Merge remote-tracking branch 'origin/master' into feature/gui-welcome-copy-v2

packages/bash/bash-sandbox/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/bash/bash-sandbox/README.md
-README.md: ca77a9c626784b29145712535d69de4afbd3a697
-README.zh.md: 4ecc8d533f7af373bdacd133d44a8def6d265868
+README.md: 035a8ad2401ca608d264049d454359eda7b2b9a7
+README.zh.md: cee27a9baaa539ba07eb1d730ea9bef2004fbeeb

packages/bash/bash-sandbox/README.md

@@ -18,7 +18,7 @@ Semantics:
 
 - **Denials are result facts.** A failed run whose stderr carries the selected backend's own denial dialect — the signatures the provider stamps on every wrap (EROFS text under bwrap, EACCES under Landlock, EPERM under Seatbelt) — is reported as `BashRunResult.sandbox.denied: true` (conservative classification, read from the collected stderr tail); every CONFINED run also carries the mode it executed under (`result.sandbox.mode`) and the provider's enforcement completeness (`result.sandbox.enforcement`: `full`, or `partial` on an older Landlock ABI).
 - **Runner failures are sandbox failures, never command failures.** Foreground execution throws `SANDBOX_UNAVAILABLE`; a settled background process stamps `process.sandbox.runnerFailed`, which the bash producer renders through generic `task_output`. Spawn failures also pass through settlement, so confined background handles retain their mode/enforcement facts and release per-process accounting.
-- **Deployment fallback, per-call policy.** [`ctx.sandboxPolicy`](../../sandbox/sandbox-policy/) resolves a complete `SandboxExecutionPolicy` for every tool call: the calling session supplies its mode override and immutable cwd root, while deployment config supplies the fallbacks for agentless calls. An approved escalation changes only that policy's mode; its session root stays attached. `resolve()` carries the policy onto the spec, so overlapping commands from different projects run, classify, and report under their own roots and modes. The capability fact `ctx.bash.sandboxMode` reports the configured default so the tool layer advertises escalation only when this executor is mounted. The model learns of the sandbox only through result facts — the static bash tool description explains the denial marker; there is no current-mode statement in the system prompt.
+- **Deployment fallback, per-call policy.** [`ctx.sandboxPolicy`](../../sandbox/sandbox-policy/) resolves a complete `SandboxExecutionPolicy` for every tool call: the calling session supplies its mode override and immutable cwd root, while deployment config supplies the fallbacks for agentless calls. An approved escalation changes only that policy's mode; its session root stays attached. `resolve()` carries the policy onto the spec, so overlapping commands from different projects run, classify, and report under their own roots and modes. The capability fact `ctx.bash.sandboxMode` reports the configured default so the tool layer advertises escalation only when this executor is mounted; the static bash tool description separately owns denial and escalation guidance.
 - **File effects only.** Network and process visibility are deliberately not restricted — the mode vocabulary does not pretend to cover what the backend does not enforce.
 - Process mechanics (spawn, process-group kills, output collection/spill, background handles, credential scrub) are inherited from [`dsh-bash-local`](../bash-local/); runner selection lives in [`dsh-sandbox-local`](../../sandbox/sandbox-local/).
 
@@ -44,15 +44,15 @@ The keyless consumer-integration proofs are `tests/bwrap.e2e.ts`, `tests/landloc
 
 #### What the model sees
 
-The generated [`dsh-tool-bash` schemas](../../../docs/tool-catalog.md#deepseek-aidsh-tool-bash) are the baseline. By advertising a confining `sandboxMode`, this backend augments `bash` with `sandbox_permissions` using enum `workspace-write` | `danger-full-access` and with `justification`. The backend adds no prompt prose, and the session's effective mode remains unstated.
+The generated [`dsh-tool-bash` schemas](../../../docs/tool-catalog.md#deepseek-aidsh-tool-bash) are the baseline. By advertising a confining `sandboxMode`, this backend augments `bash` with `sandbox_permissions` using enum `workspace-write` | `danger-full-access` and with `justification`. The policy owner separately contributes the current capability-neutral `sandbox:policy` context.
 
 #### Token effect
 
-Small fixed schema increment on requests where `bash` is visible; mode switches add no context tokens.
+Small fixed schema increment on requests where `bash` is visible, plus the current-policy clause owned by `dsh-sandbox-policy`.
 
 #### KV Cache effect
 
-Prefix-stable while the executor advertises the same sandbox capabilities. Changing those capabilities alters the `bash` schema and may invalidate reuse from that definition; per-session mode switches do not.
+A standing-policy change appends a complete owner-rendered context snapshot after retained history, preserving the existing system/history prefix byte-for-byte. Changing executor capabilities alters the `bash` schema.
 
 ### Bash tool result, indirectly
 

packages/bash/bash-sandbox/README.zh.md

@@ -18,7 +18,7 @@
 
 - **拒绝是结果事实。** 如果一次失败运行的 stderr 包含所选后端自身的拒绝方言，即提供方在每次包装时加上的特征（bwrap 下的 EROFS 文本、Landlock 下的 EACCES、Seatbelt 下的 EPERM），则结果报告 `BashRunResult.sandbox.denied: true`（从已收集的 stderr 尾部进行保守分类）。每次受限制运行还会携带执行时模式（`result.sandbox.mode`）与提供方强制执行完整性（`result.sandbox.enforcement`：`full`，或在较旧 Landlock ABI 上为 `partial`）。
 - **Runner 失败是沙箱失败，绝不是命令失败。** 前台执行会抛出 `SANDBOX_UNAVAILABLE`；已结算的后台进程会标记 `process.sandbox.runnerFailed`，Bash 结果生成方通过通用 `task_output` 渲染它。spawn 失败也会经过结算，因此受限制的后台句柄会保留自身的模式／强制执行事实，并释放每进程计数。
-- **部署回退，每次调用策略。** [`ctx.sandboxPolicy`](../../sandbox/sandbox-policy/) 为每次工具调用解析完整的 `SandboxExecutionPolicy`：调用会话提供自身的模式覆盖与不可变 cwd 根目录，部署配置则为无 agent（智能体）调用提供回退。已批准的升权只更改该策略的模式，会话根目录仍然附着其上。`resolve()` 把策略带入 spec，因此来自不同项目的重叠命令会在各自的根目录与模式下运行、分类和报告。能力事实 `ctx.bash.sandboxMode` 报告已配置的默认值，因此工具层只在装载该执行器时才公布升权。模型只能通过结果事实了解沙箱：静态 bash 工具描述会解释拒绝标记，系统提示词中不会声明当前模式。
+- **部署回退，每次调用策略。** [`ctx.sandboxPolicy`](../../sandbox/sandbox-policy/) 为每次工具调用解析完整的 `SandboxExecutionPolicy`：调用会话提供自身的模式覆盖与不可变 cwd 根目录，部署配置则为无 agent（智能体）调用提供回退。已批准的升权只更改该策略的模式，会话根目录仍然附着其上。`resolve()` 把策略带入 spec，因此来自不同项目的重叠命令会在各自的根目录与模式下运行、分类和报告。能力事实 `ctx.bash.sandboxMode` 报告已配置的默认值，因此工具层只在装载该执行器时才公布升权；静态 bash 工具描述则单独负责拒绝与升级引导。
 - **只限制文件影响。** 设计上不限制网络与进程可见性：模式词汇不会声称覆盖后端未强制执行的范围。
 - 进程机制（spawn、进程组终止、输出收集／spill、后台句柄、凭证清理）继承自 [`dsh-bash-local`](../bash-local/)；runner 选择位于 [`dsh-sandbox-local`](../../sandbox/sandbox-local/)。
 
@@ -44,15 +44,15 @@
 
 #### 模型看到的内容
 
-基线是生成的 [`dsh-tool-bash` schema](../../../docs/tool-catalog.md#deepseek-aidsh-tool-bash)。通过公布表明启用隔离的 `sandboxMode` 能力，此后端会为 `bash` 增加 `sandbox_permissions`，其 enum 为 `workspace-write` | `danger-full-access`，并增加 `justification`。后端不添加提示词文本，会话的有效模式仍不会声明。
+基线是生成的 [`dsh-tool-bash` schema](../../../docs/tool-catalog.md#deepseek-aidsh-tool-bash)。通过公布表明启用隔离的 `sandboxMode` 能力，此后端会为 `bash` 增加 `sandbox_permissions`，其 enum 为 `workspace-write` | `danger-full-access`，并增加 `justification`。策略归属方会另行贡献当前且不区分具体能力的 `sandbox:policy` 上下文。
 
 #### Token 影响
 
-在 `bash` 可见的请求上，schema 固定增加少量内容；模式切换不增加上下文 token。
+在 `bash` 可见的请求上，schema 固定增加少量内容，另有一条由 `dsh-sandbox-policy` 负责的当前策略子句。
 
 #### KV Cache 影响
 
-执行器持续公布相同沙箱能力时，前缀保持稳定。更改这些能力会改变 `bash` schema，可能使从该定义起的复用失效；每会话模式切换不会导致失效。
+常驻策略变化会在保留的历史之后追加一份由归属方渲染的完整上下文快照，并使既有 system/history 前缀保持逐字节不变。更改执行器能力会改变 `bash` schema。
 
 ### 间接的 Bash 工具结果
 

packages/bash/bash-sandbox/src/index.ts

@@ -21,7 +21,7 @@ import { classifyDenial, classifyRunnerFailure, matchesSignature, shellQuote } f
  * Plugin config: the local executor's knobs, verbatim. The sandbox policy —
  * the default mode and fallback `workspace-write` root — is NOT here: it lives
  * on `ctx.sandboxPolicy` (`@deepseek-ai/dsh-sandbox-policy`), which resolves
- * each calling session's mode and cwd for both enforcing families. The runner
+ * each calling session's mode and cwd for every enforcing capability. The runner
  * choice is likewise the `ctx.sandbox` provider's config, not this executor's.
  */
 export type Config = LocalConfig
@@ -30,8 +30,8 @@ export type Config = LocalConfig
  * Registers as `ctx.bash` in place of the local executor and requires a
  * `ctx.sandbox` provider plus `ctx.sandboxPolicy`; the tool layer is
  * unchanged. Tool calls pass the calling session's resolved policy; direct
- * calls fall back to deployment policy. The prompt does not state the standing
- * mode; `result.sandbox` reports the mode and enforcement actually used.
+ * calls fall back to deployment policy. `result.sandbox` reports the mode and
+ * enforcement actually used.
  */
 export class SandboxBashExecutor extends LocalBashExecutor {
   static override inject = ['subprocess', 'sandbox', 'sandboxPolicy']

packages/bash/tool-bash/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/bash/tool-bash/README.md
-README.md: deb6b899c81cb8c335b4c1cffdde4797e0a8be92
-README.zh.md: c2514308fb9f234e6d191a6b1a821ac3d195378b
+README.md: 29b9fba369e1fc6a4b8bb7bdd6543b7678df627d
+README.zh.md: 31f691f7bfb8d2cb905751663151c3f6a6bc6c57

packages/bash/tool-bash/README.md

@@ -71,7 +71,7 @@ Escalating bash calls resolve `ctx.approval` before execution. `allowed-once` ap
 
 ## Per-session mode switching
 
-For sandboxing executors, each call resolves mode as one-shot escalation, then session override, then executor default. Non-sandboxing and agent-less calls carry no session override. Neither the prompt nor a switch notice announces the standing mode; denial results report the effective mode when the boundary matters. See the [`dsh-bash` fold](../bash/README.md) and [sandbox switching contract](../../../.agents/notes/implemented/feature/2026-07-06-sandbox.md).
+For sandboxing executors, each call resolves mode as one-shot escalation, then session override, then executor default. Non-sandboxing and agent-less calls carry no session override. The policy owner contributes the current capability-neutral standing mode; denial results still own the operation-specific effective mode and retry guidance. See the [`dsh-bash` fold](../bash/README.md) and [sandbox switching contract](../../../.agents/notes/implemented/feature/2026-07-06-sandbox.md).
 
 ## Model Experience
 
@@ -79,7 +79,7 @@ For sandboxing executors, each call resolves mode as one-shot escalation, then s
 
 #### What the model sees
 
-Every request in this plugin's registration scope contains the bash guidance below. A sandboxing executor adds no mode statement or switch notice. Scoped tool restrictions can hide the schemas without removing this independently registered section.
+Every request in this plugin's registration scope contains the bash guidance below. The policy owner contributes current sandbox state through its cache-safe runtime context rather than changing this section. Scoped tool restrictions can hide the schemas without removing this independently registered section.
 
 ##### Bash guidance
 

packages/bash/tool-bash/README.zh.md

@@ -71,7 +71,7 @@ overlay 根据当前 `ToolExecution` 计算，并通过专用的 `BashExecReques
 
 ## 逐会话模式切换
 
-对于启用沙箱的执行器，每次调用依次按单次升权、会话覆盖、执行器默认值解析模式。未启用沙箱以及没有 agent 的调用不携带会话覆盖。提示词和切换通知均不公布当前常驻模式；拒绝结果会在边界相关时报告有效模式。参见 [`dsh-bash` 整合](../bash/README.md)和[沙箱切换契约](../../../.agents/notes/implemented/feature/2026-07-06-sandbox.md)。
+对于启用沙箱的执行器，每次调用依次按单次升权、会话覆盖、执行器默认值解析模式。未启用沙箱以及没有 agent 的调用不携带会话覆盖。策略归属方贡献当前且不区分具体能力的常驻模式；拒绝结果仍负责操作特定的有效模式与重试引导。参见 [`dsh-bash` 整合](../bash/README.md)和[沙箱切换契约](../../../.agents/notes/implemented/feature/2026-07-06-sandbox.md)。
 
 ## 模型体验
 
@@ -79,7 +79,7 @@ overlay 根据当前 `ToolExecution` 计算，并通过专用的 `BashExecReques
 
 #### 模型看到的内容
 
-此插件注册作用域内的每个请求都包含下方 bash 指引。启用沙箱的执行器不会添加模式声明或切换通知。作用域工具限制可以隐藏 schema，但不会移除这个独立注册的段落。
+此插件注册作用域内的每个请求都包含下方 bash 指引。策略归属方通过自身的缓存安全运行时上下文贡献当前沙箱状态，而不改变此段落。作用域工具限制可以隐藏 schema，但不会移除这个独立注册的段落。
 
 ##### Bash 指引
 

packages/bash/tool-bash/src/index.ts

@@ -376,7 +376,6 @@ export function apply(ctx: Context, config: Config = {}): void {
   if (defaultMode !== undefined && sandboxPolicy === undefined) {
     throw new Error('tool-bash: the mounted bash executor confines but ctx.sandboxPolicy is missing')
   }
-
   /** Resolve the complete standing policy for this call when a confining executor is mounted. */
   const resolveSandboxPolicy = (exec: ToolExecution): SandboxExecutionPolicy | undefined =>
     sandboxPolicy?.resolve(exec.agent === undefined ? {} : { session: exec.agent.session })

packages/client/connection/src/client/fixture.ts

@@ -138,7 +138,107 @@ const TERMINAL_EXIT_STATUS: Record<string, { exitCode: number } | { signal: stri
 }
 
 /**
- * The structured `web_search` result view for fixture turn 66, authored inline
+ * Structured grep result for the search sample (turn 66): matches grouped by
+ * file, authored inline because the client-side fixture cannot import the tool
+ * that produces the canonical value. `truncated` with a larger `total` than the
+ * retained match count exercises the search card's capped indicator; the file
+ * with more than CHAT_SEARCH_MAX_LINES rows exercises its head/tail height cap.
+ */
+const SEARCH_MATCHES_FIXTURE: { path: string; matches: { lineNumber: number; line: string }[] }[] = [
+  {
+    path: 'packages/client/ui-primitives/src/SearchBlock.tsx',
+    matches: [
+      { lineNumber: 16, line: 'export const DEFAULT_SEARCH_MAX_LINES = 16' },
+      { lineNumber: 138, line: 'export function SearchBlock(props: SearchBlockProps) {' },
+      { lineNumber: 141, line: '  const [collapsed, setCollapsed] = useState<ReadonlySet<number>>(() => new Set())' },
+    ],
+  },
+  {
+    path: 'packages/client/ui-conversation/src/client/contract/search-card-model.ts',
+    matches: [
+      { lineNumber: 24, line: 'export const CHAT_SEARCH_MAX_LINES = 8' },
+      { lineNumber: 60, line: 'export function searchCardModel(block: ToolCallBlock): SearchCardModel | null {' },
+    ],
+  },
+  {
+    path: 'packages/client/ui-conversation/src/client/toolviews/search-row.tsx',
+    matches: [
+      { lineNumber: 71, line: 'export function SearchRow({ toolName, block }: ToolRowProps) {' },
+      { lineNumber: 73, line: '  const search = searchCardModel(block)' },
+      { lineNumber: 90, line: '        <SearchBlock {...search.card} maxLines={CHAT_SEARCH_MAX_LINES} className={css.search} />' },
+      { lineNumber: 113, line: "    ctx.slots.register({ name: 'conversation.chat.toolview', key: 'grep' }, SearchRow)" },
+    ],
+  },
+]
+
+/**
+ * The model-facing grep render text for the sample — what a UI without a search
+ * card shows, attached as the view's `content`. Mirrors the real grep
+ * presenter's shape (see formatGrepOutput in dsh-tool-fs-search): a
+ * `Found X of Y matches` header, the matches grouped under file headers with
+ * `Line N:` rows, then a spill-recovery footer.
+ */
+const SEARCH_MATCHES_TEXT = [
+  'Found 9 of 42 matches',
+  '',
+  ...SEARCH_MATCHES_FIXTURE.map(file =>
+    [file.path, ...file.matches.map(m => `Line ${m.lineNumber}: ${m.line}`)].join('\n')),
+  '',
+  '(Full grep result stored at: fixture://spill/grep-66. Read it to see every match.)',
+].join('\n')
+
+/**
+ * Structured glob result for the search sample (turn 67): a flat path list,
+ * truncated with a larger `total` so the path card shows its capped indicator.
+ */
+const SEARCH_PATHS_FIXTURE = [
+  'packages/client/ui-primitives/src/SearchBlock.tsx',
+  'packages/client/ui-primitives/src/SearchBlock.module.css',
+  'packages/client/ui-conversation/src/client/contract/search-card-model.ts',
+  'packages/client/ui-conversation/src/client/toolviews/search-row.tsx',
+  'packages/client/ui-conversation/src/client/toolviews/search-row.module.css',
+]
+
+/**
+ * The model-facing glob render text — the newline-joined path list plus a
+ * spill-recovery footer, mirroring the real glob presenter's shape (see
+ * formatGlobOutput in dsh-tool-fs-search).
+ */
+const SEARCH_PATHS_TEXT = [
+  ...SEARCH_PATHS_FIXTURE,
+  '',
+  '(Showing 5 of 23 paths. Full sorted result stored at: fixture://spill/glob-67. Read it to see every path.)',
+].join('\n')
+
+/**
+ * Read-card sample for the read turn: a WINDOW past an offset, so the line
+ * numbers start above 1 (the card's gutter keeps the file's own numbering) and
+ * `totalLines` exceeds the window (the card shows a "showing N of M" note). The
+ * fixture is client-side and cannot import the read tool, so the structured
+ * window is authored inline exactly as the tool would project it through
+ * `presentationMeta`. `lang` is a `ts` hint so the shiki path highlights it.
+ */
+const READ_SAMPLE_FIRST_LINE = 41
+const READ_SAMPLE_SOURCE = [
+  'export interface ReadBlockProps {',
+  '  label?: string | undefined',
+  '  lines: readonly ReadBlockLine[]',
+  '  totalLines: number',
+  '  lang?: string | undefined',
+  '  maxLines?: number | undefined',
+  '  className?: string | undefined',
+  '}',
+  '',
+  '// A windowed read keeps the file line numbers in the gutter.',
+  'const marker = "fixture read sample"',
+]
+const READ_SAMPLE_LINES = READ_SAMPLE_SOURCE.map((text, index) => ({ number: READ_SAMPLE_FIRST_LINE + index, text }))
+const READ_SAMPLE_PATH = 'packages/client/ui-primitives/src/ReadBlock.tsx'
+const READ_SAMPLE_TOTAL = 180
+const READ_SAMPLE_TEXT = READ_SAMPLE_SOURCE.map((text, index) => `${READ_SAMPLE_FIRST_LINE + index}: ${text}`).join('\n')
+
+/**
+ * The structured `web_search` result view for the web-search turn, authored inline
  * because this client-side fixture cannot import the web tool that projects it.
  * The sources exercise the citation list's features: a titled source with a
  * snippet and a date, a source with no title (its hostname labels the link) and
@@ -168,7 +268,7 @@ const WEB_SEARCH_RESULT: Omit<Extract<ToolResultView, { card: 'web'; kind: 'sear
   truncated: true,
 }
 
-/** The `web_fetch` result view for fixture turn 67, authored inline for the same reason. */
+/** The `web_fetch` result view for the web-fetch turn, authored inline for the same reason. */
 const WEB_FETCH_RESULT: Omit<Extract<ToolResultView, { card: 'web'; kind: 'fetch' }>, 'card' | 'kind'> = {
   url: 'https://www.deepseek.com/blog/harness-architecture',
   statusCode: 200,
@@ -315,8 +415,8 @@ function buildAlphaLog(): SessionEvent[] {
     const turn = 64
     const callId = `fx-call-${turn}`
     const program = 'const listing = await tools.bash({ command: "ls notes", description: "List notes" })\n'
-      + 'const demo = await tools.read({ path: "notes/demo.txt" })\n'
-      + 'await tools.read({ path: "notes/missing.txt" }).catch(() => "tolerated")\n'
+      + 'const demo = await tools.read({ file_path: "notes/demo.txt" })\n'
+      + 'await tools.read({ file_path: "notes/missing.txt" }).catch(() => "tolerated")\n'
       + 'return { listing, demo }'
     const args = JSON.stringify({ code: program, description: 'Read the notes files and summarize' })
     push({ type: 'turn/start', data: { turn, trigger: { kind: 'message', source: { kind: 'user' } } } })
@@ -341,8 +441,8 @@ function buildAlphaLog(): SessionEvent[] {
       })
     }
     dispatchPair(1, 'bash', { command: 'ls notes', description: 'List notes' }, 'demo.txt\nnew-demo.txt')
-    dispatchPair(2, 'read', { path: 'notes/demo.txt' }, 'hello fixture\n')
-    dispatchPair(3, 'read', { path: 'notes/missing.txt' }, 'Error: ENOENT: notes/missing.txt not found', true)
+    dispatchPair(2, 'read', { file_path: 'notes/demo.txt' }, 'hello fixture\n')
+    dispatchPair(3, 'read', { file_path: 'notes/missing.txt' }, 'Error: ENOENT: notes/missing.txt not found', true)
     push({
       type: 'tool/result', surfaceOp: 'append',
       data: { turn, step: 0, message: toolResultMessage(callId, text('{"listing":"demo.txt\\nnew-demo.txt","demo":"hello fixture\\n"}'), false) },
@@ -350,7 +450,7 @@ function buildAlphaLog(): SessionEvent[] {
     push({ type: 'step/end', data: { turn, step: 0 } })
     push({ type: 'turn/end', data: { turn, reason: { kind: 'completed' } } })
   }
-  // Turn 65: todo_write sample — the TodoRow toolview in the flow plus the
+  // Turn 67: todo_write sample — the TodoRow toolview in the flow plus the
   // todo/write snapshot event feeding the TodoPanel plan strip.
   const fixtureTodos = [
     { content: '梳理需求', status: 'completed' },
@@ -371,7 +471,30 @@ function buildAlphaLog(): SessionEvent[] {
   // strip empty and take the todo surfaces' own coverage with it.
   toolTurn(65, 'bash', '{"command":"pnpm run check","cwd":"/tmp/fixture/deep/nested"}', TERMINAL_OUTPUT_FIXTURE)
 
-  // Turns 66-67: the web render intent — a web_search whose result view carries
+  // Turns 66-67: the search card's two shapes. `grep` emits a `card: 'search'`
+  // `shape: 'matches'` result view (grouped-by-file matches, truncated with a
+  // larger `total`), `glob` emits `shape: 'paths'` (a flat path list, likewise
+  // truncated). Both ride the keyed SearchRow registration under their own
+  // names; the render-site fallback row is covered by the model derivation
+  // tests, since every fixture search tool has a keyed row. Ordered before the
+  // todo turn for the same standing-plan reason the bash turn is.
+  toolTurn(66, 'grep', '{"pattern":"SEARCH_MAX_LINES","path":"packages/client"}', SEARCH_MATCHES_TEXT)
+  toolTurn(67, 'glob', '{"pattern":"**/SearchBlock*","path":"packages/client"}', SEARCH_PATHS_TEXT)
+
+  // Turn 68: the read sample — a WINDOW past an offset so the card draws file
+  // line numbers starting above 1 and a "showing N of M" note (the window is
+  // shorter than READ_SAMPLE_TOTAL), with a `ts` language hint the shiki path
+  // highlights. Named `read`, so it exercises the keyed ReadRow registration.
+  // The render-site fallback ROW SHAPE (a read call on the generic flattened
+  // path) is covered by the turn 64 run_code read sub-dispatches, which
+  // session.ts folds with resultView: null; the fallback-row + read-CARD
+  // combination is pinned by the web_fetch case in read-card.spec.tsx, not by
+  // this fixture. The read render intent is result-side only, so its pending
+  // call stays a generic `kind: 'read'` card; presentResult carries the
+  // structured window.
+  toolTurn(68, 'read', `{"file_path":${JSON.stringify(READ_SAMPLE_PATH)},"offset":${READ_SAMPLE_FIRST_LINE}}`, READ_SAMPLE_TEXT)
+
+  // Turns 69-70: the web render intent — a web_search whose result view carries
   // structured sources plus an answer (the citation list, one source lacking a
   // title so its hostname labels the link, the capped indicator on), and a
   // web_fetch whose result view carries the fetched URL and its HTTP status.
@@ -380,11 +503,11 @@ function buildAlphaLog(): SessionEvent[] {
   // the real tools so they hit the keyed WebRow registration. Ordered BEFORE
   // the todo turn for the same reason turn 65 is: the standing plan retires at
   // the next turn/start, so a turn after it would empty the dock's plan strip.
-  toolTurn(66, 'web_search', '{"query":"deepseek harness architecture"}', 'Search results for deepseek harness architecture.')
-  toolTurn(67, 'web_fetch', '{"url":"https://www.deepseek.com/blog/harness-architecture"}', '# Harness architecture\n\nEverything is a plugin.')
+  toolTurn(69, 'web_search', '{"query":"deepseek harness architecture"}', 'Search results for deepseek harness architecture.')
+  toolTurn(70, 'web_fetch', '{"url":"https://www.deepseek.com/blog/harness-architecture"}', '# Harness architecture\n\nEverything is a plugin.')
 
   const todoArgs = JSON.stringify({ todos: fixtureTodos })
-  toolTurn(68, 'todo_write', todoArgs, 'Updated todo list: 1 pending, 1 in progress, 1 completed.')
+  toolTurn(71, 'todo_write', todoArgs, 'Updated todo list: 1 pending, 1 in progress, 1 completed.')
   // The real tool appends the snapshot mid-execution — between tool/call and
   // tool/result — so the fixture reproduces that exact ordering (the last
   // toolTurn events run ... tool/call, tool/result, step/end, turn/end).
@@ -419,6 +542,12 @@ function presentCall(name: string, argsRaw: string): ToolCallView | undefined {
         card: 'diff', title: `Write ${str(args.path)}`,
         diffs: [{ path: str(args.path), oldText: null, newText: str(args.content) }],
       }
+    // A read pending call is a GENERIC card (kind: 'read', a follow-along
+    // location): the read render intent is result-side only, because a call
+    // carries no file content until execute returns. The rich read card arrives
+    // in presentResult.
+    case 'read':
+      return { card: 'generic', title: `Read ${str(args.file_path)}`, kind: 'read', locations: [{ path: str(args.file_path) }] }
     case 'edit':
       // The multi-hunk sample (turn 67) is keyed on its file_path, so the two
       // scattered hunks share one path header and the card draws the `⋯` gap.
@@ -440,6 +569,13 @@ function presentCall(name: string, argsRaw: string): ToolCallView | undefined {
         card: 'diff', title: `Write ${str(args.file_path)}`,
         diffs: [{ path: str(args.file_path), oldText: null, newText: str(args.content) }],
       }
+    // A search call stays a generic card (kind: 'search'): the structured
+    // matches/paths exist only after execute, so the search card is result-time
+    // only (presentResult builds it). This mirrors the real grep/glob presenters.
+    case 'grep':
+      return { card: 'generic', title: `Grep ${str(args.pattern)}`, kind: 'search', rawInput: args }
+    case 'glob':
+      return { card: 'generic', title: `Glob ${str(args.pattern)}`, kind: 'search', rawInput: args }
     // The web tools keep a GENERIC pending card and add the `web` result card
     // only at result time (the contract's result-only web shape); their pending
     // kind matches the result kind so a call and its result read as one category.
@@ -455,6 +591,28 @@ function presentCall(name: string, argsRaw: string): ToolCallView | undefined {
 function presentResult(name: string, argsRaw: string, resultText: string): ToolResultView | undefined {
   const call = presentCall(name, argsRaw)
   if (call === undefined) return undefined
+  // Search is result-time only: the call stays a generic search card, and the
+  // result view carries the structured shape the card renders. The view holds no
+  // result text — a UI without a search card falls back to the raw tool/result
+  // content — so the truncation recovery footer rides that raw content (the
+  // `toolTurn` message text), not the view. `total` exceeds the retained count so
+  // the card shows its capped indicator.
+  if (name === 'grep') {
+    return { card: 'search', shape: 'matches', files: SEARCH_MATCHES_FIXTURE, truncated: true, total: 42 }
+  }
+  if (name === 'glob') {
+    return { card: 'search', shape: 'paths', paths: SEARCH_PATHS_FIXTURE, truncated: true, total: 23 }
+  }
+  // The read result is the structured window the tool projects through
+  // `presentationMeta`; the fixture authors it inline (it cannot import the
+  // tool). Keyed on the name because the read pending call is a generic card,
+  // so `call.card` alone does not distinguish it from edit/write.
+  if (name === 'read') {
+    return {
+      card: 'read', path: READ_SAMPLE_PATH, offset: READ_SAMPLE_FIRST_LINE, lines: READ_SAMPLE_LINES,
+      totalLines: READ_SAMPLE_TOTAL, lang: 'ts', content: text(resultText),
+    }
+  }
   // The web tools keep a generic pending card, so their result card is chosen
   // by tool name rather than by the pending card tag: the structured `web` card
   // the frontend consumes. The view carries no `content` copy (per the contract

packages/client/ui-conversation/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/client/ui-conversation/README.md
-README.md: b183aab56a10c3a22eeab31f8b1b0e6b5219beba
-README.zh.md: af6427774a048de324d57bc9856f77547e9a03cc
+README.md: dea051fa608119ec1386299d48a1be8aef23aaec
+README.zh.md: ae4537b899f44badba5d7e2051434bbbddb410c3

packages/client/ui-conversation/README.md

@@ -22,6 +22,8 @@ A tool call declaring the `diff` render intent (the `write`/`edit` tools) render
 
 The chat flow projects consecutive model-retry nodes across retry turns into one stable, muted status row updated to the latest attempt; every retry event remains in the runtime snapshot and session log. Its frontend countdown anchors the scheduled delay to client receipt, avoiding host/browser clock skew, rounds remaining time up to seconds, and has a one-second floor. The latest unresolved retry uses a left-to-right text shimmer. Subsequent turn facts distinguish an attempt that started from one cancelled during backoff, while the Host running bit only controls the live animation; the row then shows a static completed or cancelled label. Normal policy rows show the finite retry maximum; always policy rows show `∞`. Activating the row reveals the latest exact retry delay and failure message. The client runtime removes each failed step's streaming tail before its retry node arrives, while the status remains visible after a later attempt succeeds.
 
+A `grep`/`glob` call declaring the `search` render intent renders its result inline, at the same render sites, through ui-primitives' `SearchBlock` — grep's matches grouped by file (each a collapsible header of `lineNumber: line` rows), glob's flat path list. `contract/search-card-model.ts` is the single derivation from the snapshot's `resultView`; unlike the terminal card it reads no `callView`, since a search has no matches or paths before `execute`, so a running search shows its summary alone. It yields null — the generic path — for any non-search result view, a `card` or `kind` this client version does not compile, and (because those ride the untrusted wire frame) a known kind whose `files`/`paths` is malformed. The keyed `SearchRow`, registered under both `grep` and `glob` since the derived `kind` decides the shape, carries the card resident below its summary; the render-site fallback keeps it behind the expand control. Both cap at `CHAT_SEARCH_MAX_LINES` (8) against the panel's 16. A capped search drops rows from the card, but the locator to the rest — grep/glob's `Full … stored at …` footer — lives only in the result text, so the derivation surfaces that as a recovery footer below the card when (and only when) the result was truncated; a settled call with no card at all (an errored search, a nested `run_code` sub-dispatch, a legacy generic result) falls back to its flattened result text so nothing is lost behind a bare summary ([decision](../../../.agents/notes/implemented/feature/2026-07-30-web-search-card.md)).
+
 Tool rows are slots too — the standalone tool ring (`ToolViewRegistry`/`ctx.toolviews`/outlet) is retired. The chat entry declares the keyed `'conversation.chat.toolview'` hole (session scope; the key space is runtime-open); its render site dispatches per row via `entryKey: toolName` with `GenericToolCard` as the call-site `fallback`. The owner payload is the uniform `ToolRowOwnerProps` (`callId`/`toolName`/`block`/`openFile`) and `ToolRowProps` pre-composes it with the session standard kit. A registrant is a plain plugin: `ctx.slots.register({ name: 'conversation.chat.toolview', key: '<tool>', inject? }, Row)` with `inject: ['slots', 'conversation']` as the load-order seam (apply mounts ConversationService after the chat registration, so the service being present guarantees the slot is declared); session differentiation happens inside the component (`useSessions` reading `parentId` — the bash sample is the third-party-posture exemplar). Trajectory/waterfall toolview slots share this shape and land with their own render sites (RendersCheck rejects a declaration nobody renders).
 
 The todo surfaces are two registrations over that shape, both plain registrant plugins with `inject: ['slots', 'conversation']`. `TodoRow` takes the `'conversation.chat.toolview'` key `todo_write` and summarizes what the call attempted (`<done>/<total> 已完成 · <active item>` parsed from its args, falling back to the generic summary on malformed or wrongly-shaped model JSON, and keeping the generic dot for non-ok execution states so a cancelled call never reads as a completed update). `TodoDock` takes the `'conversation.input.dock'` list slot at `order: -1` — above the queue rows — and is the plan strip: it reads the host-computed `todos` projection via `useProjection` (standing plan: latest `todo/write` with no later `turn/start`) and renders `TodoPanel`, which takes the plain list, hides itself while the list is empty, and starts collapsed as a header of title plus `"<done>/<total> tasks · <n> in progress"` (status glyphs are the figma check / progress / dashed-pending set). The dock adapter owns the selection so the panel stays a pure function of its props; the standing list lives here rather than in the row so the row stays one line. Anything the input-zone composer chain hides (a `conversation.composer` takeover such as ui-question's) hides the whole dock, this strip included.
@@ -47,6 +49,7 @@ None; this package neither assembles nor sends a provider request.
 - **Stats-line durations cover the in-window flow only** — LLM and tool wall times fold the snapshot's assistant `timing` and tool call/result pairs, so nodes outside the loaded event window (older history) are not counted.
 - **Details panel is the minimal form and currently has no entry point** — selected call args/result raw display; the Input/Output/Metadata switch, Prev/Next stepping, and See-in-trajectory deep link are deferred. Tool rows stopped being details-panel click targets and nothing replaced that gesture, so `ChatViewInjected.openDetails` is implemented but uncalled and the panel (including its terminal card) is unreachable in the assembled application; its rendering stays covered by mounting it with a selection directly.
 - **Assistant per-message paging is a reserved slot** — drawn in the design, not implemented. The finalized content IconActions row (copy / branch / clock) ships under the last content-text assistant of each turn only; mid-turn narration and Think-only nodes stay chrome-free. Branch forks through the turn containing that message, increments the inherited title on the client, and then opens the child, while a fork or rename failure leaves the source selected.
+- **Sent user messages cannot be edited** — the user bubble's IconActions row carries clock / copy / branch only, and branching from the message is the nearest gesture. The control returns with the capability behind it: a client mutation over a settled user message, plus the host behavior for the turn that already consumed it ([decision](../../../.agents/notes/implemented/simplification/2026-07-31-drop-user-message-edit-stub.md)).
 - **The sparkle icon for the others tool row is a hand-drawn approximation** — the design glyph's vector geometry is not exportable locally; promotion into ui-primitives waits on an exact export.
 - **The approval panel's "Always allow this type" is deferred** — durable grants need a grant-storage design; only allow-once/reject answer today.
 - **TodoPanel truncates long item text to one ellipsized line** — the figma strip has no wrap or expand affordance; full text is not readable inline.

packages/client/ui-conversation/README.zh.md

@@ -20,6 +20,8 @@
 
 聊天流会将跨重试轮次连续出现的模型重试节点投影为一个稳定的弱化状态行，并用最新一次尝试更新该行；每个重试事件仍保留在运行时快照与会话日志中。前端倒计时以客户端收到事件的时刻为计划延迟的起点，避免 Host 与浏览器的时钟偏差；剩余时间向上取整到秒，且下限为 1 秒。最近一次尚未完成的重试会显示从左到右的文字渐变动画。后续轮次事实用于区分已开始的尝试与在退避期间取消的尝试，Host 的 running 位只控制实时动画；随后该行会显示静态的已完成或已取消标签。normal 策略行显示有限重试上限；always 策略行显示 `∞`。激活该行会显示最近一次重试的精确延迟和失败消息。客户端运行时会在相应重试节点到达前移除每个失败步骤的流式输出尾部；后续某次尝试成功后，该状态仍保持可见。
 
+声明 `search` 渲染意图的 `grep`／`glob` 调用，会在同样的渲染点上通过 ui-primitives 的 `SearchBlock` 内联渲染其结果——grep 的匹配按文件分组（每个是一个可折叠的头，下辖 `lineNumber: line` 行），glob 是扁平路径列表。`contract/search-card-model.ts` 是从快照的 `resultView` 推导的唯一位置；与终端卡片不同，它不读 `callView`，因为搜索在 `execute` 前没有匹配或路径，所以运行中的搜索只显示摘要。对任何非搜索的结果视图、当前客户端版本无法编译的 `card` 或 `kind`、以及（因为这些都与不可信的 wire 帧同行）一个 `files`／`paths` 格式错误的已知 kind，它都返回 null，落回通用路径。键控的 `SearchRow` 因推导出的 `kind` 决定形态而同时注册在 `grep` 与 `glob` 下，把卡片常驻在摘要行下方；渲染点兜底行则把它保持在展开控件之后。两者上限都是 `CHAT_SEARCH_MAX_LINES`（8），面板为 16。被截断的搜索会从卡片里丢掉一些行，但通往其余部分的定位符——grep/glob 的 `Full … stored at …` 脚注——只存在于结果文本里，因此推导在（且仅在）结果被截断时把它作为恢复脚注画在卡片下方；一个完全没有卡片的已结算调用（出错的搜索、嵌套 `run_code` 子派发、旧日志的 generic 结果）则回退到其压平后的结果文本，从而不让任何内容丢失在一个光秃秃的摘要之后（[决策](../../../.agents/notes/implemented/feature/2026-07-30-web-search-card.md)）。
+
 工具行同样是 slot：独立工具环（`ToolViewRegistry`／`ctx.toolviews`／outlet）已经退役。聊天配置项声明键控的 `'conversation.chat.toolview'` 空位（Session scope；key 空间在运行时开放）；其渲染点逐行通过 `entryKey: toolName` 分发，并以 `GenericToolCard` 作为调用点 `fallback`。owner 载荷是统一的 `ToolRowOwnerProps`（`callId`／`toolName`／`block`／`openFile`），`ToolRowProps` 则预先将其与 Session 标准工具包组合。注册方只是普通插件：`ctx.slots.register({ name: 'conversation.chat.toolview', key: '<tool>', inject? }, Row)`，以 `inject: ['slots', 'conversation']` 作为加载顺序 seam（apply 在聊天注册后挂载 ConversationService，因此服务存在即可保证 slot 已声明）；Session 区分在组件内部完成（`useSessions` 读取 `parentId`，bash 示例是第三方姿态的范例）。Trajectory/waterfall 工具视图 slot 共享此形状，并随各自的渲染点落地（RendersCheck 会拒绝没有任何渲染方的声明）。
 
 审批经由本包声明的链接管编辑器：`ApprovalPanel` 注册为按选择器路由的 `'conversation.composer'` 配置项（ui-question 模式），在审批等待未决期间取代 InputBar 占据编辑器（琥珀色条、理由标题、来自运行中调用参数的配对命令行、一次性的拒绝／允许）。`contract/slots.ts` 中的 `PendingApproval` 领域面在运行时 `PendingWait` 载体之上拥有 wire 编码——带审计关联的 `ApprovalResponsePayload` 值；广播的 `approval/resolved` 帧使等待落定并恢复编辑器。侧边栏通过 manager 跟踪的 `waitingApproval` 列表位（未实例化会话同样点亮）镜像该阻塞状态，其优先级高于运行中圆环，直至问题解决。未决等待完全离开消息流：问题（ui-question）与审批（ApprovalPanel）都经编辑器接管作答，不再保留只读占位卡。编辑器底行的 Access 席位挂载 `PermissionSelect`，由 host 计算的 `permissions` 投影经标准工具包 `useProjection` 供数（key 缺席即隐藏 chip）；chip 打开 Menu 原语下拉，普通安全预设会立即经输入栏注入的 `command` 回调提交 `/permission <preset>`，而 `danger-full-access` 在界面中显示为 `Full access`，选择后先打开页面内的 Modal 风险确认。用户勾选确认项前启用按钮始终不可用；取消、Escape、关闭按钮与点击遮罩都不会提交命令。
@@ -47,6 +49,7 @@ todo 两个面就是在该形状上的两个注册项，都是普通注册方插
 - **统计行的耗时只覆盖窗口内消息流**：LLM（大语言模型）与工具墙钟时间由快照的 assistant `timing` 与工具 call/result 配对折算，落在已加载事件窗口之外的节点（更早的历史）不计入。
 - **详情面板是最小形态，且当前没有入口**：以原始形式显示已选择调用的参数／结果；Input/Output/Metadata 切换、Prev/Next 步进与 See-in-trajectory 深链接暂缓实现。工具行已不再是详情面板的点击目标，且没有任何手势接替它，因此 `ChatViewInjected.openDetails` 虽已实现却无人调用，该面板（含其终端卡片）在组装后的应用中不可达；其渲染仍由直接以选中态挂载它来覆盖。
 - **assistant 逐消息分页是预留 slot**：设计中已有图稿，尚未实现。已定稿的内容 IconActions 行（复制／分支／时钟）只挂在每个轮次中最后一条带 text 内容的 assistant 下；轮次中间的叙述与纯 Think 节点不带 chrome。分支会 fork 到包含该消息的轮次末尾，在 client 端递增继承标题后打开子会话，而 fork 或改名失败时源会话保持选中。
+- **已发送的 user 消息无法编辑**：user 气泡的 IconActions 行只有时钟／复制／分支，从该消息分支是最接近的手势。该控件要与其背后的能力一起回归：既需要针对已定稿 user 消息的 client 变更，也需要 host 侧对已经消费过它的轮次给出行为（[决策](../../../.agents/notes/implemented/simplification/2026-07-31-drop-user-message-edit-stub.md)）。
 - **others 工具行的闪光图标是手绘近似版本**：无法在本地导出设计字形的矢量几何；等到存在精确导出后再将其提升到 ui-primitives。
 - **审批面板的「始终允许此类」暂缓**：持久授权需要授权存储设计；今天只能回答允许一次／拒绝。
 - **TodoPanel 将过长条目截成单行省略号**：figma 条没有换行或展开入口，完整文本无法在行内读完。

packages/client/ui-conversation/src/client/apply.ts

@@ -20,6 +20,8 @@ import { InputBar } from './skeleton/InputBar.tsx'
 import { ChatView } from './chat/ChatView.tsx'
 import { StatsLine } from './chat/StatsLine.tsx'
 import { bashToolviewSample } from './toolviews/bash-sample.tsx'
+import { searchToolview } from './toolviews/search-row.tsx'
+import { readToolview } from './toolviews/read-row.tsx'
 import { fileMutationToolview } from './toolviews/file-mutation-row.tsx'
 import { webToolview } from './toolviews/web-row.tsx'
 import { ApprovalPanel } from './skeleton/ApprovalPanel.tsx'
@@ -320,6 +322,14 @@ export function apply(ctx: Context): void {
   // (ToolRow-matching Bash · {description} chrome; scoped badge in child sessions).
   ctx.plugin(bashToolviewSample)
 
+  // The grep/glob search row rides the same seam: one component registered
+  // under both tool names, since both declare the same search render intent.
+  ctx.plugin(searchToolview)
+
+  // The read row rides the same seam (a product registration, not a sample):
+  // Read · {path} chrome with the file's read card resident below it.
+  ctx.plugin(readToolview)
+
   // The write/edit rows ride the same seam: a file-mutation call declares the
   // diff render intent, so these rows stack the applied diff card under their
   // path-link summary (the terminal card's posture, applied to diffs).

packages/client/ui-conversation/src/client/chat/GenericToolCard.module.css

@@ -1,7 +1,8 @@
-/* The generic card grows a resident web card under its summary row when the
-   tool declares the `web` render intent but has no keyed row of its own (the
-   web_search/web_fetch rows register their own WebRow). A column around the
-   ToolRow keeps the row's own 24px height. */
+/* GenericToolCard resident cards: a read-declaring or web-declaring tool
+   without its own keyed row (e.g. web_fetch) grows a resident card under its
+   summary row. A column around the ToolRow keeps the row's own 24px height, so
+   the read card renders identically to the keyed ReadRow and the web card to
+   the web_search/web_fetch WebRow. */
 
 .card {
   display: flex;
@@ -10,6 +11,7 @@
 
 /* Row indentation matches ToolRow's expanded bodies (16px leading + 6px gap),
    and replaces the primitive's standalone vertical margin with the flow's. */
+.read,
 .web {
   margin: 4px 0 4px 22px;
 }

packages/client/ui-conversation/src/client/chat/GenericToolCard.tsx

@@ -7,9 +7,11 @@
 import type { ReactNode } from 'react'
 import {
   IconApiOutline14, IconBrowseOutline16, IconCodeOutline16, IconEditOutline16, IconSearchOutline16, IconSparkle16,
-  IconThinkOutline14, WebBlock,
+  IconThinkOutline14, ReadBlock, WebBlock,
 } from '@deepseek-ai/dsh-client-ui-primitives'
 import type { ChatViewSlotProps, ToolRowOwnerProps } from '../contract/slots.ts'
+import { searchCardModel } from '../contract/search-card-model.ts'
+import { CHAT_READ_MAX_LINES, readCardModel } from '../contract/read-card-model.ts'
 import { diffCardModel } from '../contract/diff-card-model.ts'
 import { terminalCardModel, terminalFailed } from '../contract/terminal-card-model.ts'
 import { CHAT_WEB_MAX_SOURCES, webCardModel } from '../contract/web-card-model.ts'
@@ -37,6 +39,8 @@ export interface GenericToolCardProps extends ToolRowOwnerProps {
 export function GenericToolCard({ toolName, block, cwd, openFile, inspect, t }: GenericToolCardProps) {
   const model = toolRowModel(toolName, block, cwd)
   const terminal = terminalCardModel(block, cwd)
+  const search = searchCardModel(block)
+  const read = readCardModel(block, cwd)
   const diff = diffCardModel(block)
   const web = webCardModel(block)
   // A failing exit status is the terminal card's own error signal (the call
@@ -53,8 +57,9 @@ export function GenericToolCard({ toolName, block, cwd, openFile, inspect, t }:
       icon={VARIANT_ICONS[model.variant]}
       title={model.title}
       // A terminal presenter's description is the contract's above-card text, so
-      // it outranks the args-derived summary here exactly as it does in BashRow.
-      summary={terminal?.description ?? model.summary}
+      // it outranks the args-derived summary here exactly as it does in BashRow;
+      // a search result view's replacement title outranks it the same way.
+      summary={terminal?.description ?? search?.title ?? model.summary}
       // Single-file tools never expose an args body — the path link is the only
       // args interaction. A diff card is not an args body: a write/edit row is
       // single-file AND carries a diff, so the card expands under the path link.
@@ -62,6 +67,7 @@ export function GenericToolCard({ toolName, block, cwd, openFile, inspect, t }:
       output={model.output}
       errorSummary={model.errorSummary}
       terminal={terminal}
+      search={search}
       diff={diff}
       state={state}
       filePath={model.filePath}
@@ -69,6 +75,18 @@ export function GenericToolCard({ toolName, block, cwd, openFile, inspect, t }:
       inspect={inspect}
     />
   )
+  // A read-declaring tool without its own keyed row lands here (e.g. web_fetch),
+  // so the file's read card is resident below the summary row exactly as the
+  // keyed ReadRow draws it. Only wrap when a card is present, so every other
+  // tool keeps the bare ToolRow.
+  if (read !== null) {
+    return (
+      <div className={css.card}>
+        {row}
+        <ReadBlock {...read} maxLines={CHAT_READ_MAX_LINES} className={css.read} />
+      </div>
+    )
+  }
   // A web-declaring tool without its own keyed row lands here; its card is
   // resident under the summary, mirroring WebRow (and BashRow's terminal card).
   if (web === null) return row

packages/client/ui-conversation/src/client/chat/MessageIconActions.tsx

@@ -1,10 +1,9 @@
 // Shared IconActions chrome for user and assistant messages: copy live,
-// branch wired through onBranch, date-aware clock,
-// optional edit stub.
+// branch wired through onBranch, date-aware clock.
 
