Summary

Reworks how an assistant turn is presented in the conversation flow, in six atomic commits (commits 5–6 added in response to code review):

1. Icons — adds thinking, skill, and read-file icons and remaps the tool-family icon table so each trow-block leading icon matches its tool. The thinking glyph is resized to fill the same icon box as its neighbours.
2. Reasoning folds into the trow-block — reasoning now collapses into the same summary row as tool calls (reusing BasicTool) instead of rendering its own standalone thinking collapsible. A folded reasoning row no longer paints a duplicate icon, and the dead thinking-heading derivation (heading/clean/reasoningHeading/TextReveal) is removed.
3. Interrupted-turn layout fix — an interrupted turn no longer draws a duplicate "interrupted" divider at the top, and its footer toolbar stays left-aligned like every other turn (the "interrupted" notice still shows in the footer meta line).
4. Remove the now-dead reasoning-summaries setting — once reasoning always renders as a collapsible row, the General settings "show reasoning summaries" toggle became a no-op (renderable() ignores the flag). The toggle, its i18n copy, the persisted showReasoningSummaries field plus accessor, and the whole prop chain (message-timeline → SessionTurn → AssistantParts / AssistantMessageDisplay → renderable()) are removed.
5. Render reasoning in AssistantMessageDisplay (review P2) — groupParts() folds reasoning into the trow group, but this second turn renderer's trow branch kept only tool parts and never passed renderReasoning, so a reasoning-only or reasoning+tool message silently dropped its reasoning on that path. It now mirrors AssistantParts: the trow filter accepts reasoning and renders it via renderReasoning. New grouping.test.ts cases pin the shared folding (reasoning folds into a trow, groups with adjacent tools, empty reasoning filtered).
6. Single tool-icon source (review P3) — tool icons were defined in three places (toolFamilyIcon, toolInfoForInput, and hard-coded strings in each tool component) and had already drifted for read/edit/write/apply_patch/skill. A new toolIcon() in tool-info.ts is the single source of truth; toolFamilyIcon delegates to it, toolInfoForInput returns it, and all 16 tool components pass toolIcon("…") to BasicTool. Icon values are unchanged, so rendering is identical; a reducer test pins the collapsed (trow) and expanded (tool-info) icons in lock-step so they cannot diverge again.

Why

Reasoning used to render as its own collapsible, which showed two stacked icons in the collapse bar when folded and duplicated the tool-call fold pattern with a parallel component. Folding reasoning into the existing trow-block removes the duplicate icon and the second code path. Separately, an interrupted turn drew a redundant top divider and right-shifted its footer toolbar, breaking alignment with normal turns. Finally, the fold made the "show reasoning summaries" setting meaningless — reasoning is always a collapsible row now — so the dead setting and its prop chain are cleaned up rather than left as confusing dead UI. Code review then surfaced two follow-ups, fixed in commits 5–6: a second turn renderer (AssistantMessageDisplay) was never updated when reasoning began folding into the trow, dropping reasoning on that path; and the tool icon was defined in three drifting places, so it is consolidated into one toolIcon() source.

Related Issue

None — scope and rationale are stated in Summary. Confirmed distinct from #943, which concerns a reasoning-attached retry notice, not the fold/layout presentation.

Human Review Status

Pending

Review Focus

• reduceTrowBlock / toolFamilyIcon in session-turn-trow-block.tsx: reasoning is excluded from toolCount, and a reasoning-only block falls back to the thinking leading icon.
• reasoning.tsx now renders through BasicTool — confirm a folded reasoning row shows exactly one icon.
• session-turn.tsx interrupted-divider removal: confirm the footer meta still surfaces the "interrupted" notice after dropping the top divider and data-interrupted flag.
• The removed showReasoningSummaries chain: confirm renderable() already ignored the flag (so reasoning renders unconditionally) and that nothing else read it before deletion.
• AssistantMessageDisplay trow branch (commit 5): confirm it now accepts reasoning and passes renderReasoning, matching AssistantParts, so the second renderer no longer drops reasoning.
• toolIcon() in tool-info.ts (commit 6): confirm it is the only place an icon name is chosen per tool, that toolFamilyIcon / toolInfoForInput / all tool components route through it, and that the lock-step reducer test would catch any future divergence.

