[Silthus]

Problem

With Tool calls in chat set to Zen, an agent that writes a reply and then asks a question shows the user only the question card. Zen hides every assistant message of a run that is still working, and a pending question keeps the run working. The reply stays hidden until the turn ends, so the user has to press Stop to read what the agent wrote before answering. Mobile shows the reply as a one-line "thinking" preview instead.

Steps: turn on Zen, run an agent turn that writes a reply and then calls its question tool (AskUserQuestion in Claude). Only the question card shows.

Note

🔵 Merge risk: low · two-way door: a client-side view rule, nothing persisted

• Zen users on web and desktop whose agent asks a question: the base hides the last reply before the question until the turn ends. HEAD shows it above the question card, while the question is open and after the answer, and the live preview shows only newer text. Covered by the unit tests and a live run on a thread the base wrote and one HEAD wrote.
• Unchanged: every other tool-call setting, and approvals in Zen. The new rule sits inside the zen branch; the existing hide-successful and hide-all assertions pass unchanged.
• Uncovered in a running app, unit tests only: mobile, and the secret and link request cards on web. A missed bug there leaves the reply hidden or shown twice, the base's cosmetic behavior, for Zen users who meet those cards.
• Switch-off: none: live on deploy, fix forward. A user can leave Zen in Settings.

Change

Zen keys on the feed, not on a pending-request flag: the last assistant reply before a question (user_input_request, pending or answered), a secret_request, or a link_request stays visible. Agents often run a command or a tool between the reply and the question, so only a user message ends the search. The live one-line preview stops at the request, so it never repeats the reply above it. Web (MessagesTimeline.logic.ts) and mobile (threadActivity.ts) get the same rule. Nothing changes outside zen.

Scope

Small fix of a bug in the new Zen setting (#30000, #33131), no issue. Zen's own description says "questions … always show", and the question cannot be answered without the reply it refers to.

Verification

• Web unit test (MessagesTimeline.logic.test.ts, zen case): the reply shows above a pending question, above an answered question while the run still works, above a secret request, and when a command runs between the reply and the question. Only the last reply shows, and the preview shows newer thinking only. On the base: expected [ 'user', …(2) ] to include 'final'.
• Mobile unit test (threadActivity.test.ts, zen case): [ 'user', 'final', 'question', 'thinking:' ] after the answer; on the base [ 'user', 'thinking:final' ]. Secret and link request cards keep the reply on both clients.
• vp test for apps/web/src/components/chat (479 passed) and mobile threadActivity plus features/threads (335 passed); web and mobile typecheck pass. All required CI checks pass on this head.
• Live run: a thread the base code wrote and a new thread, both loaded under this change; the reply shows with the question open, after the answer while the run works, and after the turn ends.
• Real data: I replayed the rule over the question, secret, and link requests stored in a local T3 state database. Every request with an earlier reply in its run keeps that reply. The others asked before writing anything. This replay is why the rule allows tool calls between the reply and the question.
• Manual: dev server, Zen on, the Replay (dev) question scenario with a reply chunk before the elicitation and a slow thinking stream after the answer (local fixture edit, not committed).

Before	Question open	After the answer, still working

I did not check the mobile app on a device. The mobile change is covered by the unit test only.

Decisions

Question	Answer	Who	Where
What keeps the reply visible?	The feed: the last reply before a question, secret, or link request, pending or answered. This replaces a pending-request flag, which hid the reply again after the answer.	🤖	review round 2
Do approvals reveal the narration before them?	No, approvals keep the base Zen behavior: their card carries the command.	🤖	review round 2
May work sit between the reply and the request?	Yes, any work. Only a user message ends the search, because real agents often run a command or tool before they ask.	🤖	real-data replay

Involvement: two review rounds (one reviewer, then a seven-lens panel across two model families) and a final read; UI walk and real-data replay captured.

Claude Opus 5.5 in Claude Code, run from T3 Code.

🤖 Generated with Claude Code

[coderabbitai]

No actionable comments were generated in the recent review.

Walkthrough

Mobile and web Zen visibility now retain the latest assistant reply before a user-input, secret, or link request. Live answer previews stop at these requests. Tests cover reply visibility, request ordering, intervening activity, and preview selection.

Changes: Zen request visibility

Layer / File(s)	Summary
Retain replies before requests apps/mobile/src/lib/threadActivity.ts, apps/web/src/components/chat/MessagesTimeline.logic.ts, apps/mobile/src/lib/threadActivity.test.ts, apps/web/src/components/chat/MessagesTimeline.logic.test.ts	Both implementations retain the latest assistant reply before a user-input, secret, or link request. A subsequent user message clears the candidate. Tests cover intervening activity, request ordering, and later replies.
Stop live previews at requests apps/mobile/src/lib/threadActivity.ts, apps/web/src/components/chat/MessagesTimeline.logic.ts, apps/mobile/src/lib/threadActivity.test.ts, apps/web/src/components/chat/MessagesTimeline.logic.test.ts	Live answer previews stop when they reach a user-input, secret, or link request. Tests verify that later reasoning can become the preview.

Estimated code review effort: 3 (Moderate) | ~20 minutes

Change: Bug fix

Suggested reviewers: flamboh, juliusmarminge

Merge Risk: Minimal · up to bec73

Zen now keeps the agent's last reply visible while a question, secret request, or link request is waiting on the user. The change is limited to client-side display logic and is covered by unit tests. No merge-blocking risk was found.

Pre-merge checks: 4

Check name	Status	Explanation
Title check		The title names the main change: Zen mode now shows the agent's reply while a question waits on the user. It uses the conventional commit format and matches the changeset, which keeps the last assista…
Description check		The description follows the template. It has Problem, Change, Scope, and Verification sections, and it includes the merge-risk note, the reproduction steps, the linked issue references, a list of test…
Linked Issues check		Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check		Check skipped because no linked issues were found for this pull request.

Full details: Title check

Explanation

The title names the main change: Zen mode now shows the agent's reply while a question waits on the user. It uses the conventional commit format and matches the changeset, which keeps the last assistant reply visible before question, secret, and link requests on web and mobile.

Full details: Description check

Explanation

The description follows the template. It has Problem, Change, Scope, and Verification sections, and it includes the merge-risk note, the reproduction steps, the linked issue references, a list of tests and their results, a live run, a replay over real data, and before/after screenshots. The Verification section also states what was not checked: the mobile app on a device and the secret and link cards in a running app. The Scope section cites issues #30000 and #33131 but does not link them as URLs. The PR does not state a model or harness in the required form, although the author does name Claude Opus 5.5 and Claude Code at the end.

Finishing Touches

Generate unit tests (beta)

• Create a new PR

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Recent review info

Run configuration

• Source mode: Auto
• Repository: pingdotgg/t3code
• Branch: main
• File: .coderabbit.config.ts
• Commit: a35fd81
• Layers: Repository TS config
• Review profile: CHILL
• Plan: Advanced
• Run ID: 9d695dd0-1f27-41c3-a538-83cad70313a7

Commits

Reviewing files that changed from the base of the PR and between b803f37 and bec7360.

Files selected for processing: 4

• apps/mobile/src/lib/threadActivity.test.ts
• apps/mobile/src/lib/threadActivity.ts
• apps/web/src/components/chat/MessagesTimeline.logic.test.ts
• apps/web/src/components/chat/MessagesTimeline.logic.ts

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 9 remain after this review.