 import { useCallback } from 'react'
 import {
-  IconBranchOutline16, IconCopyOutline16, IconEditOutline16, Tooltip,
+  IconBranchOutline16, IconCopyOutline16, Tooltip,
 } from '@deepseek-ai/dsh-client-ui-primitives'
 import type { ChatViewSlotProps } from '../contract/slots.ts'
 import { formatMessageClock, writeClipboard } from './message-chrome.ts'
@@ -18,8 +17,6 @@ export interface MessageIconActionsProps {
   time: number
   /** Clock before icons (user) or after (assistant). */
   clock: 'start' | 'end'
-  /** When true, append the stub edit control (user bubble). */
-  edit?: boolean | undefined
   /** Fork the session at this message. */
   onBranch?: (() => void) | undefined
   /** Parent layout class composed onto the actions row. */
@@ -30,11 +27,11 @@ export interface MessageIconActionsProps {
 
 /**
  * Copy / branch (/ clock) IconActions row shared by user and assistant chrome.
- * @param props - Copy text, event time, clock side, optional edit, branch callback, className.
+ * @param props - Copy text, event time, clock side, branch callback, className.
  * @returns The actions row element.
  */
 export function MessageIconActions({
-  text, time, clock, edit, onBranch, className, t,
+  text, time, clock, onBranch, className, t,
 }: MessageIconActionsProps) {
   const day = useCalendarDay()
   const onCopy = useCallback(() => {
@@ -58,13 +55,6 @@ export function MessageIconActions({
           <IconBranchOutline16 />
         </button>
       </Tooltip>
-      {edit === true && (
-        <Tooltip label={t('edit')} side="bottom">
-          <button type="button" className={css.action} aria-label={t('edit')}>
-            <IconEditOutline16 />
-          </button>
-        </Tooltip>
-      )}
       {clock === 'end' ? clockEl : null}
     </div>
   )

packages/client/ui-conversation/src/client/chat/MessageItem.tsx

@@ -1,5 +1,5 @@
 // MessageItem: simple chat nodes — user bubble (right-aligned, with
-// clock + copy / branch / edit IconActions), steering (badged bubble), context
+// clock + copy / branch IconActions), steering (badged bubble), context
 // injection, retry disclosure, and unknown-surface JSON rows.
 
 import { memo, useEffect, useMemo, useState } from 'react'
@@ -156,7 +156,6 @@ export const MessageItem = memo(function MessageItem({
             text={text}
             time={node.time}
             clock="start"
-            edit
             onBranch={onFork === undefined ? undefined : () => { onFork(node.seq) }}
             className={css.actions}
             t={t}

packages/client/ui-conversation/src/client/chat/ToolRow.module.css

@@ -246,17 +246,29 @@
   color: var(--dsw-alias-state-error-primary);
 }
 
-/* The two block-shaped expanded bodies: the code variant's run_code program
-   through CodeBlock (shiki-highlighted TypeScript) and a terminal card's
-   command output through TerminalBlock. Both are drawn by the shared
-   primitive, so only the row's indentation is this file's concern — the margin
-   also replaces each primitive's own standalone vertical spacing with the
-   flow's row rhythm. */
+/* The block-shaped expanded bodies: the code variant's run_code program
+   through CodeBlock (shiki-highlighted TypeScript), a terminal card's command
+   output through TerminalBlock, and a search card's grouped matches or path
+   list through SearchBlock. All are drawn by the shared primitive, so only the
+   row's indentation is this file's concern — the margin also replaces each
+   primitive's own standalone vertical spacing with the flow's row rhythm. */
 .codeBody,
-.terminalBody {
+.terminalBody,
+.searchBody {
   margin: 4px 0 4px 4px;
 }
 
+/* The recovery footer for a capped search: the result text (its `Full … stored
+   at …` locator) below the card in the muted tone, since the card holds only the
+   retained rows. Same column indent as the card body. */
+.searchRecovery {
+  margin: 4px 0 4px 4px;
+  white-space: pre-wrap;
+  overflow-wrap: anywhere;
+  font: var(--dsw-font-xs-13);
+  color: var(--dsw-alias-label-tertiary);
+}
+
 /* A write/edit diff renders through DiffBlock; like the terminal card it draws
    its own surface, so only the row indentation is this file's concern. */
 .diffBody {

packages/client/ui-conversation/src/client/chat/ToolRow.tsx

@@ -3,23 +3,25 @@
 // separator dot + FILL-truncated summary, drawn through the shared
 // DisclosureRow chrome with the whole row as the expand toggle (click /
 // Enter / Space, icon→chevron hover preview). The collapsed row is always
-// one line; every row with body, output, or terminal material is expandable;
-// the summary stays inline while open, except Think, whose body opens with
-// the same first line and would repeat it.
+// one line; every row with body, output, terminal, or search material is
+// expandable; the summary stays inline while open, except Think, whose body
+// opens with the same first line and would repeat it.
 // The expanded body — an IN/OUT gutter-labeled card (figma 1249:35657) for
-// text input/output, the run_code program through CodeBlock, or a terminal
-// card's command output through TerminalBlock — lives in a max-height scroll
-// container so a long payload scrolls internally instead of taking over the
-// message flow; Think's prose is the exception and flows uncapped like
-// message text. Expand state is component-local view state. File-tool
-// summaries are path links that open through the host (stopPropagation keeps
-// the two gestures independent); an error row's collapsed summary is the
+// text input/output, the run_code program through CodeBlock, a terminal
+// card's command output through TerminalBlock, or a search card's grouped
+// matches / path list through SearchBlock (capped at CHAT_SEARCH_MAX_LINES) —
+// lives in a max-height scroll container so a long payload scrolls internally
+// instead of taking over the message flow; Think's prose is the exception and
+// flows uncapped like message text. Expand state is component-local view state.
+// File-tool summaries are path links that open through the host (stopPropagation
+// keeps the two gestures independent); an error row's collapsed summary is the
 // failure's first line in the error color.
 
 import { useState, type MouseEvent, type ReactNode } from 'react'
 import clsx from 'clsx'
-import { CodeBlock, DiffBlock, StateDot, TerminalBlock } from '@deepseek-ai/dsh-client-ui-primitives'
+import { CodeBlock, DiffBlock, SearchBlock, StateDot, TerminalBlock } from '@deepseek-ai/dsh-client-ui-primitives'
 import type { TranslateNS } from '@deepseek-ai/dsh-client-ui-slots'
+import { CHAT_SEARCH_MAX_LINES, type SearchCardModel } from '../contract/search-card-model.ts'
 import { CHAT_DIFF_MAX_LINES, type DiffCardModel } from '../contract/diff-card-model.ts'
 import { terminalBlockLabels, type TerminalCardModel } from '../contract/terminal-card-model.ts'
 import type { ToolRowState, ToolRowVariant } from '../contract/tool-call-model.ts'
@@ -49,10 +51,17 @@ export interface ToolRowProps {
    * expandable.
    */
   terminal?: TerminalCardModel | null | undefined
+  /**
+   * Search-card material for a call whose render intent is a search card
+   * (derived by `searchCardModel`); it replaces the text body when present.
+   * Null or absent leaves the text body. A call carries at most one card kind,
+   * so `terminal`, `search`, and `diff` are never both present on the same row.
+   */
+  search?: SearchCardModel | null | undefined
   /**
    * Diff-card material for a call whose render intent is a diff card (derived by
    * `diffCardModel`); it replaces the text body when present, the same way
-   * `terminal` does. A call carries at most one card intent, so the two are
+   * `terminal` does. A call carries at most one card intent, so the cards are
    * never both set.
    */
   diff?: DiffCardModel | null | undefined
@@ -103,6 +112,7 @@ export function ToolRow({
   output,
   errorSummary,
   terminal,
+  search,
   diff,
   state,
   filePath,
@@ -111,9 +121,12 @@ export function ToolRow({
 }: ToolRowProps) {
   const [expanded, setExpanded] = useState(false)
   const terminalBody = terminal ?? null
+  const searchBody = search ?? null
   const diffBody = diff ?? null
   const outputText = output ?? null
-  const expandable = body !== null || outputText !== null || terminalBody !== null || diffBody !== null
+  // A search or diff card replaces the text body; a call carries at most one
+  // card kind, so terminal, search, and diff are never both present on a row.
+  const expandable = body !== null || outputText !== null || terminalBody !== null || searchBody !== null || diffBody !== null
   const open = expanded && expandable
   // An error row's collapsed summary IS the failure: the first error line in
   // the error color outranks both the args summary and a terminal description.
@@ -185,40 +198,51 @@ export function ToolRow({
                 className={css.terminalBody}
               />
             )
-            : diffBody !== null
-              ? <DiffBlock {...diffBody.card} maxLines={CHAT_DIFF_MAX_LINES} className={css.diffBody} />
-              : isThink
-                ? <div className={css.thinkBody}>{body}</div>
-                : (
-                  <>
-                    {variant === 'code' && body !== null && (
-                      <div className={css.bodyScroll}>
-                        <CodeBlock code={body} lang="typescript" copyLabel={t('copy')} copiedLabel={t('copied')} className={css.codeBody} />
-                      </div>
-                    )}
-                    {(cardBody !== null || outputText !== null) && (
-                      <div className={css.ioCard}>
-                        {cardBody !== null && (
-                          <div className={css.ioSection}>
-                            <span className={css.ioLabel}>IN</span>
-                            <span className={css.ioText}>{cardBody}</span>
-                          </div>
-                        )}
-                        {cardBody !== null && outputText !== null && (
-                          <span className={css.ioDivider} aria-hidden />
-                        )}
-                        {outputText !== null && (
-                          <div className={css.ioSection}>
-                            <span className={css.ioLabel}>OUT</span>
-                            <span className={css.ioText} data-error={state === 'error' || undefined}>
-                              {outputText}
-                            </span>
-                          </div>
-                        )}
-                      </div>
-                    )}
-                  </>
-                )}
+            : searchBody !== null
+              ? (
+                <>
+                  <SearchBlock {...searchBody.card} maxLines={CHAT_SEARCH_MAX_LINES} className={css.searchBody} />
+                  {/* A capped search's recovery locator lives only in the result
+                      text; show it below the card so the dropped rows survive. */}
+                  {searchBody.recovery !== undefined && (
+                    <div className={css.searchRecovery}>{searchBody.recovery}</div>
+                  )}
+                </>
+              )
+              : diffBody !== null
+                ? <DiffBlock {...diffBody.card} maxLines={CHAT_DIFF_MAX_LINES} className={css.diffBody} />
+                : isThink
+                  ? <div className={css.thinkBody}>{body}</div>
+                  : (
+                    <>
+                      {variant === 'code' && body !== null && (
+                        <div className={css.bodyScroll}>
+                          <CodeBlock code={body} lang="typescript" copyLabel={t('copy')} copiedLabel={t('copied')} className={css.codeBody} />
+                        </div>
+                      )}
+                      {(cardBody !== null || outputText !== null) && (
+                        <div className={css.ioCard}>
+                          {cardBody !== null && (
+                            <div className={css.ioSection}>
+                              <span className={css.ioLabel}>IN</span>
+                              <span className={css.ioText}>{cardBody}</span>
+                            </div>
+                          )}
+                          {cardBody !== null && outputText !== null && (
+                            <span className={css.ioDivider} aria-hidden />
+                          )}
+                          {outputText !== null && (
+                            <div className={css.ioSection}>
+                              <span className={css.ioLabel}>OUT</span>
+                              <span className={css.ioText} data-error={state === 'error' || undefined}>
+                                {outputText}
+                              </span>
+                            </div>
+                          )}
+                        </div>
+                      )}
+                    </>
+                  )}
           {inspect !== undefined && (
             <button
               type="button"

packages/client/ui-conversation/src/client/contract/read-card-model.ts

@@ -0,0 +1,76 @@
+/**
+ * Pure derivation of the read-card props from a frozen call slice: the
+ * `card:'read'` render intent the read tool declares arrives on the snapshot as
+ * the settled result node's `resultView`, and this is the one place that turns
+ * it into what {@link ReadBlock} draws. Both conversation render sites (the chat
+ * tool row's resident body and the details panel's Output section) call this, so
+ * the path, lines, total, and language they show are derived once.
+ *
+ * The read card is result-side only ([read card note](../../../../../../.agents/notes/implemented/feature/2026-07-30-web-read-card.md)):
+ * a call carries no file content until `execute` returns, so the pending call
+ * stays a generic card (`kind: 'read'`). A running read therefore has no read
+ * card, and this returns null for it — the row keeps its args-derived summary
+ * until the result arrives.
+ * @module
+ */
+import type { ReadBlockLine, ReadBlockProps } from '@deepseek-ai/dsh-client-ui-primitives'
+import { relativizeToCwd, type ToolCallBlock } from './tool-call-model.ts'
+
+/**
+ * Content lines the chat row's resident read body shows before collapsing the
+ * middle — half the primitive's own default, which the details panel keeps. A
+ * chat row is a summary surface inside the message flow: the flow must stay
+ * scannable across many calls, while the details panel is the single-call
+ * reading surface. A design constant of this UI's row geometry, not a
+ * deployment choice, so it is fixed here rather than a plugin Config field. The
+ * same split [`CHAT_TERMINAL_MAX_LINES`](./terminal-card-model.ts) draws for
+ * terminal output.
+ */
+export const CHAT_READ_MAX_LINES = 8
+
+/**
+ * The {@link ReadBlock} props this derivation owns. Picked off the primitive's
+ * props so the two stay in step; `maxLines`/`className` belong to each render
+ * site.
+ */
+export type ReadCardModel = Pick<ReadBlockProps, 'label' | 'lines' | 'totalLines' | 'lang'>
+
+/**
+ * Derive the read-card props for a tool call, or null when this call is not a
+ * read card and belongs on the generic path.
+ *
+ * The read card is result-side only, so only a settled call whose result view
+ * declares `card:'read'` produces one. Every other case is null — the
+ * documented generic-card default:
+ *
+ * - A running call: it has no result view yet, and a read carries no content at
+ *   call time.
+ * - A settled call whose result view is not a read card — including a `card`
+ *   value this UI version does not know, which arrives over the wire and cannot
+ *   be trusted to be one of the compiled variants, and the read tool's own
+ *   generic fallback for an error result or a non-envelope body.
+ *
+ * The label is the read view's `title` when the tool supplied one (the
+ * presentation contract's replacement-title rule), otherwise the file path
+ * relativized to the session workspace so a workspace-rooted absolute path
+ * displays the same short form the row summary shows.
+ * @param block - RunningToolCall or ToolResultNode off the snapshot caches.
+ * @param sessionCwd - the session workspace root; a workspace-rooted absolute
+ *   path label displays relative to it. Absent leaves the path as authored.
+ * @returns the read-card props, or null for the generic path.
+ */
+export function readCardModel(block: ToolCallBlock, sessionCwd?: string): ReadCardModel | null {
+  // Running has no result view; a read carries no content until execute returns.
+  if (!('kind' in block)) return null
+  const result = block.resultView?.card === 'read' ? block.resultView : null
+  if (result === null) return null
+  // Lines arrive frozen off the snapshot; copy into the primitive's own line
+  // shape so the card never holds a reference into the runtime's cache.
+  const lines: ReadBlockLine[] = result.lines.map(line => ({ number: line.number, text: line.text }))
+  return {
+    label: result.title ?? relativizeToCwd(result.path, sessionCwd),
+    lines,
+    totalLines: result.totalLines,
+    lang: result.lang,
+  }
+}

packages/client/ui-conversation/src/client/contract/search-card-model.ts

@@ -0,0 +1,159 @@
+/**
+ * Pure derivation of the search-card props from a frozen call slice: the
+ * `card:'search'` render intent the `grep` and `glob` tools declare arrives on
+ * the snapshot as `resultView`, and this is the one place that turns it into
+ * what {@link SearchBlock} draws. Both conversation render sites (the chat tool
+ * row's resident body and the details panel's Output section) call this, so the
+ * grouped matches or the path list they show are derived once.
+ *
+ * The search card is result-time only: a search call has no matches or paths
+ * before `execute`, so its pending state stays a `GenericCallView`
+ * ({@link module:@deepseek-ai/dsh-tools/src/presentation}). This derivation
+ * therefore reads only `resultView` and returns null for a still-running call,
+ * unlike the terminal card whose call view carries the command before
+ * execution.
+ *
+ * A capped result also carries a recovery locator (grep/glob's `Full … stored
+ * at …` footer) in the raw `tool/result` content, not in the structured
+ * matches/paths the view carries. Since both render sites replace that raw
+ * result with the card, this derivation surfaces the block's own result text as
+ * {@link SearchCardModel.recovery} so the one path to the dropped rows is not
+ * lost.
+ * @module
+ */
+import type { SearchBlockProps, SearchFileGroup } from '@deepseek-ai/dsh-client-ui-primitives'
+import type { ToolCallBlock } from './tool-call-model.ts'
+
+/**
+ * Distributive `Omit`: a plain `Omit<A | B, K>` keeps only the keys common to
+ * both members, which would drop the `files`/`paths` discriminated fields.
+ * Distributing over the naked type parameter `T` preserves each shape.
+ */
+type DistributiveOmit<T, K extends keyof T> = T extends unknown ? Omit<T, K> : never
+
+/** The {@link SearchBlockProps} union minus each render site's own fields. */
+type SearchBlockModelProps = DistributiveOmit<SearchBlockProps, 'maxLines' | 'className'>
+
+/**
+ * Result rows the chat row's resident search body shows before collapsing the
+ * middle — half the primitive's own default, which the details panel keeps. A
+ * chat row is a summary surface inside the message flow: the flow must stay
+ * scannable across many calls, while the details panel is the single-call
+ * reading surface. A design constant of this UI's row geometry, not a
+ * deployment choice, so it is fixed here rather than a plugin Config field.
+ */
+export const CHAT_SEARCH_MAX_LINES = 8
+
+/**
+ * The {@link SearchBlock} props this derivation owns. Held as a nested object
+ * (`card`) so a render site spreads exactly the primitive's own surface and can
+ * never leak a neighbouring field into it. `maxLines`/`className` belong to each
+ * render site.
+ */
+export interface SearchCardModel {
+  /**
+   * The props {@link SearchBlock} draws, minus each render site's own
+   * `maxLines`/`className`.
+   */
+  card: SearchBlockModelProps
+  /**
+   * The result view's replacement title, which the presentation contract lets a
+   * search tool set at settle time. Absent when the presenter supplied none; a
+   * row then keeps its args-derived summary.
+   */
+  title: string | undefined
+  /**
+   * The raw `tool/result` text, flattened, surfaced only when the search was
+   * capped. The card renders the retained matches or paths, but the recovery
+   * locator a capped result carries — grep/glob's `Full … stored at: <locator>`
+   * footer, the one way to reach the rows the cap dropped — lives only in the raw
+   * result text, which the card replaces. A UI that shows the card would
+   * otherwise lose it. Absent when the result was not capped (the card holds
+   * every result) or the block carries no text.
+   */
+  recovery: string | undefined
+}
+
+/**
+ * Whether every file group in a matches view is structurally valid: the wire
+ * frame carries `shape` and `card` as strings the host schema checks, but not the
+ * grouped shape, so a version mismatch or loose producer could deliver
+ * `shape: 'matches'` with a missing or malformed `files`. Rendering that would
+ * crash {@link SearchBlock} at `.reduce`/`.map`; an invalid shape falls to the
+ * generic path instead.
+ * @param files - the candidate `files` field off the untrusted result view.
+ * @returns whether `files` is a valid {@link SearchFileGroup} array.
+ */
+function isValidFiles(files: unknown): files is SearchFileGroup[] {
+  return Array.isArray(files) && files.every(file =>
+    typeof file === 'object' && file !== null
+    && typeof (file as { path?: unknown }).path === 'string'
+    && Array.isArray((file as { matches?: unknown }).matches)
+    && (file as { matches: unknown[] }).matches.every(match =>
+      typeof match === 'object' && match !== null
+      && typeof (match as { lineNumber?: unknown }).lineNumber === 'number'
+      && typeof (match as { line?: unknown }).line === 'string'))
+}
+
+/**
+ * Flatten a settled tool result's content blocks to their text, joined by
+ * newlines. The search view carries no result text — a UI without a card falls
+ * back to the raw `tool/result` content — so the truncation recovery footer is
+ * read from the block's own content here. Non-text blocks (a search result
+ * carries none) are skipped.
+ * @param content - the result node's content blocks.
+ * @returns the joined text, or undefined when empty.
+ */
+function flattenContent(content: readonly { type: string; text?: string }[]): string | undefined {
+  const text = content
+    .filter((block): block is { type: 'text'; text: string } => block.type === 'text' && typeof block.text === 'string')
+    .map(block => block.text)
+    .join('\n')
+  return text === '' ? undefined : text
+}
+
+/**
+ * Derive the search-card props for a tool call, or null when this call is not a
+ * search card and belongs on the generic path.
+ *
+ * Only the result side matters: the search card carries no call-time state, so
+ * a still-running call (no result view) is null, as is a settled call whose
+ * result view is not a search card — including a `card` value this UI version
+ * does not know, which arrives over the wire and cannot be trusted to be one of
+ * the compiled variants, a `card: 'search'` view whose `shape` is neither
+ * `matches` nor `paths` (equally untrusted wire data), and a generic result a
+ * `grep`/`glob` failure or nested `run_code` dispatch produces (its text keeps
+ * the generic path).
+ * @param block - RunningToolCall or ToolResultNode off the snapshot caches.
+ * @returns the search-card props, or null for the generic path.
+ */
+export function searchCardModel(block: ToolCallBlock): SearchCardModel | null {
+  // Running: no result view exists yet, and a search card is result-only.
+  if (!('kind' in block)) return null
+  const result = block.resultView?.card === 'search' ? block.resultView : null
+  if (result === null) return null
+  const common = { truncated: result.truncated, total: result.total }
+  // The recovery footer only matters when the tool capped the result: an
+  // uncapped card holds every match/path, so the raw text adds nothing the card
+  // does not already show. When capped, the raw result's `Full … stored at …`
+  // locator is the only path to the dropped rows, so surface it.
+  const recovery = result.truncated ? flattenContent(block.content) : undefined
+  if (result.shape === 'matches') {
+    // `files` rides the untrusted wire frame: the host schema checks `card`/`shape`
+    // strings but not the grouped shape, so validate it before SearchBlock, which
+    // would crash on a missing/malformed `files`. An invalid shape falls to generic.
+    if (!isValidFiles(result.files)) return null
+    return { title: result.title, recovery, card: { kind: 'matches', files: result.files, ...common } }
+  }
+  // `shape` rides the same untrusted wire frame as `card`, so a version mismatch
+  // or a loose protocol producer could deliver a `card: 'search'` subtype this
+  // client does not compile. Guard the paths shape explicitly: an unknown shape
+  // falls to the generic path rather than being rendered as a paths card, which
+  // would leave SearchBlock calling `.length`/`.map` on an absent `paths`.
+  // oxlint-disable-next-line typescript/no-unnecessary-condition -- shape is wire data; the compiled union cannot prove this exhaustive.
+  if (result.shape !== 'paths') return null
+  // `paths` is likewise unchecked by the wire schema; a known shape with a
+  // missing/malformed array would crash the paths card at `.map`.
+  if (!Array.isArray(result.paths) || !result.paths.every((path): path is string => typeof path === 'string')) return null
+  return { title: result.title, recovery, card: { kind: 'paths', paths: result.paths, ...common } }
+}

packages/client/ui-conversation/src/client/contract/tool-call-model.ts

@@ -133,8 +133,13 @@ const SUMMARY_KEYS: Record<ToolRowVariant, readonly string[]> = {
   others: [],
 }
 
-/** Strip the workspace root from workspace-rooted absolute paths (display only). */
-function relativizeToCwd(text: string, cwd: string | undefined): string {
+/**
+ * Strip the workspace root from a workspace-rooted absolute path (display only).
+ * @param text - the path to shorten.
+ * @param cwd - session workspace root; absent or empty leaves the path unchanged.
+ * @returns the path relative to the workspace root, or unchanged when it is not rooted there.
+ */
+export function relativizeToCwd(text: string, cwd: string | undefined): string {
   if (cwd === undefined || cwd === '') return text
   const root = cwd.replace(/[/\\]+$/, '')
   if (text.startsWith(`${root}/`) || text.startsWith(`${root}\\`)) return text.slice(root.length + 1)

packages/client/ui-conversation/src/client/contract/toolview-status.ts

@@ -0,0 +1,45 @@
+// Shared toolview-row helpers for the keyed rows whose card is resident below a
+// summary (SearchRow, FileMutationRow): the visually hidden run-state label and
+// the flattened settled-result text for the fallback arm a card cannot render.
+// Both are pure functions of a frozen call slice — no chat-domain imports — so a
+// row stays a thin ToolRowProps consumer.
+
+import type { ToolRowProps } from './slots.ts'
+import type { ToolRowState } from './tool-call-model.ts'
+
+/**
+ * Visually hidden run-state label for a row's leading `StateDot` (which is
+ * `aria-hidden`), so assistive technology still announces the state. Returns
+ * null for the settled-ok state, which needs no spoken label.
+ * @param state - the row's run state.
+ * @returns the label, or null when none is needed.
+ */
+export function rowStateStatus(state: ToolRowState): string | null {
+  switch (state) {
+    case 'running': return '运行中'
+    case 'error': return '失败'
+    case 'stopped': return '已停止'
+    default: return null
+  }
+}
+
+/**
+ * A settled result's text, flattened from its content blocks, for the fallback
+ * arm a keyed row shows when its card cannot render the result — an errored call
+ * (the tool emits no result view on error) or a settled call with no card view
+ * (a nested `run_code` sub-dispatch, a legacy generic result). The keyed row owns
+ * the render slot, so without this the model-facing text would have nowhere to
+ * go. Falls back to the error name/code when the result carries no text block.
+ * @param block - the frozen call slice.
+ * @returns the result text, or null for a running call or an empty result.
+ */
+export function rowResultText(block: ToolRowProps['block']): string | null {
+  if (!('kind' in block)) return null
+  const parts: string[] = []
+  for (const item of block.content) {
+    if (item.type === 'text') parts.push(item.text)
+  }
+  if (parts.length === 0 && block.error !== undefined) parts.push(`${block.error.name}: ${block.error.code}`)
+  const text = parts.join('\n')
+  return text === '' ? null : text
+}

packages/client/ui-conversation/src/client/skeleton/DetailsPanel.module.css

@@ -101,15 +101,27 @@
   font: var(--dsw-font-xs-13);
 }
 
-/* A card body (terminal or diff) sits directly under its section label, so it
-   drops the primitive's standalone vertical margin; the section owns the
-   spacing. Card-neutral: no terminal- or diff-specific value. */
+/* A card body (terminal, search, or diff) sits directly under its section
+   label, so it drops the primitive's standalone vertical margin; the section
+   owns the spacing. Card-neutral: no card-specific value. */
 .cardBody {
   margin: 0;
 }
 
-/* Same rule for the web card: it sits under the section label, so the section
-   owns the spacing rather than the primitive's own vertical margin. */
+/* The recovery footer for a capped search: the result text (its `Full … stored
+   at …` locator) below the card in the muted tone, since the card holds only the
+   retained rows. */
+.searchRecovery {
+  margin: 6px 0 0;
+  white-space: pre-wrap;
+  overflow-wrap: anywhere;
+  color: var(--dsw-alias-label-tertiary);
+  font: var(--dsw-font-xs-13);
+}
+
+/* The read and web cards sit directly under their section label, same as the
+   terminal card: drop the primitive's standalone vertical margin. */
+.read,
 .web {
   margin: 0;
 }

packages/client/ui-conversation/src/client/skeleton/DetailsPanel.tsx

@@ -7,10 +7,12 @@
 // share the store seat exists for) and derives the call material from the
 // session snapshot — no data of its own.
 
-import { CodeBlock, DiffBlock, TerminalBlock, WebBlock } from '@deepseek-ai/dsh-client-ui-primitives'
+import { CodeBlock, DiffBlock, ReadBlock, SearchBlock, TerminalBlock, WebBlock } from '@deepseek-ai/dsh-client-ui-primitives'
 import { shallowEqual } from '@deepseek-ai/dsh-client-runtime/client'
 import type { ConversationSnapshot, RunningToolCall, ToolResultNode } from '@deepseek-ai/dsh-client-runtime/client'
 import type { DetailsSlotProps } from '../contract/slots.ts'
+import { searchCardModel } from '../contract/search-card-model.ts'
+import { readCardModel } from '../contract/read-card-model.ts'
 import { diffCardModel } from '../contract/diff-card-model.ts'
 import { terminalBlockLabels, terminalCardModel } from '../contract/terminal-card-model.ts'
 import { webCardModel } from '../contract/web-card-model.ts'
@@ -129,11 +131,15 @@ export function DetailsPanel({ useSession, useSessions, sessionId, useStore, clo
  * The Output section's body for the selected call. A terminal-card call — a
  * shell command's call/result views — renders through the shared TerminalBlock
  * at the primitive's own full height allowance, so column-aligned output keeps
- * its alignment and scrolls sideways instead of folding. A diff-card call — a
- * write/edit's applied change — renders through the shared DiffBlock at the same
- * full height. A web-card call — a `web_search`/`web_fetch` result — renders
- * through WebBlock at its own full source-list allowance. Every other call, and
- * a running call with no card yet, keeps the flattened text form.
+ * its alignment and scrolls sideways instead of folding. A search-card call —
+ * a `grep`/`glob` result view — renders through the shared SearchBlock at the
+ * same full height allowance, with a capped search's recovery footer below it.
+ * A read-card call renders through the shared ReadBlock at that same full height,
+ * so the whole returned window is line-numbered and highlighted. A diff-card
+ * call — a write/edit's applied change — renders through the shared DiffBlock at
+ * the same full height. A web-card call — a `web_search`/`web_fetch` result —
+ * renders through WebBlock at its own full source-list allowance. Every other
+ * call, and a running call with no card yet, keeps the flattened text form.
  * @param props.material - the selected call's material from {@link materialFor}.
  * @param props.cwd - the session workspace root, resolving the terminal view's cwd.
  * @param props.t - the panel's locale seat, passed down as a plain prop.
@@ -153,6 +159,23 @@ function OutputBody({ material, cwd, t }: { material: CallMaterial; cwd: string
       </>
     )
   }
+  const search = searchCardModel(material.block)
+  if (search !== null) {
+    return (
+      <>
+        <SearchBlock {...search.card} className={css.cardBody} />
+        {/* A capped search's recovery locator lives only in the result text;
+            show it below the card so the dropped rows stay reachable. */}
+        {search.recovery !== undefined && (
+          <div className={css.searchRecovery}>{search.recovery}</div>
+        )}
+      </>
+    )
+  }
+  const read = readCardModel(material.block, cwd)
+  // The panel takes the primitive's own default cap, not the row's tighter one:
+  // it is the single-call reading surface, so the whole window is available.
+  if (read !== null) return <ReadBlock {...read} className={css.read} />
   const diff = diffCardModel(material.block)
   if (diff !== null) return <DiffBlock {...diff.card} className={css.cardBody} />
   const web = webCardModel(material.block)

packages/client/ui-conversation/src/client/toolviews/file-mutation-row.tsx

@@ -18,6 +18,7 @@ import { DiffBlock, IconEditOutline16, StateDot } from '@deepseek-ai/dsh-client-
 import type { ToolRowProps } from '../contract/slots.ts'
 import { CHAT_DIFF_MAX_LINES, diffCardModel } from '../contract/diff-card-model.ts'
 import { toolRowModel, type ToolRowState } from '../contract/tool-call-model.ts'
+import { rowResultText, rowStateStatus } from '../contract/toolview-status.ts'
 import css from './file-mutation-row.module.css'
 
 function leadingFor(state: ToolRowState) {
@@ -29,37 +30,6 @@ function leadingFor(state: ToolRowState) {
   }
 }
 
-/** Visually hidden status — StateDot is aria-hidden; AT needs a text label. */
-function stateStatus(state: ToolRowState): string | null {
-  switch (state) {
-    case 'running': return '运行中'
-    case 'error': return '失败'
-    case 'stopped': return '已停止'
-    default: return null
-  }
-}
-
-/**
- * A settled result's text, flattened from its content blocks, for the arm that
- * shows a failure the diff card cannot: write/edit return `undefined` from
- * `presentResult` on `result.isError`, so an errored mutation has no diff card,
- * and the keyed row is not a details-panel target. Without this the failure —
- * an `old_string` that did not match, a permission denial — would read as a bare
- * red dot with the model-facing error text nowhere on screen.
- * @param block - the frozen call slice.
- * @returns the result text, or null for a running call or an empty result.
- */
-function errorText(block: ToolRowProps['block']): string | null {
-  if (!('kind' in block)) return null
-  const parts: string[] = []
-  for (const item of block.content) {
-    if (item.type === 'text') parts.push(item.text)
-  }
-  if (parts.length === 0 && block.error !== undefined) parts.push(`${block.error.name}: ${block.error.code}`)
-  const text = parts.join('\n')
-  return text === '' ? null : text
-}
-
 /**
  * File-mutation row: icon + {Edit,Write} · {path} in the shared ToolRow chrome,
  * with the applied diff resident below it. The summary is a path link (a file
@@ -70,11 +40,11 @@ function errorText(block: ToolRowProps['block']): string | null {
 export function FileMutationRow({ toolName, block, cwd, openFile }: ToolRowProps) {
   const model = toolRowModel(toolName, block, cwd)
   const diff = diffCardModel(block)
-  const status = stateStatus(model.state)
+  const status = rowStateStatus(model.state)
   const filePath = model.filePath
   // An errored mutation has no diff card (presentResult returns undefined on
   // isError); surface its result text so the failure is more than a red dot.
-  const failure = diff === null && model.state === 'error' ? errorText(block) : null
+  const failure = diff === null && model.state === 'error' ? rowResultText(block) : null
   return (
     <div className={css.card}>
       <div className={css.root} data-variant={model.variant} data-state={model.state}>

packages/client/ui-conversation/src/client/toolviews/read-row.module.css

@@ -0,0 +1,119 @@
+/* Read toolview: same geometry/tokens as ToolRow (figma Read · {path}), plus
+   the read card the row stacks under its summary line. */
+
+/* Summary line over the read card; the summary row keeps its own 24px height,
+   so the card is a column around it rather than a change to it. */
+.card {
+  display: flex;
+  flex-direction: column;
+}
+
+/* Row indentation matches ToolRow's expanded bodies (16px leading + 6px gap),
+   and replaces the primitive's standalone vertical margin with the flow's. */
+.read {
+  margin: 4px 0 4px 22px;
+}
+
+.root {
+  position: relative; /* sweep-glare overlay anchor */
+  overflow: hidden;
+  display: flex;
+  align-items: center;
+  height: 24px;
+  min-width: 0;
+}
+
+/* Running sweep glare — same pattern as BashRow/ToolRow, so a running read row
+   gives the same executing feedback a running command row does. The leading
+   read icon stays static (a read has no per-step state to animate); the sweep
+   is the row-level running signal. */
+.root[data-state='running']::after {
+  content: '';
+  position: absolute;
+  top: 0;
+  bottom: 0;
+  left: 0;
+  width: 300px;
+  background: linear-gradient(
+    90deg,
+    transparent 0%,
+    color-mix(in srgb, var(--dsw-alias-bg-base) 60%, transparent) 55%,
+    transparent 100%
+  );
+  animation: dsh-read-row-sweep 2.6s ease-out infinite;
+  pointer-events: none;
+}
+
+@keyframes dsh-read-row-sweep {
+  0% { left: -300px; }
+  90%, 100% { left: 100%; }
+}
+
+.leading {
+  flex: none;
+  width: 16px;
+  height: 16px;
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  margin-right: 6px;
+  color: var(--dsw-alias-label-tertiary);
+}
+
+.title {
+  flex: none;
+  font-size: 14px;
+  line-height: 24px;
+  color: var(--dsw-alias-label-secondary);
+}
+
+.sep {
+  flex: none;
+  width: 2px;
+  height: 2px;
+  border-radius: 1px;
+  margin: 0 8px;
+  background: var(--dsw-alias-label-caption);
+}
+
+.summary {
+  flex: 1 1 auto;
+  min-width: 0;
+  overflow: hidden;
+  text-overflow: ellipsis;
+  white-space: nowrap;
+  font-size: 14px;
+  line-height: 24px;
+  color: var(--dsw-alias-label-tertiary);
+}
+
+/* File path: same geometry as .summary; hover underline + pointer. */
+.fileLink {
+  flex: 1 1 auto;
+  min-width: 0;
+  overflow: hidden;
+  text-overflow: ellipsis;
+  white-space: nowrap;
+  margin: 0;
+  padding: 0;
+  border: none;
+  background: none;
+  text-align: left;
+  font-size: 14px;
+  line-height: 24px;
+  color: var(--dsw-alias-label-tertiary);
+  cursor: pointer;
+}
+
+.fileLink:hover {
+  text-decoration: underline;
+}
+
+.visuallyHidden {
+  position: absolute;
+  width: 1px;
+  height: 1px;
+  overflow: hidden;
+  clip: rect(0 0 0 0);
+  white-space: nowrap;
+}

packages/client/ui-conversation/src/client/toolviews/read-row.tsx

@@ -0,0 +1,105 @@
+// Read toolview registrant: the keyed toolview hole for the read tool
+// (ctx.slots.register + ToolRowProps only — never imports the chat domain).
+// Product chrome matches ToolRow (figma: Read · {path}); the summary is the
+// file path as an openable link, exactly as the generic read row draws it.
+//
+// A read RESULT declares the read render intent, so this row renders the file's
+// own line-numbered, syntax-highlighted content through ReadBlock resident
+// below its summary line — the same posture BashRow gives a terminal card. The
+// card is capped at CHAT_READ_MAX_LINES (the chat flow's tighter cap over the
+// block's own default of 16) with the block's internal expander keeping a long
+// read from taking over the message flow. A running read (no result yet) and a
+// non-read result both render the summary row alone. The read intent is
+// result-side only, so there is no running-state read card to draw.
+
+import type { Context } from 'cordis'
+import { IconBrowseOutline16, ReadBlock, StateDot } from '@deepseek-ai/dsh-client-ui-primitives'
+import type { ToolRowProps } from '../contract/slots.ts'
+import { CHAT_READ_MAX_LINES, readCardModel } from '../contract/read-card-model.ts'
+import { toolRowModel, type ToolRowState } from '../contract/tool-call-model.ts'
+import css from './read-row.module.css'
+
+/** Leading-slot state substitution: the tool icon yields to the state dot
+ *  (error = red, interrupted = amber). Running keeps the icon. */
+function leadingFor(state: ToolRowState) {
+  switch (state) {
+    case 'error': return <StateDot state="error" />
+    case 'stopped': return <StateDot state="warning" />
+    default: return <IconBrowseOutline16 size={14} />
+  }
+}
+
+/** Visually hidden status — StateDot is aria-hidden; AT needs a text label. */
+function stateStatus(state: ToolRowState): string | null {
+  switch (state) {
+    case 'running': return '运行中'
+    case 'error': return '失败'
+    case 'stopped': return '已停止'
+    default: return null
+  }
+}
+
+/**
+ * Read row: icon + Read · {path} in the shared ToolRow chrome, with the file's
+ * read card resident below it. The summary path is an openable host link when
+ * the row names a single file; the card's copy and expand controls plus that
+ * link are the row's only interactions (tool rows are not details-panel
+ * targets).
+ */
+export function ReadRow({ toolName, block, sessionId, useSessions, openFile }: ToolRowProps) {
+  // Session workspace root: the read view's path relativizes against it (a
+  // workspace-rooted absolute path shows its short form), which the pure
+  // presenter cannot do.
+  const cwd = useSessions(list => list.byId[sessionId]?.cwd)
+  const model = toolRowModel(toolName, block, cwd)
+  const read = readCardModel(block, cwd)
+  const status = stateStatus(model.state)
+  const filePath = model.filePath
+  return (
+    <div className={css.card}>
+      {/* jscpd:ignore-start — the summary-line chrome (leading, status, title,
+          sep, path-link/summary) is the shared ToolRow row shape every keyed
+          toolview draws; extracting it into one component is a separate change
+          tracked for all rows at once, not this read-card PR. */}
+      <div className={css.root} data-variant="read" data-state={model.state}>
+        <span className={css.leading}>{leadingFor(model.state)}</span>
+        {status !== null && <span className={css.visuallyHidden}>{status}</span>}
+        <span className={css.title}>{model.title}</span>
+        <span className={css.sep} aria-hidden />
+        {filePath !== undefined ? (
+          <button
+            type="button"
+            className={css.fileLink}
+            onClick={() => { openFile(filePath) }}
+          >
+            {model.summary}
+          </button>
+        ) : (
+          <span className={css.summary}>{model.summary}</span>
+        )}
+      </div>
+      {/* jscpd:ignore-end */}
+      {read !== null && (
+        <ReadBlock {...read} maxLines={CHAT_READ_MAX_LINES} className={css.read} />
+      )}
+    </div>
+  )
+}
+
+/**
+ * The read row as a plain registrant plugin. `inject` carries the load-order
+ * seam: requiring the conversation service guarantees the chat entry (and with
+ * it the 'conversation.chat.toolview' declaration) is registered —
+ * ui-conversation's apply mounts the service after the chat entry.
+ */
+export const readToolview = {
+  name: 'read-toolview',
+  inject: ['slots', 'conversation'],
+  /**
+   * Register the read row into the chat view's keyed toolview hole.
+   * @param ctx - registrant context (disposal rides ctx.effect inside slots.register).
+   */
+  apply(ctx: Context): void {
+    ctx.slots.register({ name: 'conversation.chat.toolview', key: 'read' }, ReadRow)
+  },
+}

packages/client/ui-conversation/src/client/toolviews/search-row.module.css

@@ -0,0 +1,117 @@
+/* Search toolview: same geometry/tokens as ToolRow and BashRow (figma
+   Search · summary), plus the search card the row stacks resident under its
+   summary line. */
+
+/* Summary line over the search card; the summary row keeps its own 24px
+   height, so the card is a column around it rather than a change to it. */
+.card {
+  display: flex;
+  flex-direction: column;
+}
+
+/* Row indentation matches ToolRow's expanded bodies (16px leading + 6px gap),
+   and replaces the primitive's standalone vertical margin with the flow's. */
+.search {
+  margin: 4px 0 4px 22px;
+}
+
+.root {
+  position: relative; /* sweep-glare overlay anchor */
+  overflow: hidden;
+  display: flex;
+  align-items: center;
+  height: 24px;
+  min-width: 0;
+}
+
+/* Running sweep glare — same deepsuite ShimmerText pattern as ToolRow / BashRow. */
+.root[data-state='running']::after {
+  content: '';
+  position: absolute;
+  top: 0;
+  bottom: 0;
+  left: 0;
+  width: 300px;
+  background: linear-gradient(
+    90deg,
+    transparent 0%,
+    color-mix(in srgb, var(--dsw-alias-bg-base) 60%, transparent) 55%,
+    transparent 100%
+  );
+  animation: dsh-search-row-sweep 2.6s ease-out infinite;
+  pointer-events: none;
+}
+
+@keyframes dsh-search-row-sweep {
+  0% { left: -300px; }
+  90%, 100% { left: 100%; }
+}
+
+.leading {
+  flex: none;
+  width: 16px;
+  height: 16px;
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  margin-right: 6px;
+  color: var(--dsw-alias-label-tertiary);
+}
+
+.title {
+  flex: none;
+  font-size: 14px;
+  line-height: 24px;
+  color: var(--dsw-alias-label-secondary);
+}
+
+.sep {
+  flex: none;
+  width: 2px;
+  height: 2px;
+  border-radius: 1px;
+  margin: 0 8px;
+  background: var(--dsw-alias-label-caption);
+}
+
+.summary {
+  flex: 1 1 auto;
+  min-width: 0;
+  overflow: hidden;
+  text-overflow: ellipsis;
+  white-space: nowrap;
+  font-size: 14px;
+  line-height: 24px;
+  color: var(--dsw-alias-label-tertiary);
+}
+
+.visuallyHidden {
+  position: absolute;
+  width: 1px;
+  height: 1px;
+  overflow: hidden;
+  clip: rect(0 0 0 0);
+  white-space: nowrap;
+}
+
+/* The result text for an errored search, indented to the card's own column and
+   in the error tone, standing in for the search card the failure path does not
+   produce. */
+.failure {
+  margin: 4px 0 4px 22px;
+  white-space: pre-wrap;
+  overflow-wrap: anywhere;
+  font: var(--dsw-font-xs-13);
+  color: var(--dsw-alias-state-error-primary);
+}
+
+/* The recovery footer for a capped search: the model-facing result text (its
+   `Full … stored at …` locator) shown below the card in the muted tone, since
+   the card holds only the retained rows. Same column indent as the card body. */
+.recovery {
+  margin: 4px 0 4px 22px;
+  white-space: pre-wrap;
+  overflow-wrap: anywhere;
+  font: var(--dsw-font-xs-13);
+  color: var(--dsw-alias-label-tertiary);
+}

packages/client/ui-conversation/src/client/toolviews/search-row.tsx

@@ -0,0 +1,96 @@
+// Search toolview registrant: the keyed toolview hole (ctx.slots.register +
+// ToolRowProps only — never imports the chat domain). One SearchRow component
+// registered under both `grep` and `glob`, since both tools declare the same
+// `card: 'search'` render intent and render as one visual object; the row reads
+// the `kind` discriminant off the derived model to draw grouped matches or a
+// path list. Product chrome matches ToolRow / BashRow (Search · {summary}).
+//
+// A search call declares its render intent result-time only, so this row's
+// search card is resident below the summary rather than expand-gated: the row
+// itself has no expand control, and the card's own copy, per-file collapse, and
+// head/tail expand are the row's only interactions. CHAT_SEARCH_MAX_LINES is
+// passed as `maxLines` — the chat flow's tighter cap over the block's own
+// default of 16 — so a large result stays bounded in the message flow.
+
+import type { Context } from 'cordis'
+import { IconSearchOutline16, SearchBlock, StateDot } from '@deepseek-ai/dsh-client-ui-primitives'
+import type { ToolRowProps } from '../contract/slots.ts'
+import { CHAT_SEARCH_MAX_LINES, searchCardModel } from '../contract/search-card-model.ts'
+import { toolRowModel, type ToolRowState } from '../contract/tool-call-model.ts'
+import { rowResultText, rowStateStatus } from '../contract/toolview-status.ts'
+import css from './search-row.module.css'
+
+/** Leading-slot glyph substitution: the search icon yields to the terminal
+ *  state semantic (error = red, interrupted = amber). Running keeps the icon —
+ *  the row sweep carries the in-flight signal. */
+function leadingFor(state: ToolRowState) {
+  switch (state) {
+    case 'error': return <StateDot state="error" />
+    case 'stopped': return <StateDot state="warning" />
+    default: return <IconSearchOutline16 size={14} />
+  }
+}
+
+/**
+ * Search row: icon + Search · {summary} in the shared ToolRow chrome, with the
+ * completed search's card resident below it, and — when the result was capped —
+ * the recovery footer below the card. The summary row is not a details-panel
+ * control, so the card's copy, per-file collapse, and expand controls are the
+ * row's only interactions. Registered under both `grep` and `glob`; the derived
+ * model's `kind` decides the card shape.
+ */
+export function SearchRow({ toolName, block }: ToolRowProps) {
+  const model = toolRowModel(toolName, block)
+  const search = searchCardModel(block)
+  const status = rowStateStatus(model.state)
+  // A settled call with no search card — an errored search (grep/glob emit no
+  // result view on error), a successful nested run_code sub-dispatch, or a
+  // legacy generic result — has its model-facing text nowhere else to go, since
+  // the keyed SearchRow owns this render slot. Surface it as the fallback body.
+  // A running call ('kind' absent) has no result to flatten; rowResultText
+  // returns null for it, so the arm stays closed until settle.
+  const settled = 'kind' in block
+  const fallback = search === null && settled ? rowResultText(block) : null
+  return (
+    <div className={css.card}>
+      <div className={css.root} data-variant="search" data-tool={toolName} data-state={model.state}>
+        <span className={css.leading}>{leadingFor(model.state)}</span>
+        {status !== null && <span className={css.visuallyHidden}>{status}</span>}
+        <span className={css.title}>{model.title}</span>
+        <span className={css.sep} aria-hidden />
+        {/* The result view's replacement title outranks the args-derived
+            summary, matching the terminal card's description precedence. */}
+        <span className={css.summary}>{search?.title ?? model.summary}</span>
+      </div>
+      {search !== null && (
+        <SearchBlock {...search.card} maxLines={CHAT_SEARCH_MAX_LINES} className={css.search} />
+      )}
+      {/* A capped search drops rows from the card; its recovery locator (the
+          `Full … stored at …` footer) lives only in the result text, so show it
+          below the card so the one path to the dropped rows survives. */}
+      {search?.recovery !== undefined && <div className={css.recovery}>{search.recovery}</div>}
+      {fallback !== null && <div className={css.failure}>{fallback}</div>}
+    </div>
+  )
+}
+
+/**
+ * The search toolview as a plain registrant plugin. `inject` carries the
+ * load-order seam: requiring the conversation service guarantees the chat entry
+ * (and with it the 'conversation.chat.toolview' declaration) is registered.
+ * The one component registers under both keys, since `grep` and `glob` are the
+ * same visual object discriminated only by the result view's `kind`.
+ */
+export const searchToolview = {
+  name: 'search-toolview',
+  inject: ['slots', 'conversation'],
+  /**
+   * Register the search row into the chat view's keyed toolview hole under both
+   * the `grep` and `glob` tool names.
+   * @param ctx - registrant context (disposal rides ctx.effect inside slots.register).
+   */
+  apply(ctx: Context): void {
+    ctx.slots.register({ name: 'conversation.chat.toolview', key: 'grep' }, SearchRow)
+    ctx.slots.register({ name: 'conversation.chat.toolview', key: 'glob' }, SearchRow)
+  },
+}

packages/client/ui-conversation/tests/chat-apply.spec.tsx

@@ -84,14 +84,15 @@ describe('apply wiring', () => {
     await b.runtime.dispose()
   })
 
