feat(tui): /details Tab toggles apply immediately

Drop the pending state and confirm step: Tab cycles the highlighted
entry and applies at once, so the transcript behind the dialog is the
live preview; Enter/Esc/Ctrl+C just close.

packages/ui/tui/README.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write packages/ui/tui/README.md
-README.md: 7230e75535ce5710e12040e026bc77a1782ed6dd
-README.zh.md: 3079b797b670de64cfdc1faa9d002ee04a6e28dd
+README.md: b6d0ab8e143f14ea2826054f72cf5a778ffa9bbb
+README.zh.md: addce9f3b55c081208b0849bc00522fa0a22e295

packages/ui/tui/README.md

@@ -22,7 +22,7 @@ Typing `@` at a token boundary searches files and directories under the session
 
 When optional `ctx.sessionReferences` is mounted, the same `@` menu also offers metadata-only session candidates, inserts `@[label](dsh-session:<payload>)`, and prepares the selected snapshots before dispatch. Session references remain structured because the model has no filesystem-like tool for retrieving session snapshots later. Preparation disables duplicate submission and restores the editor input on failure. The TUI chooses `agent.steer()` or `agent.followup()` from the status after that asynchronous preparation, so idle follow-ups still dispatch `agent/prompt-submit` while in-turn steering joins at a checkpoint without that hook.
 
-While the agent is running, ordinary editor submissions call `agent.steer()`; otherwise they call `agent.followup()`. A slash at the start of the submitted line enters `ctx.commands` instead: known commands execute directly, unknown commands produce a warning, and neither path automatically reaches the model. A command producer may explicitly schedule agent work; [`dsh-plan-mode`](../../plan/plan-mode/README.md#model-and-human-surfaces) uses that contract for `/plan [message]`. The TUI registers `/help`, `/model`, `/clear`, `/details`, `/palette`, `/reload`, `/resume`, `/status`, and `/exit` as agent-scoped definitions; every other effective command joins autocomplete and `/help` dynamically, as do `/skill:` completions. A status line above the editor reports the turn phase the TUI derives from session events — waiting for the first token, thinking, responding, or executing tools — with the elapsed time in that phase and the running step total, refreshed each second, and ends with the `Enter sends steering, Esc cancels` hint; while steering messages wait to reach the model it inserts a `N queued ·` badge before the hint that clears as each drains. Ctrl+C or Escape cancels a running turn. Tool and injected-context cards collapse long bodies into a configurable head/tail preview; Ctrl+O cycles tool cards through collapsed preview, full output, and hidden — the hidden phase drops tool cards from the transcript entirely while context cards stay at their preview, since injected instructions are not tool traffic. The hidden phase also folds each turn's assistant steps into one message: the first step with visible text or reasoning keeps the turn's single `Assistant` header, later steps render as headerless continuations, and a step without a visible body renders nothing; leaving the hidden phase restores the per-step headers. An injected-context card renders its message as prose with the producer's outer reminder frame stripped, so neither the fold nor the frame stripping depends on the payload's syntax. Ctrl+R toggles reasoning, Ctrl+L redraws, and Ctrl+D exits while idle. `/details` names the same state those two shortcuts cycle: bare it opens a centered keyboard selector with one entry per dimension — `Tool cards` and `Reasoning` — seeded with the current values, where Tab cycles the highlighted entry's pending value (rendered as `current → pending`), Enter applies every changed dimension and closes, and Esc or Ctrl+C cancels without changing anything; `/details collapsed|expanded|hidden` jumps tool cards to that phase directly, and `/details reasoning [on|off]` sets — or bare `reasoning` toggles — reasoning-block display; arguments combine in one invocation, an unknown argument fails with the usage line, and a combined invocation applies reasoning first so its transcript rebuild never drops the card notice.
+While the agent is running, ordinary editor submissions call `agent.steer()`; otherwise they call `agent.followup()`. A slash at the start of the submitted line enters `ctx.commands` instead: known commands execute directly, unknown commands produce a warning, and neither path automatically reaches the model. A command producer may explicitly schedule agent work; [`dsh-plan-mode`](../../plan/plan-mode/README.md#model-and-human-surfaces) uses that contract for `/plan [message]`. The TUI registers `/help`, `/model`, `/clear`, `/details`, `/palette`, `/reload`, `/resume`, `/status`, and `/exit` as agent-scoped definitions; every other effective command joins autocomplete and `/help` dynamically, as do `/skill:` completions. A status line above the editor reports the turn phase the TUI derives from session events — waiting for the first token, thinking, responding, or executing tools — with the elapsed time in that phase and the running step total, refreshed each second, and ends with the `Enter sends steering, Esc cancels` hint; while steering messages wait to reach the model it inserts a `N queued ·` badge before the hint that clears as each drains. Ctrl+C or Escape cancels a running turn. Tool and injected-context cards collapse long bodies into a configurable head/tail preview; Ctrl+O cycles tool cards through collapsed preview, full output, and hidden — the hidden phase drops tool cards from the transcript entirely while context cards stay at their preview, since injected instructions are not tool traffic. The hidden phase also folds each turn's assistant steps into one message: the first step with visible text or reasoning keeps the turn's single `Assistant` header, later steps render as headerless continuations, and a step without a visible body renders nothing; leaving the hidden phase restores the per-step headers. An injected-context card renders its message as prose with the producer's outer reminder frame stripped, so neither the fold nor the frame stripping depends on the payload's syntax. Ctrl+R toggles reasoning, Ctrl+L redraws, and Ctrl+D exits while idle. `/details` names the same state those two shortcuts cycle: bare it opens a centered keyboard toggle with one entry per dimension — `Tool cards` and `Reasoning` — showing the live values, where Tab cycles the highlighted entry and applies the change immediately (the transcript behind the dialog is the preview), and Enter, Esc, or Ctrl+C closes; `/details collapsed|expanded|hidden` jumps tool cards to that phase directly, and `/details reasoning [on|off]` sets — or bare `reasoning` toggles — reasoning-block display; arguments combine in one invocation, an unknown argument fails with the usage line, and a combined invocation applies reasoning first so its transcript rebuild never drops the card notice.
 
