or tests that select rows by thinking-indicator-row or work-live:{identity} ids will no longer find them; in-tree readers in the logic and component tests are updated

^{Macroscope summarized eb18efa.}

[macroscopeapp]

Approvability

Verdict: Approved at eb18efa

Macroscope's review found this PR approvable — This is a contained chat-timeline fix that reuses one stable activity row across thinking and tool states, with focused logic and rendering tests covering the transition. It does not affect schemas, configuration defaults, deployment, or sensitive application areas.

^{You can add or adjust custom eligibility rules. Learn more.}

Dismissing prior approval to re-evaluate bfc1f1d

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 0c1bba0. Configure here.}