fix(tui): bound diff rendering work

.agents/notes/implemented/bug-fix/2026-07-31-tui-diff-context-line-accounting.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 .agents/notes/implemented/bug-fix/2026-07-31-tui-diff-context-line-accounting.md
-2026-07-31-tui-diff-context-line-accounting.md: 71593022b56d9e675025f3a7a6d1e5e3edfa9b57
-2026-07-31-tui-diff-context-line-accounting.zh.md: df374c23b73fc5667cf733b4916d9d0d2ecb9198
+2026-07-31-tui-diff-context-line-accounting.md: d1bc72ea030abd46f809ca3e746e6043f717baeb
+2026-07-31-tui-diff-context-line-accounting.zh.md: a2a1dcce1325bca68c92cb4f206bfa35671e9d86

.agents/notes/implemented/bug-fix/2026-07-31-tui-diff-context-line-accounting.md

@@ -10,7 +10,9 @@ Result-time filesystem diffs carry the applied change with three surrounding con
 
 ## Decision
 
-The TUI compares each non-create `FileDiff.oldText` and `FileDiff.newText` at render time. Added and removed rows retain their green `+` and red `-` markers; equal context rows use the recessed body tone with a neutral two-space prefix. The footer sums only the rows classified as added or removed. A create (`oldText: null`) continues to classify every non-empty new-content row as added.
+The TUI compares each `FileDiff` whose old and new text are both available. Added and removed rows retain their green `+` and red `-` markers; equal context rows use the recessed body tone with a neutral two-space prefix. The footer sums only the rows classified as added or removed. `maxDiffEditLength` bounds the exact comparison by its combined added and removed line count; the default is 1000. Exceeding the bound renders the complete old side as removed and the complete new side as added, marks the footer approximate, and caches that result so redraws do not repeat the comparison.
+
+When `oldText` is `null`, the renderer cannot distinguish a create from a pending overwrite or an argument fallback whose prior text is unavailable. It therefore shows every non-empty new-side row as added, without claiming those rows were absent from an existing file. Empty new content renders no synthetic added row.
 
 This remains a consumer-side interpretation of the existing `FileDiff` contract. Filesystem tools continue to persist contextual before/after snippets, so other consumers keep their placement context and existing session logs replay with corrected TUI presentation. The TUI uses the same maintained `diff` package as `dsh-tool-fs` instead of introducing a second line-diff implementation.
 
@@ -22,8 +24,10 @@ This remains a consumer-side interpretation of the existing `FileDiff` contract.
 
 **Match equal lines by position without a diff algorithm.** Rejected: insertions and deletions shift subsequent context, so positional pairing would misclassify valid hunks.
 
+**Run every comparison to completion.** Rejected: pending tool views can contain unrestricted model-authored old and new strings, and an unbounded Myers comparison can block the synchronous terminal renderer.
+
 ## Consequences
 
-TUI diff cards distinguish evidence-bearing context from the mutation itself, and their `+A -R` footer reports the actual line delta. Replaying an existing contextual diff gains the corrected rendering without a migration. Rendering performs one additional line comparison per non-create hunk; result-time hunks are already context-bounded, while create cards bypass the comparison.
+TUI diff cards distinguish evidence-bearing context from the mutation itself, and an exact `+A -R` footer reports the actual line delta. Replaying an existing contextual diff gains the corrected rendering without a migration. Result-time filesystem hunks are context-bounded; unrestricted pending views either complete within the configured edit-length budget or degrade to an explicitly approximate linear rendering.
 
-The focused TUI test covers neutral context and exact totals. The assembled `advanced-cards` terminal snapshots pin the neutral context style, semantic change colors, and `+1 -1` footer through collapsed and expanded card states.
+The focused TUI tests cover neutral context, exact totals, an empty create, bounded fallback, and cache reuse. The assembled `advanced-cards` terminal snapshots pin the neutral context style, semantic change colors, exact footer, and approximate fallback through collapsed and expanded card states.

.agents/notes/implemented/bug-fix/2026-07-31-tui-diff-context-line-accounting.zh.md

@@ -10,7 +10,9 @@ Status: implemented
 
 ## 决策
 
-对于每个不对应文件创建的 `FileDiff`，TUI 在渲染时比较 `FileDiff.oldText` 和 `FileDiff.newText`。新增行和删除行仍分别使用绿色 `+` 和红色 `-` 标记；相同的上下文行则使用弱化的正文色调，并带有由两个空格构成的中性前缀。页脚只汇总归类为新增或删除的行。创建操作（`oldText: null`）仍将新内容中的每个非空行归类为新增行。
+TUI 会比较每个变更前后文本均可用的 `FileDiff`。新增行和删除行仍分别使用绿色 `+` 和红色 `-` 标记；相同的上下文行则使用弱化的正文色调，并带有由两个空格构成的中性前缀。页脚只汇总归类为新增或删除的行。`maxDiffEditLength` 以新增行与删除行的合计数为精确比较设置上限，默认值为 1000。超过上限时，TUI 会把完整旧侧渲染为删除内容、把完整新侧渲染为新增内容，将页脚标记为近似结果，并缓存该结果，避免后续重绘重复比较。
+
+当 `oldText` 为 `null` 时，渲染器无法区分文件创建、待处理覆写，以及旧文本不可用的参数回退。因此，它会把新侧的每个非空行显示并计作新增行，但不会声称这些行原先不存在于已有文件中。新内容为空时，不会渲染虚构的新增行。
 