-  it('mounts the bash sample, the file-mutation rows, the web rows, and the product rows as keyed entries through the load-order seam', async () => {
+  it('mounts the bash sample, the search rows, the read row, the file-mutation rows, the web rows, and the product rows as keyed entries through the load-order seam', async () => {
     const b = await bench()
     // Every registrant plugin's inject: ['slots', 'conversation'] resolved — the
     // service being present implies the chat entry declared the hole first. The
-    // file-mutation registrant claims both write and edit for the diff card; the
-    // web rows register one component under both web tool names.
+    // one search row registers under both grep and glob; the file-mutation
+    // registrant claims both write and edit for the diff card; the web rows
+    // register one component under both web tool names.
     const entries = b.slots.entries('conversation.chat.toolview')
-    expect(entries.map(e => e.options.key)).toEqual(['bash', 'edit', 'write', 'web_search', 'web_fetch', 'todo_write', 'ask_user_question'])
+    expect(entries.map(e => e.options.key)).toEqual(['bash', 'grep', 'glob', 'read', 'edit', 'write', 'web_search', 'web_fetch', 'todo_write', 'ask_user_question'])
     // Stats stick with the composer (not inside ChatView).
     expect(b.slots.entries('conversation.composer.dock').map(e => e.options.id)).toEqual(['stats'])
     await b.runtime.dispose()

packages/client/ui-conversation/tests/chat-branch-tails.spec.tsx

@@ -27,7 +27,7 @@ afterEach(() => {
 const t: MessageItemProps['t'] = makeTranslate(zh, commonZh)
 
 describe('MessageItem arms', () => {
-  it('user bubbles expose clock / copy / branch / edit; copy writes the text', () => {
+  it('user bubbles expose clock / copy / branch and no edit; copy writes the text', () => {
     const writeText = vi.fn().mockResolvedValue(undefined)
     Object.defineProperty(navigator, 'clipboard', {
       configurable: true,
@@ -47,7 +47,7 @@ describe('MessageItem arms', () => {
     expect(screen.getByText('14:24')).toBeTruthy()
     expect(screen.getByRole('button', { name: '复制' })).toBeTruthy()
     expect(screen.getByRole('button', { name: '在新对话中分支' })).toBeTruthy()
-    expect(screen.getByRole('button', { name: '编辑' })).toBeTruthy()
+    expect(screen.queryByRole('button', { name: '编辑' })).toBeNull()
     fireEvent.click(screen.getByRole('button', { name: '复制' }))
     expect(writeText).toHaveBeenCalledWith('hello bubble')
   })

packages/client/ui-conversation/tests/read-card.spec.tsx

@@ -0,0 +1,293 @@
+// @vitest-environment jsdom
+// The read render intent on the web side: the pure readCardModel derivation
+// over the settled result view, and both conversation render sites that consume
+// it — the chat tool row (the keyed ReadRow and the GenericToolCard fallback,
+// each with the read card resident under the summary) and the details panel's
+// Output section. Also pins the keyed 'read' toolview registration.
+
+import { afterEach, describe, expect, it, vi } from 'vitest'
+import { cleanup, fireEvent, render } from '@testing-library/react'
+import { Context } from 'cordis'
+import { bindSnapshotSelector } from '@deepseek-ai/dsh-client-web-react'
+import { createSnapshotStore } from '@deepseek-ai/dsh-client-runtime/client'
+import { makeTranslate } from '@deepseek-ai/dsh-client-test-runtime'
+import { zh as commonZh } from '@deepseek-ai/dsh-client-locale/src/locales/zh.ts'
+import type {
+  ConversationSnapshot, RunningToolCall, SessionId, SessionListState, ToolResultNode, WorkspaceListState,
+} from '@deepseek-ai/dsh-client-runtime/client'
+import type { ToolResultView } from '@deepseek-ai/dsh-client-connection/client'
+import type { SelectionTarget, ToolRowProps } from '@deepseek-ai/dsh-client-ui-conversation/client'
+import { CHAT_READ_MAX_LINES, readCardModel } from '../src/client/contract/read-card-model.ts'
+import { createChatStore } from '../src/client/stores.ts'
+import { GenericToolCard, type GenericToolCardProps } from '../src/client/chat/GenericToolCard.tsx'
+import { zh } from '../src/client/locales.ts'
+import { DetailsPanel } from '../src/client/skeleton/DetailsPanel.tsx'
+import { ReadRow, readToolview } from '../src/client/toolviews/read-row.tsx'
+
+afterEach(cleanup)
+
+const SID = 's1' as SessionId
+
+/** The chat-view locale seat: this package's namespace over the common fallback. */
+const t: GenericToolCardProps['t'] = makeTranslate(zh, commonZh)
+
+// The read tool's real schema key is `file_path`; the top-level read samples
+// use it so the row exercises a production-shaped call. `web_fetch` (below) has
+// its own schema whose key is not `file_path`, so it keeps a `url`-less `path`.
+const ARGS = '{"file_path":"src/a.ts","offset":41}'
+const WEB_FETCH_ARGS = '{"path":"src/a.ts","offset":41}'
+
+/** The read block's rendered content cells, one string per row (highlighting
+ *  breaks a line across token spans, so match on the row's textContent). */
+function contentTexts(container: HTMLElement): string[] {
+  return [...container.querySelectorAll('[data-read] [class^="_content_"]')].map(cell => cell.textContent ?? '')
+}
+
+/** Three windowed lines starting at file line 41 (a read past an offset). */
+const sampleLines = [
+  { number: 41, text: 'export const a = 1' },
+  { number: 42, text: 'export const b = 2' },
+  { number: 43, text: 'export const c = 3' },
+]
+
+/** The read tool's own result view for a settled file read. */
+const resultRead = (over?: Partial<Extract<ToolResultView, { card: 'read' }>>): ToolResultView => ({
+  card: 'read', path: 'src/a.ts', offset: 41, lines: sampleLines, totalLines: 180, lang: 'ts', ...over,
+})
+
+const running = (over?: Partial<RunningToolCall>): RunningToolCall => ({
+  callId: 'c1', name: 'read', argsRaw: ARGS,
+  turn: 1, step: 1, time: 1_000, callView: { card: 'generic', title: 'Read src/a.ts', kind: 'read' }, ...over,
+})
+
+const settled = (over?: Partial<ToolResultNode>): ToolResultNode => ({
+  kind: 'tool-result', seq: 10, time: 2_000, callId: 'c1',
+  call: { name: 'read', argsRaw: ARGS },
+  callTime: 1_000,
+  content: [{ type: 'text', text: '41: export const a = 1' }], isError: false,
+  callView: { card: 'generic', title: 'Read src/a.ts', kind: 'read' }, resultView: resultRead(), ...over,
+})
+
+describe('readCardModel', () => {
+  it('derives the card from a settled read result view', () => {
+    expect(readCardModel(settled())).toEqual({
+      label: 'src/a.ts', lines: sampleLines, totalLines: 180, lang: 'ts',
+    })
+  })
+
+  it('copies the lines into the primitive shape rather than aliasing the frozen slice', () => {
+    const model = readCardModel(settled())
+    expect(model?.lines).toEqual(sampleLines)
+    expect(model?.lines).not.toBe(sampleLines)
+    expect(model?.lines[0]).not.toBe(sampleLines[0])
+  })
+
+  it('takes the result view\'s replacement title over the relativized path', () => {
+    // The presentation contract defines a result title as REPLACING the pending
+    // one, so a tool that supplies a label wins over the path here.
+    expect(readCardModel(settled({ resultView: resultRead({ title: 'Read (head) src/a.ts' }) }))?.label)
+      .toBe('Read (head) src/a.ts')
+  })
+
+  it('relativizes a workspace-rooted path label, and leaves others as authored', () => {
+    // A workspace-rooted absolute path shows its short form.
+    expect(readCardModel(settled({ resultView: resultRead({ path: '/w/app/src/a.ts' }) }), '/w/app')?.label)
+      .toBe('src/a.ts')
+    // A path outside the workspace stays as authored.
+    expect(readCardModel(settled({ resultView: resultRead({ path: '/srv/other.ts' }) }), '/w/app')?.label)
+      .toBe('/srv/other.ts')
+    // With no session cwd there is nothing to relativize against.
+    expect(readCardModel(settled({ resultView: resultRead({ path: '/w/app/src/a.ts' }) }))?.label)
+      .toBe('/w/app/src/a.ts')
+  })
+
+  it('carries an omitted language through as undefined', () => {
+    const noLang = resultRead()
+    delete (noLang as { lang?: string }).lang
+    expect(readCardModel(settled({ resultView: noLang }))?.lang).toBeUndefined()
+  })
+
+  it('returns null for a running read: the read intent is result-side only', () => {
+    // A read carries no content until execute returns, so the pending call is a
+    // generic card and there is no read card to draw yet.
+    expect(readCardModel(running())).toBeNull()
+  })
+
+  it('returns null for every non-read settled call: no view, generic view, unknown card', () => {
+    expect(readCardModel(settled({ resultView: null }))).toBeNull()
+    expect(readCardModel(settled({ resultView: { card: 'generic' } }))).toBeNull()
+    // A card tag this UI version does not know arrives over the wire; the
+    // documented generic-card default takes it, not a crash.
+    const future = { card: 'chart' } as unknown as ToolResultView
+    expect(readCardModel(settled({ resultView: future }))).toBeNull()
+  })
+})
+
+describe('GenericToolCard read body', () => {
+  const ownerProps = (block: RunningToolCall | ToolResultNode): GenericToolCardProps => ({
+    callId: 'c1', toolName: 'web_fetch', block, openFile: vi.fn(), t,
+  })
+
+  it('renders the read card resident under the summary, capped tighter than the panel', () => {
+    expect(CHAT_READ_MAX_LINES).toBeLessThan(16)
+    // web_fetch lands on the read variant without its own keyed row, so the
+    // fallback card owns the resident read block.
+    const view = render(<GenericToolCard {...ownerProps(settled({ call: { name: 'web_fetch', argsRaw: WEB_FETCH_ARGS } }))} />)
+    expect(view.container.querySelector('[data-read]')).not.toBeNull()
+    expect(contentTexts(view.container)).toContain('export const a = 1')
+    // The gutter keeps the file's own line numbers.
+    expect(view.getByText('41')).toBeTruthy()
+  })
+
+  it('a non-read tool renders the bare row with no read card', () => {
+    const view = render(<GenericToolCard {...({
+      callId: 'c1', toolName: 'echo', block: settled({
+        call: { name: 'echo', argsRaw: '{"text":"x"}' }, callView: null, resultView: null,
+      }), openFile: vi.fn(), t,
+    })} />)
+    expect(view.container.querySelector('[data-read]')).toBeNull()
+  })
+
+  it('a running read renders the summary row alone (no result view yet)', () => {
+    const view = render(<GenericToolCard {...ownerProps(running({ name: 'web_fetch' }))} />)
+    expect(view.container.querySelector('[data-read]')).toBeNull()
+  })
+})
+
+describe('ReadRow keyed toolview', () => {
+  const list = () => createSnapshotStore<SessionListState>({
+    ids: [SID],
+    byId: { [SID]: { id: SID, displayTitle: 'r', running: false, blank: false, waitingApproval: false, updatedAt: 0, cwd: '/w/app' } },
+    current: SID,
+    phase: 'ready',
+  })
+
+  const rowProps = (block: RunningToolCall | ToolResultNode): ToolRowProps => ({
+    callId: 'c1', toolName: 'read', block, openFile: vi.fn(),
+    sessionId: SID, useSessions: bindSnapshotSelector(list()),
+  } as unknown as ToolRowProps)
+
+  it('renders the file path summary and the resident read card', () => {
+    const view = render(<ReadRow {...rowProps(settled())} />)
+    expect(view.getByText('Read')).toBeTruthy()
+    // The path appears twice: the row summary link and the card's banner label.
+    expect(view.getAllByText('src/a.ts').length).toBe(2)
+    expect(view.container.querySelector('[data-read]')).not.toBeNull()
+    expect(contentTexts(view.container)).toContain('export const a = 1')
+    expect(view.getByText('显示 3 / 180 行')).toBeTruthy()
+  })
+
+  it('the path summary opens the file through the host', () => {
+    const openFile = vi.fn()
+    const view = render(<ReadRow {...{ ...rowProps(settled()), openFile }} />)
+    fireEvent.click(view.getByRole('button', { name: 'src/a.ts' }))
+    // The row derives the file path from args; the chat view resolves it against
+    // the cwd before this callback opens it, so the arg path is what arrives.
+    expect(openFile).toHaveBeenCalledWith('src/a.ts')
+  })
+
+  it('a running read renders the summary row alone, and its state', () => {
+    const view = render(<ReadRow {...rowProps(running())} />)
+    expect(view.container.querySelector('[data-variant="read"]')?.getAttribute('data-state')).toBe('running')
+    expect(view.container.querySelector('[data-read]')).toBeNull()
+  })
+
+  it('an error read result shows the error state and no read card', () => {
+    const view = render(<ReadRow {...rowProps(settled({
+      resultView: { card: 'generic' }, isError: true,
+      content: [{ type: 'text', text: 'ENOENT' }],
+    }))} />)
+    expect(view.container.querySelector('[data-variant="read"]')?.getAttribute('data-state')).toBe('error')
+    expect(view.container.querySelector('[data-read]')).toBeNull()
+  })
+
+  it('an interrupted read shows the stopped state', () => {
+    const view = render(<ReadRow {...rowProps(settled({
+      resultView: null, isError: true, error: { name: 'ToolError', code: 'interrupted' },
+    }))} />)
+    expect(view.container.querySelector('[data-variant="read"]')?.getAttribute('data-state')).toBe('stopped')
+  })
+
+  it('registers under the read key of the keyed toolview slot', () => {
+    const registered: { name: unknown; key?: unknown }[] = []
+    const ctx = { slots: { register: (options: { name: unknown; key?: unknown }) => { registered.push(options) } } } as unknown as Context
+    readToolview.apply(ctx)
+    expect(registered).toEqual([{ name: 'conversation.chat.toolview', key: 'read' }])
+    expect(readToolview.inject).toContain('conversation')
+  })
+})
+
+describe('DetailsPanel Output section (read)', () => {
+  function mount(snapshot: ConversationSnapshot, selection: SelectionTarget | null, cwd?: string) {
+    localStorage.clear()
+    const chat = createChatStore().create()
+    if (selection !== null) chat.actions.select(selection)
+    const sessions = createSnapshotStore<SessionListState>(cwd === undefined
+      ? { ids: [], byId: {}, current: undefined, phase: 'ready' }
+      : {
+        ids: [SID],
+        byId: { [SID]: { id: SID, displayTitle: 'r', running: false, blank: false, waitingApproval: false, updatedAt: 0, cwd } },
+        current: SID,
+        phase: 'ready',
+      })
+    const workspaces = createSnapshotStore<WorkspaceListState>({
+      items: [], archivedSessionIds: [], state: 'idle', phase: 'ready', error: null,
+      baselinesReady: true, recentWorkspaceId: undefined,
+    })
+    return render(
+      <DetailsPanel
+        sessionId={SID}
+        t={t}
+        useSession={bindSnapshotSelector({ getSnapshot: () => snapshot, subscribe: () => () => {} })}
+        useSessions={bindSnapshotSelector(sessions)}
+        useWorkspaces={bindSnapshotSelector(workspaces)}
+        useInput={(() => { throw new Error('unused') })}
+        inputActions={{ setDraft: () => {}, submit: () => {} }}
+        useProjection={(() => undefined)}
+        useStore={bindSnapshotSelector(chat)}
+        actions={chat.actions}
+        closeDetails={vi.fn()}
+      />,
+    )
+  }
+
+  function snapshot(over: Partial<ConversationSnapshot> = {}): ConversationSnapshot {
+    return {
+      sessionId: SID, nodes: [], foldDegraded: false, partial: null, runningCalls: [], codeDispatches: new Map(),
+      pending: [], queue: [], running: false, composerPhase: 'active', removed: false,
+      openState: 'open', openError: null, hasMore: false, loadingOlder: false,
+      promptError: null, blank: false, lastAgentError: null, ...over,
+    }
+  }
+
+  const target: SelectionTarget = { turnSeq: 10, callId: 'c1', toolName: 'read' }
+
+  it('renders the read card at full height, keeping the JSON Input section', () => {
+    const long = Array.from({ length: 20 }, (_, i) => ({ number: i + 1, text: `row-${i}` }))
+    const view = mount(snapshot({
+      nodes: [settled({ resultView: resultRead({ lines: long, totalLines: 20 }) })],
+    }), target)
+    expect(view.getByText(/"file_path"/)).toBeTruthy()
+    expect(view.container.querySelector('[data-read]')).not.toBeNull()
+    // The panel takes the primitive's own default cap (16), not the row's.
+    expect(view.getByText(`… 其余 ${20 - 16} 行`)).toBeTruthy()
+    expect(contentTexts(view.container)).toContain('row-0')
+  })
+
+  it('a non-read result keeps the flattened pre form', () => {
+    const view = mount(snapshot({
+      nodes: [settled({
+        callView: null, resultView: null,
+        content: [{ type: 'text', text: 'plain result' }],
+      })],
+    }), target)
+    expect(view.container.querySelector('[data-read]')).toBeNull()
+    expect(view.getByText('输出').closest('section')?.querySelector('pre')?.textContent).toBe('plain result')
+  })
+
+  it('a running read keeps the 运行中… placeholder (no result view)', () => {
+    const view = mount(snapshot({ runningCalls: [running()] }), target)
+    expect(view.getByText('运行中…')).toBeTruthy()
+    expect(view.container.querySelector('[data-read]')).toBeNull()
+  })
+})

packages/client/ui-conversation/tests/search-card.spec.tsx

@@ -0,0 +1,409 @@
+// @vitest-environment jsdom
+// The search render intent on the web side: the pure searchCardModel derivation
+// over resultView, and the conversation render sites that consume it — the chat
+// tool row (GenericToolCard's expand-gated body and SearchRow's resident card)
+// and the details panel's Output section. The keyed registration under both grep
+// and glob is pinned here too.
+
+import { afterEach, describe, expect, it, vi } from 'vitest'
+import { cleanup, fireEvent, render } from '@testing-library/react'
+import { bindSnapshotSelector } from '@deepseek-ai/dsh-client-web-react'
+import { createSnapshotStore } from '@deepseek-ai/dsh-client-runtime/client'
+import type {
+  ConversationSnapshot, RunningToolCall, SessionId, SessionListState, ToolResultNode, WorkspaceListState,
+} from '@deepseek-ai/dsh-client-runtime/client'
+import type { ToolResultView } from '@deepseek-ai/dsh-client-connection/client'
+import type { SelectionTarget, ToolRowProps } from '@deepseek-ai/dsh-client-ui-conversation/client'
+import { makeTranslate } from '@deepseek-ai/dsh-client-test-runtime'
+import { zh as commonZh } from '@deepseek-ai/dsh-client-locale/src/locales/zh.ts'
+import { CHAT_SEARCH_MAX_LINES, searchCardModel } from '../src/client/contract/search-card-model.ts'
+import { zh } from '../src/client/locales.ts'
+import { createChatStore } from '../src/client/stores.ts'
+import { GenericToolCard, type GenericToolCardProps } from '../src/client/chat/GenericToolCard.tsx'
+import { DetailsPanel } from '../src/client/skeleton/DetailsPanel.tsx'
+import { SearchRow, searchToolview } from '../src/client/toolviews/search-row.tsx'
+
+afterEach(cleanup)
+
+/** Conversation-locale translate stub for the render sites' `t` seat. */
+const t: GenericToolCardProps['t'] = makeTranslate(zh, commonZh)
+
+/** The rendered search card's kind attribute, so a render site cannot silently drop it. */
+function searchKindOf(container: HTMLElement): string | null {
+  return container.querySelector('[data-search]')?.getAttribute('data-search') ?? null
+}
+
+/** The rendered result rows of the search card, one string per visible row. */
+function searchRows(container: HTMLElement): string[] {
+  return [...container.querySelectorAll('[data-search] [class^="_line_"]')].map(row => row.textContent ?? '')
+}
+
+const SID = 's1' as SessionId
+
+const GREP_ARGS = '{"pattern":"foo","path":"src"}'
+const GLOB_ARGS = '{"pattern":"**/*.ts","path":"src"}'
+
+/** A grep result view: matches grouped by file. */
+const resultMatches = (over?: Partial<Extract<ToolResultView, { card: 'search'; shape: 'matches' }>>): ToolResultView => ({
+  card: 'search', shape: 'matches',
+  files: [
+    { path: 'a.ts', matches: [{ lineNumber: 12, line: 'const foo = 1' }, { lineNumber: 40, line: 'return foo' }] },
+    { path: 'b.ts', matches: [{ lineNumber: 7, line: 'foo()' }] },
+  ],
+  truncated: false, total: 3, ...over,
+})
+
+/** A glob result view: a flat path list. */
+const resultPaths = (over?: Partial<Extract<ToolResultView, { card: 'search'; shape: 'paths' }>>): ToolResultView => ({
+  card: 'search', shape: 'paths', paths: ['src/a.ts', 'src/b.ts'], truncated: false, total: 2, ...over,
+})
+
+const runningGrep = (over?: Partial<RunningToolCall>): RunningToolCall => ({
+  callId: 'c1', name: 'grep', argsRaw: GREP_ARGS,
+  turn: 1, step: 1, time: 1_000, callView: { card: 'generic', title: 'Grep foo', kind: 'search' }, ...over,
+})
+
+const settledGrep = (over?: Partial<ToolResultNode>): ToolResultNode => ({
+  kind: 'tool-result', seq: 10, time: 2_000, callId: 'c1',
+  call: { name: 'grep', argsRaw: GREP_ARGS },
+  callTime: 1_000,
+  content: [{ type: 'text', text: 'a.ts\n  Line 12: const foo = 1' }], isError: false,
+  callView: { card: 'generic', title: 'Grep foo', kind: 'search' }, resultView: resultMatches(), ...over,
+})
+
+const settledGlob = (over?: Partial<ToolResultNode>): ToolResultNode => ({
+  kind: 'tool-result', seq: 11, time: 2_000, callId: 'c2',
+  call: { name: 'glob', argsRaw: GLOB_ARGS },
+  callTime: 1_000,
+  content: [{ type: 'text', text: 'src/a.ts\nsrc/b.ts' }], isError: false,
+  callView: { card: 'generic', title: 'Glob **/*.ts', kind: 'search' }, resultView: resultPaths(), ...over,
+})
+
+describe('searchCardModel', () => {
+  it('derives a matches card from the grep result view', () => {
+    expect(searchCardModel(settledGrep())).toEqual({
+      title: undefined,
+      recovery: undefined,
+      card: {
+        kind: 'matches',
+        files: [
+          { path: 'a.ts', matches: [{ lineNumber: 12, line: 'const foo = 1' }, { lineNumber: 40, line: 'return foo' }] },
+          { path: 'b.ts', matches: [{ lineNumber: 7, line: 'foo()' }] },
+        ],
+        truncated: false, total: 3,
+      },
+    })
+  })
+
+  it('derives a paths card from the glob result view, carrying the truncation signal', () => {
+    // Empty block content isolates the truncation signal from the recovery arm.
+    expect(searchCardModel(settledGlob({ content: [], resultView: resultPaths({ truncated: true, total: 20 }) }))).toEqual({
+      title: undefined,
+      recovery: undefined,
+      card: { kind: 'paths', paths: ['src/a.ts', 'src/b.ts'], truncated: true, total: 20 },
+    })
+  })
+
+  it('carries the result view\'s replacement title when the presenter sets one', () => {
+    expect(searchCardModel(settledGrep({ resultView: resultMatches({ title: '3 matches' }) }))?.title).toBe('3 matches')
+    // Without one it is absent, so the row keeps its args-derived summary.
+    expect(searchCardModel(settledGrep())?.title).toBeUndefined()
+  })
+
+  it('returns null for every non-search call: running, no views, generic, terminal, unknown cards', () => {
+    // A search card is result-time only: a running call has no result view yet.
+    expect(searchCardModel(runningGrep())).toBeNull()
+    expect(searchCardModel(settledGrep({ callView: null, resultView: null }))).toBeNull()
+    // A generic result settles a search call as a generic card (grep/glob failure
+    // or a nested run_code dispatch), which keeps the generic path.
+    expect(searchCardModel(settledGrep({ resultView: { card: 'generic' } }))).toBeNull()
+    // A terminal result view is a different card entirely.
+    expect(searchCardModel(settledGrep({ resultView: { card: 'terminal', output: 'x' } }))).toBeNull()
+    // A card tag this UI version does not know arrives over the wire; the
+    // documented generic-card default takes it, not a crash.
+    const future = { card: 'chart' } as unknown as ToolResultView
+    expect(searchCardModel(settledGrep({ resultView: future }))).toBeNull()
+  })
+
+  it('returns null for a card:search view whose shape this version does not compile', () => {
+    // `shape` rides the same untrusted wire frame as `card`; a subtype this client
+    // does not know must fall to the generic path, never render as a paths card
+    // that would crash SearchBlock on an absent `paths`.
+    const futureShape = {
+      card: 'search', shape: 'future', truncated: false, total: 0,
+    } as unknown as ToolResultView
+    expect(searchCardModel(settledGrep({ resultView: futureShape }))).toBeNull()
+  })
+
+  it('returns null for a known shape whose structured shape is missing or malformed', () => {
+    // The host wire schema checks the `card`/`shape` strings but not the grouped
+    // shape, so a version mismatch could deliver shape:'matches' with no `files`
+    // (or shape:'paths' with no `paths`). Rendering that crashes SearchBlock at
+    // `.reduce`/`.map`; the derivation drops to the generic path instead.
+    const noFiles = { card: 'search', shape: 'matches', truncated: false, total: 0 } as unknown as ToolResultView
+    expect(searchCardModel(settledGrep({ resultView: noFiles }))).toBeNull()
+    const badFile = {
+      card: 'search', shape: 'matches', truncated: false, total: 1,
+      files: [{ path: 'a.ts', matches: [{ lineNumber: 'x', line: 1 }] }],
+    } as unknown as ToolResultView
+    expect(searchCardModel(settledGrep({ resultView: badFile }))).toBeNull()
+    const noPaths = { card: 'search', shape: 'paths', truncated: false, total: 0 } as unknown as ToolResultView
+    expect(searchCardModel(settledGlob({ resultView: noPaths }))).toBeNull()
+    const badPaths = {
+      card: 'search', shape: 'paths', truncated: false, total: 1, paths: [42],
+    } as unknown as ToolResultView
+    expect(searchCardModel(settledGlob({ resultView: badPaths }))).toBeNull()
+  })
+
+  it('surfaces the recovery text only when the result was capped', () => {
+    const recovery = 'a.ts\n  12: const foo = 1\n\n(Full grep result stored at: spill://grep-1. Read it to see every match.)'
+    // The recovery locator lives in the raw tool/result content (the view carries
+    // no text), surfaced only when the card capped the result.
+    const capped = searchCardModel(settledGrep({
+      content: [{ type: 'text', text: recovery }],
+      resultView: resultMatches({ truncated: true, total: 42 }),
+    }))
+    expect(capped?.recovery).toBe(recovery)
+    // Not capped: the card holds every match, so the raw content adds nothing and
+    // is dropped.
+    const whole = searchCardModel(settledGrep({
+      content: [{ type: 'text', text: recovery }],
+      resultView: resultMatches({ truncated: false }),
+    }))
+    expect(whole?.recovery).toBeUndefined()
+    // Capped but the block carries no text: nothing to surface.
+    const noText = searchCardModel(settledGrep({ content: [], resultView: resultMatches({ truncated: true, total: 42 }) }))
+    expect(noText?.recovery).toBeUndefined()
+  })
+})
+
+describe('chat row search body (GenericToolCard fallback)', () => {
+  const ownerProps = (block: RunningToolCall | ToolResultNode, toolName: string): GenericToolCardProps => ({
+    callId: 'c1', toolName, block, openFile: vi.fn(), t,
+  })
+  /** The whole summary row is the expand toggle (ToolRow's unified interaction). */
+  const toggleRow = (view: { container: HTMLElement }) => {
+    fireEvent.click(view.container.querySelector('[data-expandable]')!)
+  }
+
+  it('the expanded body is the grouped matches, capped tighter than the panel', () => {
+    expect(CHAT_SEARCH_MAX_LINES).toBeLessThan(16)
+    const view = render(<GenericToolCard {...ownerProps(settledGrep(), 'grep')} />)
+    // Collapsed: the one-line summary row only, no card.
+    expect(view.queryByText(/const foo = 1/)).toBeNull()
+    toggleRow(view)
+    expect(searchRows(view.container)).toContain('12: const foo = 1')
+    expect(view.getByText('a.ts')).toBeTruthy()
+    expect(searchKindOf(view.container)).toBe('matches')
+    // The args JSON body the generic path would have shown is gone.
+    expect(view.queryByText(/"pattern"/)).toBeNull()
+  })
+
+  it('the glob fallback expands to the flat path card', () => {
+    const view = render(<GenericToolCard {...ownerProps(settledGlob(), 'glob')} />)
+    toggleRow(view)
+    expect(view.getByText('src/a.ts')).toBeTruthy()
+    expect(searchKindOf(view.container)).toBe('paths')
+  })
+
+  it('a non-search result keeps the args-JSON text body', () => {
+    const view = render(<GenericToolCard {...ownerProps(settledGrep({
+      resultView: { card: 'generic' },
+    }), 'grep')} />)
+    toggleRow(view)
+    expect(view.getByText(/"pattern"/)).toBeTruthy()
+    expect(searchKindOf(view.container)).toBeNull()
+  })
+
+  it('the expanded body shows the recovery footer below a capped card', () => {
+    const recovery = 'a.ts\n  12: const foo = 1\n\n(Full grep result stored at: spill://grep-1. Read it to see every match.)'
+    const view = render(<GenericToolCard {...ownerProps(settledGrep({
+      content: [{ type: 'text', text: recovery }],
+      resultView: resultMatches({ truncated: true, total: 42 }),
+    }), 'grep')} />)
+    toggleRow(view)
+    expect(searchKindOf(view.container)).toBe('matches')
+    expect(view.getByText(/Full grep result stored at: spill:\/\/grep-1/)).toBeTruthy()
+  })
+})
+
+describe('SearchRow keyed card', () => {
+  const rowProps = (block: RunningToolCall | ToolResultNode, toolName: string): ToolRowProps => ({
+    callId: 'c1', toolName, block, openFile: vi.fn(), sessionId: SID,
+  } as unknown as ToolRowProps)
+
+  it('renders the grep card resident under the summary row, without an expand gesture', () => {
+    const view = render(<SearchRow {...rowProps(settledGrep(), 'grep')} />)
+    expect(view.getByText('Search')).toBeTruthy()
+    expect(searchRows(view.container)).toContain('12: const foo = 1')
+    expect(searchKindOf(view.container)).toBe('matches')
+    // The card's controls are the row's only interactions.
+    expect(view.getByText('复制')).toBeTruthy()
+  })
+
+  it('renders the glob path card resident', () => {
+    const view = render(<SearchRow {...rowProps(settledGlob(), 'glob')} />)
+    expect(view.getByText('src/a.ts')).toBeTruthy()
+    expect(searchKindOf(view.container)).toBe('paths')
+  })
+
+  it('agrees with the summary row about the run state', () => {
+    const runningView = render(<SearchRow {...rowProps(runningGrep(), 'grep')} />)
+    expect(runningView.container.querySelector('[data-variant="search"]')?.getAttribute('data-state')).toBe('running')
+    // No result view yet, so no resident card.
+    expect(searchKindOf(runningView.container)).toBeNull()
+    cleanup()
+    const errorView = render(<SearchRow {...rowProps(settledGrep({
+      isError: true, resultView: { card: 'generic' },
+    }), 'grep')} />)
+    expect(errorView.container.querySelector('[data-variant="search"]')?.getAttribute('data-state')).toBe('error')
+  })
+
+  it('surfaces the result text when an errored search has no card', () => {
+    // grep/glob return no presentResult on error → no card; the row shows the
+    // model-facing error text instead of a bare red dot.
+    const view = render(<SearchRow {...rowProps(settledGrep({
+      isError: true, resultView: null,
+      content: [{ type: 'text', text: 'grep: invalid regular expression' }],
+    }), 'grep')} />)
+    expect(searchKindOf(view.container)).toBeNull()
+    expect(view.getByText('grep: invalid regular expression')).toBeTruthy()
+  })
+
+  it('surfaces the result text for a settled non-error call with no card', () => {
+    // A successful nested run_code sub-dispatch (backend computes no
+    // presentationMeta, so resultView is null) or a legacy generic result settles
+    // with search === null and state ok. The keyed SearchRow owns the slot, so
+    // without the widened arm the content would be lost behind a bare summary.
+    const view = render(<SearchRow {...rowProps(settledGrep({
+      isError: false, resultView: null,
+      content: [{ type: 'text', text: 'nested run_code output line' }],
+    }), 'grep')} />)
+    expect(view.container.querySelector('[data-variant="search"]')?.getAttribute('data-state')).toBe('ok')
+    expect(searchKindOf(view.container)).toBeNull()
+    expect(view.getByText('nested run_code output line')).toBeTruthy()
+  })
+
+  it('renders the recovery footer below the card when the search was capped', () => {
+    const recovery = 'a.ts\n  12: const foo = 1\n\n(Full grep result stored at: spill://grep-1. Read it to see every match.)'
+    const view = render(<SearchRow {...rowProps(settledGrep({
+      content: [{ type: 'text', text: recovery }],
+      resultView: resultMatches({ truncated: true, total: 42 }),
+    }), 'grep')} />)
+    expect(searchKindOf(view.container)).toBe('matches')
+    expect(view.getByText(/Full grep result stored at: spill:\/\/grep-1/)).toBeTruthy()
+  })
+
+  it('shows no recovery footer for an uncapped search', () => {
+    const view = render(<SearchRow {...rowProps(settledGrep(), 'grep')} />)
+    expect(view.container.textContent).not.toMatch(/stored at/)
+  })
+
+  it('falls back to the error name/code when an errored result has no text block', () => {
+    const view = render(<SearchRow {...rowProps(settledGrep({
+      isError: true, resultView: null, content: [],
+      error: { name: 'ToolError', code: 'timeout' },
+    }), 'grep')} />)
+    expect(view.getByText('ToolError: timeout')).toBeTruthy()
+  })
+
+  it('shows the result view\'s replacement title instead of the args summary', () => {
+    const view = render(<SearchRow {...rowProps(settledGrep({
+      resultView: resultMatches({ title: '3 matches in 2 files' }),
+    }), 'grep')} />)
+    expect(view.getByText('3 matches in 2 files')).toBeTruthy()
+  })
+
+  it('keeps the args-derived summary when the result view has no title', () => {
+    const view = render(<SearchRow {...rowProps(settledGrep(), 'grep')} />)
+    expect(view.getByText('foo')).toBeTruthy()
+  })
+
+  it('registers the one row component under both grep and glob keys', () => {
+    const registered: { key: unknown; component: unknown }[] = []
+    const ctx = {
+      slots: {
+        register: (options: { name: string; key: string }, component: unknown) => {
+          registered.push({ key: options.key, component })
+        },
+      },
+    } as never
+    searchToolview.apply(ctx)
+    expect(registered.map(r => r.key)).toEqual(['grep', 'glob'])
+    // One component, two keys.
+    expect(registered[0]!.component).toBe(SearchRow)
+    expect(registered[1]!.component).toBe(SearchRow)
+    expect(searchToolview.inject).toEqual(['slots', 'conversation'])
+  })
+})
+
+describe('DetailsPanel Output section (search)', () => {
+  function mount(snapshot: ConversationSnapshot, selection: SelectionTarget | null) {
+    localStorage.clear()
+    const chat = createChatStore().create()
+    if (selection !== null) chat.actions.select(selection)
+    const sessions = createSnapshotStore<SessionListState>({ ids: [], byId: {}, current: undefined, phase: 'ready' })
+    const workspaces = createSnapshotStore<WorkspaceListState>({
+      items: [], archivedSessionIds: [], state: 'idle', phase: 'ready', error: null,
+      baselinesReady: true, recentWorkspaceId: undefined,
+    })
+    return render(
+      <DetailsPanel
+        sessionId={SID}
+        useSession={bindSnapshotSelector({ getSnapshot: () => snapshot, subscribe: () => () => {} })}
+        useSessions={bindSnapshotSelector(sessions)}
+        useWorkspaces={bindSnapshotSelector(workspaces)}
+        useInput={(() => { throw new Error('unused') })}
+        inputActions={{ setDraft: () => {}, submit: () => {} }}
+        useProjection={(() => undefined)}
+        useStore={bindSnapshotSelector(chat)}
+        actions={chat.actions}
+        closeDetails={vi.fn()}
+        t={t}
+      />,
+    )
+  }
+
+  function snapshot(over: Partial<ConversationSnapshot> = {}): ConversationSnapshot {
+    return {
+      sessionId: SID, nodes: [], foldDegraded: false, partial: null, runningCalls: [], codeDispatches: new Map(),
+      pending: [], queue: [], running: false, composerPhase: 'active', removed: false,
+      openState: 'open', openError: null, hasMore: false, loadingOlder: false,
+      promptError: null, blank: false, lastAgentError: null, ...over,
+    }
+  }
+
+  const grepTarget: SelectionTarget = { turnSeq: 10, callId: 'c1', toolName: 'grep' }
+  const globTarget: SelectionTarget = { turnSeq: 11, callId: 'c2', toolName: 'glob' }
+
+  it('renders the grep matches card at full height, keeping the JSON Input section', () => {
+    const view = mount(snapshot({ nodes: [settledGrep()] }), grepTarget)
+    expect(view.getByText(/"pattern"/)).toBeTruthy()
+    expect(searchRows(view.container)).toContain('12: const foo = 1')
+    expect(searchKindOf(view.container)).toBe('matches')
+  })
+
+  it('renders the glob path card', () => {
+    const view = mount(snapshot({ nodes: [settledGlob()] }), globTarget)
+    expect(view.getByText('src/a.ts')).toBeTruthy()
+    expect(searchKindOf(view.container)).toBe('paths')
+  })
+
+  it('renders the recovery footer below the card for a capped search', () => {
+    const recovery = 'src/a.ts\nsrc/b.ts\n\n(Showing 2 of 23 paths. Full sorted result stored at: spill://glob-7.)'
+    const view = mount(snapshot({
+      nodes: [settledGlob({ content: [{ type: 'text', text: recovery }], resultView: resultPaths({ truncated: true, total: 23 }) })],
+    }), globTarget)
+    expect(searchKindOf(view.container)).toBe('paths')
+    expect(view.getByText(/Full sorted result stored at: spill:\/\/glob-7/)).toBeTruthy()
+  })
+
+  it('a non-search result keeps the flattened pre form', () => {
+    const view = mount(snapshot({
+      nodes: [settledGrep({ callView: null, resultView: null })],
+    }), grepTarget)
+    expect(searchKindOf(view.container)).toBeNull()
+    const output = view.getByText('输出').closest('section')
+    expect(output?.querySelector('pre')?.textContent).toContain('const foo = 1')
+  })
+})

packages/client/ui-primitives/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/client/ui-primitives/README.md
-README.md: 58be01d56a85c66a144df3f8054840961e987403
-README.zh.md: 2efbec77e64d664553e93b5a8f8dcd2ec7fce49e
+README.md: 6430a789c15634538a38d6581df50a489522db55
+README.zh.md: 78249612cce3148fcececded40c529682450460a

packages/client/ui-primitives/README.md

@@ -2,7 +2,7 @@
 
 English | [中文](README.zh.md)
 
-Pure React atoms (zero cordis): StateDot, ic_ds_* icons, Button/Pill/Menu/Modal/Input, the markdown family (MessageText/MarkdownText/JsonBlock), the read-only JsonTree inspector, the `useAnchoredMaxHeight` hook that clamps a bottom-anchored overlay to the viewport space above its anchor (re-measured on resize, scroll, and a caller-supplied dependency), TerminalBlock, DiffBlock, and WebBlock. Contract: api-contracts v3 §8.
+Pure React atoms (zero cordis): StateDot, ic_ds_* icons, Button/Pill/Menu/Modal/Input, the markdown family (MessageText/MarkdownText/JsonBlock), the read-only JsonTree inspector, the `useAnchoredMaxHeight` hook that clamps a bottom-anchored overlay to the viewport space above its anchor (re-measured on resize, scroll, and a caller-supplied dependency), TerminalBlock, SearchBlock, DiffBlock, and WebBlock. Contract: api-contracts v3 §8.
 
 ## Markdown rendering
 
@@ -12,6 +12,10 @@ Pure React atoms (zero cordis): StateDot, ic_ds_* icons, Button/Pill/Menu/Modal/
 
 `TerminalBlock` renders a shell command as a terminal surface: one prompt row per line of the command (the shortened `cwd` label on the first row only, since the view knows one working directory and a `cd` moves later lines elsewhere, then that line), the command's output, a status pill for a non-zero exit code or a terminating signal, and a copy control that writes the raw `output` prop. A run-state `StateDot` marks the call once, on the first row, out of flow in a gutter the card reserves as its own left padding, so the dot sits inside the card box yet left of the prompt text. It reaches three of `StateDot`'s states — the chase while `running`, red for the same exit status that renders the pill, green otherwise — so a card states whether its command is still running rather than leaving that to be inferred from the presence of output; it carries one visually hidden text label because `StateDot` is `aria-hidden`. One dot regardless of line count is deliberate: the exit status is the whole call's, so a dot per line would claim a per-line outcome the view does not carry. Command text is `white-space: pre`, so repeated spaces, tabs, and an indented continuation render verbatim while the row stays single-line and ellipsizes. ANSI escape sequences are parsed with the `anser` runtime dependency into React spans; cursor movements replay into a per-line column buffer before inert controls are stripped, since carriage return and backspace only MOVE the cursor: `100%` + CR + `OK` alone shows `OK0%`, while the `\x1b[K` a spinner writes with its redraw erases the tail so `100%\r\x1b[KOK` shows `OK`. Erase-in-line is honored in all three parameter forms, the cursor advances by terminal columns (8-column tab stops, two for emoji and CJK, none for a combining mark), and SGR state is normalized per cell as a terminal stores it, threading across lines and closing at the state the line ended in; basic-16 foreground colors map onto `--dsw-*` tokens, while 256-palette and truecolor values pass through as literal rgb. Output keeps `white-space: pre` with horizontal scrolling, so column-aligned output holds its alignment instead of soft-wrapping, and collapses to a head slice plus a tail slice past `maxLines` (default 16, the TUI transcript's split arithmetic) behind an expand button. Rationale: [the web terminal card note](../../../.agents/notes/implemented/feature/2026-07-28-web-terminal-card.md).
 