 `/model` opens the advisory `ctx.llm` catalog as a keyboard selector: a filter box above the list narrows rows by a case-insensitive substring over each row's `provider/model` label, model name, and description, keeping the highlighted row selected when it survives the filter; Up/Down moves, Shift+Tab cycles the focused model's adapter-advertised reasoning efforts in display order, Enter selects the model and effort, and Escape clears a non-empty filter before a second Escape closes it. When an adapter does not advertise a default effort, the cycle also includes `Default`, which clears an explicit selection and preserves the provider default; models without selectable effort metadata ignore Shift+Tab. The selector renders the exact advertised effort list—including `off` when present—and does not synthesize, clamp, or transfer an effort between models. `/model <model>` still selects an unambiguous model id directly, while `/model <provider>/<model>` selects an exact target and uses its adapter default when one exists. The configured target or latest logged request header initializes the selector, and an unlisted current model remains visible because catalogs are advisory. Selection is local to this TUI session. Prompt assembly snapshots the target for one step, replaces `{{provider}}` and `{{model}}`, and applies the same provider/model/reasoning-effort target through `agent/request`; a switch during assembly therefore starts with a later step. The request header durably records targets that reach the model, while an unused selection remains process-local.
 

packages/ui/tui/README.zh.md

@@ -22,7 +22,7 @@ TUI 从活跃会话表层重建已恢复历史，渲染 Markdown 响应与 reaso
 
 挂载可选的 `ctx.sessionReferences` 后，同一个 `@` 菜单还会提供仅含元数据的会话候选项，插入 `@[label](dsh-session:<payload>)`，并在分派前准备所选快照。会话引用保持结构化，因为模型没有类似文件系统的工具可在稍后检索会话快照。准备期间会禁止重复提交，并在失败时恢复编辑器输入。TUI 会在异步准备后根据状态选择 `agent.steer()` 或 `agent.followup()`，因此空闲 followup 仍会分派 `agent/prompt-submit`，而轮次中的 steering 会在检查点加入且不触发该 hook。
 