 该行为仍然只是消费方对现有 `FileDiff` 契约的解释。文件系统工具仍会持久化带上下文的变更前后片段，因此其他消费方仍能获得定位上下文，已有会话日志在回放时也会采用修正后的 TUI 呈现。TUI 与 `dsh-tool-fs` 共用同一个受维护的 `diff` 包（package），无需引入第二套逐行 diff 实现。
 
@@ -22,8 +24,10 @@ Status: implemented
 
 **不使用 diff 算法，按位置匹配相同行。** 不予采纳：插入和删除会使后续上下文发生位移，因此按位置配对会把有效 hunk 错误分类。
 
+**让所有比较都运行至完成。** 不予采纳：待处理工具视图可能包含由模型生成且长度不受限制的新旧字符串，无界的 Myers 比较可能阻塞同步终端渲染器。
+
 ## 后果
 
-TUI diff 卡片会区分用于佐证的上下文与变更本身，其 `+A -R` 页脚报告实际的行变更量。回放已有的上下文 diff 无需迁移即可获得修正后的渲染。渲染每个不对应文件创建的 hunk 时，会额外执行一次逐行比较；结果时刻的 hunk 本就受上下文范围限制，创建卡片则会跳过比较。
+TUI diff 卡片会区分用于佐证的上下文与变更本身，精确的 `+A -R` 页脚会报告实际的行变更量。回放已有的上下文 diff 无需迁移即可获得修正后的渲染。结果时刻的文件系统 hunk 受上下文范围限制；不受限制的待处理视图要么在配置的编辑长度预算内完成比较，要么降级为明确标注为近似结果的线性渲染。
 
-聚焦的 TUI 测试覆盖中性上下文和精确合计值。组装后的 `advanced-cards` 终端快照在卡片折叠和展开状态下固定了中性上下文样式、变更行的语义色彩，以及 `+1 -1` 页脚。
+聚焦的 TUI 测试覆盖中性上下文、精确合计值、空文件创建、有界回退和缓存复用。组装后的 `advanced-cards` 终端快照在卡片折叠和展开状态下固定了中性上下文样式、变更行的语义色彩、精确结果页脚和近似回退。

docs/config-catalog.md

@@ -2015,6 +2015,8 @@ export interface TuiConfig {
   showReasoning?: boolean
   /** Maximum tool-card body lines retained in its collapsed head/tail preview. */
   maxToolOutputLines?: number
+  /** Maximum added and removed lines explored while deriving an exact line diff. */
+  maxDiffEditLength?: number
   /** Maximum options visible at once in a user-question panel. */
   maxQuestionOptions?: number
   /** Maximum models visible at once in the model selector. */
@@ -2060,7 +2062,7 @@ export interface TuiThemeConfig {
 }
 ```
 
-Source: [`packages/ui/tui/src/config.ts:117`](../packages/ui/tui/src/config.ts)
+Source: [`packages/ui/tui/src/config.ts:121`](../packages/ui/tui/src/config.ts)
 
 ## `@deepseek-ai/dsh-typert-loader`
 

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: b021789d660fd831c3fa0dad20d0bc174538eb57
-README.zh.md: b9cd7210932558a3a2feb0d5c1bfaf7e115703f6
+README.md: 837e072ec63752d5f0f1b93bcff16871d32ad615
+README.zh.md: 15f3ad49f2d5b7cdd438532d7443b066a1eac87d

packages/ui/tui/README.md

@@ -50,6 +50,7 @@ A launcher can seed a fresh session's first turn by providing `INITIAL_SKILL_KEY
 | `sessionId` | `main` | Exact shared agent/session identity driven by the terminal |
 | `showReasoning` | `true` | Render reasoning blocks |
 | `maxToolOutputLines` | `6` | Output lines retained across a collapsed tool card's head/tail preview |
+| `maxDiffEditLength` | `1000` | Maximum added and removed lines explored for an exact diff before whole-side fallback |
 | `maxQuestionOptions` | `8` | Visible options in a question panel |
 | `maxModelOptions` | `8` | Visible models in the model selector |
 | `maxResumeOptions` | `8` | Visible sessions in the resume selector |
@@ -72,6 +73,7 @@ A launcher can seed a fresh session's first turn by providing `INITIAL_SKILL_KEY
     sessionId: main-session-123
     showReasoning: true
     maxToolOutputLines: 6
+    maxDiffEditLength: 1000
     fileSearchExcludedDirectories: ['.git', 'node_modules', 'dist']
 ```
 