+## Search results
+
+`SearchBlock` renders a completed search, one component for both kinds (discriminated by `kind`). A `matches` (grep) shows each file as a bold path header with its `lineNumber: line` rows, the per-file group collapsible; a `paths` (glob) shows a flat path list. Both flatten to one row list the height cap slices head/tail over (default 16, the TerminalBlock split arithmetic), and neither soft-wraps — a long match line or path scrolls horizontally instead of folding. The banner summary folds the pre-cap total in when the tool capped the result (`显示 X / 共 N 处匹配 · K 个文件` for grep, `显示 X / 共 N 个路径` for glob), so the card never presents a capped result as complete; a copy control writes the whole structured result regardless of the cap or which groups are collapsed. Geometry mirrors CodeBlock/TerminalBlock. Rationale: [the web search card note](../../../.agents/notes/implemented/feature/2026-07-30-web-search-card.md).
+
 ## Diff rendering
 
 `DiffBlock` renders a file mutation as an inline diff surface: one bold path header per file, the removed lines (`- `, error token) above the added lines (`+ `, success token), a `⋯` gap before a same-file second hunk, and a dim `└ +A -R · N file(s)` footer. Lines are `white-space: pre` with horizontal scrolling, so a source line holds its indentation instead of soft-wrapping, and the body collapses to a head slice plus a tail slice past `maxLines` (default 16, `TerminalBlock`'s split arithmetic) behind an expand button. A create (`oldText: null`) has no removed side. The copy control writes the prefixed diff text (path headers, `- `/`+ ` lines, the gap) so a multi-file copy stays attributable, and floats in the top-right corner rather than on a banner row of its own. Geometry mirrors `CodeBlock`/`TerminalBlock`. The `+`/`-` block form mirrors the TUI transcript's diff card so a diff reads the same across front ends. Rationale: [the web diff card note](../../../.agents/notes/implemented/feature/2026-07-30-web-diff-card.md).

packages/client/ui-primitives/README.zh.md

@@ -2,7 +2,7 @@
 
 [English](README.md) | 中文
 
-纯 React 原子组件（零 cordis）：StateDot、ic_ds_* 图标、Button/Pill/Menu/Modal/Input、markdown 家族（MessageText/MarkdownText/JsonBlock）、只读 JsonTree 检查器、`useAnchoredMaxHeight` hook（把底部锚定的浮层高度收敛到锚点上方的视口空间，并在 resize、scroll 与调用方提供的依赖变化时重新测量）、TerminalBlock、DiffBlock，以及 WebBlock。契约：api-contracts v3 §8。
+纯 React 原子组件（零 cordis）：StateDot、ic_ds_* 图标、Button/Pill/Menu/Modal/Input、markdown 家族（MessageText/MarkdownText/JsonBlock）、只读 JsonTree 检查器、`useAnchoredMaxHeight` hook（把底部锚定的浮层高度收敛到锚点上方的视口空间，并在 resize、scroll 与调用方提供的依赖变化时重新测量）、TerminalBlock、SearchBlock、DiffBlock，以及 WebBlock。契约：api-contracts v3 §8。
 
 ## Markdown 渲染
 
@@ -11,6 +11,10 @@
 
 `TerminalBlock` 将一条 shell 命令渲染为终端表层：命令的每一行各占一个提示行（缩短后的 `cwd` 标签只出现在第一行，因为视图只知道一个工作目录，而一个 `cd` 就会让后面的行去到别处，标签之后是该行）、命令输出、非零退出码或终止信号对应的状态胶囊，以及写入原始 `output` prop 的复制控件。一枚运行状态 `StateDot` 为整次调用标记一次，位于第一行，以脱离文档流的方式落在卡片以自身左内边距预留的落区中，因此它位于卡片盒之内、提示文字之左。它用到 `StateDot` 的三种状态——`running` 期间为追逐动画，与渲染状态胶囊相同的退出状态为红色，其余为绿色——因此卡片直接陈述其命令是否仍在运行，而不是让人从有无输出中推断；由于 `StateDot` 是 `aria-hidden`，它携带一处视觉隐藏的文本标签。无论多少行都只有一枚状态点是有意为之：退出状态属于整次调用，因此每行一枚就会声称一个视图并不携带的逐行结果。命令文本使用 `white-space: pre`，因此重复空格、制表符与缩进续行都原样呈现，同时该行仍保持单行并以省略号截断。ANSI 转义序列通过运行时依赖 `anser` 解析为 React span；光标移动在剥除无显示意义控制符之前先重放进逐行的列缓冲，因为回车与退格**只移动**光标：单是 `100%` 加回车再加 `OK` 显示为 `OK0%`，而 spinner 随重绘写出的 `\x1b[K` 会擦掉尾巴，因此 `100%\r\x1b[KOK` 显示为 `OK`。行内擦除的三种参数形式都被遵循，光标按终端列推进（8 列制表位；emoji 与 CJK 占两列；组合标记不占列），SGR 状态按单元格归一化存储，与终端一致，并跨行延续、在行结束时的状态处收束；基础 16 色前景色映射到 `--dsw-*` token，而 256 色板与真彩色值按字面 rgb 透传。输出保持 `white-space: pre` 并支持横向滚动，因此按列对齐的输出保留其对齐而不会软换行；超过 `maxLines`（默认 16，与 TUI 转录相同的切分算法）时折叠为头部切片加尾部切片，由展开按钮控制。原理：[Web 终端卡片笔记](../../../.agents/notes/implemented/feature/2026-07-28-web-terminal-card.md)。
 
+## 搜索结果
+
+`SearchBlock` 渲染一次已完成的搜索,一个组件绘制两种 kind(由 `kind` 判别)。`matches`(grep)把每个文件渲染为粗体路径头加其 `lineNumber: line` 行,每个文件组可折叠;`paths`(glob)渲染扁平路径列表。两者都摊平成一个行列表,由高度上限做头/尾切片(默认 16,与 TerminalBlock 相同的切分算法),且都不软换行——长匹配行或路径横向滚动而非折行。当工具截断结果时,banner 摘要把截断前总数折入(grep 为 `显示 X / 共 N 处匹配 · K 个文件`,glob 为 `显示 X / 共 N 个路径`),使卡片绝不把截断结果呈现为完整;复制控件写入完整结构化结果,无论是否触及上限或哪些组被折叠。几何镜像 CodeBlock/TerminalBlock。原理:[Web 搜索卡片笔记](../../../.agents/notes/implemented/feature/2026-07-30-web-search-card.md)。
+
 ## Diff 渲染
 
 `DiffBlock` 将一次文件改动渲染为内联 diff 表层：每个文件一个粗体路径头、删除行（`- `，error token）在新增行（`+ `，success token）之上、同文件第二个 hunk 前一个 `⋯` gap，以及暗色 `└ +A -R · N file(s)` 页脚。各行使用 `white-space: pre` 并横向滚动，因此源码行保留其缩进而不软换行；超过 `maxLines`（默认 16，与 `TerminalBlock` 相同的切分算法）时折叠为头部切片加尾部切片，由展开按钮控制。新建（`oldText: null`）没有删除侧。复制控件写入带前缀的 diff 文本（路径头、`- `/`+ ` 行、gap），使多文件复制保持可归属，并浮在右上角而非占据自己的 banner 行。几何镜像 `CodeBlock`/`TerminalBlock`。`+`/`-` 块形式镜像 TUI 转录的 diff 卡片，使 diff 在两个前端读起来一致。原理：[Web diff 卡片笔记](../../../.agents/notes/implemented/feature/2026-07-30-web-diff-card.md)。

packages/client/ui-primitives/src/ReadBlock.module.css

@@ -0,0 +1,117 @@
+/* Geometry mirrors CodeBlock (12px radius, code-block surface + banner row,
+   markdown code-block font) so a read card and a fenced code block read as one
+   family. Content keeps `white-space: pre` and scrolls horizontally rather than
+   folding, because a source line's indentation is part of what a reader is
+   reading. */
+
+.block {
+  --dsl-read-radius: 12px;
+  --dsl-read-line-height: 22px;
+  /* Fixed-width gutter column for the line numbers, so the content edge stays
+     put down the whole window regardless of how wide the numbers grow. */
+  --dsl-read-gutter: 48px;
+
+  position: relative;
+  margin: 16px 0;
+  color: var(--dsw-alias-label-primary);
+  background: var(--dsw-alias-markdown-code-block);
+  border-radius: var(--dsl-read-radius);
+}
+
+.banner {
+  display: flex;
+  justify-content: space-between;
+  align-items: center;
+  gap: 12px;
+  padding: 9px 14px;
+  background: var(--dsw-alias-markdown-code-block-banner);
+  border-top-left-radius: var(--dsl-read-radius);
+  border-top-right-radius: var(--dsl-read-radius);
+}
+
+.label {
+  min-width: 0;
+  overflow: hidden;
+  text-overflow: ellipsis;
+  white-space: nowrap;
+  color: var(--dsw-alias-label-primary);
+  font-family: var(--ds-font-family-code);
+  font-size: 12px;
+  line-height: 18px;
+}
+
+.action {
+  display: flex;
+  align-items: center;
+  flex-shrink: 0;
+  gap: 12px;
+}
+
+.count {
+  color: var(--dsw-alias-label-tertiary);
+  font: var(--dsw-font-xs-13);
+}
+
+.lang {
+  color: var(--dsw-alias-label-tertiary);
+  font-family: var(--ds-font-family-code);
+  font-size: 12px;
+  line-height: 18px;
+}
+
+.copyButton {
+  background-color: transparent;
+  border: none;
+  padding: 0;
+  margin: 0;
+  color: var(--dsw-alias-label-secondary);
+  cursor: pointer;
+  font: var(--dsw-font-xs-13);
+}
+
+.body {
+  padding: 12px 0;
+  font: var(--dsw-font-markdown-code-block);
+  overflow-x: auto;
+  overflow-y: hidden;
+}
+
+/* One row per file line: a fixed gutter column, then the content. No wrapping —
+   a source line's leading whitespace is meaningful and scrolls sideways. */
+.line {
+  display: flex;
+  min-height: var(--dsl-read-line-height);
+  line-height: var(--dsl-read-line-height);
+  white-space: pre;
+}
+
+.gutter {
+  flex: none;
+  width: var(--dsl-read-gutter);
+  padding-right: 14px;
+  text-align: right;
+  color: var(--dsw-alias-label-tertiary);
+  /* The gutter is chrome, not content: keep it out of a text selection so a
+     copy of the visible rows carries the source, not the line numbers. */
+  user-select: none;
+}
+
+.content {
+  color: var(--dsw-alias-label-primary);
+}
+
+.expand {
+  display: block;
+  width: 100%;
+  padding: 0 0 0 var(--dsl-read-gutter);
+  border: none;
+  background-color: transparent;
+  color: var(--dsw-alias-label-tertiary);
+  cursor: pointer;
+  font: inherit;
+  text-align: left;
+}
+
+.expand:hover {
+  color: var(--dsw-alias-label-secondary);
+}

packages/client/ui-primitives/src/ReadBlock.tsx

@@ -0,0 +1,172 @@
+// ReadBlock: the file surface for a read tool result — a banner (label +
+// language + a "showing N of M" note when the read is a window + a copy
+// control) over line-numbered, syntax-highlighted source. Each row carries the
+// file's OWN line number in a gutter, so a windowed read past an offset keeps
+// its file numbering rather than re-counting from 1. Highlighting reuses the
+// CodeBlock shiki path (highlight.ts) at the per-line granularity a gutter
+// needs; an unknown or absent language renders plain monospace. Long content is
+// height-capped with the same head/tail arithmetic TerminalBlock uses, so the
+// two cards collapse a long body at the same place. Colors resolve through
+// --shiki-*/--dsw-* tokens.
+
+import { useCallback, useMemo, useState, useSyncExternalStore } from 'react'
+import clsx from 'clsx'
+import { writeClipboard } from './clipboard.ts'
+import {
+  grammarLoadCount,
+  highlightLines,
+  subscribeGrammarLoaded,
+  type HighlightSpan,
+} from './markdown/highlight.ts'
+import css from './ReadBlock.module.css'
+
+/**
+ * Content lines shown before the height cap collapses the middle. Matches
+ * TerminalBlock's default so a long read and a long command output cut at the
+ * same place in the same flow.
+ */
+export const DEFAULT_READ_MAX_LINES = 16
+
+/** One line of the read window: its file line number and its text (no trailing newline). */
+export interface ReadBlockLine {
+  /** 1-based line number in the file (a window past an offset keeps the file's own numbering). */
+  number: number
+  /** The line's text, already truncated to the read tool's per-line cap. */
+  text: string
+}
+
+export interface ReadBlockProps {
+  /** Banner label (the file path, or a tool-supplied replacement title); omitted draws no label. */
+  label?: string | undefined
+  /** The returned window's lines, in file order, each keeping its file line number. */
+  lines: readonly ReadBlockLine[]
+  /** Exact total line count in the file, for the "showing N of M" note when the read is a window. */
+  totalLines: number
+  /** Grammar hint (a file-extension-derived language id); unknown or absent = plain monospace. */
+  lang?: string | undefined
+  /** Height cap in content lines before the middle collapses (default {@link DEFAULT_READ_MAX_LINES}). */
+  maxLines?: number | undefined
+  /** Extra class merged onto the wrapper (callers position; this component draws). */
+  className?: string | undefined
+}
+
+/**
+ * Render one line's highlighted runs. The css-variables theme colors every run,
+ * so each run is a styled span; a line with no highlighting at all takes the
+ * bare-text path in the caller instead (an unknown or absent language).
+ * @param spans - the line's styled runs.
+ * @returns the line's children.
+ */
+function renderSpans(spans: readonly HighlightSpan[]) {
+  return spans.map((span, index) => <span key={index} style={span.style}>{span.text}</span>)
+}
+
+/**
+ * Render a read tool result as a line-numbered, optionally syntax-highlighted
+ * file view.
+ * @param props - see {@link ReadBlockProps}.
+ * @returns the read block element.
+ */
+export function ReadBlock({
+  label,
+  lines,
+  totalLines,
+  lang,
+  maxLines = DEFAULT_READ_MAX_LINES,
+  className,
+}: ReadBlockProps) {
+  // The raw text the copy control writes and the highlighter tokenizes: the
+  // window's lines joined by newlines, without the file numbers or any chrome.
+  // Highlighting the whole window in one call (not line by line) keeps grammar
+  // context across lines — a multi-line string or comment stays one construct.
+  const raw = useMemo(() => lines.map(line => line.text).join('\n'), [lines])
+  // Re-render when a lazy grammar finishes loading, so a read card that showed
+  // plain text while its language's grammar imported picks up highlighting. The
+  // snapshot value is opaque; only its change across renders drives the memo.
+  const loaded = useSyncExternalStore(subscribeGrammarLoaded, grammarLoadCount, grammarLoadCount)
+  // Per-line highlighted runs aligned 1:1 with `lines`; undefined for an
+  // unknown/absent (or not-yet-loaded) language, when every line renders as
+  // bare text.
+  const highlighted = useMemo(() => highlightLines(raw, lang), [raw, lang, loaded])
+  const [expanded, setExpanded] = useState(false)
+  const [copied, setCopied] = useState(false)
+
+  const onCopy = useCallback(() => {
+    if (copied) return
+    // The window's raw text, never the rendered tree: the gutter numbers and the
+    // banner are chrome the file does not contain.
+    void writeClipboard(raw).then((ok) => {
+      if (!ok) return
+      setCopied(true)
+      window.setTimeout(() => { setCopied(false) }, 1000)
+    })
+  }, [copied, raw])
+
+  const onToggle = useCallback(() => { setExpanded(value => !value) }, [])
+
+  const hidden = lines.length - maxLines
+  const capped = hidden > 0 && !expanded
+  // Same split arithmetic as TerminalBlock's height cap, so a long read and a
+  // long command output slice their head and tail at the same place.
+  const headLines = Math.ceil(maxLines / 2)
+  const tailLines = maxLines - headLines
+  // A read is a window when its returned lines are fewer than the file's total;
+  // the note states that so a reader is not misled that the file ends here.
+  const windowed = lines.length < totalLines
+
+  /**
+   * Render a slice of the line array as gutter-numbered rows.
+   * @param slice - the lines to draw, each with its aligned run array.
+   * @returns the row elements.
+   */
+  const rows = (slice: readonly (readonly [ReadBlockLine, readonly HighlightSpan[] | undefined])[]) =>
+    slice.map(([line, spans]) => (
+      <div key={line.number} className={css.line}>
+        <span className={css.gutter} aria-hidden>{line.number}</span>
+        <span className={css.content}>{spans === undefined ? line.text : renderSpans(spans)}</span>
+      </div>
+    ))
+
+  // Pair each line with its aligned run array up front, so head/tail slicing
+  // keeps the two in step without re-indexing.
+  const paired = lines.map((line, index): readonly [ReadBlockLine, readonly HighlightSpan[] | undefined] =>
+    [line, highlighted?.[index]])
+
+  return (
+    <div className={clsx(css.block, className)} data-read="">
+      <div className={css.banner}>
+        <div className={css.label}>{label ?? ''}</div>
+        <div className={css.action}>
+          {windowed && (
+            <span className={css.count}>{`显示 ${lines.length} / ${totalLines} 行`}</span>
+          )}
+          <span className={css.lang}>{lang ?? ''}</span>
+          {/* Hide copy on an empty window, matching TerminalBlock's empty-output
+              guard: a successful read of an empty file returns lines: [] with
+              card:'read', so this branch is reachable, and copying then would
+              wipe the clipboard with an empty string. */}
+          {lines.length > 0 && (
+            <button type="button" className={css.copyButton} onClick={onCopy}>
+              {copied ? '复制成功' : '复制'}
+            </button>
+          )}
+        </div>
+      </div>
+      <div className={css.body}>
+        {rows(capped ? paired.slice(0, headLines) : paired)}
+        {hidden > 0 && (
+          <button
+            type="button"
+            className={css.expand}
+            aria-expanded={expanded}
+            aria-label={expanded ? '收起内容' : `展开其余 ${hidden} 行`}
+            onClick={onToggle}
+          >
+            {expanded ? '收起' : `… 其余 ${hidden} 行`}
+          </button>
+        )}
+        {capped && rows(paired.slice(paired.length - tailLines))}
+      </div>
+    </div>
+  )
+}

packages/client/ui-primitives/src/SearchBlock.module.css

@@ -0,0 +1,120 @@
+/* Geometry mirrors CodeBlock and TerminalBlock (12px radius, code-block
+   surface + banner row, markdown code-block font) so a search card reads as one
+   family with them. The deliberate divergence they share: the result rows keep
+   `white-space: pre` and scroll horizontally, because folding a long match line
+   or path destroys the alignment a reader scans by. */
+
+.block {
+  --dsl-search-radius: 12px;
+  --dsl-search-line-height: 22px;
+
+  position: relative;
+  margin: 16px 0;
+  color: var(--dsw-alias-label-primary);
+  background: var(--dsw-alias-markdown-code-block);
+  border-radius: var(--dsl-search-radius);
+}
+
+/* The banner: result summary on the left, the copy control holding its
+   intrinsic width on the right. */
+.header {
+  display: flex;
+  align-items: center;
+  gap: 12px;
+  padding: 9px 14px;
+  background: var(--dsw-alias-markdown-code-block-banner);
+  border-top-left-radius: var(--dsl-search-radius);
+  border-top-right-radius: var(--dsl-search-radius);
+}
+
+.summary {
+  flex: 1;
+  min-width: 0;
+  overflow: hidden;
+  text-overflow: ellipsis;
+  white-space: nowrap;
+  font: var(--dsw-font-xs-13);
+  color: var(--dsw-alias-label-secondary);
+}
+
+.copyButton {
+  flex: none;
+  background-color: transparent;
+  border: none;
+  padding: 0;
+  margin: 0;
+  color: var(--dsw-alias-label-secondary);
+  cursor: pointer;
+  font: var(--dsw-font-xs-13);
+}
+
+.body {
+  padding: 8px 14px 12px 0;
+  font: var(--dsw-font-markdown-code-block);
+  overflow-x: auto;
+  overflow-y: hidden;
+}
+
+/* No wrapping: a match line or a path keeps its content on one row and scrolls
+   sideways instead of folding. */
+.line {
+  min-height: var(--dsl-search-line-height);
+  padding-left: 14px;
+  white-space: pre;
+}
+
+/* The 1-based line number ahead of a grep match line, dimmed so the match text
+   stays the salient content. */
+.lineNumber {
+  color: var(--dsw-alias-label-tertiary);
+}
+
+/* A file group's header: a bold path label plus its match count, the whole row
+   the collapse control. */
+.fileHeader {
+  display: flex;
+  align-items: baseline;
+  gap: 8px;
+  width: 100%;
+  min-height: var(--dsl-search-line-height);
+  padding: 0 14px;
+  border: none;
+  background-color: transparent;
+  cursor: pointer;
+  font: inherit;
+  text-align: left;
+}
+
+.filePath {
+  min-width: 0;
+  font-weight: 600;
+  color: var(--dsw-alias-label-primary);
+  white-space: pre;
+}
+
+.fileCount {
+  flex: none;
+  color: var(--dsw-alias-label-tertiary);
+}
+
+.expand {
+  display: block;
+  width: 100%;
+  padding: 0 14px;
+  border: none;
+  background-color: transparent;
+  color: var(--dsw-alias-label-tertiary);
+  cursor: pointer;
+  font: inherit;
+  text-align: left;
+}
+
+.expand:hover {
+  color: var(--dsw-alias-label-secondary);
+}
+
+.empty {
+  padding: 12px 14px;
+  font: var(--dsw-font-markdown-code-block);
+  color: var(--dsw-alias-label-tertiary);
+}

packages/client/ui-primitives/src/SearchBlock.tsx

@@ -0,0 +1,277 @@
+// SearchBlock: the search surface for a completed content or path search — a
+// banner (result summary that folds the pre-cap total in when the tool capped
+// the result, plus a copy control), then either grep matches grouped by file
+// (each file a bold
+// path header with its `lineNumber: line` rows, the group collapsible) or a
+// flat glob path list. Both shapes flatten to one list of rows the height cap
+// slices head/tail over, and neither soft-wraps: a long match line or path
+// scrolls horizontally instead of folding. Geometry mirrors CodeBlock and
+// TerminalBlock so a search card reads as one family with them.
+
+import { useCallback, useState, type ReactNode } from 'react'
+import clsx from 'clsx'
+import { headTailCap } from './head-tail-cap.ts'
+import { useCopyFeedback } from './use-copy-feedback.ts'
+import css from './SearchBlock.module.css'
+
+/**
+ * Result rows shown before the height cap collapses the middle. Matches
+ * {@link DEFAULT_TERMINAL_MAX_LINES} so a search card and a terminal card cut a
+ * long result at the same place.
+ */
+export const DEFAULT_SEARCH_MAX_LINES = 16
+
+/** One matched line inside a {@link SearchFileGroup}: its 1-based line number and text. */
+export interface SearchBlockLineMatch {
+  /** 1-based line number of the match within its file. */
+  lineNumber: number
+  /** The matched line text, as the tool surfaced it. */
+  line: string
+}
+
+/** One file's grouped matches, in first-seen file order. */
+export interface SearchFileGroup {
+  /** The file the matches belong to (the display path). */
+  path: string
+  /** The file's matched lines, in output order. */
+  matches: SearchBlockLineMatch[]
+}
+
+/** Fields both search shapes carry (the render site positions; this component draws). */
+interface SearchBlockCommon {
+  /**
+   * Whether the tool capped the inline result: the shape carries only the
+   * retained results, not every result the search found. The banner summary
+   * folds the pre-cap `total` in (`显示 X / 共 N …`) so the card never presents a
+   * capped result as complete.
+   */
+  truncated: boolean
+  /** Total results the search found before capping (equals the retained count when not `truncated`). */
+  total: number
+  /** Height cap in rows before the middle collapses (default {@link DEFAULT_SEARCH_MAX_LINES}). */
+  maxLines?: number | undefined
+  /** Extra class merged onto the wrapper. */
+  className?: string | undefined
+}
+
+/** Props for the grouped-matches (`grep`) shape. */
+export interface SearchMatchesBlockProps extends SearchBlockCommon {
+  kind: 'matches'
+  /** Matched lines grouped by file, in first-seen file order. */
+  files: SearchFileGroup[]
+}
+
+/** Props for the flat-path (`glob`) shape. */
+export interface SearchPathsBlockProps extends SearchBlockCommon {
+  kind: 'paths'
+  /** The discovered paths, in the tool's result order (the retained page when `truncated`). */
+  paths: string[]
+}
+
+/** {@link SearchBlock} props: one card, two `kind`-discriminated shapes. */
+export type SearchBlockProps = SearchMatchesBlockProps | SearchPathsBlockProps
+
+/**
+ * One flattened render row. A matches card produces a `file` header row per
+ * group followed by a `match` row per retained line while the group is
+ * expanded; a paths card produces one `path` row per path. The height cap
+ * counts these rows uniformly, so a file header costs one row exactly as a
+ * match line or a path does.
+ */
+type SearchRow =
+  | { type: 'file'; path: string; count: number; index: number; collapsed: boolean }
+  | { type: 'match'; lineNumber: number; line: string; key: string; fileIndex: number }
+  | { type: 'path'; path: string }
+
+/**
+ * The plain-text form the copy control writes: the whole structured result
+ * regardless of the height cap or which groups are collapsed, so the clipboard
+ * carries the result rather than what the card happens to be showing.
+ * @param props - the card's props.
+ * @returns the copyable text, or the empty string for an empty result.
+ */
+function copyText(props: SearchBlockProps): string {
+  if (props.kind === 'paths') return props.paths.join('\n')
+  return props.files
+    .map(file => [file.path, ...file.matches.map(m => `${m.lineNumber}: ${m.line}`)].join('\n'))
+    .join('\n\n')
+}
+
+/**
+ * Number of retained results the card holds: the matched-line count across all
+ * files for a matches card, the path count for a paths card. This is the count
+ * the banner summary reports against `total` when the result was capped.
+ * @param props - the card's props.
+ * @returns the retained result count.
+ */
+function shownCount(props: SearchBlockProps): number {
+  return props.kind === 'paths'
+    ? props.paths.length
+    : props.files.reduce((sum, file) => sum + file.matches.length, 0)
+}
+
+/**
+ * The banner summary. When the search was capped it reads `显示 X / 共 N …` so
+ * the retained count and the pre-cap total sit in one clause (mirroring the read
+ * card's `显示 X / Y 行`); when it was not capped it is a plain count of what the
+ * card holds. The unit — `处匹配 · K 个文件` for grep, `个路径` for glob — trails
+ * the count either way.
+ * @param props - the card's props.
+ * @param shown - the retained result count from {@link shownCount}.
+ * @param truncated - whether the search was capped.
+ * @param total - the pre-cap total the truncation clause reports.
+ * @returns the summary text.
+ */
+function summaryText(props: SearchBlockProps, shown: number, truncated: boolean, total: number): string {
+  const count = truncated ? `显示 ${shown} / 共 ${total}` : `${shown}`
+  return props.kind === 'paths'
+    ? `${count} 个路径`
+    : `${count} 处匹配 · ${props.files.length} 个文件`
+}
+
+/**
+ * Flatten a card's shape into its render rows, dropping a collapsed file
+ * group's match rows.
+ * @param props - the card's props.
+ * @param collapsed - the set of collapsed file-group indices (matches only).
+ * @returns the flattened rows in output order.
+ */
+function toRows(props: SearchBlockProps, collapsed: ReadonlySet<number>): SearchRow[] {
+  if (props.kind === 'paths') return props.paths.map((path): SearchRow => ({ type: 'path', path }))
+  const rows: SearchRow[] = []
+  props.files.forEach((file, index) => {
+    const isCollapsed = collapsed.has(index)
+    rows.push({ type: 'file', path: file.path, count: file.matches.length, index, collapsed: isCollapsed })
+    if (isCollapsed) return
+    for (const match of file.matches) {
+      rows.push({ type: 'match', lineNumber: match.lineNumber, line: match.line, key: `${index}:${match.lineNumber}`, fileIndex: index })
+    }
+  })
+  return rows
+}
+
+/**
+ * A stable React key for a flattened render row: the group-scoped match key, a
+ * file-index-scoped header key, or the path itself. Rows of different types
+ * never collide, since each key carries its type prefix or the group index.
+ * @param row - the flattened row.
+ * @returns the key.
+ */
+function rowKey(row: SearchRow): string {
+  switch (row.type) {
+    case 'match': return `match:${row.key}`
+    case 'file': return `file:${row.index}`
+    case 'path': return `path:${row.path}`
+  }
+}
+
+/**
+ * Render a completed search as a grouped-matches or flat-path card.
+ * @param props - see {@link SearchBlockProps}.
+ * @returns the search block element.
+ */
+export function SearchBlock(props: SearchBlockProps) {
+  const { truncated, total, maxLines = DEFAULT_SEARCH_MAX_LINES, className } = props
+  const [expanded, setExpanded] = useState(false)
+  const [collapsed, setCollapsed] = useState<ReadonlySet<number>>(() => new Set())
+
+  // `props` is a fresh object each render, so memoizing on it never hits; the
+  // flatten is cheap, so it runs inline keyed on the collapse set instead.
+  const rows = toRows(props, collapsed)
+  const shown = shownCount(props)
+  const empty = rows.length === 0
+  const { copied, onCopy } = useCopyFeedback(copyText(props))
+
+  const onToggle = useCallback(() => { setExpanded(value => !value) }, [])
+
+  const toggleFile = useCallback((index: number) => {
+    setCollapsed((prev) => {
+      const next = new Set(prev)
+      if (next.has(index)) next.delete(index)
+      else next.add(index)
+      return next
+    })
+  }, [])
+
+  const { hidden, capped, headLines, tailLines } = headTailCap(rows.length, maxLines, expanded)
+  const head = capped ? rows.slice(0, headLines) : rows
+  const naturalTail = capped ? rows.slice(rows.length - tailLines) : []
+  // When the tail slice begins inside a file's matches, its own header sits
+  // above the cut and is not shown, so those rows could not be attributed to a
+  // file. Restore the owning header at the top of the tail — unless the head
+  // slice already carries it (a single large file), where it would duplicate.
+  const tailLead = naturalTail[0]
+  const tailHeader = tailLead?.type === 'match'
+    && !head.some(row => row.type === 'file' && row.index === tailLead.fileIndex)
+    ? rows.find((row): row is Extract<SearchRow, { type: 'file' }> =>
+      row.type === 'file' && row.index === tailLead.fileIndex)
+    : undefined
+  // The restored header is itself a row. Left extra it would push the card to
+  // maxLines + 1 and overstate `hidden` by one, so it consumes a tail slot: drop
+  // the tail's first row (the match whose header this is) for it. Visible rows
+  // hold at maxLines and `hidden` stays exact; the dropped match joins the
+  // hidden middle.
+  const tail = tailHeader === undefined ? naturalTail : naturalTail.slice(1)
+
+  const renderRow = (row: SearchRow): ReactNode => {
+    if (row.type === 'path') return <div className={css.line}>{row.path}</div>
+    if (row.type === 'match') {
+      return (
+        <div className={css.line}>
+          <span className={css.lineNumber}>{row.lineNumber}: </span>
+          {row.line}
+        </div>
+      )
+    }
+    return (
+      <button
+        type="button"
+        className={css.fileHeader}
+        aria-expanded={!row.collapsed}
+        onClick={() => { toggleFile(row.index) }}
+      >
+        <span className={css.filePath}>{row.path}</span>
+        <span className={css.fileCount}>{row.count}</span>
+      </button>
+    )
+  }
+
+  return (
+    <div className={clsx(css.block, className)} data-search={props.kind}>
+      <div className={css.header}>
+        <span className={css.summary}>{summaryText(props, shown, truncated, total)}</span>
+        {!empty && (
+          <button type="button" className={css.copyButton} onClick={onCopy}>
+            {copied ? '复制成功' : '复制'}
+          </button>
+        )}
+      </div>
+      {empty
+        ? <div className={css.empty}>无结果</div>
+        : (
+          <div className={css.body}>
+            {head.map(row => (
+              <div key={rowKey(row)}>{renderRow(row)}</div>
+            ))}
+            {hidden > 0 && (
+              <button
+                type="button"
+                className={css.expand}
+                aria-expanded={expanded}
+                aria-label={expanded ? '收起结果' : `展开其余 ${hidden} 行结果`}
+                onClick={onToggle}
+              >
+                {expanded ? '收起' : `… 其余 ${hidden} 行`}
+              </button>
+            )}
+            {tailHeader !== undefined && (
+              <div key={`tailHeader:${rowKey(tailHeader)}`}>{renderRow(tailHeader)}</div>
+            )}
+            {tail.map(row => (
+              <div key={rowKey(row)}>{renderRow(row)}</div>
+            ))}
+          </div>
+        )}
+    </div>
+  )
+}

packages/client/ui-primitives/src/TerminalBlock.tsx

@@ -8,7 +8,8 @@
 import { useCallback, useMemo, useState } from 'react'
 import clsx from 'clsx'
 import { parseAnsiLines, type AnsiLine } from './ansi.ts'
-import { writeClipboard } from './clipboard.ts'
+import { headTailCap } from './head-tail-cap.ts'
+import { useCopyFeedback } from './use-copy-feedback.ts'
 import { Pill } from './Pill.tsx'
 import { StateDot, type StateDotState } from './StateDot.tsx'
 import css from './TerminalBlock.module.css'
@@ -202,18 +203,9 @@ export function TerminalBlock({
     return terminated ? parsed.slice(0, -1) : parsed
   }, [text])
   const [expanded, setExpanded] = useState(false)
-  const [copied, setCopied] = useState(false)
-
-  const onCopy = useCallback(() => {
-    if (copied) return
-    // The raw output, never the rendered tree: the prompt line and the status
-    // pill are chrome the user did not run.
-    void writeClipboard(text).then((ok) => {
-      if (!ok) return
-      setCopied(true)
-      window.setTimeout(() => { setCopied(false) }, 1000)
-    })
-  }, [copied, text])
+  // The raw output, never the rendered tree: the prompt line and the status pill
+  // are chrome the user did not run.
+  const { copied, onCopy } = useCopyFeedback(text)
 
   const onToggle = useCallback(() => { setExpanded(value => !value) }, [])
 
@@ -232,12 +224,7 @@ export function TerminalBlock({
   // the raw text drew an output box of blank rows plus a copy control for
   // invisible bytes, and hid the placeholder that belongs there.
   const empty = lines.every(line => line.every(span => span.text.trim() === ''))
-  const hidden = lines.length - maxLines
-  const capped = hidden > 0 && !expanded
-  // Same split arithmetic as the TUI transcript's collapsed tool card, so a
-  // command's head and tail slices agree between the two front ends.
-  const headLines = Math.ceil(maxLines / 2)
-  const tailLines = maxLines - headLines
+  const { hidden, capped, headLines, tailLines } = headTailCap(lines.length, maxLines, expanded)
 
   return (
     <div className={clsx(css.block, className)} data-terminal="" data-running={running ? '' : undefined}>

packages/client/ui-primitives/src/head-tail-cap.ts

@@ -0,0 +1,33 @@
+// Head/tail height-cap arithmetic shared by the block primitives (TerminalBlock,
+// SearchBlock) and matching the TUI transcript's collapsed tool card, so a long
+// result's head and tail slices agree across every surface. The split is
+// `ceil(maxLines / 2)` head rows and the remainder as tail rows; a result within
+// the cap shows every row and hides none.
+
+/** The head/tail split metrics for a capped list. */
+export interface HeadTailCap {
+  /** Rows beyond the cap (list length − maxLines); ≤ 0 means nothing is hidden. */
+  hidden: number
+  /** Whether the list is over the cap and not expanded, so it shows a head/tail slice. */
+  capped: boolean
+  /** Head-slice row count: `ceil(maxLines / 2)`. */
+  headLines: number
+  /** Tail-slice row count: the remainder after the head. */
+  tailLines: number
+}
+
+/**
+ * Compute the head/tail cap metrics for a list of `total` rows against `maxLines`,
+ * given whether the surface is expanded. Pure arithmetic; the caller slices its
+ * own rows with `headLines`/`tailLines` so a block can layer its own concerns
+ * (SearchBlock restores a tail file header) on top.
+ * @param total - the list's row count.
+ * @param maxLines - the collapsed-height cap in rows.
+ * @param expanded - whether the surface is expanded (uncaps the list).
+ * @returns the split metrics.
+ */
+export function headTailCap(total: number, maxLines: number, expanded: boolean): HeadTailCap {
+  const hidden = total - maxLines
+  const headLines = Math.ceil(maxLines / 2)
+  return { hidden, capped: hidden > 0 && !expanded, headLines, tailLines: maxLines - headLines }
+}

packages/client/ui-primitives/src/index.ts

@@ -24,6 +24,12 @@ export { JsonTree } from './JsonTree.tsx'
 export type { JsonTreeProps, JsonTreeLabels } from './JsonTree.tsx'
 export { TerminalBlock, DEFAULT_TERMINAL_MAX_LINES } from './TerminalBlock.tsx'
 export type { TerminalBlockProps, TerminalBlockLabels } from './TerminalBlock.tsx'
+export { SearchBlock, DEFAULT_SEARCH_MAX_LINES } from './SearchBlock.tsx'
+export type {
+  SearchBlockProps, SearchMatchesBlockProps, SearchPathsBlockProps, SearchFileGroup, SearchBlockLineMatch,
+} from './SearchBlock.tsx'
+export { ReadBlock, DEFAULT_READ_MAX_LINES } from './ReadBlock.tsx'
+export type { ReadBlockProps, ReadBlockLine } from './ReadBlock.tsx'
 export { DiffBlock, DEFAULT_DIFF_MAX_LINES } from './DiffBlock.tsx'
 export type { DiffBlockProps, DiffHunk } from './DiffBlock.tsx'
 export { WebBlock, DEFAULT_WEB_MAX_SOURCES } from './WebBlock.tsx'

packages/client/ui-primitives/src/markdown/CodeBlock.tsx

@@ -4,10 +4,10 @@
 // plain fallback for everything else. Chrome (language banner + copy) matches
 // deepsuite `@deepseek/md` code blocks; token colors stay on `--shiki-*`.
 
-import { useCallback, useMemo, useRef, useState } from 'react'
+import { useCallback, useMemo, useRef, useState, useSyncExternalStore } from 'react'
 import clsx from 'clsx'
 import { writeClipboard } from '../clipboard.ts'
-import { highlightToHtml } from './highlight.ts'
+import { grammarLoadCount, highlightToHtml, subscribeGrammarLoaded } from './highlight.ts'
 import css from './CodeBlock.module.css'
 
 export interface CodeBlockProps {
@@ -25,7 +25,11 @@ export interface CodeBlockProps {
 
 export function CodeBlock({ code, lang, className, copyLabel = '复制', copiedLabel = '复制成功' }: CodeBlockProps) {
   const trimmed = code.endsWith('\n') ? code.slice(0, -1) : code
-  const html = useMemo(() => highlightToHtml(trimmed, lang), [trimmed, lang])
+  // Re-render when a lazy grammar finishes loading, so a fence that showed plain
+  // text while its language's grammar imported picks up highlighting. The
+  // snapshot value is opaque; only its change across renders drives the memo.
+  const loaded = useSyncExternalStore(subscribeGrammarLoaded, grammarLoadCount, grammarLoadCount)
+  const html = useMemo(() => highlightToHtml(trimmed, lang), [trimmed, lang, loaded])
   const rootRef = useRef<HTMLDivElement>(null)
   const [copied, setCopied] = useState(false)
 

packages/client/ui-primitives/src/markdown/highlight.ts

@@ -5,10 +5,17 @@
  * theme package's token sheets as `--shiki-*` custom properties (light and
  * dark blocks), never here — the repo's tokens-only styling rule.
  *
- * Grammars are the set the harness actually renders: TypeScript programs
- * (`run_code` bodies; TS pulls in JS via grammar embedding), shell commands,
- * and JSON payloads. An unknown or absent language falls back to plain text
- * (no highlighting, still monospace) — never an error.
+ * Only the three markdown-fence and `run_code` grammars (TypeScript, shell,
+ * JSON) load into the singleton at boot — the set every session renders. The
+ * read card's wider extension set (the file-extension language hints the read
+ * tool's `langFromPath` emits — `packages/fs/tool-fs`: python, rust, yaml,
+ * markup, …) is imported lazily and registered the first time such a language
+ * is requested, so a session that never opens a read card in one of those
+ * languages pays neither the ~1.6 MB of grammar modules nor their synchronous
+ * init. The first render of a lazy language falls back to plain text while its
+ * grammar loads, then {@link onGrammarLoaded} notifies subscribers to re-render
+ * with highlighting. An unknown or absent language falls back to plain text (no
+ * highlighting, still monospace) — never an error.
  */
 
 import { createHighlighterCoreSync, createCssVariablesTheme } from 'shiki/core'
@@ -17,12 +24,69 @@ import langTs from '@shikijs/langs/typescript'
 import langBash from '@shikijs/langs/shellscript'
 import langJson from '@shikijs/langs/json'
 import type { HighlighterCore } from 'shiki/core'
+import type { CSSProperties } from 'react'
+
+/** A shiki grammar module's default export (a `LanguageRegistration[]`), taken
+ *  from a boot grammar so no direct `@shikijs/types` dependency is needed. */
+type LangModule = { default: typeof langTs }
 
 /**
- * Language ids (and aliases) the singleton registers; everything else renders
+ * Grammars the singleton loads at boot; each entry's own `name` is the id
+ * `codeToTokens`/`codeToHtml` resolve. The JS-family aliases (js/jsx/ts/tsx)
+ * resolve to the TypeScript grammar rather than a separate one: it tokenizes
+ * plain TS/JS exactly, and JSX/TSX approximately (shiki's TS grammar is not the
+ * dedicated TSX grammar, so JSX elements tokenize imperfectly) — an accepted
+ * trade to keep the boot set to one JS-family grammar. The read card's wider
+ * set loads lazily through {@link LAZY_GRAMMARS}.
+ */
+const LANGS = [langTs, langBash, langJson]
+
+/**
+ * The read card's extension grammars, each behind a dynamic import so its
+ * module stays out of the boot chunk until a read of that language renders.
+ * Keyed by the grammar id (`LanguageRegistration.name`) the aliases resolve to.
+ * `@shikijs/langs`' default export is a `LanguageRegistration[]`; the loader
+ * hands the whole array to `loadLanguageSync`, which registers each entry
+ * (including embedded sub-grammars). The three boot grammars are absent —
+ * already loaded, so no alias value ever points at a missing entry here.
+ */
+const LAZY_GRAMMARS = new Map<string, () => Promise<LangModule>>([
+  ['python', () => import('@shikijs/langs/python')],
+  ['ruby', () => import('@shikijs/langs/ruby')],
+  ['go', () => import('@shikijs/langs/go')],
+  ['rust', () => import('@shikijs/langs/rust')],
+  ['java', () => import('@shikijs/langs/java')],
+  ['c', () => import('@shikijs/langs/c')],
+  ['cpp', () => import('@shikijs/langs/cpp')],
+  ['csharp', () => import('@shikijs/langs/csharp')],
+  ['kotlin', () => import('@shikijs/langs/kotlin')],
+  ['swift', () => import('@shikijs/langs/swift')],
+  ['php', () => import('@shikijs/langs/php')],
+  ['yaml', () => import('@shikijs/langs/yaml')],
+  ['toml', () => import('@shikijs/langs/toml')],
+  ['ini', () => import('@shikijs/langs/ini')],
+  ['markdown', () => import('@shikijs/langs/markdown')],
+  ['mdx', () => import('@shikijs/langs/mdx')],
+  ['html', () => import('@shikijs/langs/html')],
+  ['css', () => import('@shikijs/langs/css')],
+  ['scss', () => import('@shikijs/langs/scss')],
+  ['less', () => import('@shikijs/langs/less')],
+  ['sql', () => import('@shikijs/langs/sql')],
+  ['xml', () => import('@shikijs/langs/xml')],
+  ['lua', () => import('@shikijs/langs/lua')],
+])
+
+/**
+ * Language ids (and aliases) the highlighter accepts; everything else renders
  * plain. A Map, not an object: fence info strings are assistant-authored, so
  * a label like `constructor` or `__proto__` must miss instead of resolving an
- * inherited property and crashing the renderer inside shiki.
+ * inherited property and crashing the renderer inside shiki. Keys cover both
+ * the markdown-fence aliases `CodeBlock` uses and the file-extension hint ids
+ * the read tool's `langFromPath` emits, so both callers resolve the same
+ * grammars. The JS family maps to the TypeScript grammar (see {@link LANGS} for
+ * the JSX/TSX approximation), unchanged from when this was the only
+ * non-shell/JSON grammar. A value not in {@link LANGS} names a
+ * {@link LAZY_GRAMMARS} entry loaded on first use.
  */
 const LANG_ALIASES = new Map<string, string>([
   ['typescript', 'typescript'],
@@ -30,6 +94,7 @@ const LANG_ALIASES = new Map<string, string>([
   ['tsx', 'typescript'],
   ['javascript', 'typescript'],
   ['js', 'typescript'],
+  ['jsx', 'typescript'],
   ['shellscript', 'shellscript'],
   ['bash', 'shellscript'],
   ['sh', 'shellscript'],
@@ -37,6 +102,35 @@ const LANG_ALIASES = new Map<string, string>([
   ['zsh', 'shellscript'],
   ['json', 'json'],
   ['jsonc', 'json'],
+  ['py', 'python'],
+  ['python', 'python'],
+  ['rb', 'ruby'],
+  ['ruby', 'ruby'],
+  ['go', 'go'],
+  ['rs', 'rust'],
+  ['rust', 'rust'],
+  ['java', 'java'],
+  ['c', 'c'],
+  ['cpp', 'cpp'],
+  ['cs', 'csharp'],
+  ['csharp', 'csharp'],
+  ['kotlin', 'kotlin'],
+  ['swift', 'swift'],
+  ['php', 'php'],
+  ['yaml', 'yaml'],
+  ['yml', 'yaml'],
+  ['toml', 'toml'],
+  ['ini', 'ini'],
+  ['md', 'markdown'],
+  ['markdown', 'markdown'],
+  ['mdx', 'mdx'],
+  ['html', 'html'],
+  ['css', 'css'],
+  ['scss', 'scss'],
+  ['less', 'less'],
+  ['sql', 'sql'],
+  ['xml', 'xml'],
+  ['lua', 'lua'],
 ])
 
 /** All token colors resolve through `--shiki-*` custom properties (theme package sheets). */
@@ -52,12 +146,68 @@ let singleton: HighlighterCore | undefined
 function highlighter(): HighlighterCore {
   singleton ??= createHighlighterCoreSync({
     themes: [cssVariablesTheme],
-    langs: [langTs, langBash, langJson],
+    langs: LANGS,
     engine: createJavaScriptRegexEngine({ forgiving: true }),
   })
   return singleton
 }
 