-Agent 运行时，普通编辑器提交会调用 `agent.steer()`；其他时候调用 `agent.followup()`。提交行以斜杠开头时会改为进入 `ctx.commands`：已知命令直接执行，未知命令产生警告，两条路径都不会自动到达模型。命令生产方可以显式调度 agent 工作；[`dsh-plan-mode`](../../plan/plan-mode/README.md#model-and-human-surfaces) 使用该契约实现 `/plan [message]`。TUI 将 `/help`、`/model`、`/clear`、`/details`、`/palette`、`/reload`、`/resume`、`/status` 和 `/exit` 注册为 agent 作用域定义；其他所有有效命令都会动态加入自动补全与 `/help`，`/skill:` 补全也相同。编辑器上方的状态行会报告 TUI 从会话事件派生的轮次阶段，包括等待首个 token、思考、响应或执行工具；它显示该阶段已经过时间和运行中的步骤总数，每秒刷新，并以 `Enter sends steering, Esc cancels` 提示结尾。Steering 消息等待到达模型期间，会在提示前插入 `N queued ·` 徽标，每条消息排空后随即清除。Ctrl+C 或 Escape 会取消运行中的轮次。工具卡片与注入上下文卡片都把长主体折叠为可配置的头尾预览；Ctrl+O 让工具卡片在折叠预览、完整输出、隐藏三种状态间循环——隐藏阶段把工具卡片从 transcript 中完全去掉，而上下文卡片保持预览，因为注入的指令不属于工具流量。隐藏阶段还会把每个轮次的 assistant 步骤折叠为一条消息：第一个有可见文本或 reasoning 的步骤保留该轮次唯一的 `Assistant` 标题，之后的步骤渲染为无标题的续段，没有可见正文的步骤则不渲染任何内容；离开隐藏阶段会恢复每步各自的标题。注入上下文卡片把消息渲染为文本，并去掉生产方的外层提醒外框，因此折叠与去外框都不依赖载荷的语法。Ctrl+R 切换 reasoning，Ctrl+L 重绘，Ctrl+D 在空闲时退出。`/details` 命名的正是这两个快捷键循环的同一份状态：不带参数时打开一个居中的键盘选择器，每个维度一个条目——`Tool cards` 与 `Reasoning`——以当前值为初始，Tab 循环高亮条目的待定值（渲染为 `current → pending`），Enter 应用所有已改变的维度并关闭，Esc 或 Ctrl+C 取消且不改变任何东西；`/details collapsed|expanded|hidden` 让工具卡片直接跳到该阶段，`/details reasoning [on|off]` 设置——或裸 `reasoning` 切换——reasoning 块显示；参数可在一次调用中组合，未知参数会以用法行报错，组合调用先应用 reasoning，使其 transcript 重建不会丢掉卡片通知。
+Agent 运行时，普通编辑器提交会调用 `agent.steer()`；其他时候调用 `agent.followup()`。提交行以斜杠开头时会改为进入 `ctx.commands`：已知命令直接执行，未知命令产生警告，两条路径都不会自动到达模型。命令生产方可以显式调度 agent 工作；[`dsh-plan-mode`](../../plan/plan-mode/README.md#model-and-human-surfaces) 使用该契约实现 `/plan [message]`。TUI 将 `/help`、`/model`、`/clear`、`/details`、`/palette`、`/reload`、`/resume`、`/status` 和 `/exit` 注册为 agent 作用域定义；其他所有有效命令都会动态加入自动补全与 `/help`，`/skill:` 补全也相同。编辑器上方的状态行会报告 TUI 从会话事件派生的轮次阶段，包括等待首个 token、思考、响应或执行工具；它显示该阶段已经过时间和运行中的步骤总数，每秒刷新，并以 `Enter sends steering, Esc cancels` 提示结尾。Steering 消息等待到达模型期间，会在提示前插入 `N queued ·` 徽标，每条消息排空后随即清除。Ctrl+C 或 Escape 会取消运行中的轮次。工具卡片与注入上下文卡片都把长主体折叠为可配置的头尾预览；Ctrl+O 让工具卡片在折叠预览、完整输出、隐藏三种状态间循环——隐藏阶段把工具卡片从 transcript 中完全去掉，而上下文卡片保持预览，因为注入的指令不属于工具流量。隐藏阶段还会把每个轮次的 assistant 步骤折叠为一条消息：第一个有可见文本或 reasoning 的步骤保留该轮次唯一的 `Assistant` 标题，之后的步骤渲染为无标题的续段，没有可见正文的步骤则不渲染任何内容；离开隐藏阶段会恢复每步各自的标题。注入上下文卡片把消息渲染为文本，并去掉生产方的外层提醒外框，因此折叠与去外框都不依赖载荷的语法。Ctrl+R 切换 reasoning，Ctrl+L 重绘，Ctrl+D 在空闲时退出。`/details` 命名的正是这两个快捷键循环的同一份状态：不带参数时打开一个居中的键盘开关，每个维度一个条目——`Tool cards` 与 `Reasoning`——显示实时值，Tab 循环高亮条目并立即应用变更（对话框背后的 transcript 即是预览），Enter、Esc 或 Ctrl+C 关闭；`/details collapsed|expanded|hidden` 让工具卡片直接跳到该阶段，`/details reasoning [on|off]` 设置——或裸 `reasoning` 切换——reasoning 块显示；参数可在一次调用中组合，未知参数会以用法行报错，组合调用先应用 reasoning，使其 transcript 重建不会丢掉卡片通知。
 