@@ -83,7 +85,7 @@ Every SGR code the TUI emits lives in one table, `paletteSpec` in `components/th
 
 There is one role per visual meaning: `dim` is the single recessed tone and `accent` the single emphasis color, while `success` and `error` double as a diff's added and removed lines. Colors and attributes are separately typed, so `bold(accent(x))` compiles and `accent(error(x))` does not — SGR has no color stack, so nesting one color inside another silently drops the outer color at the inner one's close. Attributes occupy independent SGR groups and compose with any color in either order. Run `/palette` to see every role as your terminal renders it, with its SGR pair.
 
-Grouped regions (user prompts, assistant replies, tool cards) are separated by a bold, underlined role header in the role color and blank-line spacing rather than a filled block or a per-line prefix, so a mouse drag-select copies the message text without any leading bar or indent; a tool card's status (pending, error, success) shows in its colored, underlined title glyph and title. Inside a tool card, the whole body — presenter title, a terminal `$` command and cwd, and the tool's own output — renders in one dim tone, so only the status-colored header carries color and the body reads as one recessed block instead of a run of competing shades; an injected-context card's prose is the same tone as its header. A diff card colors and counts only added `+` and removed `-` lines; unchanged context stays dim and uncounted. A `[signal …]` marker remains colored because there the color is the meaning rather than emphasis. The question panel emphasizes its active row with bold accent text, while selectors use reverse video. These treatments are foreground-only, so they never collide with the terminal background. Set `color: false` to strip all styling.
+Grouped regions (user prompts, assistant replies, tool cards) are separated by a bold, underlined role header in the role color and blank-line spacing rather than a filled block or a per-line prefix, so a mouse drag-select copies the message text without any leading bar or indent; a tool card's status (pending, error, success) shows in its colored, underlined title glyph and title. Inside a tool card, the whole body — presenter title, a terminal `$` command and cwd, and the tool's own output — renders in one dim tone, so only the status-colored header carries color and the body reads as one recessed block instead of a run of competing shades; an injected-context card's prose is the same tone as its header. A diff card with both sides available colors and counts exact added `+` and removed `-` lines, while unchanged context stays dim and uncounted. If exact comparison exceeds `maxDiffEditLength`, the card renders each old-side row as removed and each new-side row as added, marks the footer approximate, and caches that fallback for later redraws. When `oldText` is unavailable, including pending writes and replay fallbacks as well as creates, every non-empty new-side row is shown and counted as added; that count does not prove the rows were absent from an existing file. Empty new content produces no synthetic `+ ` row. A `[signal …]` marker remains colored because there the color is the meaning rather than emphasis. The question panel emphasizes its active row with bold accent text, while selectors use reverse video. These treatments are foreground-only, so they never collide with the terminal background. Set `color: false` to strip all styling.
 