+/** Grammar ids whose lazy import is in flight or done, so it is requested once. */
+const requested = new Set<string>()
+/** Subscribers re-rendered after a lazy grammar registers (React callers). */
+const listeners = new Set<() => void>()
+/** Bumped on each lazy-grammar load; the `useSyncExternalStore` snapshot. */
+let loadCount = 0
+
+/**
+ * Subscribe to lazy-grammar load completions; `listener` fires after a
+ * {@link LAZY_GRAMMARS} grammar finishes registering on the singleton, so a
+ * caller that rendered its plain fallback while the grammar loaded can
+ * re-highlight. Shaped as a `useSyncExternalStore` subscribe: pair it with
+ * {@link grammarLoadCount} as the snapshot. Returns an unsubscribe function.
+ * @param listener - invoked (no args) on each grammar-load completion.
+ * @returns a disposer that removes the listener.
+ */
+export function subscribeGrammarLoaded(listener: () => void): () => void {
+  listeners.add(listener)
+  return () => { listeners.delete(listener) }
+}
+
+/**
+ * The lazy-grammar load counter — a value that changes on every load, so a
+ * `useSyncExternalStore` snapshot re-renders the subscriber when a grammar
+ * registers. Opaque: only its identity across renders matters.
+ * @returns the current load count.
+ */
+export function grammarLoadCount(): number {
+  return loadCount
+}
+
+/**
+ * Ensure the grammar `resolved` names is registered. A boot grammar (not in
+ * {@link LAZY_GRAMMARS}) and an already-loaded lazy grammar report ready
+ * synchronously; a lazy grammar not yet loaded starts its import (once) and
+ * reports not-ready, so the caller renders plain until a
+ * {@link subscribeGrammarLoaded} listener fires.
+ * @param resolved - the grammar id an alias resolved to.
+ * @returns whether the grammar is registered and ready to tokenize now.
+ */
+function ensureGrammar(resolved: string): boolean {
+  const load = LAZY_GRAMMARS.get(resolved)
+  // A boot grammar (already registered) has no lazy loader; it is always ready.
+  if (load === undefined) return true
+  if (highlighter().getLoadedLanguages().includes(resolved)) return true
+  if (!requested.has(resolved)) {
+    requested.add(resolved)
+    void load().then((mod) => {
+      highlighter().loadLanguageSync(mod.default)
+      loadCount += 1
+      for (const listener of listeners) listener()
+    })
+  }
+  return false
+}
+
 // Engine + grammar construction costs a long task (~120-175ms); building it
 // during the first finalized fence's render would jank exactly when a stream
 // completes. Warm the singleton in a deferred task at module load (= plugin
@@ -70,13 +220,59 @@ const warmupTimer = setTimeout(() => { highlighter() }, 0)
 /**
  * Highlight `code` into shiki's HTML (a single `<pre class="shiki">` tree)
  * when `lang` maps to a registered grammar; `undefined` means the caller
- * renders its plain fallback.
+ * renders its plain fallback. A lazy grammar not yet loaded returns `undefined`
+ * for this call and loads in the background; subscribe with
+ * {@link onGrammarLoaded} to re-highlight once it registers.
  * @param code - the source text.
  * @param lang - the language hint (a markdown fence info string or a fixed caller id).
- * @returns the highlighted HTML, or `undefined` for unknown languages.
+ * @returns the highlighted HTML, or `undefined` for unknown or not-yet-loaded languages.
  */
 export function highlightToHtml(code: string, lang: string | undefined): string | undefined {
   const resolved = lang === undefined ? undefined : LANG_ALIASES.get(lang.toLowerCase())
   if (resolved === undefined) return undefined
+  if (!ensureGrammar(resolved)) return undefined
   return highlighter().codeToHtml(code, { lang: resolved, theme: 'css-variables' })
 }
+
+/**
+ * One highlighted run of a line: the text and the inline style shiki assigned
+ * it. The css-variables theme colors every run through a `--shiki-*` custom
+ * property, so `style.color` is always present; it is held as a style object
+ * rather than a bare color so a run spreads onto a `<span style>` uniformly.
+ */
+export interface HighlightSpan {
+  text: string
+  style: CSSProperties
+}
+
+/**
+ * Tokenize `code` into per-line highlighted runs when `lang` maps to a
+ * registered grammar; `undefined` means the caller renders its plain fallback.
+ * A line-numbered view needs the token runs split per line (one gutter number
+ * per line), which the single-`<pre>` {@link highlightToHtml} does not expose,
+ * so this returns shiki's own 2D line/token structure narrowed to what a run
+ * renders. Each run's color is a `--shiki-*` custom property, keeping token
+ * colors on the theme package's sheets exactly as the HTML path does; the
+ * css-variables theme carries no font-style bits, matching that path's
+ * color-only output. The trailing newline shiki appends as a final empty line
+ * is dropped so the run count matches the caller's own line array.
+ * @param code - the source text.
+ * @param lang - the language hint (a file-extension-derived language id).
+ * @returns one entry per source line (each an array of runs), or `undefined` for unknown or not-yet-loaded languages.
+ */
+export function highlightLines(code: string, lang: string | undefined): HighlightSpan[][] | undefined {
+  const resolved = lang === undefined ? undefined : LANG_ALIASES.get(lang.toLowerCase())
+  if (resolved === undefined) return undefined
+  if (!ensureGrammar(resolved)) return undefined
+  const { tokens } = highlighter().codeToTokens(code, { lang: resolved, theme: 'css-variables' })
+  // shiki tokenizes `a\nb` into two lines; a trailing newline (`a\n`) adds a
+  // third, empty line the caller's own line array does not carry. Drop that
+  // one terminator line so the two structures stay in step. The explicit
+  // `last !== undefined` (over `tokens[...]?.length`) keeps a single branch for
+  // per-file coverage, matching TerminalBlock's terminator check.
+  const last = tokens[tokens.length - 1]
+  const lines = tokens.length > 1 && last !== undefined && last.length === 0
+    ? tokens.slice(0, -1)
+    : tokens
+  return lines.map(line => line.map(token => ({ text: token.content, style: { color: token.color } })))
+}

packages/client/ui-primitives/src/use-copy-feedback.ts

@@ -0,0 +1,37 @@
+// The copy-to-clipboard-with-feedback hook shared by the block primitives
+// (TerminalBlock, SearchBlock): write the given text, and on success flip a
+// transient `copied` flag that the caller renders as a "复制成功" label for one
+// second. A refused write leaves the flag untouched, so the control never claims
+// a copy the host declined.
+
+import { useCallback, useState } from 'react'
+import { writeClipboard } from './clipboard.ts'
+
+/** How long the `copied` flag stays true after a successful write, in ms. */
+const COPIED_FEEDBACK_MS = 1000
+
+/** The copy-feedback hook's return: the transient flag and the copy handler. */
+export interface CopyFeedback {
+  /** True for {@link COPIED_FEEDBACK_MS} after a successful write; render the success label off it. */
+  copied: boolean
+  /** Copy the hook's text; no-op while `copied` is still true, silent on a refused write. */
+  onCopy: () => void
+}
+
+/**
+ * Copy `text` to the clipboard with one-second success feedback.
+ * @param text - the text to write on copy.
+ * @returns the `copied` flag and the `onCopy` handler.
+ */
+export function useCopyFeedback(text: string): CopyFeedback {
+  const [copied, setCopied] = useState(false)
+  const onCopy = useCallback(() => {
+    if (copied) return
+    void writeClipboard(text).then((ok) => {
+      if (!ok) return
+      setCopied(true)
+      window.setTimeout(() => { setCopied(false) }, COPIED_FEEDBACK_MS)
+    })
+  }, [copied, text])
+  return { copied, onCopy }
+}

packages/client/ui-primitives/tests/code-block.spec.tsx

@@ -31,6 +31,24 @@ describe('highlightToHtml', () => {
     expect(highlightToHtml('x', 'cobol')).toBeUndefined()
     expect(highlightToHtml('x', undefined)).toBeUndefined()
   })
+
+  // Every read-tool language hint whose grammar loads lazily (the boot set —
+  // ts/js/bash/sh/json — is covered above). Touching each one drives its own
+  // dynamic import thunk, so the whole LAZY_GRAMMARS table is exercised.
+  const LAZY_ALIASES = [
+    'py', 'rb', 'go', 'rs', 'java', 'c', 'cpp', 'cs', 'kotlin', 'swift', 'php',
+    'yaml', 'toml', 'ini', 'md', 'mdx', 'html', 'css', 'scss', 'less', 'sql',
+    'xml', 'lua',
+  ]
+
+  it('lazily loads every read-card grammar: plain first, highlighted after load', async () => {
+    // First touch returns the plain fallback (undefined) and starts the import.
+    for (const alias of LAZY_ALIASES) expect(highlightToHtml('x', alias)).toBeUndefined()
+    // Once every grammar has registered, the same call highlights.
+    await vi.waitFor(() => {
+      for (const alias of LAZY_ALIASES) expect(highlightToHtml('x', alias)).toContain('shiki')
+    })
+  })
 })
 
 describe('CodeBlock', () => {

packages/client/ui-primitives/tests/read-block.spec.tsx

@@ -0,0 +1,241 @@
+// @vitest-environment jsdom
+// ReadBlock + the highlightLines token path: the banner (label, language, the
+// "showing N of M" note only when the read is a window, copy control), the
+// gutter-numbered rows keeping the file's own line numbers, the shiki per-line
+// highlighting resolved to css-variables token spans with an identical-geometry
+// plain fallback for an unknown/absent language, the head/tail height cap and
+// its expand control, and the copy control writing the raw window text on both
+// the accepted and refused clipboard paths.
+
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest'
+import { act, cleanup, fireEvent, render, screen } from '@testing-library/react'
+import { DEFAULT_READ_MAX_LINES, ReadBlock, type ReadBlockLine } from '../src/index.ts'
+import { grammarLoadCount, highlightLines, subscribeGrammarLoaded } from '../src/markdown/highlight.ts'
+
+afterEach(cleanup)
+
+beforeEach(() => {
+  vi.useRealTimers()
+})
+
+/** `count` lines starting at `first`, each with distinct text. */
+function lines(count: number, first = 1): ReadBlockLine[] {
+  return Array.from({ length: count }, (_value, index) => ({ number: first + index, text: `line ${first + index}` }))
+}
+
+/** The rendered rows as `<gutter><content>` strings (CSS-module class prefix). */
+function rowTexts(container: HTMLElement): string[] {
+  return [...container.querySelectorAll('[class^="_line_"]')].map(row => row.textContent ?? '')
+}
+
+/** The gutter numbers of the rendered rows, in order. */
+function gutters(container: HTMLElement): string[] {
+  return [...container.querySelectorAll('[class^="_gutter_"]')].map(cell => cell.textContent ?? '')
+}
+
+describe('highlightLines', () => {
+  it('tokenizes a registered grammar into per-line css-variables runs', () => {
+    const result = highlightLines('const x = 1\n// c', 'ts')
+    expect(result).not.toBeUndefined()
+    expect(result).toHaveLength(2)
+    // The keyword run carries a color style through a --shiki-* custom property.
+    const keyword = result![0]!.find(span => span.text === 'const')
+    expect(keyword?.style?.color).toContain('var(--shiki-')
+    // Whitespace between tokens is a run of its own; the comment is line two.
+    expect(result![0]!.map(span => span.text).join('')).toBe('const x = 1')
+    expect(result![1]!.map(span => span.text).join('')).toBe('// c')
+  })
+
+  it('colors every run through a --shiki-* custom property', () => {
+    // The css-variables theme colors even the whitespace run (as the foreground
+    // token), so every run is a styled span; the plain fallback is the whole
+    // unknown-language path, not a per-run one.
+    const result = highlightLines('const x = 1', 'ts')
+    for (const span of result!) for (const run of span) expect(run.style.color).toContain('var(--shiki-')
+  })
+
+  it('drops the trailing terminator line so the run count matches the source lines', () => {
+    // `a\n` tokenizes to two lines in shiki (the second empty); the caller's own
+    // line array has one entry, so the terminator line is dropped.
+    const result = highlightLines('const a = 1\n', 'ts')
+    expect(result).toHaveLength(1)
+  })
+
+  it('keeps a genuinely blank final line when the source ends in two newlines', () => {
+    const result = highlightLines('a\n\n', 'ts')
+    expect(result).toHaveLength(2)
+    expect(result![1]).toEqual([])
+  })
+
+  it('returns undefined for an unknown or absent language', () => {
+    expect(highlightLines('x', 'cobol')).toBeUndefined()
+    expect(highlightLines('x', undefined)).toBeUndefined()
+  })
+
+  it('loads a lazy grammar on first use: plain first, highlighted after it registers', async () => {
+    // A boot grammar (ts) is ready synchronously; a lazy grammar (python) is
+    // not, so the first call renders plain and imports the grammar, and a
+    // subscriber fires once it registers, after which the same call highlights.
+    let notified = 0
+    const stop = subscribeGrammarLoaded(() => { notified += 1 })
+    // First touch: grammar not loaded yet, so plain fallback while it imports.
+    expect(highlightLines('def f(): pass', 'py')).toBeUndefined()
+    // The import + loadLanguageSync resolve on a microtask; wait for the notify.
+    await vi.waitFor(() => { expect(notified).toBeGreaterThan(0) })
+    expect(grammarLoadCount()).toBeGreaterThan(0)
+    const result = highlightLines('def f(): pass', 'py')
+    expect(result).not.toBeUndefined()
+    // `def` is a python keyword and carries a --shiki-* color once highlighted.
+    const keyword = result!.flat().find(span => span.text === 'def')
+    expect(keyword?.style?.color).toContain('var(--shiki-')
+    stop()
+  })
+})
+
+describe('ReadBlock rows', () => {
+  it('renders one gutter-numbered row per line, keeping the file line numbers', () => {
+    const view = render(<ReadBlock label="a.ts" lines={lines(3, 41)} totalLines={3} />)
+    expect(gutters(view.container)).toEqual(['41', '42', '43'])
+    expect(rowTexts(view.container)).toEqual(['41line 41', '42line 42', '43line 43'])
+  })
+
+  it('highlights the content for a known language into token spans', () => {
+    const view = render(
+      <ReadBlock label="a.ts" lang="ts" lines={[{ number: 1, text: 'const a = 1' }]} totalLines={1} />,
+    )
+    const content = view.container.querySelector('[class^="_content_"]')
+    expect(content?.querySelectorAll('span[style]').length).toBeGreaterThan(1)
+    expect(content?.textContent).toBe('const a = 1')
+  })
+
+  it('renders the content as bare text with no span wrappers for an unknown language', () => {
+    const view = render(
+      <ReadBlock label="a.cob" lang="cobol" lines={[{ number: 1, text: 'IDENT DIVISION.' }]} totalLines={1} />,
+    )
+    const content = view.container.querySelector('[class^="_content_"]')
+    expect(content?.querySelectorAll('span').length).toBe(0)
+    expect(content?.textContent).toBe('IDENT DIVISION.')
+  })
+
+  it('renders bare text when no language is given', () => {
+    const view = render(<ReadBlock label="x" lines={[{ number: 1, text: 'plain' }]} totalLines={1} />)
+    const content = view.container.querySelector('[class^="_content_"]')
+    expect(content?.querySelectorAll('span').length).toBe(0)
+    expect(view.getByText('plain')).toBeTruthy()
+  })
+})
+
+describe('ReadBlock banner', () => {
+  it('shows the label, the language, and the count note when the read is a window', () => {
+    const view = render(<ReadBlock label="src/a.ts" lang="ts" lines={lines(3, 41)} totalLines={180} />)
+    expect(view.getByText('src/a.ts')).toBeTruthy()
+    expect(view.getByText('ts')).toBeTruthy()
+    expect(view.getByText('显示 3 / 180 行')).toBeTruthy()
+  })
+
+  it('omits the count note when the window is the whole file', () => {
+    const view = render(<ReadBlock label="a.ts" lines={lines(3)} totalLines={3} />)
+    expect(view.queryByText(/显示/u)).toBeNull()
+  })
+
+  it('draws an empty label and empty language when neither is given', () => {
+    const view = render(<ReadBlock lines={lines(1)} totalLines={1} />)
+    expect(view.container.querySelector('[class^="_label_"]')?.textContent).toBe('')
+    expect(view.container.querySelector('[class^="_lang_"]')?.textContent).toBe('')
+  })
+})
+
+describe('ReadBlock height cap', () => {
+  it('renders every line and no expand control under the cap', () => {
+    const view = render(<ReadBlock label="a" lines={lines(4)} totalLines={4} maxLines={4} />)
+    expect(rowTexts(view.container)).toHaveLength(4)
+    expect(view.container.querySelector('[aria-expanded]')).toBeNull()
+  })
+
+  it('slices head and tail over the cap and expands on click', () => {
+    const view = render(<ReadBlock label="a" lines={lines(10)} totalLines={10} maxLines={4} />)
+    // maxLines 4: head = ceil(4/2) = 2, tail = 4 - 2 = 2, 6 hidden.
+    expect(gutters(view.container)).toEqual(['1', '2', '9', '10'])
+    const toggle = view.getByRole('button', { name: '展开其余 6 行' })
+    expect(toggle.getAttribute('aria-expanded')).toBe('false')
+    expect(toggle.textContent).toBe('… 其余 6 行')
+
+    fireEvent.click(toggle)
+    expect(rowTexts(view.container)).toHaveLength(10)
+    const collapse = view.getByRole('button', { name: '收起内容' })
+    expect(collapse.getAttribute('aria-expanded')).toBe('true')
+    expect(collapse.textContent).toBe('收起')
+
+    fireEvent.click(collapse)
+    expect(gutters(view.container)).toEqual(['1', '2', '9', '10'])
+  })
+
+  it('renders the head slice alone when the cap leaves no tail', () => {
+    const view = render(<ReadBlock label="a" lines={lines(5)} totalLines={5} maxLines={1} />)
+    expect(gutters(view.container)).toEqual(['1'])
+    expect(view.getByRole('button', { name: '展开其余 4 行' })).toBeTruthy()
+  })
+
+  it('caps at the documented default when maxLines is absent', () => {
+    const view = render(
+      <ReadBlock label="a" lines={lines(DEFAULT_READ_MAX_LINES + 1)} totalLines={DEFAULT_READ_MAX_LINES + 1} />,
+    )
+    expect(rowTexts(view.container)).toHaveLength(DEFAULT_READ_MAX_LINES)
+    expect(view.getByRole('button', { name: '展开其余 1 行' })).toBeTruthy()
+  })
+})
+
+describe('ReadBlock copy', () => {
+  it('copies the raw window text, joined by newlines, never the gutter numbers', async () => {
+    vi.useFakeTimers()
+    const writeText = vi.fn().mockResolvedValue(undefined)
+    Object.defineProperty(navigator, 'clipboard', { configurable: true, value: { writeText } })
+    render(<ReadBlock label="a" lines={lines(3, 41)} totalLines={180} />)
+    fireEvent.click(screen.getByRole('button', { name: '复制' }))
+    expect(writeText).toHaveBeenCalledWith('line 41\nline 42\nline 43')
+    await act(async () => {
+      await Promise.resolve()
+    })
+    expect(screen.getByRole('button', { name: '复制成功' })).toBeTruthy()
+    // While the ok label is showing, further clicks are no-ops.
+    fireEvent.click(screen.getByRole('button', { name: '复制成功' }))
+    expect(writeText).toHaveBeenCalledTimes(1)
+    await vi.advanceTimersByTimeAsync(1000)
+    expect(screen.getByRole('button', { name: '复制' })).toBeTruthy()
+  })
+
+  it('copies the whole window while the height cap hides its middle', async () => {
+    const writeText = vi.fn().mockResolvedValue(undefined)
+    Object.defineProperty(navigator, 'clipboard', { configurable: true, value: { writeText } })
+    render(<ReadBlock label="a" lines={lines(10)} totalLines={10} maxLines={4} />)
+    fireEvent.click(screen.getByRole('button', { name: '复制' }))
+    expect(writeText).toHaveBeenCalledWith(lines(10).map(line => line.text).join('\n'))
+    expect(await screen.findByRole('button', { name: '复制成功' })).toBeTruthy()
+  })
+
+  it('does not claim success when the host refuses the write', async () => {
+    Object.defineProperty(navigator, 'clipboard', {
+      configurable: true,
+      value: { writeText: vi.fn().mockRejectedValue(new Error('denied')) },
+    })
+    render(<ReadBlock label="a" lines={lines(1)} totalLines={1} />)
+    fireEvent.click(screen.getByRole('button', { name: '复制' }))
+    await act(async () => {
+      await Promise.resolve()
+    })
+    expect(screen.getByRole('button', { name: '复制' })).toBeTruthy()
+    expect(screen.queryByRole('button', { name: '复制成功' })).toBeNull()
+  })
+
+  it('merges className onto the wrapper', () => {
+    const view = render(<ReadBlock className="x" label="a" lines={lines(1)} totalLines={1} />)
+    expect(view.container.firstElementChild?.classList.contains('x')).toBe(true)
+  })
+
+  it('hides the copy control for an empty window so it cannot wipe the clipboard', () => {
+    // A successful read of an empty file settles to lines: [] with card:'read',
+    // so this branch is reachable; copying then would clear the clipboard.
+    const view = render(<ReadBlock label="empty.ts" lines={[]} totalLines={0} />)
+    expect(view.queryByRole('button', { name: '复制' })).toBeNull()
+  })
+})

packages/client/ui-primitives/tests/search-block.spec.tsx

@@ -0,0 +1,214 @@
+// @vitest-environment jsdom
+// SearchBlock: both kinds (grouped grep matches and a flat glob path list), the
+// folded truncation summary, the empty arm, per-file collapse/expand, the
+// head/tail height cap and its expand control, the tail slice restoring its
+// owning file header, and the copy control writing the whole structured
+// result on both the accepted and refused clipboard paths.
+
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest'
+import { act, cleanup, fireEvent, render, screen } from '@testing-library/react'
+import { DEFAULT_SEARCH_MAX_LINES, SearchBlock } from '../src/index.ts'
+import type { SearchFileGroup } from '../src/index.ts'
+
+afterEach(cleanup)
+
+beforeEach(() => {
+  vi.useRealTimers()
+})
+
+/** The rendered result rows, one string per visible row (CSS-module class prefix). */
+function lines(container: HTMLElement): string[] {
+  return [...container.querySelectorAll('[class^="_line_"]')].map(row => row.textContent ?? '')
+}
+
+/** The file-group header rows, one string per header (path + count concatenated). */
+function fileHeaders(container: HTMLElement): string[] {
+  return [...container.querySelectorAll('[class^="_fileHeader_"]')].map(row => row.textContent ?? '')
+}
+
+/** `count` numbered match lines under one file, without a terminating newline. */
+function group(path: string, count: number, from = 1): SearchFileGroup {
+  return {
+    path,
+    matches: Array.from({ length: count }, (_v, i) => ({ lineNumber: from + i, line: `hit ${from + i}` })),
+  }
+}
+
+describe('SearchBlock matches kind', () => {
+  it('renders each file as a header group with its matched lines', () => {
+    const view = render(<SearchBlock kind="matches" truncated={false} total={3} files={[
+      { path: 'a.ts', matches: [{ lineNumber: 12, line: 'const a = 1' }, { lineNumber: 40, line: 'return a' }] },
+      { path: 'b.ts', matches: [{ lineNumber: 7, line: 'const b = 2' }] },
+    ]} />)
+    expect(fileHeaders(view.container)).toEqual(['a.ts2', 'b.ts1'])
+    expect(lines(view.container)).toEqual(['12: const a = 1', '40: return a', '7: const b = 2'])
+    // The summary counts matches and files, with no folded pre-cap total below the cap.
+    expect(view.getByText('3 处匹配 · 2 个文件')).toBeTruthy()
+    expect(view.queryByText(/显示|共/u)).toBeNull()
+  })
+
+  it('collapses and re-expands a single file group without touching the others', () => {
+    const view = render(<SearchBlock kind="matches" truncated={false} total={3} files={[
+      { path: 'a.ts', matches: [{ lineNumber: 1, line: 'x' }] },
+      { path: 'b.ts', matches: [{ lineNumber: 2, line: 'y' }] },
+    ]} />)
+    const [headerA] = view.container.querySelectorAll('[class^="_fileHeader_"]')
+    expect(headerA!.getAttribute('aria-expanded')).toBe('true')
+    fireEvent.click(headerA!)
+    // a.ts collapsed: its match row is gone, b.ts's stays.
+    expect(headerA!.getAttribute('aria-expanded')).toBe('false')
+    expect(lines(view.container)).toEqual(['2: y'])
+    fireEvent.click(headerA!)
+    expect(lines(view.container)).toEqual(['1: x', '2: y'])
+  })
+
+  it('folds the pre-cap total into the summary when truncated', () => {
+    const view = render(<SearchBlock kind="matches" truncated total={99} files={[group('a.ts', 2)]} />)
+    expect(view.getByText('显示 2 / 共 99 处匹配 · 1 个文件')).toBeTruthy()
+  })
+})
+
+describe('SearchBlock paths kind', () => {
+  it('renders a flat path list with a path-count summary', () => {
+    const view = render(<SearchBlock kind="paths" truncated={false} total={2} paths={['src/a.ts', 'src/b.ts']} />)
+    expect(lines(view.container)).toEqual(['src/a.ts', 'src/b.ts'])
+    expect(view.getByText('2 个路径')).toBeTruthy()
+    // No file-group headers in the paths shape.
+    expect(fileHeaders(view.container)).toEqual([])
+  })
+
+  it('folds the pre-cap total into the paths summary when truncated', () => {
+    const view = render(<SearchBlock kind="paths" truncated total={50} paths={['a', 'b']} />)
+    expect(view.getByText('显示 2 / 共 50 个路径')).toBeTruthy()
+  })
+})
+
+describe('SearchBlock empty arm', () => {
+  it('shows the placeholder and no copy control for an empty matches result', () => {
+    const view = render(<SearchBlock kind="matches" truncated={false} total={0} files={[]} />)
+    expect(view.getByText('无结果')).toBeTruthy()
+    expect(view.queryByText('复制')).toBeNull()
+    expect(view.getByText('0 处匹配 · 0 个文件')).toBeTruthy()
+  })
+
+  it('shows the placeholder for an empty paths result', () => {
+    const view = render(<SearchBlock kind="paths" truncated={false} total={0} paths={[]} />)
+    expect(view.getByText('无结果')).toBeTruthy()
+    expect(view.queryByText('复制')).toBeNull()
+  })
+})
+
+describe('SearchBlock height cap', () => {
+  it('renders every row and no expand control under the cap', () => {
+    const view = render(<SearchBlock kind="paths" truncated={false} total={4}
+      paths={['a', 'b', 'c', 'd']} maxLines={4} />)
+    expect(lines(view.container)).toHaveLength(4)
+    expect(view.container.querySelector('[aria-label^="展开"]')).toBeNull()
+  })
+
+  it('slices head and tail over the cap and expands on click', () => {
+    const paths = Array.from({ length: 10 }, (_v, i) => `p${i + 1}`)
+    const view = render(<SearchBlock kind="paths" truncated={false} total={10} paths={paths} maxLines={4} />)
+    // maxLines 4: head = ceil(4/2) = 2, tail = 2, 6 hidden.
+    expect(lines(view.container)).toEqual(['p1', 'p2', 'p9', 'p10'])
+    const toggle = view.getByRole('button', { name: '展开其余 6 行结果' })
+    expect(toggle.textContent).toBe('… 其余 6 行')
+    fireEvent.click(toggle)
+    expect(lines(view.container)).toHaveLength(10)
+    const collapse = view.getByRole('button', { name: '收起结果' })
+    expect(collapse.textContent).toBe('收起')
+    fireEvent.click(collapse)
+    expect(lines(view.container)).toEqual(['p1', 'p2', 'p9', 'p10'])
+  })
+
+  it('counts a file header as one capped row alongside its matches', () => {
+    // One file with 10 matches → 11 rows (header + 10). Cap 4: head 2, tail 2.
+    const view = render(<SearchBlock kind="matches" truncated={false} total={10}
+      files={[group('a.ts', 10)]} maxLines={4} />)
+    // Head takes the header then the first match; tail takes the last two matches.
+    expect(lines(view.container)).toEqual(['1: hit 1', '9: hit 9', '10: hit 10'])
+    expect(fileHeaders(view.container)).toEqual(['a.ts10'])
+    expect(view.getByRole('button', { name: '展开其余 7 行结果' })).toBeTruthy()
+  })
+
+  it('renders the head slice alone when the cap leaves no tail', () => {
+    const view = render(<SearchBlock kind="paths" truncated={false} total={5}
+      paths={['a', 'b', 'c', 'd', 'e']} maxLines={1} />)
+    expect(lines(view.container)).toEqual(['a'])
+    expect(view.getByRole('button', { name: '展开其余 4 行结果' })).toBeTruthy()
+  })
+
+  it('restores the owning file header above a tail slice that begins mid-file', () => {
+    // Two files of 10 matches each → 22 rows. Cap 8: head 4 (a.ts header + 3
+    // matches), tail 4. The tail begins mid-b.ts, so its header is restored —
+    // and, being a row itself, it consumes one tail slot rather than pushing the
+    // card to 9 rows: the tail keeps its last 3 matches, total visible = 8.
+    const view = render(<SearchBlock kind="matches" truncated={false} total={20} maxLines={8} files={[
+      group('a.ts', 10), group('b.ts', 10, 11),
+    ]} />)
+    expect(fileHeaders(view.container)).toEqual(['a.ts10', 'b.ts10'])
+    expect(lines(view.container)).toEqual([
+      '1: hit 1', '2: hit 2', '3: hit 3',
+      '18: hit 18', '19: hit 19', '20: hit 20',
+    ])
+    // Visible rows hold at maxLines (2 headers + 6 matches = 8), so the hidden
+    // count stays exact: 22 − 8 = 14.
+    expect(view.getByRole('button', { name: '展开其余 14 行结果' })).toBeTruthy()
+  })
+
+  it('caps at the documented default when maxLines is absent', () => {
+    const paths = Array.from({ length: DEFAULT_SEARCH_MAX_LINES + 1 }, (_v, i) => `p${i}`)
+    const view = render(<SearchBlock kind="paths" truncated={false} total={paths.length} paths={paths} />)
+    expect(lines(view.container)).toHaveLength(DEFAULT_SEARCH_MAX_LINES)
+    expect(view.getByRole('button', { name: '展开其余 1 行结果' })).toBeTruthy()
+  })
+})
+
+describe('SearchBlock copy', () => {
+  it('copies the whole structured matches result, not the collapsed or capped view', async () => {
+    vi.useFakeTimers()
+    const writeText = vi.fn().mockResolvedValue(undefined)
+    Object.defineProperty(navigator, 'clipboard', { configurable: true, value: { writeText } })
+    const view = render(<SearchBlock kind="matches" truncated total={9} maxLines={2} files={[
+      { path: 'a.ts', matches: [{ lineNumber: 1, line: 'x' }, { lineNumber: 2, line: 'y' }] },
+      { path: 'b.ts', matches: [{ lineNumber: 3, line: 'z' }] },
+    ]} />)
+    // Collapse a group and leave the cap in place: the clipboard still gets it all.
+    fireEvent.click(view.container.querySelector('[class^="_fileHeader_"]')!)
+    fireEvent.click(screen.getByRole('button', { name: '复制' }))
+    expect(writeText).toHaveBeenCalledWith('a.ts\n1: x\n2: y\n\nb.ts\n3: z')
+    await act(async () => { await Promise.resolve() })
+    expect(screen.getByRole('button', { name: '复制成功' })).toBeTruthy()
+    // A second click while the ok label shows is a no-op.
+    fireEvent.click(screen.getByRole('button', { name: '复制成功' }))
+    expect(writeText).toHaveBeenCalledTimes(1)
+    await vi.advanceTimersByTimeAsync(1000)
+    expect(screen.getByRole('button', { name: '复制' })).toBeTruthy()
+  })
+
+  it('copies the newline-joined path list for the paths shape', async () => {
+    const writeText = vi.fn().mockResolvedValue(undefined)
+    Object.defineProperty(navigator, 'clipboard', { configurable: true, value: { writeText } })
+    render(<SearchBlock kind="paths" truncated={false} total={2} paths={['src/a.ts', 'src/b.ts']} />)
+    fireEvent.click(screen.getByRole('button', { name: '复制' }))
+    expect(writeText).toHaveBeenCalledWith('src/a.ts\nsrc/b.ts')
+    expect(await screen.findByRole('button', { name: '复制成功' })).toBeTruthy()
+  })
+
+  it('does not claim success when the host refuses the write', async () => {
+    Object.defineProperty(navigator, 'clipboard', {
+      configurable: true, value: { writeText: vi.fn().mockRejectedValue(new Error('denied')) },
+    })
+    render(<SearchBlock kind="paths" truncated={false} total={1} paths={['a']} />)
+    fireEvent.click(screen.getByRole('button', { name: '复制' }))
+    await act(async () => { await Promise.resolve() })
+    expect(screen.getByRole('button', { name: '复制' })).toBeTruthy()
+    expect(screen.queryByRole('button', { name: '复制成功' })).toBeNull()
+  })
+
+  it('merges className onto the wrapper and tags the wrapper with the kind', () => {
+    const view = render(<SearchBlock kind="paths" truncated={false} total={0} paths={[]} className="x" />)
+    expect(view.container.firstElementChild?.classList.contains('x')).toBe(true)
+    expect(view.container.firstElementChild?.getAttribute('data-search')).toBe('paths')
+  })
+})

packages/compact/compact-basic/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/compact/compact-basic/README.md
-README.md: 775355f1ac1a7c79c16f66a5b2489d73df7b960d
-README.zh.md: bfa139596b5ef61c23d29575bdea5534fa82b158
+README.md: b35e5dc110e908047338054337b309a77b7e0f68
+README.zh.md: 9c5f83d987b584d58ffacadce4e469bb6ba81aa2

packages/compact/compact-basic/README.md

@@ -136,7 +136,7 @@ Output EXACTLY the Markdown structure below: keep every section, in order. Use t
 - [decisions and their rationale, constraints, user preferences, open questions, data needed to continue]
 
 Rules:
-- Preserve exact file paths, commands, error strings, identifiers, and function signatures.
+- Write concise English engineering prose. Preserve exact file paths, commands, error strings, identifiers, numeric values, function signatures, and syntax fragments.
 - Capture user feedback and explicit instructions faithfully, especially corrections.
 - Do NOT mention this summarization request or that the context was compacted.
 - Output only the checkpoint text: do not call any tool or take any other action.

packages/compact/compact-basic/README.zh.md

@@ -136,7 +136,7 @@ Output EXACTLY the Markdown structure below: keep every section, in order. Use t
 - [decisions and their rationale, constraints, user preferences, open questions, data needed to continue]
 
 Rules:
-- Preserve exact file paths, commands, error strings, identifiers, and function signatures.
+- Write concise English engineering prose. Preserve exact file paths, commands, error strings, identifiers, numeric values, function signatures, and syntax fragments.
 - Capture user feedback and explicit instructions faithfully, especially corrections.
 - Do NOT mention this summarization request or that the context was compacted.
 - Output only the checkpoint text: do not call any tool or take any other action.

packages/compact/compact-basic/src/summarizer.ts

@@ -58,7 +58,7 @@ const COMPACTION_INSTRUCTION = [
   '- [decisions and their rationale, constraints, user preferences, open questions, data needed to continue]',
   '',
   'Rules:',
-  '- Preserve exact file paths, commands, error strings, identifiers, and function signatures.',
+  '- Write concise English engineering prose. Preserve exact file paths, commands, error strings, identifiers, numeric values, function signatures, and syntax fragments.',
   '- Capture user feedback and explicit instructions faithfully, especially corrections.',
   '- Do NOT mention this summarization request or that the context was compacted.',
   '- Output only the checkpoint text: do not call any tool or take any other action.',

packages/compact/compact-basic/tests/compact-basic.spec.ts

@@ -1224,7 +1224,8 @@ describe('default one-shot summarizer', () => {
     expect(messages[0]).toEqual(prefix)
     const last = messages.at(-1)?.content[0]
     const lastText = last?.type === 'text' ? last.text : ''
-    expect(lastText).toContain('Condense the conversation ABOVE')
+    expect(lastText).toContain('Write concise English engineering prose.')
+    expect(lastText).toContain('numeric values, function signatures, and syntax fragments.')
     expect(lastText).toContain('## Primary Request and Intent')
   })
 

packages/compact/compact-basic/tests/compact-loop-repro.spec.ts

@@ -342,6 +342,11 @@ describe('context-overflow recovery across the real loop and compact-basic', ()
 
         expect(adapter.conversationRequests).toHaveLength(2)
         expect(adapter.summaryRequests).toHaveLength(1)
+        const instruction = adapter.summaryRequests[0]!.messages.at(-1)?.content
+          .map(block => (block.type === 'text' ? block.text : ''))
+          .join('') ?? ''
+        expect(instruction).toContain('Write concise English engineering prose.')
+        expect(instruction).toContain('numeric values, function signatures, and syntax fragments.')
         expect(JSON.stringify(adapter.conversationRequests[0]!.messages)).toContain('OLD HISTORY SENTINEL')
         const retry = JSON.stringify(adapter.conversationRequests[1]!.messages)
         expect(retry).toContain('RECOVERY CHECKPOINT')

packages/cordis/tool-cordis/src/api-catalog.ts

@@ -916,6 +916,10 @@ export const SERVICE_API: readonly ServiceApiEntry[] = [
         signature: 'section(section: PromptSection): () => void',
         jsDoc: '/**\n * Register an ordered prompt section in the calling context\'s scope. A scoped\n * section shadows a global section with the same name; duplicates within one\n * layer and non-finite orders throw. Registration and disposal emit\n * `system-prompt/change`.\n * @param section - the section to register.\n * @returns the exact Cordis effect disposer.\n */',
       },
+      {
+        signature: 'context(context: PromptContext): () => void',
+        jsDoc: '/**\n * Register ordered cache-safe dynamic context in the calling context\'s scope.\n * A scoped context shadows a global context with the same name; duplicates\n * within one layer and non-finite orders throw. Registration and disposal\n * emit `system-prompt/change`.\n * @param context - the context contribution to register.\n * @returns the exact Cordis effect disposer.\n */',
+      },
       {
         signature: 'tools(provider: (context: AssembleContext) => ToolProviderResult): () => void',
         jsDoc: '/**\n * Register a tool-schema provider in the calling context\'s scope. Global and\n * matching scoped providers both contribute; returning the reserved\n * {@link TOOL_ORDER_REST} name makes assembly fail.\n * @param provider - evaluated for each assembly with its context.\n * @returns the exact Cordis effect disposer.\n */',
@@ -1446,8 +1450,8 @@ export const EVENT_API: readonly EventApiEntry[] = [
     name: 'system-prompt/assemble',
     mode: 'waterfall',
     signature: '\'system-prompt/assemble\'(this: Scoped<SystemPrompt>, assembly: PromptAssembly, context: AssembleContext, next: () => Promise<PromptAssembly>): Promise<PromptAssembly>',
-    jsDoc: '/**\n * Expert waterfall over the assembled sections, tools, and variables.\n * Scope-filtered dispatch (`@deepseek-ai/dsh-scope`): scoped listeners\n * receive only that scope\'s assemblies. The returned value is authoritative.\n * A supplied signal controls only this explicit assembly request and must not\n * be retained to control later turns.\n * @param assembly - the mutable assembly built from registered providers.\n * @param context - the caller\'s per-assembly context.\n * @mode waterfall\n */',
-    summary: 'Expert waterfall over the assembled sections, tools, and variables.',
+    jsDoc: '/**\n * Expert waterfall over the assembled sections, contexts, tools, and variables.\n * Scope-filtered dispatch (`@deepseek-ai/dsh-scope`): scoped listeners\n * receive only that scope\'s assemblies. The returned value is authoritative.\n * A supplied signal controls only this explicit assembly request and must not\n * be retained to control later turns.\n * @param assembly - the mutable assembly built from registered providers.\n * @param context - the caller\'s per-assembly context.\n * @mode waterfall\n */',
+    summary: 'Expert waterfall over the assembled sections, contexts, tools, and variables.',
   },
   {
     name: 'system-prompt/change',
@@ -1619,6 +1623,10 @@ export const TYPE_API: readonly TypeApiEntry[] = [
     name: 'AssembleContext',
     declaration: 'export interface AssembleContext {\n    scope?: ScopeKey;\n    signal?: AbortSignal;\n}',
   },
+  {
+    name: 'AssembledContext',
+    declaration: 'export interface AssembledContext {\n    name: string;\n    text: string;\n}',
+  },
   {
     name: 'AssembledSection',
     declaration: 'export interface AssembledSection {\n    name: string;\n    text: string;\n}',
@@ -2121,7 +2129,11 @@ export const TYPE_API: readonly TypeApiEntry[] = [
   },
   {
     name: 'PromptAssembly',
-    declaration: 'export interface PromptAssembly {\n    sections: AssembledSection[];\n    tools: ToolSchema[];\n    variables: Record<string, string | undefined>;\n}',
+    declaration: 'export interface PromptAssembly {\n    sections: AssembledSection[];\n    contexts: AssembledContext[];\n    tools: ToolSchema[];\n    variables: Record<string, string | undefined>;\n}',
+  },
+  {
+    name: 'PromptContext',
+    declaration: 'export interface PromptContext {\n    readonly name: string;\n    readonly order: number;\n    readonly text: string | ((context: AssembleContext) => string);\n}',
   },
   {
     name: 'PromptSection',
@@ -2287,6 +2299,26 @@ export const TYPE_API: readonly TypeApiEntry[] = [
     name: 'ScopeKey',
     declaration: 'export type ScopeKey = object;',
   },
+  {
+    name: 'SearchFileMatches',
+    declaration: 'export interface SearchFileMatches {\n    path: string;\n    matches: SearchLineMatch[];\n}',
+  },
+  {
+    name: 'SearchLineMatch',
+    declaration: 'export interface SearchLineMatch {\n    lineNumber: number;\n    line: string;\n}',
+  },
+  {
+    name: 'SearchMatchesResultView',
+    declaration: 'export interface SearchMatchesResultView {\n    card: \'search\';\n    shape: \'matches\';\n    title?: string;\n    files: SearchFileMatches[];\n    truncated: boolean;\n    total: number;\n}',
+  },
+  {
+    name: 'SearchPathsResultView',
+    declaration: 'export interface SearchPathsResultView {\n    card: \'search\';\n    shape: \'paths\';\n    title?: string;\n    paths: string[];\n    truncated: boolean;\n    total: number;\n}',
+  },
+  {
+    name: 'SearchResultView',
+    declaration: 'export type SearchResultView = SearchMatchesResultView | SearchPathsResultView;',
+  },
   {
     name: 'SendOptions',
     declaration: 'export interface SendOptions {\n    target: SendTarget;\n    wakeup: boolean;\n}',
@@ -2857,7 +2889,7 @@ export const TYPE_API: readonly TypeApiEntry[] = [
   },
   {
     name: 'ToolResultView',
-    declaration: 'export type ToolResultView = GenericResultView | TerminalResultView | DiffResultView | ReadResultView | WebResultView;',
+    declaration: 'export type ToolResultView = GenericResultView | TerminalResultView | DiffResultView | SearchResultView | ReadResultView | WebResultView;',
   },
   {
     name: 'ToolRunContext',

packages/core/agent-loop/src/agent.ts

@@ -34,6 +34,7 @@ import {
   LlmError,
   assertNever,
   createAssistantMessage,
+  createUserMessage,
   deepFreeze,
   errorChain,
   freezeMessage,
@@ -45,7 +46,7 @@ import {
 import type { GenerateOptions, LlmCallConfig, LlmFailure, Message, PreparedLlmCall, ResolvedRetryPolicy } from '@deepseek-ai/dsh-llm'
 import { canonicalHeader, headerEquals } from '@deepseek-ai/dsh-session'
 import type { AssistantMessage, EpochHeader, Session, SessionId, TurnEndReason, TurnTrigger, UserMessage } from '@deepseek-ai/dsh-session'
-import { renderPrompt } from '@deepseek-ai/dsh-system-prompt'
+import { renderContextSnapshot, renderPrompt } from '@deepseek-ai/dsh-system-prompt'
 import type {} from '@deepseek-ai/dsh-tools'
 import { executeToolCalls } from './tool-calls.ts'
 
@@ -54,6 +55,47 @@ type StepOutcome =
   | { kind: 'completed'; continueTurn: boolean; concluded: boolean; maxTokens: boolean }
   | { kind: 'request-failed'; error: RequestError; failure: LlmFailure; retryPolicy: ResolvedRetryPolicy | undefined }
 
+const RUNTIME_CONTEXT_SOURCE = '@deepseek-ai/dsh-system-prompt'
+/** Clearing marker kept distinct from every prefixed {@link renderContextSnapshot} result. */
+const CLEARED_RUNTIME_CONTEXT = 'Current runtime context: none. Earlier runtime-context snapshots no longer apply.'
+
+/** Whether one user message is owned by runtime-context materialization. */
+function isRuntimeContextMessage(message: UserMessage): boolean {
+  return message.source.kind === 'plugin' && message.source.plugin === RUNTIME_CONTEXT_SOURCE
+}
+
+/** Latest retained runtime-context snapshot; `found` distinguishes malformed content from absence. */
+function retainedRuntimeContext(session: Session): { found: boolean; text: string | undefined } {
+  const events = session.events
+  const nodes = session.surface.nodes
+  for (let index = nodes.length - 1; index >= 0; index -= 1) {
+    const event = events[nodes[index] as number]
+    if (event?.type !== 'user/message' || !isRuntimeContextMessage(event.data)) continue
+    const [block] = event.data.content
+    return {
+      found: true,
+      text: event.data.content.length === 1 && block?.type === 'text' ? block.text : undefined,
+    }
+  }
+  return { found: false, text: undefined }
+}
+
+/** Append a full current snapshot only when it changed or compaction removed it. */
+function materializeRuntimeContext(session: Session, current: string): void {
+  const previous = retainedRuntimeContext(session)
+  if (!previous.found && current.length === 0) {
+    const compactedPriorSnapshot = session.surface.replaceGeneration > 0
+      && session.events.some(event => event.type === 'user/message' && isRuntimeContextMessage(event.data))
+    if (!compactedPriorSnapshot) return
+  }
+  const snapshot = current.length === 0 ? CLEARED_RUNTIME_CONTEXT : current
+  if (previous.text === snapshot) return
+  session.append('user/message', createUserMessage({
+    content: [{ type: 'text', text: snapshot }],
+    source: { kind: 'plugin', plugin: RUNTIME_CONTEXT_SOURCE },
+  }), { surfaceOp: 'append' })
+}
+
 /** Remove adapter-derived values before plugins propose the next request config. */
 function requestProposal(header: EpochHeader): LlmCallConfig {
   if (header.adapterDefaults === undefined) return header.config
@@ -520,10 +562,13 @@ export class ReactLoopAgent implements Agent {
     // this request together.
     this.drainOutbox(turn)
 
-    // Assemble the system prompt fresh each step (it may depend on log state).
+    // Assemble request-owned prompt inputs fresh each step. Dynamic context is
+    // committed at the tail before deriving history once, preserving the stable
+    // system/history cache prefix while keeping every model-visible byte logged.
     const assembly = await this.loopCtx.systemPrompt.assemble(assembleContextFor(this, signal))
     signal.throwIfAborted()
     const system = renderPrompt(assembly)
+    materializeRuntimeContext(session, renderContextSnapshot(assembly))
 
     // Snapshot the exact log prefix: the reconstruction boundary. Appends
     // after this synchronous snapshot join the next request.

packages/core/agent-loop/tests/loop.spec.ts

@@ -254,7 +254,7 @@ describe('agent loop', () => {
     // NO system field at all (not an empty string).
     const adapter = new MockAdapter([textResponse('ok')])
     const ctx = await harness(adapter)
-    ctx.on('system-prompt/assemble', async () => ({ sections: [], tools: [], variables: {} }))
+    ctx.on('system-prompt/assemble', async () => ({ sections: [], contexts: [], tools: [], variables: {} }))
     const agent = ctx.agentLoop.create(SessionId('a-no-system'), { provider: 'mock', model: 'mock' })
 
     send(agent, 'hi')
@@ -264,6 +264,178 @@ describe('agent loop', () => {
     expect('system' in adapter.requests[0]!).toBe(false)
   })
 
+  it('materializes changed runtime context at the history tail without rewriting the system header', async () => {
+    const adapter = new MockAdapter([
+      textResponse('one'),
+      textResponse('two'),
+      textResponse('three'),
+      textResponse('four'),
+      textResponse('five'),
+    ])
+    const ctx = await harness(adapter)
+    let mode = 'read-only'
+    const dispose = ctx.systemPrompt.context({ name: 'policy', order: 0, text: () => `Mode: ${mode}.` })
+    const agent = ctx.agentLoop.create(SessionId('a-runtime-context'), { provider: 'mock', model: 'mock' })
+    const contextEvents = () => agent.session.events.flatMap(event =>
+      event.type === 'user/message'
+        && event.data.source.kind === 'plugin'
+        && event.data.source.plugin === '@deepseek-ai/dsh-system-prompt'
+        ? [event]
+        : [])
+
+    send(agent, 'first')
+    await waitForIdle(ctx, agent)
+    expect(contextEvents()).toHaveLength(1)
+    expect(contextEvents()[0]?.data.content).toEqual([{
+      type: 'text',
+      text: 'Current runtime context. This snapshot supersedes earlier runtime-context snapshots.\n\nMode: read-only.',
+    }])
+
+    send(agent, 'unchanged')
+    await waitForIdle(ctx, agent)
+    expect(contextEvents()).toHaveLength(1)
+
+    mode = 'danger-full-access'
+    send(agent, 'changed')
+    await waitForIdle(ctx, agent)
+    expect(contextEvents()).toHaveLength(2)
+    const changedBlock = contextEvents()[1]?.data.content[0]
+    expect(changedBlock?.type).toBe('text')
+    if (changedBlock?.type !== 'text') throw new Error('changed runtime context is not text')
+    expect(changedBlock.text).toContain('danger-full-access')
+
+    dispose()
+    send(agent, 'cleared')
+    await waitForIdle(ctx, agent)
+    expect(contextEvents()).toHaveLength(3)
+    expect(contextEvents()[2]?.data.content).toEqual([{
+      type: 'text',
+      text: 'Current runtime context: none. Earlier runtime-context snapshots no longer apply.',
+    }])
+
+    send(agent, 'still clear')
+    await waitForIdle(ctx, agent)
+    expect(contextEvents()).toHaveLength(3)
+    expect(adapter.requests.map(request => request.system)).toEqual(Array(5).fill(adapter.requests[0]?.system))
+    expect(agent.session.events.filter(event => event.type === 'request/header')).toHaveLength(1)
+  })
+
+  it('re-emits unchanged runtime context when a surface replacement removed the retained snapshot', async () => {
+    const adapter = new MockAdapter([textResponse('one'), textResponse('two')])
+    const ctx = await harness(adapter)
+    ctx.systemPrompt.context({ name: 'policy', order: 0, text: 'Mode: read-only.' })
+    const agent = ctx.agentLoop.create(SessionId('a-runtime-context-compacted'), { provider: 'mock', model: 'mock' })
+
+    send(agent, 'first')
+    await waitForIdle(ctx, agent)
+    const contextEvent = agent.session.events.find(event =>
+      event.type === 'user/message'
+      && event.data.source.kind === 'plugin'
+      && event.data.source.plugin === '@deepseek-ai/dsh-system-prompt')
+    if (contextEvent?.type !== 'user/message') throw new Error('first turn did not materialize runtime context')
+    agent.session.append('user/message', createUserMessage({
+      content: [{ type: 'text', text: 'compacted summary' }],
+      source: { kind: 'plugin', plugin: 'test-compaction' },
+    }), {
+      surfaceOp: { op: 'replace', start: contextEvent.seq, end: contextEvent.seq },
+      sourceEventSeqs: [contextEvent.seq],
+    })
+
+    send(agent, 'after compaction')
+    await waitForIdle(ctx, agent)
+    const runtimeContexts = agent.session.events.flatMap(event =>
+      event.type === 'user/message'
+        && event.data.source.kind === 'plugin'
+        && event.data.source.plugin === '@deepseek-ai/dsh-system-prompt'
+        ? [event]
+        : [])
+    expect(runtimeContexts).toHaveLength(2)
+    expect(adapter.requests[1]?.messages.some(message =>
+      message.source.kind === 'plugin'
+      && message.source.plugin === '@deepseek-ai/dsh-system-prompt')).toBe(true)
+  })
+
+  it('clears compacted runtime context after the active set becomes empty', async () => {
+    const adapter = new MockAdapter([textResponse('one'), textResponse('two')])
+    const ctx = await harness(adapter)
+    const dispose = ctx.systemPrompt.context({ name: 'policy', order: 0, text: 'Mode: read-only.' })
+    const agent = ctx.agentLoop.create(SessionId('a-runtime-context-compacted-clear'), { provider: 'mock', model: 'mock' })
+
+    send(agent, 'first')
+    await waitForIdle(ctx, agent)
+    const contextEvent = agent.session.events.find(event =>
+      event.type === 'user/message'
+      && event.data.source.kind === 'plugin'
+      && event.data.source.plugin === '@deepseek-ai/dsh-system-prompt')
+    if (contextEvent?.type !== 'user/message') throw new Error('first turn did not materialize runtime context')
+    agent.session.append('user/message', createUserMessage({
+      content: [{ type: 'text', text: 'summary retaining old mode: read-only' }],
+      source: { kind: 'plugin', plugin: 'test-compaction' },
+    }), {
+      surfaceOp: { op: 'replace', start: contextEvent.seq, end: contextEvent.seq },
+      sourceEventSeqs: [contextEvent.seq],
+    })
+    dispose()
+
+    send(agent, 'after compaction')
+    await waitForIdle(ctx, agent)
+    const clearing = adapter.requests[1]?.messages.find(message =>
+      message.source.kind === 'plugin'
+      && message.source.plugin === '@deepseek-ai/dsh-system-prompt')
+    expect(clearing?.content).toEqual([{
+      type: 'text',
+      text: 'Current runtime context: none. Earlier runtime-context snapshots no longer apply.',
+    }])
+  })
+
+  it('does not clear runtime context after an unrelated replacement', async () => {
+    const adapter = new MockAdapter([textResponse('ok')])
+    const ctx = await harness(adapter)
+    const agent = ctx.agentLoop.create(SessionId('a-runtime-context-unrelated-compaction'), { provider: 'mock', model: 'mock' })
+    const original = agent.session.append('user/message', createUserMessage({
+      content: [{ type: 'text', text: 'old context' }],
+      source: { kind: 'plugin', plugin: 'test-context' },
+    }), { surfaceOp: 'append' })
+    agent.session.append('user/message', createUserMessage({
+      content: [{ type: 'text', text: 'compacted summary' }],
+      source: { kind: 'plugin', plugin: 'test-compaction' },
+    }), {
+      surfaceOp: { op: 'replace', start: original.seq, end: original.seq },
+      sourceEventSeqs: [original.seq],
+    })
+
+    send(agent, 'after compaction')
+    await waitForIdle(ctx, agent)
+    expect(adapter.requests[0]?.messages.some(message =>
+      message.source.kind === 'plugin'
+      && message.source.plugin === '@deepseek-ai/dsh-system-prompt')).toBe(false)
+  })
+
+  it('replaces a malformed retained runtime-context message with the current complete snapshot', async () => {
+    const adapter = new MockAdapter([textResponse('ok')])
+    const ctx = await harness(adapter)
+    ctx.systemPrompt.context({ name: 'policy', order: 0, text: 'Mode: read-only.' })
+    const agent = ctx.agentLoop.create(SessionId('a-runtime-context-malformed'), { provider: 'mock', model: 'mock' })
+    agent.session.append('user/message', createUserMessage({
+      content: [{ type: 'text', text: 'broken' }, { type: 'text', text: 'snapshot' }],
+      source: { kind: 'plugin', plugin: '@deepseek-ai/dsh-system-prompt' },
+    }), { surfaceOp: 'append' })
+
+    send(agent, 'repair context')
+    await waitForIdle(ctx, agent)
+    const runtimeContexts = agent.session.events.flatMap(event =>
+      event.type === 'user/message'
+        && event.data.source.kind === 'plugin'
+        && event.data.source.plugin === '@deepseek-ai/dsh-system-prompt'
+        ? [event]
+        : [])
+    expect(runtimeContexts).toHaveLength(2)
+    expect(runtimeContexts[1]?.data.content).toEqual([{
+      type: 'text',
+      text: 'Current runtime context. This snapshot supersedes earlier runtime-context snapshots.\n\nMode: read-only.',
+    }])
+  })
+
   it('records raw chunks for replay as assistant/chunk session events', async () => {
     const adapter = new MockAdapter([textResponse('abc')])
     const ctx = await harness(adapter)

packages/core/system-prompt/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/core/system-prompt/README.md
-README.md: 23bc0e8177ad2a778df9522e254bfd5e03a9871f
-README.zh.md: 1fd4febc1c15acda19e7abfca94079b9585c1972
+README.md: d4e0f69323b7326fc7575834bf48a5aeeec0777e
+README.zh.md: 47290335d725083fc46ef4f2ee09b09263276788

packages/core/system-prompt/README.md

@@ -2,7 +2,7 @@
 
 English | [中文](README.zh.md)
 
-System prompt assembly registry. Plugins contribute ordered sections, tool schemas, and named variables. The loop assembles once per step and renders the result as the complete model prompt. This plugin owns the static harness identity and global deployment persona; an agent-scoped persona shadows the global default.
+Model-input assembly registry. Plugins contribute ordered stable system sections, cache-safe dynamic context, tool schemas, and named variables. The loop assembles once per step, renders stable sections as the system prompt, and appends a durable full dynamic-context snapshot only when its text changes or compaction removed the retained snapshot. This plugin owns the static harness identity and global deployment persona; an agent-scoped persona shadows the global default.
 