 `/model` 将建议性的 `ctx.llm` catalog 打开为键盘选择器：列表上方设有一个过滤框，按对每行 `provider/model` 标签、模型名称和描述的大小写不敏感子串匹配来缩小行集，并在高亮行仍通过过滤时保持其选中状态；Up/Down 移动，Shift+Tab 按显示顺序循环切换适配器为焦点模型公布的推理强度，Enter 选择模型和推理强度，Escape 会先清除非空过滤内容，再次按下才关闭选择器。适配器未公布默认推理强度时，循环还会包含 `Default`，该项会清除显式选择并保留提供方默认行为；没有可选推理强度元数据的模型会忽略 Shift+Tab。选择器会原样呈现公布的推理强度列表（包括存在时的 `off`），不会合成、自动调整或在模型之间转移推理强度。`/model <model>` 仍可直接选择无歧义的模型 id，`/model <provider>/<model>` 则选择精确目标，并在存在时使用其适配器默认值。已配置目标或最新记录的请求 header 会初始化选择器；由于 catalog 仅提供建议，未列出的当前模型仍会显示。选择仅对本 TUI 会话有效。提示词组装会为一个步骤建立目标快照，替换 `{{provider}}` 和 `{{model}}`，并通过 `agent/request` 应用同一个提供方／模型／推理强度目标；因此组装期间的切换会从后续步骤开始生效。请求 header 会持久记录真正到达模型的目标，未使用的选择则只存在于进程本地。
 

packages/ui/tui/src/components/dialogs.ts

@@ -433,7 +433,7 @@ export class ModelDialog implements Component {
   }
 }
 
-/** Both transcript-detail dimensions the details selector applies on confirm. */
+/** Both transcript-detail dimensions, applied immediately on each Tab. */
 export interface DetailsSelection {
   readonly visibility: ToolCardVisibility
   readonly showReasoning: boolean
@@ -442,62 +442,47 @@ export interface DetailsSelection {
 const TOOL_CARD_PHASES: readonly ToolCardVisibility[] = ['collapsed', 'expanded', 'hidden']
 
 /**
- * Keyboard selector over the two transcript-detail entries — tool-card
- * visibility and reasoning display. Tab cycles the highlighted entry's pending
- * value, Enter applies both pending values and closes, Esc or Ctrl+C closes
- * without changing anything. A pending value renders as `current → pending`.
+ * Keyboard toggle over the two transcript-detail entries — tool-card
+ * visibility and reasoning display. Tab cycles the highlighted entry's value
+ * and applies it immediately, so the transcript behind the dialog is the live
+ * preview; Enter, Esc, or Ctrl+C closes.
  */
 export class DetailsDialog implements Component {
   private readonly list: SelectList
   private readonly toolsItem: SelectItem
   private readonly reasoningItem: SelectItem
-  private pendingVisibility: ToolCardVisibility
-  private pendingReasoning: boolean
 