 ## Model Experience
 

packages/ui/tui/README.zh.md

@@ -50,6 +50,7 @@ Footer 将会话报告的用量汇总为 `↑<uncached input> ↓<output>`；任
 | `sessionId` | `main` | 由终端驱动的精确共享 agent／会话身份 |
 | `showReasoning` | `true` | 渲染 reasoning 块 |
 | `maxToolOutputLines` | `6` | 折叠工具卡片的头尾预览所保留的输出行数 |
+| `maxDiffEditLength` | `1000` | 回退到整侧展示前，精确 diff 最多探索的新增与删除行总数 |
 | `maxQuestionOptions` | `8` | 问题面板中可见的选项数 |
 | `maxModelOptions` | `8` | 模型选择器中可见的模型数 |
 | `maxResumeOptions` | `8` | 恢复选择器中可见的会话数 |
@@ -72,6 +73,7 @@ Footer 将会话报告的用量汇总为 `↑<uncached input> ↓<output>`；任
     sessionId: main-session-123
     showReasoning: true
     maxToolOutputLines: 6
+    maxDiffEditLength: 1000
     fileSearchExcludedDirectories: ['.git', 'node_modules', 'dist']
 ```
 
@@ -83,7 +85,7 @@ TUI 发出的所有 SGR 代码都集中在一个表中，即 `components/theme.t
 
 每种视觉语义只对应一个角色：`dim` 是唯一的弱化色调，`accent` 是唯一的强调色，`success` 和 `error` 还分别充当 diff 的新增行与删除行。颜色和属性分属不同类型，因此 `bold(accent(x))` 可以通过编译，`accent(error(x))` 则不行——SGR 没有颜色栈；在一种颜色内嵌套另一种颜色时，内层颜色闭合时会静默丢弃外层颜色。各属性占用彼此独立的 SGR 组，可以按任一顺序与任何颜色组合。运行 `/palette` 可查看每个角色在你的终端上的实际渲染效果及其 SGR 码对。
 
-成组区域（用户提示词、assistant 回复、工具卡片）通过以角色色渲染的粗体带下划线角色标题和空行分隔，而非填充背景块或逐行前缀，因此用鼠标框选复制时不会带上任何左侧竖条或缩进；工具卡片的状态（进行中、错误、成功）由其彩色带下划线的标题字形与标题体现。在工具卡片内部，整个正文——presenter 标题、终端 `$` 命令与 cwd，以及工具自身的输出——统一以同一种暗色渲染，因此只有带状态色的表头携带颜色，正文读作一个整体弱化的区块，而不是一串互相竞争的色调；注入上下文卡片的正文与其表头也是同一种色调。diff 卡片只为新增的 `+` 行和删除的 `-` 行着色并计数；未变更的上下文保持暗色且不纳入计数。`[signal …]` 标记仍保留颜色，因为那里的颜色本身就是语义，而非强调。问题面板使用粗体强调色文本突出活跃行，选择器则使用反色。所有效果都只作用于前景色，因此不会与终端背景冲突。设置 `color: false` 可移除所有样式。
+成组区域（用户提示词、assistant 回复、工具卡片）通过以角色色渲染的粗体带下划线角色标题和空行分隔，而非填充背景块或逐行前缀，因此用鼠标框选复制时不会带上任何左侧竖条或缩进；工具卡片的状态（进行中、错误、成功）由其彩色带下划线的标题字形与标题体现。在工具卡片内部，整个正文——presenter 标题、终端 `$` 命令与 cwd，以及工具自身的输出——统一以同一种暗色渲染，因此只有带状态色的表头携带颜色，正文读作一个整体弱化的区块，而不是一串互相竞争的色调；注入上下文卡片的正文与其表头也是同一种色调。当前后两侧文本均可用时，diff 卡片会为精确识别出的新增 `+` 行和删除 `-` 行着色并计数；未变更的上下文保持暗色且不纳入计数。如果精确比较超出 `maxDiffEditLength`，卡片会把旧侧每一行渲染为删除行、把新侧每一行渲染为新增行，将页脚标记为近似结果，并缓存该回退结果供后续重绘使用。当 `oldText` 不可用时（包括待处理写入、回放回退以及文件创建），新侧的每个非空行都会显示并计作新增行；该计数不能证明这些行原先不存在于已有文件中。新内容为空时，不会补出虚构的 `+ ` 行。`[signal …]` 标记仍保留颜色，因为那里的颜色本身就是语义，而非强调。问题面板使用粗体强调色文本突出活跃行，选择器则使用反色。所有效果都只作用于前景色，因此不会与终端背景冲突。设置 `color: false` 可移除所有样式。
 
 ## 模型体验
 

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

@@ -57,6 +57,7 @@ interface RenderedDiff {
   lines: string[]
   added: number
   removed: number
+  approximate: boolean
 }
 
 /** Split one diff change into display rows without counting its trailing line terminator. */
@@ -66,8 +67,12 @@ function diffValueLines(value: string): string[] {
   return (safe.endsWith('\n') ? safe.slice(0, -1) : safe).split('\n')
 }
 
-/** A file diff whose unchanged context stays neutral and does not affect change totals. */
-function renderDiff(diff: FileDiff, palette: Palette): RenderedDiff {
+/**
+ * A file diff whose unchanged context stays neutral and does not affect exact
+ * change totals. Comparisons beyond the edit-distance budget fall back to
+ * whole-side rendering so a model-authored pending edit cannot stall the TUI.
+ */
+function renderDiff(diff: FileDiff, maxDiffEditLength: number, palette: Palette): RenderedDiff {
   // The card header is a fixed `Tool / <name>` frame that never names a file, so
   // each hunk always carries its own path header (no redundancy to suppress).
   const lines = [palette.bold(displayText(diff.path))]
@@ -77,9 +82,20 @@ function renderDiff(diff: FileDiff, palette: Palette): RenderedDiff {
     const newLines = diffValueLines(diff.newText)
     added = newLines.length
     for (const line of newLines) lines.push(palette.success(`+ ${line}`))
-    return { lines, added, removed }
+    return { lines, added, removed, approximate: false }
+  }
+  const changes = compareLines(diff.oldText, diff.newText, { maxEditLength: maxDiffEditLength })
+  if (changes === undefined) {
+    const oldLines = diffValueLines(diff.oldText)
+    const newLines = diffValueLines(diff.newText)
+    lines.push(palette.dim(`[exact line diff omitted: >${maxDiffEditLength} changed lines]`))
+    removed = oldLines.length
+    added = newLines.length
+    for (const line of oldLines) lines.push(palette.error(`- ${line}`))
+    for (const line of newLines) lines.push(palette.success(`+ ${line}`))
+    return { lines, added, removed, approximate: true }
   }
-  for (const change of compareLines(diff.oldText, diff.newText)) {
+  for (const change of changes) {
     const changedLines = diffValueLines(change.value)
     if (change.added) {
       added += changedLines.length
@@ -91,7 +107,7 @@ function renderDiff(diff: FileDiff, palette: Palette): RenderedDiff {
