fix(web): keep codex restart responses continuous

[maria-rcks]

Note

Written by gpt-5.6-sol on behalf of Maria

Promptless Codex restart continuations currently split one visible response into two timers and fold the pre-restart work while the replacement turn is still running. This groups consecutive provider turns after the same user message into one active visual response while keeping live tool state scoped to the actual running turn. Verified with the focused timeline suite, web typecheck and lint, plus the same simulated restart fixture in dark and light desktop layouts.

Before:

After:

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

---

Note

Low Risk
Scoped to chat timeline presentation logic in MessagesTimeline.logic.ts with unit test coverage; no auth, persistence, or API changes.

Overview
Promptless Codex restarts no longer split one user-facing reply into separate working timers or fold pre-restart activity while the replacement turn is still running.

Timeline derivation now builds an active visual response set: all provider turn IDs after the latest user message (plus the unsettled turn) stay unfolded and treated as one in-progress assistant response until work settles. Turn folding skips every ID in that set instead of only the single unsettled turn. The working row’s createdAt anchors to the user message when multiple turns share that visual response, and assistant meta/copy/streaming flags follow the broader set so pre-restart commentary stays “live” until the continuation finishes. Live tool activity still keys off the actual unsettled turn so only the post-restart in-progress tool drives the shared activity row.

A regression test covers the restart fixture (no turn-fold, single working indicator, stale pre-restart tools not live, new turn’s tool active).

^{Reviewed by Cursor Bugbot for commit 4c47567. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add test for promptless codex restart rendering in MessagesTimeline

Adds a test case to deriveMessagesTimelineRows in MessagesTimeline.logic.test.ts that verifies a promptless provider restart stays as one active visual response. The scenario covers a user message, work and commentary from one provider turn, then replacement work from a second turn without a new user message. Assertions confirm no turn-fold row, one working indicator before the earlier work, the user-message timestamp, provisional commentary, only the replacement work active, and no thinking row.

^{Macroscope summarized 4c47567.}

[macroscopeapp]

Approvability

Verdict: Approved at 4c47567

Macroscope's review found this PR approvable — This is a focused chat-timeline bug fix that keeps promptless provider restarts visually continuous, with targeted regression coverage. Its runtime impact is confined to response grouping and activity presentation, with no API, persistence, deployment, security, or default-setting changes.

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