 ## Config
 
@@ -17,6 +17,7 @@ System prompt assembly registry. Plugins contribute ordered sections, tool schem
 ### Public API
 
 - `ctx.systemPrompt.section(section: PromptSection): () => void` Contribute a section. The layer is the calling context's scope: `agent.ctx` contributes to that agent alone, shadowing a same-named global section there. Duplicate names within one layer and non-finite orders throw. Disposed with the calling fiber.
+- `ctx.systemPrompt.context(context: PromptContext): () => void` Contribute cache-safe dynamic model context. Contexts are ordered independently from system sections; scoped contributions shadow same-named globals. The agent loop materializes the complete current set as one sourced user-role snapshot after retained history, only when changed or missing. Duplicate names within one layer and non-finite orders throw. Disposed with the calling fiber.
 - `ctx.systemPrompt.tools(provider: (context: AssembleContext) => ToolProviderResult): () => void` Contribute tool schemas, evaluated at each assembly with that assembly's context. `ToolProviderResult` = `{ schemas, knownNames? }`: `schemas` is the post-restriction visible set; `knownNames` is the pre-restriction universe used by `toolOrder`. A provider must not return a schema named `TOOL_ORDER_REST`. Scoped providers are consulted only for their scope's assemblies. Disposed with the calling fiber.
 - `ctx.systemPrompt.variable(name: string, provider: (context) => string | undefined): () => void` Contribute a prompt variable, referenced from section text as `{{name}}`. Scoped variables shadow a same-named global for that agent. Duplicate-in-layer or unreferenceable names throw; `undefined` means "no value for this assembly". Disposed with the calling fiber.
 - `ctx.systemPrompt.assemble(context?: AssembleContext): Promise<PromptAssembly>` Assemble the prompt for one caller: the global layer merged with `context.scope`'s layer, with tool schemas detached before the transform seam. Runs through the scope-filtered `system-prompt/assemble` waterfall and returns its authoritative result. An optional `context.signal` explicitly controls this assembly request; providers and listeners may cooperate with it but must not retain it for another turn. Rejects when a configured `toolOrder` names a tool outside the providers' `knownNames` universe, or when a provider returns the reserved rest-entry name.
@@ -29,14 +30,17 @@ System prompt assembly registry. Plugins contribute ordered sections, tool schem
 
 - `AssembleContext` — what one `assemble()` call is FOR. Merge-extensible; declares `scope?: ScopeKey` (the layer selector) and `signal?: AbortSignal` (the explicit request control capability) here, while `dsh-agent` declares `agent?: Agent` (the typed DX field — never set without `scope`; use `assembleContextFor(agent, signal)`). Providers must tolerate absent fields because a bare `assemble()` carries an empty, scope-less, signal-less context. `signal` is a request value, not part of the ambient Agent execution frame.
 - `PromptSection` — `{ name, order, text }`. Sections are concatenated in ascending `order`. Order bands: `-100` is the harness identity, `0` the deployment persona, tool guidance uses `100–199`.
-- `PromptAssembly` — `{ sections: AssembledSection[], tools: ToolSchema[], variables: Record<string, string | undefined> }`. Section texts arrive resolved but not yet interpolated; `variables` holds every registered variable resolved against the context. Tool schemas are part of the assembly by design: "what the model is told it can do" is one coherent thing, even though adapters transmit schemas as a separate wire field.
+- `PromptContext` — `{ name, order, text }`. Contexts carry changing current facts that must not rewrite the cached system/history prefix; they use the same per-assembly provider and strict-variable contracts as sections.
+- `PromptAssembly` — `{ sections: AssembledSection[], contexts: AssembledContext[], tools: ToolSchema[], variables: Record<string, string | undefined> }`. Section and context texts arrive resolved but not yet interpolated; `variables` holds every registered variable resolved against the context. Tool schemas are part of the assembly by design: "what the model is told it can do" is one coherent thing, even though adapters transmit schemas as a separate wire field.
 - `renderPrompt(assembly)` — interpolates `{{variable}}` references in each section, drops empty sections, joins with blank lines. STRICT: an unknown reference (`Object.hasOwn` lookup — prototype names like `{{constructor}}` are unknown), a registered-but-valueless reference, a malformed complete `{{…}}` group, or a `{{` that opens no complete group while a `}}` still follows (`{{{model}}}`) throws — fail loud beats shipping a malformed prompt. A lone `{{` with no `}}` anywhere after it passes through verbatim; substituted values are never re-scanned.
+- `renderContextSnapshot(assembly)` — applies the same strict interpolation to contexts, drops empty entries, and emits one full snapshot with an explicit supersession statement. An empty active set returns `''`; the loop emits one clearing snapshot when previously visible context disappears.
 
 Merge-extensible: plugins can declare extra fields on `PromptAssembly` and `AssembleContext` via declaration merging.
 
 ### Extension points
 
 - Section providers: tool packages own their cross-call guidance (`tool:bash`, `tool:read`, …); this plugin owns `harness:identity` and `deployment:persona`.
+- Context providers: policy and other changing-state owners contribute complete current facts without mutating the stable system prompt.
 - Variable providers: the agent loop registers `model` and `cwd`; any plugin can register the facts it owns (a future `date`, git state, …).
 - Tool schema providers: `ToolRegistry` registers itself as a tool provider automatically.
 - The [`system-prompt/assemble` waterfall](#live-events): cooperatively mutate or replace the assembly per caller.
@@ -65,6 +69,20 @@ Identity is a fixed per-request cost when enabled. Persona and plugin text are r
 
 Prefix-stable while identity, persona, variables, section text, and order render identically. Any change may invalidate reuse from the first changed system-prompt token.
 
+### Dynamic runtime context
+
+#### What the model sees
+
+Active contexts are joined in deterministic order after strict interpolation and logged as one sourced user-role message immediately before the request that first needs that snapshot. The message begins `Current runtime context. This snapshot supersedes earlier runtime-context snapshots.` A changed snapshot is appended after retained history; an unchanged retained snapshot adds nothing. If compaction removes it, the current full snapshot is emitted again. Removing the last context emits one explicit clearing snapshot.
+
+#### Token effect
+
+One concise message on the first request, on an effective context change, after compaction removed the retained snapshot, or when the active set becomes empty. Unchanged steps add no duplicate tokens.
+
+#### KV Cache effect
+
+Append-only after retained history. A context change preserves the previously cached system and conversation prefix instead of rewriting the first wire message.
+
 ### Tool schemas
 
 #### What the model sees

packages/core/system-prompt/README.zh.md

@@ -2,7 +2,7 @@
 
 [English](README.md) | 中文
 
-系统提示词组装注册表。插件贡献有序段、工具 schema 和具名变量。循环在每个步骤组装一次，并将结果渲染为完整的模型提示词。此插件拥有静态 harness 身份和全局部署 persona；agent（智能体）作用域的 persona 会遮蔽全局默认值。
+模型输入组装注册表。插件贡献有序且稳定的系统段、缓存安全的动态上下文、工具 schema 和具名变量。循环在每个步骤组装一次，将稳定段渲染为系统提示词，并且仅在文本变化或压缩（compaction）移除了保留的快照时，追加一份持久的完整动态上下文快照。此插件拥有静态 harness 身份和全局部署 persona；agent（智能体）作用域的 persona 会遮蔽全局默认值。
 
 ## 配置
 
@@ -17,6 +17,7 @@
 ### 公开 API
 
 - `ctx.systemPrompt.section(section: PromptSection): () => void`：贡献一个段。层由调用上下文的作用域决定：`agent.ctx` 只为该 agent 贡献，并在该处遮蔽同名全局段。同一层中的重复名称和非有限顺序会抛出。随调用 fiber 一并 dispose（资源释放）。
+- `ctx.systemPrompt.context(context: PromptContext): () => void`：贡献缓存安全的动态模型上下文。上下文与系统段分别排序；带作用域的贡献会遮蔽同名全局项。仅在完整当前集合变化或缺失时，agent loop（智能体循环）会在保留的历史后将其具体化为一份带来源的 user 角色快照。同一层中的重复名称和非有限顺序会抛出。随调用 fiber 一并 dispose。
 - `ctx.systemPrompt.tools(provider: (context: AssembleContext) => ToolProviderResult): () => void`：贡献工具 schema；每次组装时使用该次组装的上下文求值。`ToolProviderResult` = `{ schemas, knownNames? }`：`schemas` 是限制后的可见集合；`knownNames` 是限制前由 `toolOrder` 使用的全集。提供方不得返回名为 `TOOL_ORDER_REST` 的 schema。带作用域提供方只在其作用域的组装中查询。随调用 fiber 一并 dispose。
 - `ctx.systemPrompt.variable(name: string, provider: (context) => string | undefined): () => void`：贡献提示词变量，在段文本中以 `{{name}}` 引用。带作用域变量会为该 agent 遮蔽同名全局变量。同层重复或无法引用的名称会抛出；`undefined` 表示「本次组装没有值」。随调用 fiber 一并 dispose。
 - `ctx.systemPrompt.assemble(context?: AssembleContext): Promise<PromptAssembly>`：为一个调用方组装提示词：将全局层与 `context.scope` 的层合并，并在变换 seam 前分离工具 schema。它经过按作用域筛选的 `system-prompt/assemble` waterfall，并返回其权威结果。可选的 `context.signal` 显式控制本次组装请求；提供方与监听器可以配合该信号，但不得将它保留给另一轮次。当已配置的 `toolOrder` 指名提供方 `knownNames` 全集以外的工具，或提供方返回保留的其余项名称时，调用会被拒绝。
@@ -29,15 +30,18 @@
 
 - `AssembleContext`：说明一次 `assemble()` 调用的用途。它可通过合并扩展；此处声明 `scope?: ScopeKey`（层选择器）与 `signal?: AbortSignal`（显式请求控制能力），而 `dsh-agent` 声明 `agent?: Agent`（类型化 DX 字段；绝不能在没有 `scope` 时设置，应使用 `assembleContextFor(agent, signal)`）。提供方必须容忍字段缺席，因为裸 `assemble()` 携带的是无作用域、无信号的空上下文。`signal` 是请求值，不是环境 Agent 执行 frame 的一部分。
 - `PromptSection`：`{ name, order, text }`。各段按 `order` 升序拼接。顺序区间：`-100` 是 harness 身份，`0` 是部署 persona，工具引导使用 `100–199`。
-- `PromptAssembly`：`{ sections: AssembledSection[], tools: ToolSchema[], variables: Record<string, string | undefined> }`。段文本到达时已解析，但尚未插值；`variables` 包含对上下文解析后的每个已注册变量。工具 schema 按设计属于组装结果：「模型获知自己能做什么」是一个连贯整体，尽管适配器把 schema 作为独立 wire 字段传输。
+- `PromptContext`：`{ name, order, text }`。上下文承载不断变化的当前事实，这些事实不能改写已缓存的系统／历史前缀；上下文与段使用相同的逐组装提供方契约和严格变量契约。
+- `PromptAssembly`：`{ sections: AssembledSection[], contexts: AssembledContext[], tools: ToolSchema[], variables: Record<string, string | undefined> }`。段与上下文文本到达时已解析，但尚未插值；`variables` 包含对上下文解析后的每个已注册变量。工具 schema 按设计属于组装结果：「模型获知自己能做什么」是一个连贯整体，尽管适配器把 schema 作为独立 wire 字段传输。
 - `renderPrompt(assembly)`：插值每个段中的 `{{variable}}` 引用，删除空段，并用空行连接。严格规则：未知引用（使用 `Object.hasOwn` 查找，因此 `{{constructor}}` 等原型名称未知）、已注册但无值的引用、格式错误的完整 `{{…}}` 组，或一个起始 `{{` 没有打开完整组、但后面仍有 `}}`（`{{{model}}}`），都会抛出；明确失败胜过交付格式错误的提示词。孤立的 `{{` 如果后面任何位置都没有 `}}`，会按字面量通过；替换值绝不再次扫描。
+- `renderContextSnapshot(assembly)`：对上下文执行同样严格的插值，删除空条目，并发出一份带显式取代声明的完整快照。活动集合为空时返回 `''`；先前可见的上下文消失时，循环会发出一份清除快照。
 
 可通过合并扩展：插件可以借助声明合并，为 `PromptAssembly` 和 `AssembleContext` 声明额外字段。
 
 ### 扩展点
 
 - 段提供方：工具包（package）拥有跨调用引导（`tool:bash`、`tool:read` 等）；此插件拥有 `harness:identity` 与 `deployment:persona`。
-- 变量提供方：agent loop（智能体循环）注册 `model` 与 `cwd`；任何插件都可以注册自己拥有的事实（未来的 `date`、git 状态等）。
+- 上下文提供方：策略及其他变化状态的归属方贡献完整的当前事实，而不改变稳定的系统提示词。
+- 变量提供方：agent loop 注册 `model` 与 `cwd`；任何插件都可以注册自己拥有的事实（未来的 `date`、git 状态等）。
 - 工具 schema 提供方：`ToolRegistry` 自动将自身注册为工具提供方。
 - [`system-prompt/assemble` waterfall](#live-events)：按调用方协作式修改或替换组装结果。
 
@@ -65,6 +69,20 @@ You are an AI agent powered by the DeepSeek Harness SDK.
 
 只要身份、persona、变量、段文本与顺序的渲染完全相同，前缀就保持稳定。任何变更都可能从第一个变化的系统提示词 token 起使复用失效。
 
+### 动态运行时上下文
+
+#### 模型看到的内容
+
+活动上下文经过严格插值后按确定顺序连接，并在首次需要该快照的请求之前立即记录为一条带来源的 user 角色消息。消息以 `Current runtime context. This snapshot supersedes earlier runtime-context snapshots.` 开头。变化后的快照会追加到保留的历史之后；保留的快照未变时不会增加内容。如果压缩移除了它，当前完整快照会再次发出。移除最后一项上下文时会发出一份显式清除快照。
+
+#### Token 影响
+
+首次请求、上下文实际变化、压缩移除保留的快照或活动集合变空时，会增加一条简洁消息。未变化的步骤不会增加重复 token。
+
+#### KV Cache 影响
+
+在保留的历史之后仅追加。上下文变化会保留先前缓存的系统与对话前缀，而不会改写第一条 wire 消息。
+
 ### 工具 schema
 
 #### 模型看到的内容

packages/core/system-prompt/src/index.ts

@@ -1,5 +1,5 @@
 /**
- * Registry for ordered prompt sections, tool schemas, and prompt variables.
+ * Registry for ordered system sections, cache-safe context, tool schemas, and prompt variables.
  *
  * @module @deepseek-ai/dsh-system-prompt
  */
@@ -17,7 +17,7 @@ declare module 'cordis' {
 
   interface Events {
     /**
-     * Expert waterfall over the assembled sections, tools, and variables.
+     * Expert waterfall over the assembled sections, contexts, tools, and variables.
      * Scope-filtered dispatch (`@deepseek-ai/dsh-scope`): scoped listeners
      * receive only that scope's assemblies. The returned value is authoritative.
      * A supplied signal controls only this explicit assembly request and must not
@@ -65,6 +65,20 @@ export interface PromptSection {
   readonly text: string | ((context: AssembleContext) => string)
 }
 
+/**
+ * One dynamic model-context contribution. Unlike a {@link PromptSection}, its
+ * rendered text is materialized as a durable user-role snapshot at the request
+ * tail, so changing runtime state preserves the stable system/history prefix.
+ */
+export interface PromptContext {
+  /** Unique name — a duplicate registration throws (see {@link SystemPrompt.context}). */
+  readonly name: string
+  /** Contexts are joined in ascending order, independently of system-section order. */
+  readonly order: number
+  /** Static text or a provider evaluated for each assembly. Empty text contributes nothing. */
+  readonly text: string | ((context: AssembleContext) => string)
+}
+
 /** One section of an assembly: {@link PromptSection} with its text resolved. */
 export interface AssembledSection {
   /** The contributing section's unique name. */
@@ -73,6 +87,14 @@ export interface AssembledSection {
   text: string
 }
 
+/** One dynamic context contribution with its text resolved. */
+export interface AssembledContext {
+  /** The contributing context's unique name. */
+  name: string
+  /** The resolved (but not yet interpolated) context text. */
+  text: string
+}
+
 /** Tool schemas visible in one assembly and their pre-restriction name set. */
 export interface ToolProviderResult {
   /** The schemas this provider contributes to THIS assembly. */
@@ -82,11 +104,13 @@ export interface ToolProviderResult {
 }
 
 /**
- * Merge-extensible assembled prompt. Sections remain uninterpolated until
- * {@link renderPrompt}; tools are already in canonical model-facing order.
+ * Merge-extensible assembled model input. Sections and contexts remain
+ * uninterpolated until their renderers; tools are already in canonical
+ * model-facing order.
  */
 export interface PromptAssembly {
   sections: AssembledSection[]
+  contexts: AssembledContext[]
   tools: ToolSchema[]
   variables: Record<string, string | undefined>
 }
@@ -170,14 +194,35 @@ export interface Config {
  */
 export function renderPrompt(assembly: PromptAssembly): string {
   return assembly.sections
-    .map(section => interpolate(section, assembly.variables))
+    .map(section => interpolate(section, assembly.variables, 'section'))
     .filter(text => text.length > 0)
     .join('\n\n')
 }
 
-/** Interpolate one section's `{{variable}}` references (see {@link renderPrompt}). */
-function interpolate(section: AssembledSection, variables: Record<string, string | undefined>): string {
-  const text = section.text
+/**
+ * Render the complete current dynamic context snapshot. The agent loop appends
+ * a new durable snapshot only when this text changes or is no longer retained
+ * after compaction; the explicit supersession clause makes older snapshots in
+ * history harmless.
+ * @param assembly - the assembly whose contexts and variables to render.
+ * @returns the current full snapshot, or `''` when no context is active.
+ */
+export function renderContextSnapshot(assembly: PromptAssembly): string {
+  const body = assembly.contexts
+    .map(context => interpolate(context, assembly.variables, 'context'))
+    .filter(text => text.length > 0)
+    .join('\n\n')
+  if (body.length === 0) return ''
+  return `Current runtime context. This snapshot supersedes earlier runtime-context snapshots.\n\n${body}`
+}
+
+/** Interpolate one section or context and attribute diagnostics to its owning input. */
+function interpolate(
+  input: AssembledSection | AssembledContext,
+  variables: Record<string, string | undefined>,
+  kind: 'section' | 'context',
+): string {
+  const text = input.text
   let result = ''
   let last = 0
   for (let open = text.indexOf('{{'); open >= 0; open = text.indexOf('{{', last)) {
@@ -185,7 +230,7 @@ function interpolate(section: AssembledSection, variables: Record<string, string
     if (group === null) {
       // A later closing brace makes this malformed; otherwise it is literal prose.
       if (text.indexOf('}}', open + 2) >= 0) {
-        throw new Error(`malformed prompt variable reference at "${text.slice(open, open + 16)}…" in section "${section.name}" (references are complete simple {{name}} groups)`)
+        throw new Error(`malformed prompt variable reference at "${text.slice(open, open + 16)}…" in ${kind} "${input.name}" (references are complete simple {{name}} groups)`)
       }
       result += text.slice(last, open + 2)
       last = open + 2
@@ -194,16 +239,16 @@ function interpolate(section: AssembledSection, variables: Record<string, string
     // `{{}}` yields an empty name and follows the malformed-reference path.
     const name = group[0].slice(2, -2)
     if (!VARIABLE_NAME.test(name)) {
-      throw new Error(`malformed prompt variable reference "{{${name}}}" in section "${section.name}" (variable names match ${String(VARIABLE_NAME)})`)
+      throw new Error(`malformed prompt variable reference "{{${name}}}" in ${kind} "${input.name}" (variable names match ${String(VARIABLE_NAME)})`)
     }
     // Do not resolve unregistered names through Object.prototype.
     if (!Object.hasOwn(variables, name)) {
       const known = Object.keys(variables)
-      throw new Error(`unknown prompt variable "{{${name}}}" in section "${section.name}"; registered variables: ${known.length > 0 ? known.join(', ') : '(none)'}`)
+      throw new Error(`unknown prompt variable "{{${name}}}" in ${kind} "${input.name}"; registered variables: ${known.length > 0 ? known.join(', ') : '(none)'}`)
     }
     const value = variables[name]
     if (value === undefined) {
-      throw new Error(`prompt variable "{{${name}}}" has no value for this assembly (section "${section.name}")`)
+      throw new Error(`prompt variable "{{${name}}}" has no value for this assembly (${kind} "${input.name}")`)
     }
     result += text.slice(last, open) + value
     last = open + group[0].length
@@ -220,6 +265,7 @@ type VariableProvider = (context: AssembleContext) => string | undefined
 /** All prompt registrations owned by one global or scoped layer. */
 class PromptLayer implements ScopeLayer {
   readonly sections: NamedEntries<PromptSection>
+  readonly contexts: NamedEntries<PromptContext>
   readonly toolProviders = new AnonymousEntries<ToolProvider>()
   readonly variables: NamedEntries<VariableProvider>
 
@@ -231,6 +277,9 @@ class PromptLayer implements ScopeLayer {
     this.sections = new NamedEntries(name => new Error(scope === undefined
       ? `prompt section "${name}" is already registered (for a per-agent override, register through that agent's \`agent.ctx\` instead)`
       : `prompt section "${name}" is already registered in this scope`))
+    this.contexts = new NamedEntries(name => new Error(scope === undefined
+      ? `prompt context "${name}" is already registered (for a per-agent override, register through that agent's \`agent.ctx\` instead)`
+      : `prompt context "${name}" is already registered in this scope`))
     this.variables = new NamedEntries(name => new Error(scope === undefined
       ? `prompt variable "${name}" is already registered (for a per-agent value, register through that agent's \`agent.ctx\` instead)`
       : `prompt variable "${name}" is already registered in this scope`))
@@ -239,6 +288,7 @@ class PromptLayer implements ScopeLayer {
   /** @returns whether this layer owns no prompt registrations. */
   isEmpty(): boolean {
     return this.sections.isEmpty()
+      && this.contexts.isEmpty()
       && this.toolProviders.isEmpty()
       && this.variables.isEmpty()
   }
@@ -297,6 +347,25 @@ export class SystemPrompt extends Service {
     )
   }
 
+  /**
+   * Register ordered cache-safe dynamic context in the calling context's scope.
+   * A scoped context shadows a global context with the same name; duplicates
+   * within one layer and non-finite orders throw. Registration and disposal
+   * emit `system-prompt/change`.
+   * @param context - the context contribution to register.
+   * @returns the exact Cordis effect disposer.
+   */
+  context(context: PromptContext): () => void {
+    if (!Number.isFinite(context.order)) {
+      throw new TypeError(`prompt context "${context.name}" order must be a finite number`)
+    }
+    return this.layers.effect(
+      this.ctx,
+      layer => layer.contexts.insert(context.name, context),
+      { label: 'systemPrompt.context()' },
+    )
+  }
+
   /**
    * Register a tool-schema provider in the calling context's scope. Global and
    * matching scoped providers both contribute; returning the reserved
@@ -352,6 +421,7 @@ export class SystemPrompt extends Service {
     }
     // Scoped sections shadow globals before the stable order sort.
     const sectionByName = this.layers.merge(scope, layer => layer.sections)
+    const contextByName = this.layers.merge(scope, layer => layer.contexts)
     // Validate order against pre-restriction names while collecting visible schemas.
     const providers = [
       ...this.layers.global.toolProviders.values(),
@@ -377,6 +447,12 @@ export class SystemPrompt extends Service {
           name: section.name,
           text: typeof section.text === 'function' ? section.text(context) : section.text,
         })),
+      contexts: [...contextByName.values()]
+        .sort((a, b) => a.order - b.order)
+        .map(entry => ({
+          name: entry.name,
+          text: typeof entry.text === 'function' ? entry.text(context) : entry.text,
+        })),
       tools: orderTools(collected, this.toolOrder, knownNames),
       variables,
     }

packages/core/system-prompt/src/invariant.ts

@@ -22,6 +22,14 @@ function validateAssembly(assembly: PromptAssembly, fail: InvariantFailure): voi
     if (typeof section.text !== 'string') fail(`assembled section ${JSON.stringify(section.name)} text must be a string`)
   }
 
+  const contextNames = new Set<string>()
+  for (const context of assembly.contexts) {
+    if (context.name.length === 0) fail('assembled context names must be non-empty')
+    if (contextNames.has(context.name)) fail(`assembled context name ${JSON.stringify(context.name)} is duplicated`)
+    contextNames.add(context.name)
+    if (typeof context.text !== 'string') fail(`assembled context ${JSON.stringify(context.name)} text must be a string`)
+  }
+
   for (const tool of assembly.tools) {
     if (tool.name.length === 0) fail('assembled tool names must be non-empty')
   }

packages/core/system-prompt/tests/invariant.spec.ts

@@ -13,6 +13,7 @@ async function setup(): Promise<Context> {
 
 const valid = (): PromptAssembly => ({
   sections: [{ name: 'identity', text: 'prompt' }],
+  contexts: [{ name: 'policy', text: 'current policy' }],
   tools: [{ name: 'echo', description: 'Echo', parameters: {} }],
   variables: { cwd: '/repo', optional: undefined },
 })
@@ -34,6 +35,9 @@ describe('system-prompt invariants', () => {
     [{ ...valid(), sections: [{ name: '', text: 'x' }] }, /section names must be non-empty/],
     [{ ...valid(), sections: [{ name: 'x', text: 'a' }, { name: 'x', text: 'b' }] }, /section name "x" is duplicated/],
     [{ ...valid(), sections: [{ name: 'x', text: 1 as never }] }, /section "x" text must be a string/],
+    [{ ...valid(), contexts: [{ name: '', text: 'x' }] }, /context names must be non-empty/],
+    [{ ...valid(), contexts: [{ name: 'x', text: 'a' }, { name: 'x', text: 'b' }] }, /context name "x" is duplicated/],
+    [{ ...valid(), contexts: [{ name: 'x', text: 1 as never }] }, /context "x" text must be a string/],
     [{ ...valid(), tools: [{ name: '', description: 'x', parameters: {} }] }, /tool names must be non-empty/],
     [{ ...valid(), variables: { Bad: 'x' } }, /variable name "Bad" is invalid/],
     [{ ...valid(), variables: { value: 1 as never } }, /variable "value" must be a string or undefined/],

packages/core/system-prompt/tests/scoped.spec.ts

@@ -2,7 +2,7 @@ import { describe, expect, it, vi } from 'vitest'
 import { Context } from 'cordis'
 import { createScope, scopeOf } from '@deepseek-ai/dsh-scope'
 import type { Scope, ScopeKey } from '@deepseek-ai/dsh-scope'
-import SystemPrompt, { TOOL_ORDER_REST, renderPrompt } from '@deepseek-ai/dsh-system-prompt'
+import SystemPrompt, { TOOL_ORDER_REST, renderContextSnapshot, renderPrompt } from '@deepseek-ai/dsh-system-prompt'
 import type { Config, PromptAssembly } from '@deepseek-ai/dsh-system-prompt'
 
 async function mount(config: Config = {}): Promise<Context> {
@@ -125,6 +125,25 @@ describe('scoped variables', () => {
   })
 })
 
+describe('scoped cache-safe context', () => {
+  it('shadows a global context for one scope and cleans up with that scope', async () => {
+    const ctx = await mount()
+    const scope = await mintScope(ctx, 'child-context')
+    ctx.systemPrompt.context({ name: 'policy', order: 1, text: 'global policy' })
+    scope.ctx.systemPrompt.context({ name: 'policy', order: 1, text: 'scoped policy' })
+    expect(() => scope.ctx.systemPrompt.context({ name: 'policy', order: 2, text: 'duplicate' }))
+      .toThrow('prompt context "policy" is already registered in this scope')
+
+    expect(renderContextSnapshot(await ctx.systemPrompt.assemble({ scope: scopeKeyOf(scope) })))
+      .toContain('scoped policy')
+    expect(renderContextSnapshot(await ctx.systemPrompt.assemble())).toContain('global policy')
+
+    await scope.dispose()
+    expect(renderContextSnapshot(await ctx.systemPrompt.assemble({ scope: scopeKeyOf(scope) })))
+      .toContain('global policy')
+  })
+})
+
 describe('scoped tool providers and toolOrder × restriction', () => {
   it('scoped providers are consulted only for their scope', async () => {
     const ctx = await mount()

packages/core/system-prompt/tests/system-prompt.spec.ts

@@ -1,6 +1,6 @@
 import { describe, expect, it } from 'vitest'
 import { Context } from 'cordis'
-import SystemPrompt, { AssembleContext, PromptAssembly, renderPrompt } from '@deepseek-ai/dsh-system-prompt'
+import SystemPrompt, { AssembleContext, PromptAssembly, renderContextSnapshot, renderPrompt } from '@deepseek-ai/dsh-system-prompt'
 
 /**
  * Every assembly carries the plugin's own built-ins — `harness:identity`
@@ -64,14 +64,21 @@ describe('SystemPrompt', () => {
 
     ctx.systemPrompt.section({ name: 'cwd', order: 20, text: () => 'cwd: /tmp' })
     ctx.systemPrompt.section({ name: 'rules', order: 10, text: 'Be precise.' })
+    ctx.systemPrompt.context({ name: 'later', order: 20, text: () => 'context 2' })
+    ctx.systemPrompt.context({ name: 'earlier', order: 10, text: 'context 1' })
     ctx.systemPrompt.tools(() => ({ schemas: [{ name: 'echo', description: 'echo back', parameters: {} }] }))
 
     const assembly = await ctx.systemPrompt.assemble()
     expect(assembly.sections.map(s => s.name)).toEqual(['harness:identity', 'deployment:persona', 'rules', 'cwd'])
     expect(assembly.sections.map(s => s.text)).toEqual([IDENTITY, 'You are DeepSeek Harness SDK.', 'Be precise.', 'cwd: /tmp'])
+    expect(assembly.contexts).toEqual([
+      { name: 'earlier', text: 'context 1' },
+      { name: 'later', text: 'context 2' },
+    ])
     expect(assembly.tools).toEqual([{ name: 'echo', description: 'echo back', parameters: {} }])
     expect(assembly.variables).toEqual({})
     expect(renderPrompt(assembly)).toBe(`${IDENTITY}\n\nYou are DeepSeek Harness SDK.\n\nBe precise.\n\ncwd: /tmp`)
+    expect(renderContextSnapshot(assembly)).toBe('Current runtime context. This snapshot supersedes earlier runtime-context snapshots.\n\ncontext 1\n\ncontext 2')
   })
 
   it('resolves section text providers against the assemble context, at each assemble call', async () => {
@@ -96,16 +103,19 @@ describe('SystemPrompt', () => {
 
     const fiber = await ctx.plugin(Object.assign((inner: Context) => {
       inner.systemPrompt.section({ name: 'scoped', order: 0, text: 'scoped section' })
+      inner.systemPrompt.context({ name: 'scoped-context', order: 0, text: 'scoped context' })
       inner.systemPrompt.tools(() => ({ schemas: [{ name: 'scoped-tool', description: '', parameters: {} }] }))
       inner.systemPrompt.variable('scoped_var', () => 'v')
     }, { inject: ['systemPrompt'] }))
 
     const before = await ctx.systemPrompt.assemble()
     expect(contributed(before)).toHaveLength(1)
+    expect(before.contexts).toHaveLength(1)
     expect(before.variables).toEqual({ scoped_var: 'v' })
     await fiber.dispose()
     const assembly = await ctx.systemPrompt.assemble()
     expect(contributed(assembly)).toHaveLength(0)
+    expect(assembly.contexts).toHaveLength(0)
     // The built-ins belong to the service fiber, so they survive the plugin's disposal.
     expect(assembly.sections.map(s => s.name)).toEqual(BUILT_IN)
     expect(assembly.tools).toHaveLength(0)
@@ -131,6 +141,17 @@ describe('SystemPrompt', () => {
     expect(contributed(await ctx.systemPrompt.assemble())).toEqual([])
   })
 
+  it('rejects duplicate and non-finite context registrations without leaking', async () => {
+    const ctx = new Context()
+    await ctx.plugin(SystemPrompt)
+    ctx.systemPrompt.context({ name: 'policy', order: 1, text: 'first' })
+    expect(() => ctx.systemPrompt.context({ name: 'policy', order: 2, text: 'second' }))
+      .toThrow('prompt context "policy" is already registered')
+    expect(() => ctx.systemPrompt.context({ name: 'bad', order: Number.NaN, text: 'x' }))
+      .toThrow('prompt context "bad" order must be a finite number')
+    expect((await ctx.systemPrompt.assemble()).contexts).toEqual([{ name: 'policy', text: 'first' }])
+  })
+
   it('rolls back a section when a system-prompt/change listener throws (P1-1)', async () => {
     const ctx = new Context()
     await ctx.plugin(SystemPrompt)
@@ -236,7 +257,7 @@ describe('SystemPrompt', () => {
     ctx.systemPrompt.section({ name: 'real', order: 0, text: 'real' })
 
     ctx.on('system-prompt/assemble', async () => {
-      return { sections: [], tools: [], variables: {} } satisfies PromptAssembly
+      return { sections: [], contexts: [], tools: [], variables: {} } satisfies PromptAssembly
     })
 
     const assembly = await ctx.systemPrompt.assemble()
@@ -252,6 +273,7 @@ describe('SystemPrompt', () => {
     const first = await ctx.systemPrompt.assemble()
     first.sections[0]!.name = 'mutated'
     first.sections[0]!.text = 'mutated'
+    first.contexts.push({ name: 'mutated', text: 'mutated' })
     first.tools[0]!.description = 'mutated'
     const firstParameters = first.tools[0]!.parameters as { properties: Record<string, unknown> }
     firstParameters.properties['leak'] = { type: 'string' }
@@ -259,6 +281,7 @@ describe('SystemPrompt', () => {
     const second = await ctx.systemPrompt.assemble()
     expect(second.sections.map(section => section.name)).toEqual(['harness:identity', 'deployment:persona', 'base'])
     expect(second.sections[0]!.text).toBe(IDENTITY)
+    expect(second.contexts).toEqual([])
     expect(second.tools).toEqual([{ name: 't', description: 'tool', parameters: { type: 'object', properties: {} } }])
   })
 
@@ -268,12 +291,33 @@ describe('SystemPrompt', () => {
         { name: 'empty', text: '' },
         { name: 'real', text: 'content' },
       ],
+      contexts: [],
       tools: [],
       variables: {},
     })
     expect(result).toBe('content')
   })
 
+  it('filters empty context, interpolates variables, and returns empty without active context', async () => {
+    const ctx = new Context()
+    await ctx.plugin(SystemPrompt)
+    ctx.systemPrompt.context({ name: 'empty', order: 0, text: '' })
+    expect(renderContextSnapshot(await ctx.systemPrompt.assemble())).toBe('')
+    ctx.systemPrompt.variable('mode', () => 'read-only')
+    ctx.systemPrompt.context({ name: 'policy', order: 1, text: 'Mode: {{mode}}.' })
+    expect(renderContextSnapshot(await ctx.systemPrompt.assemble()))
+      .toBe('Current runtime context. This snapshot supersedes earlier runtime-context snapshots.\n\nMode: read-only.')
+  })
+
+  it('attributes context interpolation failures to the contributing context', () => {
+    expect(() => renderContextSnapshot({
+      sections: [],
+      contexts: [{ name: 'policy', text: 'Mode: {{missing}}.' }],
+      tools: [],
+      variables: {},
+    })).toThrow('unknown prompt variable "{{missing}}" in context "policy"; registered variables: (none)')
+  })
+
   it('emits system-prompt/change when a tool provider is registered and disposed', async () => {
     const ctx = new Context()
     await ctx.plugin(SystemPrompt)
@@ -290,6 +334,17 @@ describe('SystemPrompt', () => {
     expect(changeCount).toBe(2)
   })
 
+  it('emits system-prompt/change when a context is registered and disposed', async () => {
+    const ctx = new Context()
+    await ctx.plugin(SystemPrompt)
+    let changeCount = 0
+    ctx.on('system-prompt/change', () => void changeCount++)
+    const dispose = ctx.systemPrompt.context({ name: 'policy', order: 0, text: 'current' })
+    expect(changeCount).toBe(1)
+    dispose()
+    expect(changeCount).toBe(2)
+  })
+
   it('cleans up tool providers on fiber dispose', async () => {
     const ctx = new Context()
     await ctx.plugin(SystemPrompt)
@@ -404,13 +459,14 @@ describe('SystemPrompt', () => {
     })
 
     it('names "(none)" when no variables are registered at all', () => {
-      expect(() => renderPrompt({ sections: [{ name: 's', text: '{{x}}' }], tools: [], variables: {} }))
+      expect(() => renderPrompt({ sections: [{ name: 's', text: '{{x}}' }], contexts: [], tools: [], variables: {} }))
         .toThrow('unknown prompt variable "{{x}}" in section "s"; registered variables: (none)')
     })
 
     it('throws when a referenced variable has no value for this assembly', () => {
       expect(() => renderPrompt({
         sections: [{ name: 'persona', text: 'in {{cwd}}' }],
+        contexts: [],
         tools: [],
         variables: { cwd: undefined },
       })).toThrow('prompt variable "{{cwd}}" has no value for this assembly (section "persona")')
@@ -419,6 +475,7 @@ describe('SystemPrompt', () => {
     it('throws on a malformed complete reference, e.g. inner spaces', () => {
       expect(() => renderPrompt({
         sections: [{ name: 's', text: 'on {{ model }}' }],
+        contexts: [],
         tools: [],
         variables: { model: 'm' },
       })).toThrow('malformed prompt variable reference "{{ model }}" in section "s"')
@@ -427,6 +484,7 @@ describe('SystemPrompt', () => {
     it('leaves a lone {{ verbatim only when NO }} follows anywhere after it', () => {
       const text = renderPrompt({
         sections: [{ name: 's', text: 'shell ${X:-{{fallback} stays' }],
+        contexts: [],
         tools: [],
         variables: {},
       })
@@ -439,6 +497,7 @@ describe('SystemPrompt', () => {
     ])('throws on a mangled reference with a }} still following ($label)', ({ text }) => {
       expect(() => renderPrompt({
         sections: [{ name: 's', text }],
+        contexts: [],
         tools: [],
         variables: { model: 'm' },
       })).toThrow('malformed prompt variable reference at')
@@ -449,6 +508,7 @@ describe('SystemPrompt', () => {
       // source into the prompt; Object.hasOwn must reject it instead.
       expect(() => renderPrompt({
         sections: [{ name: 's', text: 'on {{constructor}}' }],
+        contexts: [],
         tools: [],
         variables: { model: 'm' },
       })).toThrow('unknown prompt variable "{{constructor}}"')
@@ -465,6 +525,7 @@ describe('SystemPrompt', () => {
     it('never re-scans substituted values (a value containing {{sneaky}} stays literal)', () => {
       const text = renderPrompt({
         sections: [{ name: 's', text: 'v = {{model}}!' }],
+        contexts: [],
         tools: [],
         variables: { model: 'literal {{sneaky}} inside' },
       })

packages/core/tools/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/core/tools/README.md
-README.md: dcce455f9551318f3871e3df84c29789078fef7c
-README.zh.md: 63eaa2e0b66797c74a1d4845c29ca970b63f5f99
+README.md: 15fc5839a3b0e3fa2d20c5a9cc50577e9807ffda
+README.zh.md: 8547ee4a796dcd93945dfa40373c14c10d7d0c8a

packages/core/tools/README.md

@@ -108,7 +108,7 @@ Optional `isConcurrencySafe(args)` receives typed, softly validated arguments. E
 Tools optionally own pure `presentCall()` and `presentResult()` render intents, so UIs do not special-case tool names:
 
 - Call views are `{ card: 'generic', title, kind?, rawInput?, content?, locations? }`, `{ card: 'terminal', title, description?, cwd? }`, or `{ card: 'diff', title, diffs, locations? }`.
-- Result views are `{ card: 'generic', title?, content? }`, `{ card: 'terminal', title?, output?, exitCode?, signal? }`, `{ card: 'diff', title?, diffs }`, `{ card: 'read', title?, path, offset, lines, totalLines, lang?, content? }` (a completed file read → a line-numbered, optionally syntax-highlighted code view; `offset` is the 1-based first line the window requested, kept even when `lines` is empty; `lines` is `{ number, text }[]` keeping each file line number, and `content` is the envelope-stripped text a UI without read support falls back to), or `{ card: 'web', kind: 'search' | 'fetch', title?, … }` (a completed web retrieval; the `kind` arms carry the structured search sources or the fetch summary, and a UI without the `web` capability falls back to the raw result content).
+- Result views are `{ card: 'generic', title?, content? }`, `{ card: 'terminal', title?, output?, exitCode?, signal? }`, `{ card: 'diff', title?, diffs }`, `{ card: 'search', shape, title?, truncated, total, … }` (a completed discovery search — grouped-by-file matches for `shape: 'matches'` (grep) or a flat path list for `shape: 'paths'` (glob), with `truncated`/`total` so a UI never presents a capped result as complete; the view carries no result text and a search has no `card: 'search'` call-time analogue), `{ card: 'read', title?, path, offset, lines, totalLines, lang?, content? }` (a completed file read → a line-numbered, optionally syntax-highlighted code view; `offset` is the 1-based first line the window requested, kept even when `lines` is empty; `lines` is `{ number, text }[]` keeping each file line number, and `content` is the envelope-stripped text a UI without read support falls back to), or `{ card: 'web', kind: 'search' | 'fetch', title?, … }` (a completed web retrieval; the `kind` arms carry the structured search sources or the fetch summary, and a UI without the `web` capability falls back to the raw result content).
 