       for (const line of changedLines) lines.push(palette.dim(`  ${line}`))
     }
   }
-  return { lines, added, removed }
+  return { lines, added, removed, approximate: false }
 }
 
 /**
@@ -354,12 +370,14 @@ export class ToolCardComponent implements Component {
   private visibility: ToolCardVisibility = 'collapsed'
   private callView: ToolCallView
   private resultView: ToolResultView | undefined
+  private diffBodyCache: { view: ToolCallView | ToolResultView; body: CardBody } | undefined
 
   constructor(
     private readonly name: string,
     private readonly parsed: ParsedArguments,
     private readonly definition: ToolDefinition | undefined,
     private readonly maxOutputLines: number,
+    private readonly maxDiffEditLength: number,
     private readonly palette: Palette,
     private readonly mdTheme: MarkdownTheme,
   ) {
@@ -530,21 +548,27 @@ export class ToolCardComponent implements Component {
       return { prelude: prelude.filter(Boolean), lines: lines.filter(Boolean) }
     }
     if (view.card === 'diff') {
+      if (this.diffBodyCache?.view === view) return this.diffBodyCache.body
       // The header no longer names the file, so each diff keeps its own path
       // header. A trailing footer summarizes the change (`+A -R · N file(s)`).
-      let added = 0
-      let removed = 0
-      const hunks = view.diffs.flatMap((diff, index) => {
-        const rendered = renderDiff(diff, this.palette)
-        added += rendered.added
-        removed += rendered.removed
+      const renderedDiffs = view.diffs.map(diff =>
+        renderDiff(diff, this.maxDiffEditLength, this.palette),
+      )
+      const added = renderedDiffs.reduce((total, rendered) => total + rendered.added, 0)
+      const removed = renderedDiffs.reduce((total, rendered) => total + rendered.removed, 0)
+      const approximate = renderedDiffs.some(rendered => rendered.approximate)
+      const hunks = renderedDiffs.flatMap((rendered, index) => {
         return [...index > 0 ? [''] : [], ...rendered.lines]
       })
       const files = view.diffs.length
-      const footer = this.palette.dim(`└ +${added} -${removed} · ${files} file${files === 1 ? '' : 's'}`)
+      const footer = this.palette.dim(
+        `└ +${added} -${removed} · ${files} file${files === 1 ? '' : 's'}${approximate ? ' · approximate' : ''}`,
+      )
       // A diff's own `+`/`-` colors carry its meaning, so it renders verbatim
       // rather than under the dim result-output color.
-      return { prelude: [...hunks, footer], lines: [] }
+      const body = { prelude: [...hunks, footer], lines: [] }
+      this.diffBodyCache = { view, body }
+      return body
     }
     // The web card carries no `content` copy, so a `web` result view falls back
     // to the raw result content here (`view.card === 'generic'` narrows the

packages/ui/tui/src/config.ts

@@ -34,6 +34,8 @@ export interface TuiConfig {
   showReasoning?: boolean
   /** Maximum tool-card body lines retained in its collapsed head/tail preview. */
   maxToolOutputLines?: number
+  /** Maximum added and removed lines explored while deriving an exact line diff. */
+  maxDiffEditLength?: number
   /** Maximum options visible at once in a user-question panel. */
   maxQuestionOptions?: number
   /** Maximum models visible at once in the model selector. */
@@ -64,6 +66,7 @@ export interface TuiConfig {
 
 const showReasoningSchema = z.boolean().default(true)
 const maxToolOutputLinesSchema = z.number().step(1).min(1).default(6)
+const maxDiffEditLengthSchema = z.number().step(1).min(1).default(1000)
 const maxQuestionOptionsSchema = z.number().step(1).min(1).default(8)
 const maxModelOptionsSchema = z.number().step(1).min(1).default(8)
 const maxResumeOptionsSchema = z.number().step(1).min(1).default(8)
@@ -95,6 +98,7 @@ const titleSchema = z.string().default('DeepSeek Harness')
 const tuiConfigSchemaFields = {
   showReasoning: showReasoningSchema,
   maxToolOutputLines: maxToolOutputLinesSchema,
+  maxDiffEditLength: maxDiffEditLengthSchema,
   maxQuestionOptions: maxQuestionOptionsSchema,
   maxModelOptions: maxModelOptionsSchema,
   maxResumeOptions: maxResumeOptionsSchema,
@@ -135,6 +139,7 @@ export const Config: z<Config> = z.object({
   initialSkill: z.string(),
   showReasoning: tuiConfigSchemaFields.showReasoning,
   maxToolOutputLines: tuiConfigSchemaFields.maxToolOutputLines,
+  maxDiffEditLength: tuiConfigSchemaFields.maxDiffEditLength,
   maxQuestionOptions: tuiConfigSchemaFields.maxQuestionOptions,