   constructor(
-    private readonly visibility: ToolCardVisibility,
-    private readonly showReasoning: boolean,
+    private visibility: ToolCardVisibility,
+    private showReasoning: boolean,
     private readonly palette: Palette,
-    done: (selection: DetailsSelection) => void,
-    private readonly cancel: () => void,
+    private readonly apply: (selection: DetailsSelection) => void,
+    private readonly close: () => void,
   ) {
-    this.pendingVisibility = visibility
-    this.pendingReasoning = showReasoning
-    this.toolsItem = { value: 'tools', label: 'Tool cards', description: this.describeTools() }
-    this.reasoningItem = { value: 'reasoning', label: 'Reasoning', description: this.describeReasoning() }
+    this.toolsItem = { value: 'tools', label: 'Tool cards', description: visibility }
+    this.reasoningItem = { value: 'reasoning', label: 'Reasoning', description: this.reasoningLabel() }
     this.list = new SelectList([this.toolsItem, this.reasoningItem], 2, dialogSelectTheme(palette))
-    this.list.onSelect = () => {
-      done({ visibility: this.pendingVisibility, showReasoning: this.pendingReasoning })
-    }
+    this.list.onSelect = close
   }
 
-  /** `current → pending` when Tab moved the value, otherwise the current value. */
-  private static pendingLabel(current: string, pending: string): string {
-    return pending === current ? current : `${current} → ${pending}`
+  private reasoningLabel(): string {
+    return this.showReasoning ? 'shown' : 'hidden'
   }
 
-  private describeTools(): string {
-    return DetailsDialog.pendingLabel(this.visibility, this.pendingVisibility)
-  }
-
-  private describeReasoning(): string {
-    const label = (show: boolean): string => show ? 'shown' : 'hidden'
-    return DetailsDialog.pendingLabel(label(this.showReasoning), label(this.pendingReasoning))
-  }
-
-  /** Cycle the highlighted entry's pending value one step. */
-  private cyclePending(): void {
+  /** Cycle the highlighted entry one step and apply the new state. */
+  private cycle(): void {
     const selected = this.list.getSelectedItem()
     /* v8 ignore next -- the two-entry list always has a selection. */
     if (selected === null) return
     if (selected.value === 'tools') {
-      const index = TOOL_CARD_PHASES.indexOf(this.pendingVisibility)
-      this.pendingVisibility = TOOL_CARD_PHASES[(index + 1) % TOOL_CARD_PHASES.length] as ToolCardVisibility
-      this.toolsItem.description = this.describeTools()
+      const index = TOOL_CARD_PHASES.indexOf(this.visibility)
+      this.visibility = TOOL_CARD_PHASES[(index + 1) % TOOL_CARD_PHASES.length] as ToolCardVisibility
+      this.toolsItem.description = this.visibility
     } else {
-      this.pendingReasoning = !this.pendingReasoning
-      this.reasoningItem.description = this.describeReasoning()
+      this.showReasoning = !this.showReasoning
+      this.reasoningItem.description = this.reasoningLabel()
     }
+    this.apply({ visibility: this.visibility, showReasoning: this.showReasoning })
   }
 
   invalidate(): void {
@@ -505,8 +490,8 @@ export class DetailsDialog implements Component {
   }
 
   handleInput(data: string): void {
-    if (matchesKey(data, Key.escape) || matchesKey(data, Key.ctrl('c'))) this.cancel()
-    else if (matchesKey(data, Key.tab)) this.cyclePending()
+    if (matchesKey(data, Key.escape) || matchesKey(data, Key.ctrl('c'))) this.close()
+    else if (matchesKey(data, Key.tab)) this.cycle()
     else this.list.handleInput(data)
     this.invalidate()
   }
@@ -516,7 +501,7 @@ export class DetailsDialog implements Component {
     return renderDialog('Transcript details', [
       ...this.list.render(innerWidth),
       '',
-      this.palette.dim('↑/↓ move • Tab cycle • Enter apply • Esc cancel'),
+      this.palette.dim('↑/↓ move • Tab toggle • Enter/Esc close'),
     ], width, this.palette)
   }
 }