Risk Notes

Presentation-only changes; no data, persistence schema, or API surface touched. The removed setting (showReasoningSummaries) was already a no-op after commit 2 — any value still in a user's persisted store is simply ignored on load (the withFallback accessor is gone), so no migration is needed. Commit 6 is a pure refactor: every tool's icon value is unchanged (the 16 components and toolFamilyIcon now resolve the same icon through toolIcon()), so rendering is identical and snapshots need no regeneration.

Skipped conditional checklist item:

• Platform (macOS/Windows) — left unticked: no platform, packaging, updater, signing, path, shell, or permission surface was touched.

How To Verify

packages/ui typecheck (tsgo --noEmit): ok, no errors
packages/app typecheck (tsgo -b): ok, no errors (exit 0)
ui unit suite (bun test, 56 files): 707 pass / 0 fail
focused: trow-block reducer + grouping + typography-semantics = pass (toolFamilyIcon↔tool-info icon lock-step, reasoning folds into trow + groups with tools + empty filtered, reasoning-body line-height, renderable)
tool-icon: 16 tool components + toolFamilyIcon + toolInfoForInput all route through toolIcon() (grep: no hard-coded icon= in tools/)
no residual references to showReasoningSummaries / reasoningSummaries in app+ui source (grep clean)
snap session-trow (playwright): 1 passed — reasoning-only (1 trow-summary-icon), reasoning-with-tools, expanded reasoning body (non-mono); rendering unchanged by the setting removal

Screenshots or Recordings

Visual states are covered by the session-trow snapshot grid, regenerated in this change at docs/design/preview/screenshots/session-trow.png (via bun run snap session-trow), including the reasoning-only and reasoning-with-tools rows in collapsed and expanded form. The snapshot is a repo artifact and is not inlined here because the body cannot embed local files. The removed General setting is a one-row deletion in Settings → General with no dedicated snapshot.

Checklist

How to use this checklist:

• Tick a box by replacing [ ] with [x]. Do not edit, add, or remove items.
• The bot-applied label items can only be honestly ticked AFTER the PR is opened and the labeler / priority-triage bots have run — return to the PR description and tick them then.
• Most items are required. The few that are conditional are explicitly marked (conditional); for those, leave unticked if they truly do not apply and explain why in Risk Notes. All other items must be ticked before requesting human review.

• Type label — this PR carries exactly one of bug, enhancement, task, documentation. Type labels are author-added; the labeler bot does NOT assign them. Add the label in the GitHub UI, then tick this.
• Routing labels — this PR carries at least one of app, ui, platform, harness, ci. The labeler bot assigns these on PR open based on changed paths. Confirm the bot's choice (or override if wrong), then tick this.
• Priority label — this PR carries exactly one of P0, P1, P2, P3. The priority-triage bot suggests one on PR open. Confirm or override, then tick this.
• Human Review Status above is set to Pending, Approved by @<reviewer>, or Not required: <reason> (default is Pending; "not required" is restricted to bot-authored low-risk PRs).
• I linked the related issue, or stated in Summary why there is no issue.
• I described the review focus and any meaningful risks.
• I replaced the example block in How To Verify with the real verification steps and the key result for each.
• I did not introduce unrelated refactors, dependencies, generated files, or file changes beyond the stated scope.
• (conditional) I manually checked visible UI or copy changes when needed, with screenshots or recordings. Leave unticked only if no visible UI or copy changed.
• (conditional) I considered macOS and Windows impact for platform, packaging, updater, signing, paths, shell, or permissions changes. Leave unticked only if no platform/packaging surface was touched.
• (conditional) I called out docs, release notes, dependencies, permissions, credentials, deletion behavior, generated content, or local file changes when relevant. Leave unticked only if none of those surfaces was touched.
• I reviewed the final diff for unrelated changes and suspicious dependency changes.
• I am targeting dev, and my PR title and commit messages use Conventional Commits in English.