   maxModelOptions: tuiConfigSchemaFields.maxModelOptions,
   maxResumeOptions: tuiConfigSchemaFields.maxResumeOptions,
@@ -164,6 +169,7 @@ export interface ResolvedTuiThemeConfig {
 export interface ResolvedTuiConfig {
   showReasoning: boolean
   maxToolOutputLines: number
+  maxDiffEditLength: number
   maxQuestionOptions: number
   maxModelOptions: number
   maxResumeOptions: number
@@ -189,6 +195,7 @@ export function resolveTuiConfig(config: TuiConfig | undefined): ResolvedTuiConf
   return {
     showReasoning: config?.showReasoning ?? true,
     maxToolOutputLines: config?.maxToolOutputLines ?? 6,
+    maxDiffEditLength: config?.maxDiffEditLength ?? 1000,
     maxQuestionOptions: config?.maxQuestionOptions ?? 8,
     maxModelOptions: config?.maxModelOptions ?? 8,
     maxResumeOptions: config?.maxResumeOptions ?? 8,

packages/ui/tui/src/index.ts

@@ -605,6 +605,7 @@ export function createTuiChat(
       parsed,
       ctx.tools.get(event.data.name, agent),
       resolved.maxToolOutputLines,
+      resolved.maxDiffEditLength,
       palette,
       mdTheme,
     )
@@ -748,7 +749,15 @@ export function createTuiChat(
         const callId = event.data.message.source.callId
         let card = toolCards.get(callId)
         if (card === undefined) {
-          card = new ToolCardComponent('tool', { value: {}, valid: true }, undefined, resolved.maxToolOutputLines, palette, mdTheme)
+          card = new ToolCardComponent(
+            'tool',
+            { value: {}, valid: true },
+            undefined,
+            resolved.maxToolOutputLines,
+            resolved.maxDiffEditLength,
+            palette,
+            mdTheme,
+          )
           card.setVisibility(toolsVisibility)
           chat.addChild(card)
           allToolCards.add(card)

packages/ui/tui/tests/snapshots/advanced-cards-collapsed.expected.txt

@@ -1,7 +1,7 @@
-terminal 100x40 buffer=normal length=40 base=0 viewport=0
+terminal 100x40 buffer=normal length=41 base=1 viewport=1
 lifecycle started=1 stopped=0 progress=inactive
 title "DSH snapshot"
-cursor hidden column=7 viewportRow=34 bufferRow=34
+cursor hidden column=7 viewportRow=39 bufferRow=40
 buffer
 0| " DEEPSEEK HARNESS"
   style 1-8 fg=bright-magenta bold
@@ -58,17 +58,27 @@ buffer
   style 0-99 dim
 30| "Loaded review instructions.                                                                         "
   style 0-99 dim
-31| "Model wait 0.0s                                                                                     "
+31| <blank>
+32| "● Tool / large_edit"
+  style 0-18 fg=green
+33| "src/large.ts                                                                                        "
+  style 0-11 bold
+34| "[exact line diff omitted: >2 changed lines]                                                         "
+  style 0-42 dim
+35| "… +6 lines (Ctrl+O to expand)                                                                       "
+  style 0-28 dim
+36| "└ +3 -3 · 1 file · approximate                                                                      "
+  style 0-29 dim
+37| "Model wait 0.0s                                                                                     "
   style 0-14 dim
-32| <blank>
-33| "/workspace/project (tui-staging)  deepseek-v4-flash  ↑0 ↓0  0% context"
+38| <blank>
+39| "/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
-34| " dsh >                                                                                              "
+40| " dsh >                                                                                              "
   style 1-3 fg=bright-magenta bold
   style 5-6 dim
   style 7-7 inverse
-35-39| <blank>

packages/ui/tui/tests/snapshots/advanced-cards-expanded.expected.txt

@@ -1,7 +1,7 @@
-terminal 100x40 buffer=normal length=42 base=2 viewport=2
+terminal 100x40 buffer=normal length=53 base=13 viewport=13
 lifecycle started=1 stopped=0 progress=inactive
 title "DSH snapshot"
-cursor hidden column=7 viewportRow=39 bufferRow=41
+cursor hidden column=7 viewportRow=39 bufferRow=52
 buffer
 0| " DEEPSEEK HARNESS"
   style 1-8 fg=bright-magenta bold
@@ -70,19 +70,40 @@ buffer
   style 0-99 dim
 35| "Loaded review instructions.                                                                         "
   style 0-99 dim
-36| "Model wait 0.0s                                                                                     "
+36| <blank>
+37| "● Tool / large_edit"
+  style 0-18 fg=green