packages/ui/tui/src/index.ts

@@ -1049,9 +1049,8 @@ export function createTuiChat(
         toolsVisibility,
         showReasoning,
         palette,
+        // Each Tab applies immediately; one dimension changes per call.
         (selection: DetailsSelection) => {
-          void session.close()
-          // Reasoning first: its transcript rebuild would drop the card notice.
           if (selection.showReasoning !== showReasoning) setReasoning(selection.showReasoning)
           if (selection.visibility !== toolsVisibility) setToolsVisibility(selection.visibility)
         },

packages/ui/tui/tests/snapshots/details-selector.expected.txt

@@ -20,41 +20,53 @@ buffer
 8| "You                                                                                                 "
   style 0-2 fg=bright-magenta bold underline
 9| "Inspect the renderer.                                                                               "
-10| "Model wait 0.0s · Completed 2026-07-30 18:00:00                                                     "
-  style 0-46 dim
-11| <blank>
-12| "Reasoning blocks hidden.                                                                            "
+10| <blank>
+11| "Assistant                                                                                           "
+  style 0-8 fg=bright-magenta bold underline
+12| <blank>
+13| "● Tool / bash / Run the coverage gate"
+  style 0-36 fg=green
+14| "$ pnpm run test:coverage                                                                            "
   style 0-23 dim
-13| <blank>
-14| "Tool cards hidden.                                                                                  "
+15| "/workspace/project                                                                                  "
   style 0-17 dim
-15| <blank>
-16| "/workspace/project (tui-staging)  deepseek-v4-flash  ↑0 ↓0  0% context"
-  style 0-17 fg=bright-magenta bold
-  style 18-31 dim
-  style 34-50 dim
-  style 53-57 dim
-  style 60-69 dim
-17| " dsh >        ╭ Transcript details ──────────────────────────────────────────────────╮              "
-  style 1-3 fg=bright-magenta bold
-  style 5-6 dim
-  style 7-7 inverse
+16| "… +4 lines (Ctrl+O to expand)                                                                       "
+  style 0-28 dim
+17| "[exit 0]      ╭ Transcript details ──────────────────────────────────────────────────╮              "
+  style 0-7 dim
   style 14-85 fg=bright-magenta
-18| "              │ → Tool cards                      hidden → collapsed                 │              "
+18| "Model wait 0.0│ → Tool cards                      collapsed                          │              "
+  style 0-13 dim
   style 14-14 fg=bright-magenta
-  style 16-67 fg=bright-magenta inverse
+  style 16-58 fg=bright-magenta inverse
   style 85-85 fg=bright-magenta
 19| "              │   Reasoning                       hidden                             │              "
   style 14-14 fg=bright-magenta
   style 27-55 dim
   style 85-85 fg=bright-magenta
-20| "              │                                                                      │              "
+20| "Reasoning bloc│                                                                      │              "
+  style 0-13 dim
   style 14-14 fg=bright-magenta
   style 85-85 fg=bright-magenta
-21| "              │ ↑/↓ move • Tab cycle • Enter apply • Esc cancel                      │              "
+21| "              │ ↑/↓ move • Tab toggle • Enter/Esc close                              │              "
   style 14-14 fg=bright-magenta
-  style 16-62 dim
+  style 16-54 dim
   style 85-85 fg=bright-magenta
-22| "              ╰──────────────────────────────────────────────────────────────────────╯              "
+22| "Tool cards hid╰──────────────────────────────────────────────────────────────────────╯              "
+  style 0-13 dim
   style 14-85 fg=bright-magenta
-23-39| <blank>
+23| <blank>
+24| "Tool and context cards collapsed.                                                                   "
+  style 0-32 dim
+25| <blank>
+26| "/workspace/project (tui-staging)  deepseek-v4-flash  ↑0 ↓0  0% context"
+  style 0-17 fg=bright-magenta bold
+  style 18-31 dim
+  style 34-50 dim
+  style 53-57 dim
+  style 60-69 dim
+27| " dsh >                                                                                              "
+  style 1-3 fg=bright-magenta bold
+  style 5-6 dim
+  style 7-7 inverse
+28-39| <blank>

