fix(web,mobile): replies to mid-run messages stay visible when a run folds#28171

Problem

When the user sends a message while a run is working, it becomes a steer inside that run. Once the run settles, the turn fold keeps only the run's last assistant message visible, so the agent's reply to the mid-run question is folded away and the user sees no answer. Reproduced on main (79f1fadfc6) in the web app with a real Claude run (one steer) and a real Codex run (two steers): after settling, each steer is followed only by a "Continued for Ns" row, and the replies ("I'd use 'Multiplies x by N'…", "No—each function is exported under its own name", "No—this task only removes the specified comments…") are hidden inside it.

Change

In a settled, folded run, the first assistant message after each steer stays visible, the same way the run's terminal message already does. Back-to-back steers count once. Everything else still folds: work, other progress notes, the "Worked for" and "Continued for" rows and their durations. The rule is applied in both fold implementations: web/desktop deriveTurnFolds (MessagesTimeline.logic.ts) and mobile deriveThreadFeedRunFolds (threadActivity.ts). Each tracks the runs awaiting a steer reply while it walks the timeline and leaves that message out of hiddenEntryIds.

Scope and approval

Fixes pingdotgg/t3code#28161

Triaged via the automated issue flow.

Verification

• MessagesTimeline.logic.test.ts "keeps the agent's reply to each steer visible after the settled turn folds" and threadActivity.test.ts "keeps the agent's reply to each steer visible after the settled run folds": prompt → progress note → work → steer → reply → progress note → work → second steer → reply → work → final. They fail on main because the replies are hidden. They pass here, and the progress notes and work stay folded. A mutation that keeps every message after a steer visible makes both fail.

• The existing "shows a created chat after a steer only when that steer's segment is open" test got one extra progress message, so its steer segment still has something to fold now that the reply stays visible.

• Full files: web 92/92, mobile 44/44. vp run check:changed passes (format, lint, knip, typecheck, tests, React compiler).

• Web app on the same database, before (79f1fadfc6) and after (bfb4e88bf2), 1280×800, light theme:

  Thread	Before	After
  Claude, one steer	Reply hidden under "Continued for 1m 12s"; only the final summary visible	Reply visible below "Continued for 1m 12s", then the final summary
  Codex, two steers	Both replies hidden under "Continued for 22s" and "Continued for 39s"	Both replies visible below their rows; progress notes and commands still folded
  Codex, "Continued for 22s" expanded (after only)	—	Reasoning, reply, progress notes and commands in chronological order

  The screenshots of these states are not attached yet.

• Not checked: the native mobile client in the app (covered by the mobile unit test above), the recordings of expand/collapse and of a steer settling, and providers other than Claude and Codex.

Model and harness: Claude Opus 5.5 (lead, tests, implementation, review) and GPT-6.1 (review) in Claude Code and Codex, through T3 Code.

🤖 Generated with Claude Code