+38| "src/large.ts                                                                                        "
+  style 0-11 bold
+39| "[exact line diff omitted: >2 changed lines]                                                         "
+  style 0-42 dim
+40| "- old one                                                                                           "
+  style 0-8 fg=red
+41| "- old two                                                                                           "
+  style 0-8 fg=red
+42| "- old three                                                                                         "
+  style 0-10 fg=red
+43| "+ new one                                                                                           "
+  style 0-8 fg=green
+44| "+ new two                                                                                           "
+  style 0-8 fg=green
+45| "+ new three                                                                                         "
+  style 0-10 fg=green
+46| "└ +3 -3 · 1 file · approximate                                                                      "
+  style 0-29 dim
+47| "Model wait 0.0s                                                                                     "
   style 0-14 dim
-37| <blank>
-38| "Tool and context cards expanded.                                                                    "
+48| <blank>
+49| "Tool and context cards expanded.                                                                    "
   style 0-31 dim
-39| <blank>
-40| "/workspace/project (tui-staging)  deepseek-v4-flash  ↑0 ↓0  0% context"
+50| <blank>
+51| "/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
-41| " dsh >                                                                                              "
+52| " dsh >                                                                                              "
   style 1-3 fg=bright-magenta bold
   style 5-6 dim
   style 7-7 inverse

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

@@ -269,13 +269,32 @@ const ADVANCED_CARD_TOOLS: Record<string, ToolDefinition> = {
   edit: visualTool(
     'edit',
     () => ({ card: 'diff', title: 'Edit src/view.ts', diffs: [{ path: 'src/view.ts', oldText: 'old line', newText: 'new line' }] }),
-    // The real edit/write tools produce exactly one diff whose path the title
-    // already names, so the card omits the redundant per-file header.
+    // The fixed tool header never names a path, so the hunk retains its path.
     (): ToolResultView => ({
       card: 'diff',
       diffs: [{ path: 'src/view.ts', oldText: 'old line\nkeep', newText: 'new line\nkeep' }],
     }),
   ),
+  large_edit: visualTool(
+    'large_edit',
+    () => ({
+      card: 'diff',
+      title: 'Edit src/large.ts',
+      diffs: [{
+        path: 'src/large.ts',
+        oldText: 'old one\nold two\nold three',
+        newText: 'new one\nnew two\nnew three',
+      }],
+    }),
+    (): ToolResultView => ({
+      card: 'diff',
+      diffs: [{
+        path: 'src/large.ts',
+        oldText: 'old one\nold two\nold three',
+        newText: 'new one\nnew two\nnew three',
+      }],
+    }),
+  ),
   subagent: visualTool('subagent', args => ({
     card: 'generic',
     title: 'Delegate renderer audit',
@@ -585,7 +604,7 @@ describe('TUI terminal-state snapshots', () => {
   it('pins terminal, diff, subagent, task, skill, collapsed, and expanded cards', async () => {
     const harness = await setupSnapshot({
       tools: ADVANCED_CARD_TOOLS,
-      config: { maxToolOutputLines: 3 },
+      config: { maxToolOutputLines: 3, maxDiffEditLength: 2 },
     }, { columns: 100, rows: 40 })
     const calls = [
       { id: 'advanced-1', name: 'bash', arguments: { command: 'pnpm run test:coverage' } },
@@ -593,6 +612,7 @@ describe('TUI terminal-state snapshots', () => {
       { id: 'advanced-3', name: 'subagent', arguments: { prompt: 'Review renderer ownership and report only gaps.' } },
       { id: 'advanced-4', name: 'task_output', arguments: { task_id: 'subagent-7', wait: true } },
       { id: 'advanced-5', name: 'skill', arguments: { name: 'dsh-code-review' } },
+      { id: 'advanced-6', name: 'large_edit', arguments: { file_path: 'src/large.ts' } },
     ]
     await renderAfter(harness, () => {
       appendToolCalls(harness.session, calls)
@@ -601,6 +621,7 @@ describe('TUI terminal-state snapshots', () => {
       appendToolResult(harness.session, 'advanced-3', [{ type: 'text', text: 'The renderer has explicit lifecycle ownership.' }])