packages/ui/tui/tests/tui.snapshot.ts

@@ -663,9 +663,10 @@ describe('TUI terminal-state snapshots', () => {
       harness.terminal.send('\r')
     })
     await checkpoint('details-command', harness.terminal, { includeScrollback: true })
-    // Bare /details opens the two-entry selector seeded with the current
-    // hidden/reasoning-off state; one Tab renders the tool-card entry's
-    // pending cycle as `hidden → collapsed`.
+    // Bare /details opens the two-entry toggle seeded with the current
+    // hidden/reasoning-off state; one Tab immediately cycles tool cards
+    // hidden -> collapsed, so the frame pins the applied notice, the restored
+    // tool card behind the dialog, and the updated entry value together.
     await renderAfter(harness, () => {
       harness.terminal.send('/details')
       harness.terminal.send('\r')

packages/ui/tui/tests/tui.spec.ts

@@ -2544,7 +2544,7 @@ describe('pi-tui chat lifecycle and transcript', () => {
     await dispose(result)
   })
 
-  it('bare /details opens the transcript-details selector and applies the confirmed state', async () => {
+  it('bare /details opens the transcript-details toggle and Tab applies immediately', async () => {
     const result = await setup()
     const open = async (): Promise<number> => {
       const from = result.terminal.output.length
@@ -2563,50 +2563,40 @@ describe('pi-tui chat lifecycle and transcript', () => {
     await result.ctx.commands.execute(result.agent, '/details', new AbortController().signal)
     await tick()
 
-    // Esc cancels without touching the state.
-    const cancelOutput = result.terminal.output.length
-    result.terminal.send('\x1b')
-    await tick()
-    expect(result.terminal.output.slice(cancelOutput)).not.toContain('Tool and context cards')
-
-    // Tab cycles the highlighted entry's pending value; Enter applies it.
-    const cycled = await open()
+    // Each Tab applies one step immediately while the dialog stays open:
+    // collapsed -> expanded -> hidden -> collapsed (wraparound).
     result.terminal.send('\t')
     await tick()
-    expect(result.terminal.output.slice(cycled)).toContain('collapsed → expanded')
-    result.terminal.send('\r')
-    await tick()
     expect(result.terminal.output).toContain('Tool and context cards expanded.')
-
-    // Both entries apply in one confirm: cycle tool cards through the
-    // wraparound back to collapsed and toggle reasoning off.
-    const reopened = await open()
-    result.terminal.send('\t')
     result.terminal.send('\t')
     await tick()
-    expect(result.terminal.output.slice(reopened)).toContain('expanded → collapsed')
+    expect(result.terminal.output).toContain('Tool cards hidden.')
+    result.terminal.send('\t')
+    await tick()
+    expect(result.terminal.output).toContain('Tool and context cards collapsed.')
+
+    // The reasoning entry toggles the same way.
     result.terminal.send('\x1b[B')
     result.terminal.send('\t')
     await tick()
-    expect(result.terminal.output.slice(reopened)).toContain('shown → hidden')
-    result.terminal.send('\r')
-    await tick()
     expect(result.terminal.output).toContain('Reasoning blocks hidden.')
-    expect(result.terminal.output).toContain('Tool and context cards collapsed.')
 
-    // Enter with no pending change closes without a notice.
-    const unchanged = await open()
+    // Enter closes without further changes.
+    const entered = result.terminal.output.length
     result.terminal.send('\r')
     await tick()
-    expect(result.terminal.output.slice(unchanged)).not.toContain('Tool and context cards')
-    expect(result.terminal.output.slice(unchanged)).not.toContain('Reasoning blocks')
+    expect(result.terminal.output.slice(entered)).not.toContain('Reasoning blocks')
 
-    // Ctrl+C also cancels.
-    const ctrlCOutput = result.terminal.output.length
-    await open()
+    // Esc and Ctrl+C also close; the reopened dialog shows the live values.
+    const reopened = await open()
+    expect(result.terminal.output.slice(reopened)).toContain('collapsed')
+    expect(result.terminal.output.slice(reopened)).toContain('hidden')
+    result.terminal.send('\x1b')
+    await tick()
+    const ctrlCOutput = await open()
     result.terminal.send('\x03')
     await tick()
-    expect(result.terminal.output.slice(ctrlCOutput)).not.toContain('Reasoning blocks shown.')
+    expect(result.terminal.output.slice(ctrlCOutput)).not.toContain('Reasoning blocks')
 
     await dispose(result)
   })