 Returning `undefined` selects generic fallback. Presenters depend only on their arguments and the durable result because UIs call them during live streaming and log replay. `output.presentationMeta(args, value)` derives JSON metadata for direct surface calls; that metadata persists with `tool/result` and returns to `presentResult`, while the canonical value itself remains execution-local and is never replayed. Nested Code dispatches do not compute metadata. `defineTool` soft-validates older logged arguments and falls back instead of crashing replay. `dsh-tool-bash` and `dsh-tool-fs` are the reference implementations; the [canonical-output Agent Note](../../../.agents/notes/implemented/architecture/2026-07-20-canonical-tool-output-contract.md) owns the value/presentation split and the [render-intent Agent Note](../../../.agents/notes/implemented/architecture/2026-07-02-tool-render-intent-union.md) owns card vocabulary.
 

packages/core/tools/README.zh.md

@@ -108,7 +108,7 @@ ctx.tools.register(defineTool({
 工具可以选择拥有纯 `presentCall()` 和 `presentResult()` 呈现意图，使 UI 无需特殊处理工具名称：
 
 - 调用视图为 `{ card: 'generic', title, kind?, rawInput?, content?, locations? }`、`{ card: 'terminal', title, description?, cwd? }` 或 `{ card: 'diff', title, diffs, locations? }`。
-- 结果视图为 `{ card: 'generic', title?, content? }`、`{ card: 'terminal', title?, output?, exitCode?, signal? }`、`{ card: 'diff', title?, diffs }`、`{ card: 'read', title?, path, offset, lines, totalLines, lang?, content? }`（已完成的文件读取→带行号、可选语法高亮的代码视图；`offset` 是窗口请求的 1-based 起始行，即使 `lines` 为空也保留；`lines` 是 `{ number, text }[]`，保留每一行的文件行号，`content` 是无读取能力的 UI 回退时使用的去信封文本）或 `{ card: 'web', kind: 'search' | 'fetch', title?, … }`（已完成的 web 检索；`kind` 各分支携带结构化的搜索来源或抓取摘要，不具备 `web` 能力的 UI 回退到原始结果内容）。
+- 结果视图为 `{ card: 'generic', title?, content? }`、`{ card: 'terminal', title?, output?, exitCode?, signal? }`、`{ card: 'diff', title?, diffs }`、`{ card: 'search', shape, title?, truncated, total, … }`（已完成的发现型搜索——`shape: 'matches'`（grep）为按文件分组的匹配，`shape: 'paths'`（glob）为扁平路径列表，配 `truncated`/`total` 使 UI 永不把被截断的结果当作完整结果呈现；该视图不携带结果文本，且搜索没有 `card: 'search'` 的调用时对应视图）、`{ card: 'read', title?, path, offset, lines, totalLines, lang?, content? }`（已完成的文件读取→带行号、可选语法高亮的代码视图；`offset` 是窗口请求的 1-based 起始行，即使 `lines` 为空也保留；`lines` 是 `{ number, text }[]`，保留每一行的文件行号，`content` 是无读取能力的 UI 回退时使用的去信封文本）或 `{ card: 'web', kind: 'search' | 'fetch', title?, … }`（已完成的 web 检索；`kind` 各分支携带结构化的搜索来源或抓取摘要，不具备 `web` 能力的 UI 回退到原始结果内容）。
 
 返回 `undefined` 会选择通用回退。呈现器只依赖其参数和持久结果，因为 UI 会在实时流式输出和日志回放期间调用它们。`output.presentationMeta(args, value)` 为直接接口调用派生 JSON 元数据；该元数据随 `tool/result` 持久化并传回 `presentResult`，而规范值本身仍只存在于执行局部，绝不会回放。嵌套 Code 分发不会计算元数据。`defineTool` 会软验证较旧的日志参数并回退，而不会使回放崩溃。`dsh-tool-bash` 与 `dsh-tool-fs` 是参考实现；[规范输出 Agent Note](../../../.agents/notes/implemented/architecture/2026-07-20-canonical-tool-output-contract.md) 规定值／呈现拆分，[呈现意图 Agent Note](../../../.agents/notes/implemented/architecture/2026-07-02-tool-render-intent-union.md) 规定卡片词汇。
 

packages/core/tools/src/index.ts

@@ -83,6 +83,11 @@ export type {
   GenericResultView,
   TerminalResultView,
   DiffResultView,
+  SearchResultView,
+  SearchMatchesResultView,
+  SearchPathsResultView,
+  SearchFileMatches,
+  SearchLineMatch,
   ReadResultView,
   WebResultView,
   WebSearchResultView,

packages/core/tools/src/presentation.ts

@@ -137,7 +137,7 @@ export interface ReadFileLine {
  * `ToolDefinition.presentResult`; omitting the method keeps the pending
  * title and renders the raw result content.
  */
-export type ToolResultView = GenericResultView | TerminalResultView | DiffResultView | ReadResultView | WebResultView
+export type ToolResultView = GenericResultView | TerminalResultView | DiffResultView | SearchResultView | ReadResultView | WebResultView
 
 /**
  * The default completed card: an optional replacement title and reformatted
@@ -189,6 +189,83 @@ export interface DiffResultView {
   diffs: FileDiff[]
 }
 
+/** One matched line inside a {@link SearchFileMatches} group: its 1-based line number and text. */
+export interface SearchLineMatch {
+  /** 1-based line number of the match within its file. */
+  lineNumber: number
+  /** The matched line text, as the tool surfaced it (the per-line preview budget already applied). */
+  line: string
+}
+
+/** One file's grouped content matches for a {@link SearchMatchesResultView}, in first-seen file order. */
+export interface SearchFileMatches {
+  /** The file the matches belong to (the model-facing display path). */
+  path: string
+  /** The file's matched lines, in output order. */
+  matches: SearchLineMatch[]
+}
+
+/**
+ * A completed content search (`grep`) rendered as a search card whose matches are
+ * grouped by file, so a capable UI can list each file as an expandable group of
+ * its matched lines. `shape: 'matches'` discriminates this variant from the path
+ * variant ({@link SearchPathsResultView}) within {@link SearchResultView}. The
+ * discriminant is `shape`, not `kind`, so it never collides with the
+ * {@link ToolCallKind} `kind` an icon-picking bridge reads off a call view.
+ */
+export interface SearchMatchesResultView {
+  card: 'search'
+  shape: 'matches'
+  /** Replacement title for the completed call. Omit to keep the pending-state title. */
+  title?: string
+  /** Matched lines grouped by file, in first-seen file order. */
+  files: SearchFileMatches[]
+  /**
+   * Whether the tool capped the inline result: `files` carries only the retained
+   * matches, not every match the search found. A UI shows a capped indicator so it
+   * never presents a partial group as complete.
+   */
+  truncated: boolean
+  /** Total matches the search found before capping (equals the retained count when not `truncated`). */
+  total: number
+}
+
+/**
+ * A completed path search (`glob`) rendered as a search card whose result is a flat
+ * path list. `shape: 'paths'` discriminates this variant from the grouped-matches
+ * variant ({@link SearchMatchesResultView}) within {@link SearchResultView}.
+ */
+export interface SearchPathsResultView {
+  card: 'search'
+  shape: 'paths'
+  /** Replacement title for the completed call. Omit to keep the pending-state title. */
+  title?: string
+  /** The discovered paths, in the tool's result order (the retained page when `truncated`). */
+  paths: string[]
+  /**
+   * Whether the tool capped the inline result: `paths` carries only the retained
+   * page, not every path the search found. A UI shows a capped indicator so it
+   * never presents a partial list as complete.
+   */
+  truncated: boolean
+  /** Total paths the search found before capping (equals `paths.length` when not `truncated`). */
+  total: number
+}
+
+/**
+ * A completed search rendered as a search card, the result-time view a discovery
+ * tool (`grep`, `glob`) returns from `presentResult`. One `card: 'search'` view
+ * with two `shape`-discriminated variants: grouped-by-file content matches
+ * ({@link SearchMatchesResultView}) and a flat path list
+ * ({@link SearchPathsResultView}). Both carry a `truncated`/`total` signal so a UI
+ * never presents a capped result as complete. The view carries no result text: a
+ * UI without a search card falls back to the raw `tool/result` content. There is
+ * no call-time analogue: a search call stays a {@link GenericCallView}
+ * (`kind: 'search'`) because the pending state has no matches or paths to show —
+ * the structured shape exists only after `execute`.
+ */
+export type SearchResultView = SearchMatchesResultView | SearchPathsResultView
+
 /**
  * A completed file read rendered as a line-numbered, optionally syntax-highlighted
  * code view by a capable UI. Set by a tool whose call reads file text (e.g.

packages/fs/fs-sandbox/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/fs/fs-sandbox/README.md
-README.md: 790444a4184b9bcccd3a0798cf0c09cb6f1b166e
-README.zh.md: d54bdcbe65673b6892ebd1d539dd066f66d68cb6
+README.md: d6070f4971e7531e929e659b6b5ed476a672dc5d
+README.zh.md: c051b240163f229f7fbc583939c9f4da5095e9aa

packages/fs/fs-sandbox/README.md

@@ -22,11 +22,19 @@ A denial is a structured `FsError` (`FS_SANDBOX_DENIED`, carrying the effective
 
 ## Model Experience
 
-Indirectly, through `dsh-tool-fs`, which renders this backend's `FS_SANDBOX_DENIED` refusals as the `[sandbox: file access denied under <mode> mode]` marker plus the same-turn escalation hint.
+### Filesystem policy and refusals
+
+#### What the model sees
+
+The policy owner contributes capability-neutral `sandbox:policy` context. Indirectly, `dsh-tool-fs` renders this backend's `FS_SANDBOX_DENIED` refusals as the `[sandbox: file access denied under <mode> mode]` marker plus the same-turn escalation hint.
+
+#### Token effect
+
+The current-policy clause adds a small runtime-context message while this backend is mounted; a denial adds the bounded marker and escalation hint to conversation history.
 
 #### KV Cache effect
 
-No direct invalidation; the named consumer owns any request-prefix changes.
+A standing-policy change appends an owner-rendered superseding runtime-context snapshot after retained history; operation results remain append-only.
 
 ## Known Limitations and Deferred Work
 

packages/fs/fs-sandbox/README.zh.md

@@ -22,11 +22,19 @@
 
 ## 模型体验
 
-通过 `dsh-tool-fs` 间接产生影响；该消费方把本后端的 `FS_SANDBOX_DENIED` 拒绝渲染为 `[sandbox: file access denied under <mode> mode]` 标记和同轮次升级提示。
+### 文件系统策略与拒绝
+
+#### 模型看到的内容
+
+策略归属方会贡献与具体能力无关的 `sandbox:policy` 上下文。作为间接影响，`dsh-tool-fs` 会把本后端的 `FS_SANDBOX_DENIED` 拒绝渲染为 `[sandbox: file access denied under <mode> mode]` 标记和同轮次升级提示。
+
+#### Token 影响
+
+该后端挂载期间，当前策略条款会增加一条简短的运行时上下文消息；拒绝则会把有界标记和升级提示追加到对话历史。
 
 #### KV Cache 影响
 
-不会直接使缓存失效；上述消费方负责请求前缀的任何变化。
+常驻策略发生变化时，会在保留的历史之后追加一份由归属方渲染、取代先前状态的运行时上下文快照；操作结果保持仅追加。
 
 ## 已知限制与暂缓事项
 

packages/fs/fs-sandbox/src/index.ts

@@ -44,7 +44,7 @@ import { isPathUnder } from './containment.ts'
  * Plugin config: the local backend's knobs, verbatim (only `cwd`, the resolve
  * base for relative paths). The sandbox default (mode + `workspace-write`
  * fallback root) is NOT here — `ctx.sandboxPolicy` resolves each calling
- * session for both enforcing families.
+ * session for every enforcing capability.
  */
 export type Config = LocalConfig
 

packages/fs/tool-fs-search/src/glob.ts

@@ -11,11 +11,12 @@
 import type { Context } from 'cordis'
 import { sep } from 'node:path'
 import { defineTool } from '@deepseek-ai/dsh-tools'
-import type { GenericCallView } from '@deepseek-ai/dsh-tools'
+import type { GenericCallView, SearchResultView, ToolResult } from '@deepseek-ai/dsh-tools'
 import type { SpillRef } from '@deepseek-ai/dsh-spill'
 import type {} from '@deepseek-ai/dsh-bash'
 import type {} from '@deepseek-ai/dsh-system-prompt'
 import { runRipgrep, toWorkdirRelative, trySaveFormattedResult } from './search-core.ts'
+import { globSearchMeta, searchViewFromMeta } from './presentation.ts'
 import { singleQuote } from './shell-quote.ts'
 import { acceptedSurfaceValue } from './surface.ts'
 
@@ -43,6 +44,8 @@ export interface GlobToolCaps {
   sampleOverCapGlobResults: boolean
   /** Max paths retained inline; later paths go to the formatted spill file. */
   maxResults: number
+  /** Max bytes of serialized `presentationMeta`; trailing paths drop past it. */
+  maxMetaBytes: number
   /** Cap on the complete raw `rg` stdout the tool will parse. */
   rawOutputMaxBytes: number
   /** Cooperative tool-call budget (ms) attached as `ToolDefinition.timeoutMs`. */
@@ -231,6 +234,24 @@ function renderGlobPaths(paths: string[], caps: GlobToolCaps, root: string, spil
   return formatGlobOutput(sampleAcrossTopLevel(paths, caps.maxResults, root), paths.length, spillRef)
 }
 
+/**
+ * The inline page of paths a completed `glob` card shows, computed the SAME way
+ * {@link renderGlobPaths} computes its model-facing page so the card and the text
+ * agree on which paths survived the cap. A result within the cap is shown whole;
+ * an over-cap result is either the modification-time head or the top-level sample,
+ * matching the deployment's `sampleOverCapGlobResults`.
+ *
+ * @param paths - the complete discovered path list, in modification-time order.
+ * @param caps - the resolved glob caps (the inline cap and the sampling switch).
+ * @param root - the search root in the same display-path space as `paths`.
+ * @returns the inline page and whether the complete result was capped.
+ */
+function globCardPage(paths: string[], caps: GlobToolCaps, root: string): { items: string[]; truncated: boolean } {
+  if (paths.length <= caps.maxResults) return { items: paths, truncated: false }
+  if (!caps.sampleOverCapGlobResults) return { items: paths.slice(0, caps.maxResults), truncated: true }
+  return { items: sampleAcrossTopLevel(paths, caps.maxResults, root).items, truncated: true }
+}
+
 /**
  * Pending-call presentation: a search card titled by the pattern (and root).
  *
@@ -242,6 +263,24 @@ export function presentGlobCall(args: { pattern: string; path?: string }): Gener
   return { card: 'generic', title: `Glob ${args.pattern}${where}`, kind: 'search', rawInput: args.pattern }
 }
 
+/**
+ * Completed-call presentation: the search card projected from the result's
+ * `presentationMeta` (the discovered path list, with the truncation signal). A UI
+ * without a search card falls back to the raw `tool/result` content, so the view
+ * carries no result text of its own. Malformed or absent metadata (an obsolete or
+ * hand-edited replayed log) falls back to the generic card.
+ *
+ * @param _args - the raw tool arguments; unused, the view derives from the result.
+ * @param result - the final model-facing tool result carrying the projected metadata.
+ * @returns the search card view, or `undefined` for the generic fallback.
+ */
+export function presentGlobResult(_args: { pattern: string; path?: string }, result: ToolResult): SearchResultView | undefined {
+  if (result.isError) return undefined
+  const view = searchViewFromMeta(result.meta)
+  if (view === undefined || view.shape !== 'paths') return undefined
+  return view
+}
+
 /**
  * Register the `glob` tool and its system-prompt guidance.
  *
@@ -289,6 +328,10 @@ export function applyGlobTool(ctx: Context, caps: GlobToolCaps): void {
         },
       },
       render: (_args, value) => [{ type: 'text', text: renderGlobPaths(value.paths, caps, value.root) }],
+      presentationMeta: (_args, value) => {
+        const page = globCardPage(value.paths, caps, value.root)
+        return globSearchMeta({ items: page.items, truncated: page.truncated, seen: value.paths.length }, caps.maxMetaBytes)
+      },
     },
     async execute(args, exec) {
       const input = parseGlobArgs(args)
@@ -305,6 +348,7 @@ export function applyGlobTool(ctx: Context, caps: GlobToolCaps): void {
       return { root, paths: all }
     },
     presentCall: presentGlobCall,
+    presentResult: presentGlobResult,
   })
   ctx.tools.register(tool)
 

packages/fs/tool-fs-search/src/grep.ts

@@ -12,13 +12,14 @@
 
 import type { Context } from 'cordis'
 import { defineTool } from '@deepseek-ai/dsh-tools'
-import type { GenericCallView } from '@deepseek-ai/dsh-tools'
-import { ItemRetainer, TextRetainer } from '@deepseek-ai/dsh-retention'
+import type { GenericCallView, SearchResultView, ToolResult } from '@deepseek-ai/dsh-tools'
 import type { RetainedItems } from '@deepseek-ai/dsh-retention'
 import type { SpillRef } from '@deepseek-ai/dsh-spill'
 import type {} from '@deepseek-ai/dsh-bash'
 import type {} from '@deepseek-ai/dsh-system-prompt'
-import { SearchError, runRipgrep, toWorkdirRelative, trySaveFormattedResult } from './search-core.ts'
+import type { GrepMatch } from './search-core.ts'
+import { SearchError, previewLine, retainGrepMatches, runRipgrep, toWorkdirRelative, trySaveFormattedResult } from './search-core.ts'
+import { grepSearchMeta, searchViewFromMeta } from './presentation.ts'
 import { singleQuote } from './shell-quote.ts'
 import { acceptedSurfaceValue } from './surface.ts'
 
@@ -41,6 +42,8 @@ export interface GrepToolCaps {
   maxMatches: number
   /** Max bytes retained per matched-line preview. */
   maxLineBytes: number
+  /** Max bytes of serialized `presentationMeta`; trailing file groups drop past it. */
+  maxMetaBytes: number
   /** Cap on the complete raw `rg` stdout the tool will parse. */
   rawOutputMaxBytes: number
   /** Cooperative tool-call budget (ms) attached as `ToolDefinition.timeoutMs`. */
@@ -54,13 +57,6 @@ export interface GrepInput {
   include?: string
 }
 
-/** One parsed match: the file, the 1-based line number, and the (possibly previewed) line text. */
-export interface GrepMatch {
-  path: string
-  lineNumber: number
-  line: string
-}
-
 /**
  * Reject an `include` that is not ONE positive glob filter: blank strings,
  * negated patterns (`!…`), and comma-separated lists. A comma inside a brace
@@ -177,22 +173,6 @@ export function parseGrepMatches(stdout: string): GrepMatch[] {
   return matches
 }
 
-/**
- * Bound one matched-line preview to `maxBytes` (UTF-8 boundary preserved) and
- * mark the cut. The cap is a per-line budget fact; the complete line stays in
- * the searched file for `read`.
- *
- * @param line - the matched line text (trailing newline already stripped).
- * @param maxBytes - the preview budget in bytes.
- * @returns the preview, suffixed with ` (line truncated)` when bytes were cut.
- */
-export function previewLine(line: string, maxBytes: number): string {
-  const retainer = new TextRetainer({ kind: 'head', maxBytes })
-  retainer.push(line)
-  const kept = retainer.finish()
-  return kept.truncated ? `${kept.text} (line truncated)` : kept.text
-}
-
 /** `match` / `matches` for a count. */
 function matchNoun(count: number): string {
   return count === 1 ? 'match' : 'matches'
@@ -241,18 +221,10 @@ export function formatGrepOutput(retained: RetainedItems<GrepMatch>, spillRef: S
   return `${header}\n\n${body}\n\n(${recovery})`
 }
 
-/** Apply the Native per-line preview budget without changing the canonical matches. */
-function previewGrepMatches(matches: GrepMatch[], maxLineBytes: number): GrepMatch[] {
-  return matches.map(match => ({ ...match, line: previewLine(match.line, maxLineBytes) }))
-}
-
-/** Retain and format one canonical match list for the Native surface. */
-function renderGrepMatches(matches: GrepMatch[], maxMatches: number, maxLineBytes: number, spillRef?: SpillRef): string {
-  if (matches.length === 0) return 'No matches found'
-  const previewed = previewGrepMatches(matches, maxLineBytes)
-  const retainer = new ItemRetainer<GrepMatch>({ kind: 'head', maxItems: maxMatches })
-  for (const match of previewed) retainer.push(match)
-  return formatGrepOutput(retainer.finish(), spillRef)
+/** Format one already-retained match list for the Native surface. */
+function formatRetainedGrep(retained: RetainedItems<GrepMatch>, spillRef?: SpillRef): string {
+  if (retained.seen === 0) return 'No matches found'
+  return formatGrepOutput(retained, spillRef)
 }
 
 /**
@@ -268,6 +240,27 @@ export function presentGrepCall(args: { pattern: string; path?: string; include?
   return { card: 'generic', title: `Grep ${args.pattern}${where}${filter}`, kind: 'search', rawInput: args.pattern }
 }
 
+/**
+ * Completed-call presentation: the search card projected from the result's
+ * `presentationMeta` (matches grouped by file, with the truncation signal). A UI
+ * without a search card falls back to the raw `tool/result` content, so the view
+ * carries no result text of its own. Malformed or absent metadata (an obsolete or
+ * hand-edited replayed log) falls back to the generic card.
+ *
+ * @param _args - the raw tool arguments; unused, the view derives from the result.
+ * @param result - the final model-facing tool result carrying the projected metadata.
+ * @returns the search card view, or `undefined` for the generic fallback.
+ */
+export function presentGrepResult(
+  _args: { pattern: string; path?: string; include?: string },
+  result: ToolResult,
+): SearchResultView | undefined {
+  if (result.isError) return undefined
+  const view = searchViewFromMeta(result.meta)
+  if (view === undefined || view.shape !== 'matches') return undefined
+  return view
+}
+
 /**
  * Register the `grep` tool and its system-prompt guidance.
  *
@@ -315,8 +308,10 @@ export function applyGrepTool(ctx: Context, caps: GrepToolCaps): void {
       },
       render: (_args, value) => [{
         type: 'text',
-        text: renderGrepMatches(value.matches, caps.maxMatches, caps.maxLineBytes),
+        text: formatRetainedGrep(retainGrepMatches(value.matches, caps.maxMatches, caps.maxLineBytes)),
       }],
+      presentationMeta: (_args, value) =>
+        grepSearchMeta(retainGrepMatches(value.matches, caps.maxMatches, caps.maxLineBytes), caps.maxMetaBytes),
     },
     async execute(args, exec) {
       const input = parseGrepArgs(args)
@@ -335,6 +330,7 @@ export function applyGrepTool(ctx: Context, caps: GrepToolCaps): void {
       return { matches: all }
     },
     presentCall: presentGrepCall,
+    presentResult: presentGrepResult,
   })
   ctx.tools.register(tool)
 
@@ -344,17 +340,20 @@ export function applyGrepTool(ctx: Context, caps: GrepToolCaps): void {
     if (value === undefined) return decision
     const matches = value.matches
     if (matches.length <= caps.maxMatches) return decision
+    // The spill artifact holds the COMPLETE result: preview each line, but keep
+    // every match (no inline cap), so the recovery file is the full search.
+    const previewedAll = matches.map(match => ({ ...match, line: previewLine(match.line, caps.maxLineBytes) }))
     const spillRef = await trySaveFormattedResult(
       ctx,
       exec,
       'grep-results.txt',
-      `Found ${matches.length} ${matchNoun(matches.length)}\n\n${formatGrepMatches(previewGrepMatches(matches, caps.maxLineBytes))}`,
+      `Found ${matches.length} ${matchNoun(matches.length)}\n\n${formatGrepMatches(previewedAll)}`,
     )
     return {
       kind: 'accept',
       content: [{
         type: 'text',
-        text: renderGrepMatches(matches, caps.maxMatches, caps.maxLineBytes, spillRef),
+        text: formatRetainedGrep(retainGrepMatches(matches, caps.maxMatches, caps.maxLineBytes), spillRef),
       }],
       ...decision.additionalContexts !== undefined ? { additionalContexts: decision.additionalContexts } : {},
     }

packages/fs/tool-fs-search/src/index.ts

@@ -31,9 +31,9 @@ import type { Context } from 'cordis'
 import z from 'schemastery'
 import { GLOB_MAX_RESULTS, applyGlobTool } from './glob.ts'
 import { GREP_MAX_LINE_BYTES, GREP_MAX_MATCHES, applyGrepTool } from './grep.ts'
-import { RAW_OUTPUT_MAX_BYTES, SEARCH_TIMEOUT_MS } from './search-core.ts'
+import { RAW_OUTPUT_MAX_BYTES, SEARCH_META_MAX_BYTES, SEARCH_TIMEOUT_MS } from './search-core.ts'
 
-export { GLOB_MAX_RESULTS, GLOB_VCS_EXCLUDES, applyGlobTool, buildGlobCommand, formatGlobOutput, parseGlobArgs, presentGlobCall, sampleAcrossTopLevel } from './glob.ts'
+export { GLOB_MAX_RESULTS, GLOB_VCS_EXCLUDES, applyGlobTool, buildGlobCommand, formatGlobOutput, parseGlobArgs, presentGlobCall, presentGlobResult, sampleAcrossTopLevel } from './glob.ts'
 export type { GlobInput, GlobSample, GlobToolCaps } from './glob.ts'
 export {
   GREP_MAX_LINE_BYTES,
@@ -45,11 +45,20 @@ export {
   parseGrepArgs,
   parseGrepMatches,
   presentGrepCall,
-  previewLine,
+  presentGrepResult,
 } from './grep.ts'
-export type { GrepInput, GrepMatch, GrepToolCaps } from './grep.ts'
-export { RAW_OUTPUT_MAX_BYTES, SEARCH_TIMEOUT_MS, SearchError, runRipgrep, toWorkdirRelative, trySaveFormattedResult } from './search-core.ts'
-export type { RipgrepRun, SearchErrorCode } from './search-core.ts'
+export type { GrepInput, GrepToolCaps } from './grep.ts'
+export {
+  RAW_OUTPUT_MAX_BYTES,
+  SEARCH_META_MAX_BYTES,
+  SEARCH_TIMEOUT_MS,
+  SearchError,
+  previewLine,
+  runRipgrep,
+  toWorkdirRelative,
+  trySaveFormattedResult,
+} from './search-core.ts'
+export type { GrepMatch, RipgrepRun, SearchErrorCode } from './search-core.ts'
 export { singleQuote } from './shell-quote.ts'
 
 /** Cordis plugin name used by loader diagnostics. */
@@ -68,6 +77,8 @@ export interface Config {
   grepMaxMatches?: number
   /** Max bytes retained for one matched-line preview (the cut preserves UTF-8 boundaries). */
   grepMaxLineBytes?: number
+  /** Max bytes of one search's serialized `presentationMeta`; trailing groups/paths drop past it so the persisted card stays bounded. */
+  searchMetaMaxBytes?: number
   /** Max complete raw `rg` stdout bytes a search will parse; larger raw output fails with `SEARCH_RAW_OUTPUT_OVERFLOW`. */
   rawOutputMaxBytes?: number
   /** Cooperative tool-call timeout budget (ms) on both tools, enforced by `@deepseek-ai/dsh-timeout-policy` through `exec.signal`. */
@@ -79,6 +90,7 @@ export const Config: z<Config> = z.object({
   globMaxResults: z.number().default(GLOB_MAX_RESULTS),
   grepMaxMatches: z.number().default(GREP_MAX_MATCHES),
   grepMaxLineBytes: z.number().default(GREP_MAX_LINE_BYTES),
+  searchMetaMaxBytes: z.number().default(SEARCH_META_MAX_BYTES),
   rawOutputMaxBytes: z.number().default(RAW_OUTPUT_MAX_BYTES),
   timeoutMs: z.number().default(SEARCH_TIMEOUT_MS),
 })
@@ -133,6 +145,7 @@ export async function apply(ctx: Context, config: Config): Promise<void> {
   assertPositiveInteger('globMaxResults', resolved.globMaxResults)
   assertPositiveInteger('grepMaxMatches', resolved.grepMaxMatches)
   assertPositiveInteger('grepMaxLineBytes', resolved.grepMaxLineBytes)
+  assertPositiveInteger('searchMetaMaxBytes', resolved.searchMetaMaxBytes)
   assertPositiveInteger('rawOutputMaxBytes', resolved.rawOutputMaxBytes)
   assertPositiveInteger('timeoutMs', resolved.timeoutMs)
   if (!await ripgrepAvailable(ctx)) {
@@ -142,12 +155,14 @@ export async function apply(ctx: Context, config: Config): Promise<void> {
   applyGlobTool(ctx, {
     sampleOverCapGlobResults: resolved.sampleOverCapGlobResults,
     maxResults: resolved.globMaxResults,
+    maxMetaBytes: resolved.searchMetaMaxBytes,
     rawOutputMaxBytes: resolved.rawOutputMaxBytes,
     timeoutMs: resolved.timeoutMs,
   })
   applyGrepTool(ctx, {
     maxMatches: resolved.grepMaxMatches,
     maxLineBytes: resolved.grepMaxLineBytes,
+    maxMetaBytes: resolved.searchMetaMaxBytes,
     rawOutputMaxBytes: resolved.rawOutputMaxBytes,
     timeoutMs: resolved.timeoutMs,
   })

packages/fs/tool-fs-search/src/presentation.ts

@@ -0,0 +1,205 @@
+/**
+ * Result-time search-card presentation for `grep` and `glob`. Both tools land on
+ * one `card: 'search'` render intent ({@link SearchResultView}) with two
+ * `shape`-discriminated variants: `grep` projects its matches grouped by file
+ * ({@link SearchMatchesResultView}), `glob` projects a flat path list
+ * ({@link SearchPathsResultView}). This module owns the value→`presentationMeta`
+ * projection each tool declares and the defensive `meta`→view narrowing each
+ * tool's `presentResult` reads back on replay.
+ *
+ * The canonical value never crosses the wire — only the model-facing render text
+ * and this JSON `meta` do — so the structured shape a UI renders MUST ride in
+ * `meta`. Each projection consumes the SAME retained matches/paths the
+ * model-facing render consumes ({@link module:@deepseek-ai/dsh-tool-fs-search/search-core}
+ * `retainGrepMatches`/`retainGlobPaths`), so text and card agree about which
+ * results survived the inline cap, and reports `total` (every result found) and
+ * `truncated`, so a UI never presents a capped result as complete.
+ *
+ * A second, independent cap bounds the JSON `meta` itself: the retained matches
+ * of a broad search (hundreds of long lines) can still serialize to hundreds of
+ * kilobytes, and `meta` is persisted with the session log and re-sent on every
+ * request. {@link capMetaBytes} drops trailing groups/paths until the serialized
+ * `meta` fits `maxMetaBytes` and marks the result `truncated`; a deployment's
+ * final output budget (`dsh-spill-policy`) only shrinks `content`, never `meta`,
+ * so this projection owns keeping `meta` bounded.
+ *
+ * @module @deepseek-ai/dsh-tool-fs-search/presentation
+ */
+
+import type {
+  SearchFileMatches,
+  SearchLineMatch,
+  SearchResultView,
+} from '@deepseek-ai/dsh-tools'
+import type { RetainedItems } from '@deepseek-ai/dsh-retention'
+import type { GrepMatch } from './search-core.ts'
+
+/**
+ * The retention fields a meta projection reads: the retained page, whether the
+ * complete result was capped, and the pre-cap total. Both a full
+ * {@link RetainedItems} (from `retainGrepMatches`) and `glob`'s sampled page
+ * satisfy this structural subset, so a projection consumes either without a fake
+ * `kept`/`omitted`.
+ */
+type RetainedPage<T> = Pick<RetainedItems<T>, 'items' | 'truncated' | 'seen'>
+
+/**
+ * The `grep`/`glob` tools' private `tool/result` `meta` payload: the capped,
+ * structured search result. Attached opaquely (as `JsonValue`) on the tool result
+ * and persisted with the session log, so `presentResult` reproduces the search
+ * card on replay. The `matches` shape carries the by-file groups; the `paths`
+ * shape carries the flat list. Both carry the pre-cap `total` and the `truncated`
+ * flag. The producing tool owns and narrows this opaque shape.
+ *
+ * The member shapes use object-literal `type` aliases rather than the
+ * {@link SearchFileMatches}/{@link SearchLineMatch} interfaces because only a type
+ * alias is assignable to the `JsonValue` index signature `presentationMeta`
+ * returns; the two are structurally identical, so the projected value still reads
+ * back as a {@link SearchResultView}.
+ */
+export type SearchMeta =
+  | { shape: 'matches'; files: MetaFileMatches[]; truncated: boolean; total: number }
+  | { shape: 'paths'; paths: string[]; truncated: boolean; total: number }
+
+/** One matched line in {@link SearchMeta} (the JSON-assignable form of {@link SearchLineMatch}). */
+type MetaLineMatch = { lineNumber: number; line: string }
+
+/** One file's grouped matches in {@link SearchMeta} (the JSON-assignable form of {@link SearchFileMatches}). */
+type MetaFileMatches = { path: string; matches: MetaLineMatch[] }
+
+/**
+ * Group flat matches by file (first-seen order) into the structured by-file shape
+ * a UI renders as expandable per-file groups. The grouping matches the
+ * model-facing text grouping
+ * ({@link module:@deepseek-ai/dsh-tool-fs-search/grep} `formatGrepMatches`), so
+ * card and text agree about file order and membership.
+ *
+ * @param matches - the retained matches to group, in output order.
+ * @returns one entry per file, in first-seen order.
+ */
+export function groupMatchesByFile(matches: GrepMatch[]): MetaFileMatches[] {
+  const byFile = new Map<string, MetaLineMatch[]>()
+  for (const match of matches) {
+    const entry: MetaLineMatch = { lineNumber: match.lineNumber, line: match.line }
+    const group = byFile.get(match.path)
+    if (group !== undefined) group.push(entry)
+    else byFile.set(match.path, [entry])
+  }
+  return Array.from(byFile, ([path, fileMatches]) => ({ path, matches: fileMatches }))
+}
+
+/** The serialized UTF-8 byte size of one meta payload (the size persisted and re-sent). */
+function metaBytes(meta: SearchMeta): number {
+  return Buffer.byteLength(JSON.stringify(meta), 'utf8')
+}
+
+/**
+ * Drop trailing top-level items (file groups or paths) until the serialized meta
+ * fits `maxMetaBytes`, marking the result `truncated` when anything was dropped.
+ * `total` is preserved (it counts what the search found, not what meta retains).
+ * A single item too large to fit on its own is kept: the invariant is a bounded
+ * payload wherever droppable, never an empty card that hides a real result.
+ *
+ * @param meta - the projected meta, already capped to the inline item count.
+ * @param maxMetaBytes - the serialized-meta byte budget.
+ * @returns the same meta when it fits, else a byte-bounded copy marked `truncated`.
+ */
+function capMetaBytes(meta: SearchMeta, maxMetaBytes: number): SearchMeta {
+  if (metaBytes(meta) <= maxMetaBytes) return meta
+  if (meta.shape === 'matches') {
+    const files = [...meta.files]
+    while (files.length > 1 && metaBytes({ ...meta, files, truncated: true }) > maxMetaBytes) files.pop()
+    return { ...meta, files, truncated: true }
+  }
+  const paths = [...meta.paths]
+  while (paths.length > 1 && metaBytes({ ...meta, paths, truncated: true }) > maxMetaBytes) paths.pop()
+  return { ...meta, paths, truncated: true }
+}
+
+/**
+ * Project the retained `grep` matches into {@link SearchMeta} for the search
+ * card. Consumes the same {@link RetainedItems} the model-facing render consumes
+ * (preview budget and inline match cap already applied), groups the retained
+ * matches by file, reports `total` (every parsed match) and `truncated`, then
+ * bounds the serialized meta to `maxMetaBytes`.
+ *
+ * @param retained - the retention outcome over every parsed match (previewed, capped).
+ * @param maxMetaBytes - the serialized-meta byte budget.
+ * @returns the `matches`-shaped search metadata.
+ */
+export function grepSearchMeta(retained: RetainedPage<GrepMatch>, maxMetaBytes: number): SearchMeta {
+  const meta: SearchMeta = {
+    shape: 'matches',
+    files: groupMatchesByFile(retained.items),
+    truncated: retained.truncated,
+    total: retained.seen,
+  }
+  return capMetaBytes(meta, maxMetaBytes)
+}
+
+/**
+ * Project the retained `glob` paths into {@link SearchMeta} for the search card.
+ * Consumes the same {@link RetainedItems} the model-facing render consumes (inline
+ * path cap already applied), reports `total` (every discovered path) and
+ * `truncated`, then bounds the serialized meta to `maxMetaBytes`.
+ *
+ * @param retained - the retention outcome over every discovered path (capped).
+ * @param maxMetaBytes - the serialized-meta byte budget.
+ * @returns the `paths`-shaped search metadata.
+ */
+export function globSearchMeta(retained: RetainedPage<string>, maxMetaBytes: number): SearchMeta {
+  const meta: SearchMeta = {
+    shape: 'paths',
+    paths: retained.items,
+    truncated: retained.truncated,
+    total: retained.seen,
+  }
+  return capMetaBytes(meta, maxMetaBytes)
+}
+
+/** Whether `value` is a valid {@link SearchLineMatch} (defensive narrowing from opaque `meta`). */
+function isSearchLineMatch(value: unknown): value is SearchLineMatch {
+  if (typeof value !== 'object' || value === null || Array.isArray(value)) return false
+  const { lineNumber, line } = value as Record<string, unknown>
+  return typeof lineNumber === 'number' && typeof line === 'string'
+}
+
+/** Whether `value` is a valid {@link SearchFileMatches} (defensive narrowing from opaque `meta`). */
+function isSearchFileMatches(value: unknown): value is SearchFileMatches {
+  if (typeof value !== 'object' || value === null || Array.isArray(value)) return false
+  const { path, matches } = value as Record<string, unknown>
+  return typeof path === 'string' && Array.isArray(matches) && matches.every(isSearchLineMatch)
+}
+
+/**
+ * Narrow opaque live or replayed result metadata to a {@link SearchResultView}.
+ * Malformed metadata returns `undefined` so `presentResult` can fall back to the
+ * generic card instead of throwing during replay of an older or hand-edited log.
+ * The view carries no result text: a UI without a search card falls back to the
+ * raw `tool/result` content.
+ *
+ * A zero-result meta (`files: []` / `paths: []`) narrows to a valid empty card —
+ * unlike the mirrored `diffsFromMeta`, which rejects empty diffs, because a
+ * zero-match grep is a legitimate result a UI shows as "no matches", not an
+ * absent projection.
+ *
+ * @param meta - result metadata (the {@link SearchMeta} the tool projected).
+ * @returns the search view, or `undefined` for absent or malformed metadata.
+ */
+export function searchViewFromMeta(meta: unknown): SearchResultView | undefined {
+  if (typeof meta !== 'object' || meta === null || Array.isArray(meta)) return undefined
+  const record = meta as Record<string, unknown>
+  const { truncated, total } = record
+  if (typeof truncated !== 'boolean' || typeof total !== 'number') return undefined
+  if (record.shape === 'matches') {
+    const { files } = record
+    if (!Array.isArray(files) || !files.every(isSearchFileMatches)) return undefined
+    return { card: 'search', shape: 'matches', files: files, truncated, total }
+  }
+  if (record.shape === 'paths') {
+    const { paths } = record
+    if (!Array.isArray(paths) || !paths.every((path): path is string => typeof path === 'string')) return undefined
+    return { card: 'search', shape: 'paths', paths, truncated, total }
+  }
+  return undefined
+}

packages/fs/tool-fs-search/src/search-core.ts

@@ -19,6 +19,8 @@
 import { isAbsolute, relative, sep } from 'node:path'
 import type { Context } from 'cordis'
 import { HarnessError } from '@deepseek-ai/dsh-llm'
+import { ItemRetainer, TextRetainer } from '@deepseek-ai/dsh-retention'
+import type { RetainedItems } from '@deepseek-ai/dsh-retention'
 import type { BashRunResult, CollectedOutput } from '@deepseek-ai/dsh-bash'
 import type { SaveTextSpill, SpillRef } from '@deepseek-ai/dsh-spill'
 import type { ToolExecution } from '@deepseek-ai/dsh-tools'
@@ -36,6 +38,18 @@ export const RAW_OUTPUT_MAX_BYTES = 20_000_000
  */
 export const SEARCH_TIMEOUT_MS = 30_000
 
+/**
+ * Default cap in bytes on one search's serialized `presentationMeta` (the
+ * `searchMetaMaxBytes` config). The inline match/path caps already bound the item
+ * COUNT, but retained matches of a broad search (many long lines) can still
+ * serialize to hundreds of kilobytes, and `meta` is persisted with the session
+ * log and re-sent on every request. A deployment's final output budget
+ * (`dsh-spill-policy`) only shrinks a result's `content`, never its `meta`, so the
+ * projection owns this cap. 64 KiB holds the full default-capped result of a
+ * typical search while bounding the pathological one.
+ */
+export const SEARCH_META_MAX_BYTES = 65_536
+
 /**
  * Stable, machine-routable codes for search failures. Package-owned (not
  * `FsErrorCode`) because these tools are bash-backed discovery, not `ctx.fs`
@@ -212,6 +226,63 @@ export function toWorkdirRelative(path: string, workdir: string): string {
   return rel
 }
 
+/** One parsed match: the file, the 1-based line number, and the (possibly previewed) line text. */
+export interface GrepMatch {
+  path: string
+  lineNumber: number
+  line: string
+}
+
+/**
+ * Bound one matched-line preview to `maxBytes` (UTF-8 boundary preserved) and
+ * mark the cut. The cap is a per-line budget fact; the complete line stays in
+ * the searched file for `read`.
+ *
+ * @param line - the matched line text (trailing newline already stripped).
+ * @param maxBytes - the preview budget in bytes.
+ * @returns the preview, suffixed with ` (line truncated)` when bytes were cut.
+ */
+export function previewLine(line: string, maxBytes: number): string {
+  const retainer = new TextRetainer({ kind: 'head', maxBytes })
+  retainer.push(line)
+  const kept = retainer.finish()
+  return kept.truncated ? `${kept.text} (line truncated)` : kept.text
+}
+
+/**
+ * Apply the shared inline cap to a canonical `grep` match list: preview each
+ * retained line to `maxLineBytes` and keep the first `maxMatches`. The single
+ * retention pass both the model-facing render ({@link module:@deepseek-ai/dsh-tool-fs-search/grep}
+ * `formatGrepOutput`) and the search-card projection
+ * ({@link module:@deepseek-ai/dsh-tool-fs-search/presentation} `grepSearchMeta`)
+ * consume, so text and card never disagree about which matches survived.
+ *
+ * @param matches - every match the search parsed (the canonical value's matches).
+ * @param maxMatches - the inline match cap (the `grepMaxMatches` config).
+ * @param maxLineBytes - the per-matched-line preview budget in bytes.
+ * @returns the retention outcome over the previewed matches.
+ */
+export function retainGrepMatches(matches: GrepMatch[], maxMatches: number, maxLineBytes: number): RetainedItems<GrepMatch> {
+  const retainer = new ItemRetainer<GrepMatch>({ kind: 'head', maxItems: maxMatches })
+  for (const match of matches) retainer.push({ ...match, line: previewLine(match.line, maxLineBytes) })
+  return retainer.finish()
+}
+
+/**
+ * Apply the shared inline cap to a canonical `glob` path list: keep the first
+ * `maxResults`. The single retention pass both the model-facing render and the
+ * search-card projection consume.
+ *
+ * @param paths - every path the search discovered (the canonical value's paths).
+ * @param maxResults - the inline path cap (the `globMaxResults` config).
+ * @returns the retention outcome over the paths.
+ */
+export function retainGlobPaths(paths: string[], maxResults: number): RetainedItems<string> {
+  const retainer = new ItemRetainer<string>({ kind: 'head', maxItems: maxResults })
+  for (const path of paths) retainer.push(path)
+  return retainer.finish()
+}
+
 /**
  * Best-effort save of one COMPLETE formatted search result through
  * `ctx.spillStore.saveText()` — the model-facing recovery path for a capped