       appendToolResult(harness.session, 'advanced-4', [{ type: 'text', text: 'audit complete\n[status: completed]' }])
       appendToolResult(harness.session, 'advanced-5', [{ type: 'text', text: 'Loaded review instructions.' }])
+      appendToolResult(harness.session, 'advanced-6', [{ type: 'text', text: 'large edit complete' }])
     })
     await checkpoint('advanced-cards-collapsed', harness.terminal, { includeScrollback: true })
 

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

@@ -178,6 +178,7 @@ describe('TUI config', () => {
     expect(resolveTuiConfig(undefined)).toEqual({
       showReasoning: true,
       maxToolOutputLines: 6,
+      maxDiffEditLength: 1000,
       maxQuestionOptions: 8,
       maxModelOptions: 8,
       maxResumeOptions: 8,
@@ -202,6 +203,7 @@ describe('TUI config', () => {
     expect(resolveTuiConfig({
       showReasoning: false,
       maxToolOutputLines: 2,
+      maxDiffEditLength: 12,
       maxQuestionOptions: 3,
       maxModelOptions: 4,
       maxResumeOptions: 5,
@@ -218,6 +220,7 @@ describe('TUI config', () => {
     })).toEqual({
       showReasoning: false,
       maxToolOutputLines: 2,
+      maxDiffEditLength: 12,
       maxQuestionOptions: 3,
       maxModelOptions: 4,
       maxResumeOptions: 5,
@@ -4651,6 +4654,98 @@ describe('tool cards and surface replay', () => {
     await dispose(result)
   })
 
+  it('renders an empty create without a synthetic added row', async () => {
+    const emptyCreate: Record<string, ToolDefinition> = {
+      emptyCreate: {
+        name: 'emptyCreate',
+        description: '',
+        parameters: {},
+        output: UNUSED_TOOL_OUTPUT,
+        execute: async () => [],
+        presentCall: () => ({
+          card: 'diff',
+          title: 'Write empty.txt',
+          diffs: [{ path: 'empty.txt', oldText: null, newText: '' }],
+        }),
+      },
+    }
+    const result = await setup({
+      tools: emptyCreate,
+      config: { maxToolOutputLines: 20, theme: { color: false } },
+    })
+    appendAssistant(result.session, [
+      { type: 'tool-call', id: 'empty-create' as never, name: 'emptyCreate', arguments: '{}' },
+    ])
+    result.session.append('tool/call', {
+      turn: 1,
+      step: 1,
+      callId: 'empty-create' as never,
+      name: 'emptyCreate',
+      arguments: '{}',
+    })
+    await tick()
+    const rows = result.terminal.output.split('\n').map(row => row.trim())
+    expect(result.terminal.output).toContain('empty.txt')
+    expect(result.terminal.output).toContain('└ +0 -0 · 1 file')
+    expect(rows).not.toContain('+')
+    await dispose(result)
+  })
+
+  it('bounds and caches exact diff comparison before whole-side fallback', async () => {
+    let oldTextReads = 0
+    const boundedDiff = {
+      path: 'bounded.txt',
+      get oldText() {
+        oldTextReads += 1
+        return 'old one\nold two'
+      },
+      newText: 'new one\nnew two',
+    }
+    const bounded: Record<string, ToolDefinition> = {
+      bounded: {
+        name: 'bounded',
+        description: '',
+        parameters: {},
+        output: UNUSED_TOOL_OUTPUT,
+        execute: async () => [],
+        presentCall: () => ({
+          card: 'diff',
+          title: 'Edit bounded.txt',
+          diffs: [boundedDiff],
+        }),
+      },
+    }
+    const result = await setup({
+      tools: bounded,
+      config: {
+        maxToolOutputLines: 20,
+        maxDiffEditLength: 1,
+        theme: { color: false },
+      },
+    })
+    appendAssistant(result.session, [
+      { type: 'tool-call', id: 'bounded-diff' as never, name: 'bounded', arguments: '{}' },
+    ])
+    result.session.append('tool/call', {
+      turn: 1,
+      step: 1,
+      callId: 'bounded-diff' as never,
+      name: 'bounded',
+      arguments: '{}',
+    })
+    await tick()
+    expect(result.terminal.output).toContain('[exact line diff omitted: >1 changed lines]')
+    expect(result.terminal.output).toContain('- old one')
+    expect(result.terminal.output).toContain('+ new one')
+    expect(result.terminal.output).toContain('└ +2 -2 · 1 file · approximate')
+    const readsAfterFirstRender = oldTextReads
+    expect(readsAfterFirstRender).toBeGreaterThan(0)
+    result.terminal.resize(87)
+    await tick()
+    expect(oldTextReads).toBe(readsAfterFirstRender)
+    await dispose(result)
+  })
+
   it('drops blank rows from a terminal card result that the dim styling wraps', async () => {
     const blankRowTools: Record<string, ToolDefinition> = {
       trailing: {