packages/fs/tool-fs-search/tests/presentation.spec.ts

@@ -0,0 +1,176 @@
+/**
+ * Unit tests for the search-card presentation layer (`src/presentation.ts`): the
+ * canonical value → `presentationMeta` projections (`grepSearchMeta`,
+ * `globSearchMeta`, `groupMatchesByFile`) and the defensive `meta` → view
+ * narrowing (`searchViewFromMeta`). These pin the by-file grouping, the
+ * `truncated`/`total` honesty over already-retained input, the serialized-meta
+ * byte cap, and the malformed-metadata fallback a replayed or hand-edited log can
+ * deliver.
+ */
+
+import { describe, expect, it } from 'vitest'
+import type { JsonValue } from '@deepseek-ai/dsh-session'
+import {
+  globSearchMeta,
+  grepSearchMeta,
+  groupMatchesByFile,
+  searchViewFromMeta,
+} from '../src/presentation.ts'
+import type { GrepMatch } from '../src/search-core.ts'
+import { retainGlobPaths, retainGrepMatches } from '../src/search-core.ts'
+
+const match = (path: string, lineNumber: number, line: string): GrepMatch => ({ path, lineNumber, line })
+
+/** A byte cap large enough that no test payload here is meta-capped. */
+const WIDE = 1_000_000
+
+describe('groupMatchesByFile', () => {
+  it('groups matches by first-seen file order, keeping line/lineNumber only', () => {
+    expect(groupMatchesByFile([
+      match('b.ts', 2, 'x'),
+      match('a.ts', 1, 'y'),
+      match('b.ts', 5, 'z'),
+    ])).toEqual([
+      { path: 'b.ts', matches: [{ lineNumber: 2, line: 'x' }, { lineNumber: 5, line: 'z' }] },
+      { path: 'a.ts', matches: [{ lineNumber: 1, line: 'y' }] },
+    ])
+  })
+
+  it('returns an empty list for no matches', () => {
+    expect(groupMatchesByFile([])).toEqual([])
+  })
+})
+
+describe('grepSearchMeta', () => {
+  it('projects grouped matches with total and a false truncation flag within the cap', () => {
+    const meta = grepSearchMeta(retainGrepMatches([match('a.ts', 1, 'one'), match('a.ts', 2, 'two')], 10, 2000), WIDE)
+    expect(meta).toEqual({
+      shape: 'matches',
+      files: [{ path: 'a.ts', matches: [{ lineNumber: 1, line: 'one' }, { lineNumber: 2, line: 'two' }] }],
+      truncated: false,
+      total: 2,
+    })
+  })
+
+  it('reports the pre-cap total and truncation from the shared retention pass', () => {
+    const meta = grepSearchMeta(retainGrepMatches([match('a.ts', 1, 'one'), match('a.ts', 2, 'two'), match('b.ts', 3, 'three')], 2, 2000), WIDE)
+    expect(meta).toEqual({
+      shape: 'matches',
+      files: [{ path: 'a.ts', matches: [{ lineNumber: 1, line: 'one' }, { lineNumber: 2, line: 'two' }] }],
+      truncated: true,
+      total: 3,
+    })
+  })
+
+  it('carries the per-line preview budget (UTF-8 boundary) the retention pass applied', () => {
+    const meta = grepSearchMeta(retainGrepMatches([match('a.txt', 1, 'aéaéaéaé')], 10, 7), WIDE)
+    expect(meta).toMatchObject({ shape: 'matches', files: [{ path: 'a.txt', matches: [{ lineNumber: 1, line: 'aéaéa (line truncated)' }] }] })
+  })
+
+  it('drops trailing file groups until the serialized meta fits the byte cap, marking it truncated', () => {
+    const retained = retainGrepMatches(
+      [match('a.ts', 1, 'x'.repeat(60)), match('b.ts', 2, 'y'.repeat(60)), match('c.ts', 3, 'z'.repeat(60))],
+      10,
+      2000,
+    )
+    // One 60-byte group serializes to ~110 bytes; a 260-byte cap holds two, not three.
+    const meta = grepSearchMeta(retained, 260)
+    expect(meta.shape).toBe('matches')
+    if (meta.shape !== 'matches') throw new Error('unreachable')
+    expect(meta.truncated).toBe(true)
+    expect(meta.total).toBe(3)
+    expect(meta.files.length).toBeLessThan(3)
+    expect(Buffer.byteLength(JSON.stringify(meta), 'utf8')).toBeLessThanOrEqual(260)
+  })
+
+  it('keeps a single oversized group rather than emit an empty card', () => {
+    const meta = grepSearchMeta(retainGrepMatches([match('a.ts', 1, 'x'.repeat(500))], 10, 2000), 50)
+    expect(meta.shape).toBe('matches')
+    if (meta.shape !== 'matches') throw new Error('unreachable')
+    expect(meta.files).toHaveLength(1)
+    expect(meta.truncated).toBe(true)
+  })
+})
+
+describe('globSearchMeta', () => {
+  it('projects the path list with total and a false truncation flag within the cap', () => {
+    expect(globSearchMeta(retainGlobPaths(['a.ts', 'b.ts'], 10), WIDE)).toEqual({ shape: 'paths', paths: ['a.ts', 'b.ts'], truncated: false, total: 2 })
+  })
+
+  it('reports the pre-cap total and truncation from the shared retention pass', () => {
+    expect(globSearchMeta(retainGlobPaths(['a.ts', 'b.ts', 'c.ts'], 2), WIDE)).toEqual({ shape: 'paths', paths: ['a.ts', 'b.ts'], truncated: true, total: 3 })
+  })
+
+  it('drops trailing paths until the serialized meta fits the byte cap, marking it truncated', () => {
+    const retained = retainGlobPaths([`${'a'.repeat(100)}.ts`, `${'b'.repeat(100)}.ts`, `${'c'.repeat(100)}.ts`], 10)
+    const meta = globSearchMeta(retained, 180)
+    expect(meta.shape).toBe('paths')
+    if (meta.shape !== 'paths') throw new Error('unreachable')
+    expect(meta.truncated).toBe(true)
+    expect(meta.total).toBe(3)
+    expect(meta.paths.length).toBeLessThan(3)
+    expect(Buffer.byteLength(JSON.stringify(meta), 'utf8')).toBeLessThanOrEqual(180)
+  })
+})
+
+describe('searchViewFromMeta (defensive narrowing)', () => {
+  // The narrowing accepts an opaque JsonValue; a malformed payload is not a
+  // statically-valid JsonValue, so route every case through one cast helper that
+  // mirrors how a hand-edited/older session log delivers arbitrary shapes.
+  const m = (value: unknown): JsonValue | undefined => value as JsonValue | undefined
+
+  it('narrows a well-formed matches payload into a matches view', () => {
+    const meta = { shape: 'matches', files: [{ path: 'a.ts', matches: [{ lineNumber: 1, line: 'x' }] }], truncated: true, total: 5 }
+    expect(searchViewFromMeta(m(meta))).toEqual({ card: 'search', ...meta })
+  })
+
+  it('narrows a well-formed paths payload into a paths view', () => {
+    const meta = { shape: 'paths', paths: ['a.ts', 'b.ts'], truncated: false, total: 2 }
+    expect(searchViewFromMeta(m(meta))).toEqual({ card: 'search', ...meta })
+  })
+
+  it('narrows a zero-result payload into a valid empty card (not a rejected projection)', () => {
+    expect(searchViewFromMeta(m({ shape: 'matches', files: [], truncated: false, total: 0 })))
+      .toEqual({ card: 'search', shape: 'matches', files: [], truncated: false, total: 0 })
+    expect(searchViewFromMeta(m({ shape: 'paths', paths: [], truncated: false, total: 0 })))
+      .toEqual({ card: 'search', shape: 'paths', paths: [], truncated: false, total: 0 })
+  })
+
+  it('rejects undefined / non-object / array meta', () => {
+    expect(searchViewFromMeta(undefined)).toBeUndefined()
+    expect(searchViewFromMeta(null)).toBeUndefined()
+    expect(searchViewFromMeta(m('nope'))).toBeUndefined()
+    expect(searchViewFromMeta(m([]))).toBeUndefined()
+  })
+
+  it('rejects a payload with a missing / mistyped truncated or total field', () => {
+    expect(searchViewFromMeta(m({ shape: 'paths', paths: [], total: 0 }))).toBeUndefined()
+    expect(searchViewFromMeta(m({ shape: 'paths', paths: [], truncated: 'no', total: 0 }))).toBeUndefined()
+    expect(searchViewFromMeta(m({ shape: 'paths', paths: [], truncated: false }))).toBeUndefined()
+    expect(searchViewFromMeta(m({ shape: 'paths', paths: [], truncated: false, total: '0' }))).toBeUndefined()
+  })
+
+  it('rejects an unknown or missing shape discriminant', () => {
+    expect(searchViewFromMeta(m({ shape: 'other', truncated: false, total: 0 }))).toBeUndefined()
+    expect(searchViewFromMeta(m({ truncated: false, total: 0 }))).toBeUndefined()
+  })
+
+  it('rejects a matches payload with a malformed files array', () => {
+    const base = { shape: 'matches', truncated: false, total: 1 }
+    expect(searchViewFromMeta(m({ ...base, files: 'x' }))).toBeUndefined()
+    expect(searchViewFromMeta(m({ ...base, files: [null] }))).toBeUndefined()
+    expect(searchViewFromMeta(m({ ...base, files: ['x'] }))).toBeUndefined()
+    expect(searchViewFromMeta(m({ ...base, files: [[]] }))).toBeUndefined()
+    expect(searchViewFromMeta(m({ ...base, files: [{ path: 1, matches: [] }] }))).toBeUndefined()
+    expect(searchViewFromMeta(m({ ...base, files: [{ path: 'a', matches: 'x' }] }))).toBeUndefined()
+    expect(searchViewFromMeta(m({ ...base, files: [{ path: 'a', matches: [null] }] }))).toBeUndefined()
+    expect(searchViewFromMeta(m({ ...base, files: [{ path: 'a', matches: [{ lineNumber: '1', line: 'x' }] }] }))).toBeUndefined()
+    expect(searchViewFromMeta(m({ ...base, files: [{ path: 'a', matches: [{ lineNumber: 1, line: 2 }] }] }))).toBeUndefined()
+  })
+
+  it('rejects a paths payload with a non-array or non-string-element paths field', () => {
+    const base = { shape: 'paths', truncated: false, total: 1 }
+    expect(searchViewFromMeta(m({ ...base, paths: 'x' }))).toBeUndefined()
+    expect(searchViewFromMeta(m({ ...base, paths: [1] }))).toBeUndefined()
+  })
+})

packages/fs/tool-fs-search/tests/tools.spec.ts

@@ -27,7 +27,9 @@ import {
   formatGrepMatches,
   parseGrepMatches,
   presentGlobCall,
+  presentGlobResult,
   presentGrepCall,
+  presentGrepResult,
   previewLine,
   sampleAcrossTopLevel,
   toWorkdirRelative,
@@ -987,6 +989,73 @@ describe('presentation', () => {
     expect(presentGrepCall({ pattern: 'todo' })).toMatchObject({ card: 'generic', title: 'Grep todo', kind: 'search' })
     expect(presentGrepCall({ pattern: 'todo', path: 'src', include: '*.ts' }).title).toBe('Grep todo in src (*.ts)')
   })
+
+  it('grep projects a search card from a real execute, grouped by file with total and truncation', async () => {
+    const { ctx, bash } = await setup({ config: { grepMaxMatches: 2 } })
+    bash.handler = () => runResult([
+      matchLine('a.ts', 1, 'one'),
+      matchLine('a.ts', 2, 'two'),
+      matchLine('b.ts', 3, 'three'),
+      '',
+    ].join('\n'))
+    const result = await call(ctx, 'grep', { pattern: 'e' }, { agent: agent('/w') })
+    if (result.isError) throw new Error('expected grep success')
+    // The presentationMeta projection rides the result meta (a surface call).
+    expect(result.meta).toEqual({
+      shape: 'matches',
+      files: [{ path: 'a.ts', matches: [{ lineNumber: 1, line: 'one' }, { lineNumber: 2, line: 'two' }] }],
+      truncated: true,
+      total: 3,
+    })
+    const view = presentGrepResult({ pattern: 'e' }, result)
+    expect(view).toEqual({
+      card: 'search',
+      shape: 'matches',
+      files: [{ path: 'a.ts', matches: [{ lineNumber: 1, line: 'one' }, { lineNumber: 2, line: 'two' }] }],
+      truncated: true,
+      total: 3,
+    })
+  })
+
+  it('glob projects a search card from a real execute, a flat path list with total and truncation', async () => {
+    const { ctx, bash } = await setup({ config: { globMaxResults: 2 } })
+    bash.handler = () => runResult('a.ts\nb.ts\nc.ts\n')
+    const result = await call(ctx, 'glob', { pattern: '*.ts' }, { agent: agent('/w') })
+    if (result.isError) throw new Error('expected glob success')
+    expect(result.meta).toEqual({ shape: 'paths', paths: ['a.ts', 'b.ts'], truncated: true, total: 3 })
+    const view = presentGlobResult({ pattern: '*.ts' }, result)
+    expect(view).toEqual({ card: 'search', shape: 'paths', paths: ['a.ts', 'b.ts'], truncated: true, total: 3 })
+  })
+
+  it('nested Code dispatch computes no meta, so presentResult falls back to the generic card', async () => {
+    const { ctx, bash } = await setup()
+    bash.handler = () => runResult(`${matchLine('a.ts', 1, 'one')}\n`)
+    const result = await call(ctx, 'grep', { pattern: 'o' }, {
+      agent: agent('/w'),
+      parent: Symbol('run_code') as ToolExecutionToken,
+    })
+    if (result.isError) throw new Error('expected grep success')
+    expect(result.meta).toBeUndefined()
+    expect(presentGrepResult({ pattern: 'o' }, result)).toBeUndefined()
+  })
+
+  it('presentResult returns undefined for a failed result and for the other tool’s meta shape', () => {
+    const errorResult = { content: [{ type: 'text' as const, text: 'boom' }], isError: true }
+    expect(presentGrepResult({ pattern: 'x' }, errorResult)).toBeUndefined()
+    expect(presentGlobResult({ pattern: '*' }, errorResult)).toBeUndefined()
+    // A grep result carrying a paths-shaped meta (and vice versa) is not this
+    // tool's shape: each presenter narrows to its own shape and otherwise falls back.
+    const pathsResult = { content: [], isError: false, meta: { shape: 'paths', paths: ['a.ts'], truncated: false, total: 1 } }
+    const matchesResult = { content: [], isError: false, meta: { shape: 'matches', files: [], truncated: false, total: 0 } }
+    expect(presentGrepResult({ pattern: 'x' }, pathsResult)).toBeUndefined()
+    expect(presentGlobResult({ pattern: '*' }, matchesResult)).toBeUndefined()
+  })
+
+  it('presentResult falls back to the generic card on malformed replayed meta', () => {
+    const malformed = { content: [], isError: false, meta: { shape: 'matches', files: 'nope', truncated: false, total: 0 } }
+    expect(presentGrepResult({ pattern: 'x' }, malformed)).toBeUndefined()
+    expect(presentGlobResult({ pattern: '*' }, { content: [], isError: false, meta: 42 })).toBeUndefined()
+  })
 })
 
 describe('helpers', () => {

packages/fs/tool-fs/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/fs/tool-fs/README.md
-README.md: c00b59fed06249e6d9479c4a809cdf7d78f93239
-README.zh.md: f90fbb36391c1388ab0f6836daa2a9061d046be6
+README.md: a695d0ba8fb1d600689d2b68763e8423d1591da5
+README.zh.md: 5c600ab70b46da640637aec64efc1c0f0d0d54c0

packages/fs/tool-fs/README.md

@@ -2,7 +2,7 @@
 
 English | [中文](README.zh.md)
 
-The **model-facing filesystem tools** — `read`, `write`, `edit` — and their **executor**. This is the consumer layer of the filesystem stack: it owns tool names, JSON schemas, argument validation, prompt sections, **read windowing**, and result formatting. It reads/writes/edits through the `ctx.fs` provider seam ([`@deepseek-ai/dsh-fs`](../fs)) **directly** — it injects `fs` (plus `tools`/`systemPrompt`), **not** a policy service. The freshness/observation policy is contributed by a separate plugin ([`@deepseek-ai/dsh-fs-policy`](../fs-policy)) through the `fs/*` event gate; the tool is not method-coupled to it.
+The **model-facing filesystem tools** — `read`, `write`, `edit` — and their **executor**. This is the consumer layer of the filesystem stack: it owns tool names, JSON schemas, argument validation, prompt sections, **read windowing**, and result formatting. It reads/writes/edits through the `ctx.fs` provider seam ([`@deepseek-ai/dsh-fs`](../fs)) **directly**. The freshness/observation policy is contributed by a separate plugin ([`@deepseek-ai/dsh-fs-policy`](../fs-policy)) through the `fs/*` event gate; the tool is not method-coupled to it. Under a confining provider, the shared sandbox-policy service is required for per-session execution and the tool exposes escalation for filesystem mutations.
 
 ```ts ignore-check
 // Default deployment: a ctx.fs provider, the policy plugin, then the tools.
@@ -46,6 +46,8 @@ The tools do **not** inject a policy service or inspect any cache. Each tool res
 
 The tool passes `exec` (the tool-execution context) as the opaque `actor` on every dispatch. The default thunks return `undefined` (the unconstrained bare provider). When `@deepseek-ai/dsh-fs-policy` is loaded it occupies the single decision slot — returning `createIfAbsent`/`replaceIfVersion`/`{ version }` or throwing `FS_NOT_OBSERVED` — and records on `fs/observed`. Backend errors (`FsError`) and a thrown `FS_NOT_OBSERVED` flow through `ToolRegistry.execute()` and become `isError` tool results with their `{ name, code }` attached.
 
+When `ctx.fs.sandboxMode` reports confinement, write/edit advertise `sandbox_permissions` and `justification` and resolve approved retries through `ctx.approval`. The policy owner contributes capability-neutral standing policy; the tool results retain operation-specific denial and retry guidance.
+
 ## `fs/observed` is fire-and-forget
 
 `fs/observed` fires AFTER the read/write/edit already succeeded, via a plain `ctx.emit`. A listener is contractually a synchronous, side-effect-only recorder (`@deepseek-ai/dsh-fs-policy`'s is a `WeakMap.set`); the tool does not guard the emit, so a listener that throws would surface as the tool's `isError` result — async or fallible observation does not belong on this event.

packages/fs/tool-fs/README.zh.md

@@ -2,7 +2,7 @@
 
 [English](README.md) | 中文
 
-**面向模型的文件系统工具**（`read`、`write`、`edit`）及其**执行器**。这是文件系统栈的消费方层：拥有工具名称、JSON Schema、参数校验、提示词段、**读取窗口逻辑**和结果格式化。它**直接**通过 `ctx.fs` 提供方 seam（[`@deepseek-ai/dsh-fs`](../fs)）读取/写入/编辑：注入 `fs`（以及 `tools`/`systemPrompt`），**不**注入策略服务。新鲜度/观察策略由独立插件（[`@deepseek-ai/dsh-fs-policy`](../fs-policy)）通过 `fs/*` 事件门禁贡献；工具不与其方法耦合。
+**面向模型的文件系统工具**（`read`、`write`、`edit`）及其**执行器**。这是文件系统栈的消费方层：拥有工具名称、JSON Schema、参数校验、提示词段、**读取窗口逻辑**和结果格式化。它**直接**通过 `ctx.fs` 提供方 seam（[`@deepseek-ai/dsh-fs`](../fs)）读取／写入／编辑。新鲜度／观察策略由独立插件（[`@deepseek-ai/dsh-fs-policy`](../fs-policy)）通过 `fs/*` 事件门禁贡献；工具不与其方法耦合。使用施加沙箱限制的提供方时，逐会话执行需要共享沙箱策略服务，工具还会为文件系统变更提供升权路径。
 
 ```ts ignore-check
 // Default deployment: a ctx.fs provider, the policy plugin, then the tools.
@@ -46,6 +46,8 @@ await ctx.plugin(ToolFs)                                  // this package — re
 
 工具在每次分派中把 `exec`（工具执行上下文）作为不透明 `actor` 传入。默认 thunk 返回 `undefined`（不受约束的裸提供方）。加载 `@deepseek-ai/dsh-fs-policy` 后，它会占用单个决策槽：返回 `createIfAbsent`/`replaceIfVersion`/`{ version }` 或抛出 `FS_NOT_OBSERVED`，并在 `fs/observed` 时记录。后端错误（`FsError`）和抛出的 `FS_NOT_OBSERVED` 会流经 `ToolRegistry.execute()`，变成 `isError` 工具结果，并附带 `{ name, code }`。
 
+当 `ctx.fs.sandboxMode` 表明提供方施加沙箱限制时，write/edit 会公开 `sandbox_permissions` 与 `justification`，并通过 `ctx.approval` 解析经批准的重试。策略归属方会贡献与具体能力无关的常驻策略；工具结果仍保留操作特定的拒绝与重试引导。
+
 ## `fs/observed` 发后即忘
 
 `fs/observed` 在读取/写入/编辑已经成功之后，通过普通 `ctx.emit` 发出。监听器的契约是同步且只有副作用的记录器（`@deepseek-ai/dsh-fs-policy` 使用 `WeakMap.set`）；工具不保护这次发出，因此监听器抛出会作为工具的 `isError` 结果出现。异步或可能失败的观察不属于该事件。

packages/pty/pty-local/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/pty/pty-local/README.md
-README.md: 6f243a6edf3ab8bc228cfda3f6b3b774dabadcbb
-README.zh.md: 0cf2981332fa41dab22a496de5ae302066fbdba8
+README.md: ba05495318127b63b3d2a6a60ec743e1ff1c5821
+README.zh.md: 81987ea0685d761507b535b7ed6eefa0888fbd54

packages/pty/pty-local/README.md

@@ -6,7 +6,7 @@ Local Linux/macOS `node-pty` backend for `ctx.pty`; loading it on another platfo
 
 ## Plugin (`pty-local`)
 
-The plugin injects `pty`, `sandbox`, and `sandboxPolicy`, then registers the configured backend type (`shell`). `danger-full-access` starts the shell directly; confined modes wrap the exact shell argv through `ctx.sandbox`. The effective session mode is resolved at spawn. A change to a different effective mode is rejected before its `sandbox/mode` event commits while that owner has an open PTY or a spawn in progress; the fence is attached to the exact owner and therefore outlives a local-provider reload that retains existing sessions. Wait for creation to settle and close the sessions before changing modes, so a terminal opened with wider access cannot survive a downgrade.
+The plugin injects `pty`, `sandbox`, and `sandboxPolicy`, then registers the configured backend type (`shell`). `danger-full-access` starts the shell directly; confined modes wrap the exact shell argv through `ctx.sandbox`. At spawn, one `ctx.sandboxPolicy.resolve({ session })` call supplies both the effective mode and the session workspace root; the same root is the default shell cwd when the caller omits one. A change to a different effective mode is rejected before its `sandbox/mode` event commits while that owner has an open PTY or a spawn in progress; the fence is attached to the exact owner and therefore outlives a local-provider reload that retains existing sessions. Wait for creation to settle and close the sessions before changing modes, so a terminal opened with wider access cannot survive a downgrade.
 
 Linux readiness combines a foreground-verified private bash prompt marker, foreground-process-group syscall inspection, silence fallback, and absolute timeout. macOS uses the verified prompt marker plus silence/timeout because it has no `/proc` syscall surface. A marker is not ready until printable prompt text arrives, including when the OSC marker and `PS1` are split across data callbacks; when bash prints the marker before the kernel publishes its return to the foreground process group, polling retains the candidate for `handoffGraceMs` past the ordinary silence bound so a coincident handoff can win; that grace must cover at least one `pollIntervalMs` and is rejected at load otherwise. An interactive child that inherits `PROMPT_COMMAND` therefore cannot suppress inferred-idle readiness until the absolute timeout. Unrecognized or unreadable process state is never a positive exact-idle signal. A foreground group's stdin wait that already existed before a send is likewise not post-write readiness: the same group must be observed outside that wait before a later wait can settle the send, while a changed foreground group is new evidence. During unpublished startup, a fallback requires observed output; zero-output silence cannot publish an empty session, and timeout rejects the spawn. Cancellation closes the unpublished shell and rejects with the caller's exact abort reason even when its foreground process group is not observable yet; if that close fails, `PtyBackendCleanupError` separately preserves the cleanup failure for registry disposal. Incomplete terminal-control sequences are bounded by `maxReadBytes` and discarded through their terminator after crossing that limit; a trailing carriage return is carried across callbacks so split CRLF becomes one newline.
 
@@ -14,19 +14,19 @@ Send cancellation resolves the current foreground process group and delivers a r
 
 ## Model Experience
 
-### Indirect consumer
+### Current file policy and indirect consumer
 
 #### What the model sees
 
-Nothing directly. Through `@deepseek-ai/dsh-tool-pty`, the model may receive bounded MOTD, send deltas, scrollback pages, readiness reasons, and cleanup errors.
+The policy owner contributes capability-neutral `sandbox:policy` context. Through `@deepseek-ai/dsh-tool-pty` or another PTY consumer, the model may also receive bounded MOTD, send deltas, scrollback pages, readiness reasons, and cleanup errors.
 
 #### Token effect
 
-None until a consumer returns bounded backend output. Retained PTY scrollback is not placed in model history by this package.
+The current-policy clause is present while this backend is mounted. Retained PTY scrollback is not placed in model history until a consumer returns bounded output.
 
 #### KV Cache effect
 
-No direct invalidation; the consumer owns prompts, schemas, and appended results.
+A standing-policy change appends an owner-rendered superseding runtime-context snapshot after retained history; consumer results remain append-only.
 
 ## Known Limitations and Deferred Work
 

packages/pty/pty-local/README.zh.md

@@ -6,7 +6,7 @@
 
 ## 插件（`pty-local`）
 
-该插件注入 `pty`、`sandbox` 和 `sandboxPolicy`，然后注册所配置的后端类型（`shell`）。`danger-full-access` 会直接启动 shell；受限模式则通过 `ctx.sandbox` 包装确切的 shell argv。系统在 spawn 时解析会话的实际模式。当某个所有者存在开放的 PTY 或正在进行 spawn 时，如果配置变更会得到不同的实际模式，系统会在对应 `sandbox/mode` 事件提交前拒绝该变更。该限制绑定到确切所有者，因此即使本地提供方重新加载并保留现有会话，它仍然有效。更改模式前，请等待创建完成并关闭会话，避免以更宽权限打开的终端在权限降级后继续存在。
+该插件注入 `pty`、`sandbox` 和 `sandboxPolicy`，然后注册所配置的后端类型（`shell`）。`danger-full-access` 会直接启动 shell；受限模式则通过 `ctx.sandbox` 包装确切的 shell argv。Spawn 时，一次 `ctx.sandboxPolicy.resolve({ session })` 调用会同时给出实际模式与会话工作区根目录；调用方省略 cwd 时，同一根目录也是 shell 的默认 cwd。当某个所有者存在开放的 PTY 或正在进行 spawn 时，如果配置变更会得到不同的实际模式，系统会在对应 `sandbox/mode` 事件提交前拒绝该变更。该限制绑定到确切所有者，因此即使本地提供方重新加载并保留现有会话，它仍然有效。更改模式前，请等待创建完成并关闭会话，避免以更宽权限打开的终端在权限降级后继续存在。
 
 Linux 的就绪检测结合以下机制：由前台状态验证的私有 bash 提示符标记、前台进程组 syscall 检查、静默回退和绝对超时。macOS 没有 `/proc` syscall 接口，因此使用经过验证的提示符标记以及静默／超时。当可打印的提示符文本尚未到达时，即使 OSC 标记和 `PS1` 被拆到多个数据回调中，系统也不会把标记视为就绪。如果 bash 在内核发布其重新取得前台进程组的状态前打印标记，轮询会在普通静默上限之后再保留该候选状态 `handoffGraceMs`，使恰好同时发生的前台交接有机会胜出；该宽限至少要覆盖一个 `pollIntervalMs`，否则加载时即被拒绝。因此，继承 `PROMPT_COMMAND` 的交互式子进程无法一直抑制推断空闲就绪直至绝对超时。无法识别或读取的进程状态绝不会作为精确空闲的正向信号。同样，一次 send 之前就已存在的前台进程组 stdin 等待并不代表写入后就绪：必须先观察到同一进程组脱离该等待，之后再次进入等待才能使该次 send 完成；前台进程组发生变化则构成新的证据。尚未发布的启动过程中，回退路径要求已经观察到输出；零输出静默不能发布空会话，超时则拒绝 spawn。取消操作会关闭尚未发布的 shell，并以调用方提供的确切中止原因拒绝，即使当时还无法观察其前台进程组。如果关闭失败，`PtyBackendCleanupError` 会单独保留清理失败，供注册表 dispose（资源释放）时处理。未完成的终端控制序列受 `maxReadBytes` 限制；超过上限后，系统会丢弃内容直到其终止符。末尾的回车会跨回调保留，使拆分的 CRLF 合并为一个换行。
 
@@ -14,19 +14,19 @@ Linux 的就绪检测结合以下机制：由前台状态验证的私有 bash
 
 ## 模型体验
 
-### 间接消费方
+### 当前文件策略与间接消费方
 
 #### 模型看到的内容
 
-没有直接可见内容。模型通过 `@deepseek-ai/dsh-tool-pty` 可能收到有界的 MOTD、发送增量、scrollback 页、就绪原因和清理错误。
+策略归属方会贡献与具体能力无关的 `sandbox:policy` 上下文。模型通过 `@deepseek-ai/dsh-tool-pty` 或其他 PTY 消费方还可能收到有界的 MOTD、发送增量、scrollback 页、就绪原因和清理错误。
 
 #### Token 影响
 
-消费方返回有界的后端输出前没有影响。此包（package）不会把保留的 PTY scrollback 放入模型历史。
+装载该后端期间，当前策略子句会一直存在。消费方返回有界输出前，保留的 PTY scrollback 不会进入模型历史。
 
 #### KV Cache 影响
 
-不会直接使 KV Cache 失效；提示词、schema 与追加结果由消费方负责。
+常驻策略发生变化时，会在保留的历史之后追加一份由归属方渲染、取代先前状态的运行时上下文快照；消费方结果保持仅追加。
 
 ## 已知限制与暂缓事项
 

packages/pty/pty-local/src/index.ts

@@ -12,7 +12,7 @@ import type { Session, SessionEvent } from '@deepseek-ai/dsh-session'
 import { PtyBackendCleanupError } from '@deepseek-ai/dsh-pty'
 import { scrubbedParentEnv } from '@deepseek-ai/dsh-subprocess'
 import type { PtyBackend, PtyBackendSpawnSpec } from '@deepseek-ai/dsh-pty'
-import type { SandboxMode } from '@deepseek-ai/dsh-sandbox'
+import type { SandboxExecutionPolicy } from '@deepseek-ai/dsh-sandbox'
 import { effectiveSandboxMode } from '@deepseek-ai/dsh-sandbox-policy'
 import { type Config, type ResolvedConfig, validateConfig } from './config.ts'
 import { createProcessInspector } from './process-inspector.ts'
@@ -71,14 +71,11 @@ function childEnvironment(spec: PtyBackendSpawnSpec): NodeJS.ProcessEnv {
   }
 }
 
-function spawnArgv(ctx: Context, config: ResolvedConfig, spec: PtyBackendSpawnSpec): string[] {
+function spawnArgv(ctx: Context, config: ResolvedConfig, policy: SandboxExecutionPolicy): string[] {
   const argv = [config.shellPath, ...config.shellArgs]
-  const mode: SandboxMode = effectiveSandboxMode(spec.owner.session.events) ?? ctx.sandboxPolicy.defaultMode
-  if (mode === 'danger-full-access') return argv
-  return ctx.sandbox.confine(argv, {
-    mode: mode,
-    workspaceRoot: ctx.sandboxPolicy.workspaceRoot,
-  }).argv
+  if (policy.mode === 'danger-full-access') return argv
+  // Re-state the discriminant because object spread does not preserve its narrowed type.
+  return ctx.sandbox.confine(argv, { ...policy, mode: policy.mode }).argv
 }
 
 /** Local shell backend registered under the configured type. */
@@ -102,14 +99,15 @@ export class LocalPtyBackend implements PtyBackend {
   async spawn(spec: PtyBackendSpawnSpec): Promise<LocalPtySession> {
     spec.signal?.throwIfAborted()
     ensureSandboxModeFence(this.ctx, spec.owner)
-    const argv = spawnArgv(this.ctx, this.config, spec)
+    const policy = this.ctx.sandboxPolicy.resolve({ session: spec.owner.session })
+    const argv = spawnArgv(this.ctx, this.config, policy)
     const file = argv[0]
     if (file === undefined) throw new Error('pty-local: sandbox returned empty argv')
     const options: IPtyForkOptions = {
       name: 'dumb',
       cols: this.config.cols,
       rows: this.config.rows,
-      cwd: spec.cwd ?? this.ctx.sandboxPolicy.workspaceRoot,
+      cwd: spec.cwd ?? policy.workspaceRoot,
       env: childEnvironment(spec),
     }
     const terminal = this.spawnTerminal(file, argv.slice(1), options)

packages/pty/pty-local/tests/index.spec.ts

@@ -38,10 +38,13 @@ function config(): ResolvedConfig {
   }
 }
 
-function agent(ctx: Context): Agent {
+function agent(ctx: Context, cwd?: string): Agent {
   const id = SessionId('agent')
   return {
-    id, options: {}, session: new Session(id), status: 'idle', acceptsNextStep: false, ctx,
+    id,
+    options: {},
+    session: new Session(id, undefined, { version: 0, id, createdAt: 0, ...cwd === undefined ? {} : { cwd } }),
+    status: 'idle', acceptsNextStep: false, ctx,
     followup: () => {}, steer: () => {}, inject: () => {}, send: () => {}, updateInbox: () => 'not-found', cancel() {}, whenIdle: () => Promise.resolve(),
   }
 }
@@ -125,10 +128,10 @@ describe('LocalPtyBackend startup rollback', () => {
     } satisfies Partial<PtyBackendCleanupError>))
   })
 
-  it('wraps confined argv, scrubs the environment, and returns initialized sessions', async () => {
+  it('resolves session mode and root together before wrapping the shell', async () => {
     const ctx = new Context()
     await ctx.plugin(RecordingSandbox)
-    await ctx.plugin(SandboxPolicyService, { mode: 'workspace-write', workspaceRoot: '/workspace' })
+    await ctx.plugin(SandboxPolicyService, { mode: 'read-only', workspaceRoot: '/deployment-fallback' })
     const terminal = {} as IPty
     let spawned: { file: string; args: string[]; options: IPtyForkOptions } | undefined
     const spawnTerminal = ((file: string, args: string[], options: IPtyForkOptions) => {
@@ -146,8 +149,10 @@ describe('LocalPtyBackend startup rollback', () => {
     )
     const previous = process.env.PTY_TEST_SECRET
     process.env.PTY_TEST_SECRET = 'must-not-leak'
+    const owner = agent(ctx, '/session-workspace')
+    setSandboxMode(owner.session, 'workspace-write')
     try {
-      expect(await backend.spawn({ ...spec(agent(ctx)), cwd: '/work' })).toBe(session)
+      expect(await backend.spawn(spec(owner))).toBe(session)
     } finally {
       if (previous === undefined) delete process.env.PTY_TEST_SECRET
       else process.env.PTY_TEST_SECRET = previous
@@ -157,7 +162,7 @@ describe('LocalPtyBackend startup rollback', () => {
       file: '/sandbox',
       args: ['--', '/bin/bash', '-i'],
       options: {
-        name: 'dumb', cols: 80, rows: 24, cwd: '/work',
+        name: 'dumb', cols: 80, rows: 24, cwd: '/session-workspace',
         env: {
           TERM: 'dumb', PAGER: 'cat', GIT_PAGER: 'cat', PS1: 'dsh> ', BASH_SILENCE_DEPRECATION_WARNING: '1',
           DSH_SHELL: '1', DSH_SESSION_ID: 'agent', DSH_PTY_SESSION_ID: 'pty-1',
@@ -166,6 +171,10 @@ describe('LocalPtyBackend startup rollback', () => {
     })
     expect(spawned?.options.env?.PTY_TEST_SECRET).toBeUndefined()
     expect(initialized).toHaveBeenCalledWith(undefined)
+    expect((ctx.sandbox as RecordingSandbox).calls).toEqual([{
+      argv: ['/bin/bash', '-i'],
+      policy: { mode: 'workspace-write', workspaceRoot: '/session-workspace' },
+    }])
   })
 
   it('composes the default local session around a spawned terminal', async () => {

packages/sandbox/sandbox-policy/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/sandbox/sandbox-policy/README.md
-README.md: dca54330bc888af9ecac21aa92019d8a2b0140bd
-README.zh.md: a201d48c81f563fc3d85495e964bb67432517a3c
+README.md: b2512790c5cf5b3a06523cf91b50dd5d288b1522
+README.zh.md: dda98b7aa4775ac2453704295df87c3c92025093

packages/sandbox/sandbox-policy/README.md

@@ -2,11 +2,11 @@
 
 English | [中文](README.zh.md)
 
-The single owner of sandbox-policy resolution: the deployment's default [`SandboxMode`](../sandbox/README.md) and fallback root, plus each session's durable mode override and immutable workspace root. Every enforcing capability family receives one resolved mode-and-root policy per call.
+The single owner of sandbox-policy resolution: the deployment's default [`SandboxMode`](../sandbox/README.md) and fallback root, plus each session's durable mode override and immutable workspace root. Every enforcing capability receives one resolved mode-and-root policy per call; before each request, the model receives the current policy without a separate capability inventory.
 
 ## Why a shared home
 
-Two families enforce the same mode vocabulary: the sandboxed bash executor (`@deepseek-ai/dsh-bash-sandbox`) and the sandboxed filesystem provider (`@deepseek-ai/dsh-fs-sandbox`). If each resolved its own `mode` + `workspaceRoot`, the two could drift into a split world — bash confined to one root while fs fences another, exactly what [the sandbox RFC](../../../.agents/notes/implemented/feature/2026-07-06-sandbox.md) warns against. Both tool layers resolve policy through `ctx.sandboxPolicy`, and both enforcing backends consume that complete per-call result. The [cross-family fs sandbox RFC](../../../.agents/notes/implemented/feature/2026-07-14-cross-family-fs-sandbox.md) records the shared-policy decision.
+Filesystem tools, one-shot bash commands, and terminal sessions may enforce the same mode vocabulary in different combinations. If each resolved its own `mode` + `workspaceRoot`, they could drift into a split world, exactly what [the sandbox Agent Note](../../../.agents/notes/implemented/feature/2026-07-06-sandbox.md) warns against. Each enforcing backend consumes the complete owner-resolved policy, while the current context describes only what that policy means for any available operation the DSH file sandbox enforces. The [cross-family fs sandbox Agent Note](../../../.agents/notes/implemented/feature/2026-07-14-cross-family-fs-sandbox.md) records the shared-policy decision.
 
 ## Config
 
@@ -17,25 +17,53 @@ Two families enforce the same mode vocabulary: the sandboxed bash executor (`@de
 
 - `ctx.sandboxPolicy.resolve({ session?, mode? })` — resolves one complete per-call policy. An explicit approved mode outranks the session's last `sandbox/mode` event, which outranks `defaultMode`; the session's immutable `cwd` is canonicalized with filesystem semantics before becoming `workspaceRoot`, otherwise the configured fallback applies. Canonicalization precedes lexical normalization so `symlink/..` agrees with process working-directory resolution.
 - `ctx.sandboxPolicy.defaultMode` / `ctx.sandboxPolicy.workspaceRoot` — the deployment default and fallback root used by `resolve()`.
+- `sandbox:policy` — a request-time cache-safe context contribution derived directly from `resolve({ session })`. It states the mode's capability-neutral file-effect contract and the canonical session workspace under `workspace-write`; tool owners retain operation-specific denial and escalation guidance.
 - `effectiveSandboxMode(events)` — the pure fold of a session's `sandbox/mode` events (the last switch wins, or `undefined`), used inside `resolve()`.
 - `setSandboxMode(session, mode)` — THE write path for a per-session override: appends exactly one `sandbox/mode` event. The switch IS its event; nothing mutates the mode out of band.
 - `SANDBOX_MODES` — every mode, for option advertisement and runtime validation.
 
-The optional `./invariant` companion rejects a forged durable `sandbox/mode` event whose value falls outside that closed vocabulary; Session and its companion own the surrounding storage and core execution-enclosure rules.
+The optional `./invariant` companion rejects a forged durable `sandbox/mode` event whose value falls outside that closed vocabulary; Session and its companion own the surrounding storage and core execution-enclosure rules. The agent loop logs the assembled full runtime-context snapshot as a sourced `user/message`, so exact policy input remains reconstructable without an in-memory “last told” mirror.
 
 ## The per-session store
 
-A runtime switch is one log-only `sandbox/mode` event on the session it applies to. `effective = explicit grant ?? fold(events) ?? deployment default`, so an override survives restart by replay and two sessions never see each other's state. Workspace identity does not need another event: the immutable `SessionHeader.cwd` recorded at creation is the root for every call in that session. The event is log-only (the `approval/*` precedent): the model learns the mode from the enforcing tools' denial markers, never from the event.
+A runtime switch is one log-only `sandbox/mode` event on the session it applies to. `effective = explicit grant ?? fold(events) ?? deployment default`, so an override survives restart by replay and two sessions never see each other's state. Workspace identity does not need another event: the immutable `SessionHeader.cwd` recorded at creation is the root for every call in that session. The event stays log-only; before the next request, the owner contributes the current fact to the full runtime-context snapshot.
 
 ## Model Experience
 
-Indirectly, through `dsh-tool-bash` and `dsh-tool-fs`, which render the effective mode this service holds in their `[sandbox: …]` denial markers and escalation prompts; the `sandbox/mode` event itself never reaches the model.
+### Current file sandbox policy
+
+#### What the model sees
+
+One `sandbox:policy` contribution in the current runtime-context snapshot for every agent session. It does not enumerate mounted capabilities. Tool plugins retain operation and escalation guidance, approval policy contributes separately to the same snapshot, and plan guidance remains `dsh-plan-mode`'s system section.
+
+##### Read-only
+
+```markdown
+Current DSH file policy: read-only. Any available operation enforced by the DSH file sandbox cannot modify files in the standing mode. Do not refuse a required modification from this policy alone: try an available tool normally and follow any denial and escalation guidance it returns.
+```
+
+##### Workspace-write
+
+```markdown
+Current DSH file policy: workspace-write. Any available operation enforced by the DSH file sandbox may modify files under the session workspace: "<workspace root>". Some platform temporary areas may also be writable.
+```
+
+##### Danger-full-access
+
+```markdown
+Current DSH file policy: danger-full-access. The DSH file sandbox does not restrict file modifications by available operations.
+```
+
+#### Token effect
+
+One concise durable context message on the first request and each effective policy change; unchanged requests add nothing. `workspace-write` carries only the canonical session workspace path; platform-specific temporary paths are summarized without adding host-dependent bytes.
 
 #### KV Cache effect
 
-No direct invalidation; the named consumers own any request-prefix changes, and the mode is deliberately absent from the prompt.
+The stable system prompt remains byte-identical across mode changes. A changed full context snapshot is appended after retained history, preserving the prior cached prefix; subsequent unchanged requests reuse that retained snapshot.
 
 ## Known Limitations and Deferred Work
 
 - **One primary workspace root per session** — policy resolves `SessionHeader.cwd`; extra writable roots are not part of `SandboxExecutionPolicy`.
 - **File-effect modes only** — `SandboxMode` governs file effects; network and process policy are outside its vocabulary, so no knob here restricts them.
+- **Temporary areas are deliberately summarized** — enforcing backends grant different platform temporary areas, which are selected after policy resolution and therefore cannot be enumerated truthfully in the current context.

packages/sandbox/sandbox-policy/README.zh.md

@@ -2,11 +2,11 @@
 
 [English](README.md) | 中文
 
-沙箱策略解析的唯一归属位置：部署默认 [`SandboxMode`](../sandbox/README.md) 与回退根目录，加上每个会话的持久模式覆盖和不可变工作区根目录。每个强制执行策略的能力家族在每次调用时都会收到一项解析完成的模式与根目录策略。
+沙箱策略解析的唯一归属位置：部署默认 [`SandboxMode`](../sandbox/README.md) 与回退根目录，加上每个会话的持久模式覆盖和不可变工作区根目录。每项负责强制执行的能力在每次调用时都会收到一项解析完成的模式与根目录策略；模型在每次请求前会收到当前策略，而不会另收一份能力清单。
 
 ## 为何需要共享归属位置
 
-两个家族强制执行同一套模式词汇：沙箱化 bash 执行器（`@deepseek-ai/dsh-bash-sandbox`）与沙箱化文件系统提供方（`@deepseek-ai/dsh-fs-sandbox`）。如果两者各自解析 `mode` + `workspaceRoot`，就可能漂移成分裂世界：bash 限制在一个根目录，fs 却隔离另一个根目录，正是[沙箱 RFC](../../../.agents/notes/implemented/feature/2026-07-06-sandbox.md)所警告的情况。两个工具层都通过 `ctx.sandboxPolicy` 解析策略，两个强制执行后端也都消费完整的逐调用结果。[跨家族 fs 沙箱 RFC](../../../.agents/notes/implemented/feature/2026-07-14-cross-family-fs-sandbox.md)记录了共享策略决策。
+文件系统工具、一次性 bash 命令和终端会话可以用不同组合强制执行同一套模式词汇。如果各自解析 `mode` + `workspaceRoot`，就可能漂移成分裂世界，正是[沙箱 Agent Note](../../../.agents/notes/implemented/feature/2026-07-06-sandbox.md)所警告的情况。每个强制执行后端都会消费归属方解析出的完整策略，而当前上下文只说明该策略对于任何受 DSH 文件沙箱强制执行的可用操作有何含义。[跨家族 fs 沙箱 Agent Note](../../../.agents/notes/implemented/feature/2026-07-14-cross-family-fs-sandbox.md)记录了共享策略决策。
 
 ## 配置
 
@@ -17,25 +17,53 @@
 
 - `ctx.sandboxPolicy.resolve({ session?, mode? })`：解析一项完整的逐调用策略。显式批准的模式优先于会话最后一条 `sandbox/mode` 事件，后者又优先于 `defaultMode`；会话不可变的 `cwd` 会先按文件系统语义规范化，再成为 `workspaceRoot`，否则使用配置的回退值。规范化先于词法归一化，因此 `symlink/..` 与进程工作目录解析保持一致。
 - `ctx.sandboxPolicy.defaultMode`／`ctx.sandboxPolicy.workspaceRoot`：`resolve()` 使用的部署默认值与回退根目录。
+- `sandbox:policy`：直接派生自 `resolve({ session })` 的请求时缓存安全上下文贡献。它说明该模式中与具体能力无关的文件效果契约，以及 `workspace-write` 下规范化的会话工作区；工具归属方仍负责操作特定的拒绝与升权引导。
 - `effectiveSandboxMode(events)`：会话 `sandbox/mode` 事件的纯 fold（最后一次切换胜出，没有则为 `undefined`），在 `resolve()` 内使用。
 - `setSandboxMode(session, mode)`：逐会话覆盖的唯一写入路径：恰好追加一条 `sandbox/mode` 事件。切换本身就是事件；不会在带外修改模式。
 - `SANDBOX_MODES`：所有模式，用于选项展示与运行时验证。
 
-可选的 `./invariant` 配套组件会拒绝伪造的持久 `sandbox/mode` 事件，只要其值不在该封闭词汇中；Session 与其配套组件负责相关存储与核心执行封闭规则。
+可选的 `./invariant` 配套组件会拒绝伪造的持久 `sandbox/mode` 事件，只要其值不在该封闭词汇中；Session 与其配套组件负责相关存储与核心执行封闭规则。agent loop（智能体循环）会将组装后的完整运行时上下文快照记录为一条带来源的 `user/message`，因此无需内存中的「上次告知」镜像，也能重建确切的策略输入。
 
 ## 逐会话存储
 
-运行时切换是在对应会话日志中追加的一条 `sandbox/mode` 事件。`effective = explicit grant ?? fold(events) ?? deployment default`，因此覆盖会通过回放跨重启保留，两个会话也绝不会看到彼此状态。工作区标识无需另一条事件：创建时记录的不可变 `SessionHeader.cwd` 是该会话每次调用使用的根。该事件只进入日志（沿用 `approval/*` 先例）：模型通过强制执行工具的拒绝标记获知模式，绝不会从事件获知。
+运行时切换是在对应会话日志中追加的一条 `sandbox/mode` 事件。`effective = explicit grant ?? fold(events) ?? deployment default`，因此覆盖会通过回放跨重启保留，两个会话也绝不会看到彼此状态。工作区标识无需另一条事件：创建时记录的不可变 `SessionHeader.cwd` 是该会话每次调用使用的根。该事件仍只进入日志；在下一次请求前，归属方会将当前事实贡献给完整运行时上下文快照。
 
 ## 模型体验
 
-通过 `dsh-tool-bash` 和 `dsh-tool-fs` 间接影响；它们会在 `[sandbox: …]` 拒绝标记和升权提示词中渲染该服务持有的有效模式，`sandbox/mode` 事件本身绝不会到达模型。
+### 当前文件沙箱策略
+
+#### 模型看到的内容
+
+每个 agent 会话的当前运行时上下文快照中都有一项 `sandbox:policy` 贡献。它不枚举已装载的能力。工具插件继续负责操作与升级引导，批准策略单独贡献给同一份快照，计划引导仍由 `dsh-plan-mode` 的系统段落管理。
+
+##### 只读
+
+```markdown
+Current DSH file policy: read-only. Any available operation enforced by the DSH file sandbox cannot modify files in the standing mode. Do not refuse a required modification from this policy alone: try an available tool normally and follow any denial and escalation guidance it returns.
+```
+
+##### 工作区写入
+
+```markdown
+Current DSH file policy: workspace-write. Any available operation enforced by the DSH file sandbox may modify files under the session workspace: "<workspace root>". Some platform temporary areas may also be writable.
+```
+
+##### 完全访问
+
+```markdown
+Current DSH file policy: danger-full-access. The DSH file sandbox does not restrict file modifications by available operations.
+```
+
+#### Token 影响
+
+首次请求和有效策略每次变化时增加一条简洁的持久上下文消息；未变化的请求不增加内容。`workspace-write` 只携带规范化的会话工作区路径；平台特定的临时路径会以摘要表述，不会加入依赖主机的字节。
 
 #### KV Cache 影响
 
-不会直接使 KV Cache 失效；请求前缀变更由上述消费方负责，且提示词有意不包含模式。
+模式切换时，稳定的系统提示词仍逐字节相同。变化后的完整上下文快照会追加到保留的历史之后，从而保留此前已缓存的前缀；后续未变化的请求会复用该保留快照。
 
 ## 已知限制与暂缓事项
 
 - **每个会话只有一个主要工作区根目录**：策略解析 `SessionHeader.cwd`；额外可写根目录不属于 `SandboxExecutionPolicy`。
 - **仅限文件操作模式**：`SandboxMode` 管控文件操作；网络和进程策略不在其词汇中，因此这里没有限制它们的旋钮。
+- **有意概述临时区域**：强制执行后端会授予不同的平台临时区域，这些区域在策略解析后才会选定，因此无法在当前上下文中如实枚举。

packages/sandbox/sandbox-policy/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@deepseek-ai/dsh-sandbox-policy",
-  "description": "Per-call sandbox policy resolver (ctx.sandboxPolicy): deployment fallbacks plus each session's mode and workspace root, shared by every enforcing capability family",
+  "description": "Per-call sandbox policy resolver and current model context: deployment fallbacks plus each session's mode and workspace root, shared by every enforcing capability family",
   "version": "0.0.1",
   "private": true,
   "type": "module",
@@ -27,18 +27,22 @@
   ],
   "license": "BSD-3-Clause",
   "peerDependencies": {
+    "@deepseek-ai/dsh-agent": "^0.0.1",
     "@deepseek-ai/dsh-invariants": "^0.0.1",
     "@deepseek-ai/dsh-sandbox": "^0.0.1",
     "@deepseek-ai/dsh-session": "^0.0.1",
+    "@deepseek-ai/dsh-system-prompt": "^0.0.1",
     "cordis": "^4.0.0-rc.7"
   },
   "dependencies": {
     "schemastery": "^3.18.0"
   },
   "devDependencies": {
+    "@deepseek-ai/dsh-agent": "workspace:^",
     "@deepseek-ai/dsh-invariants": "workspace:^",
     "@deepseek-ai/dsh-sandbox": "workspace:^",
     "@deepseek-ai/dsh-session": "workspace:^",
+    "@deepseek-ai/dsh-system-prompt": "workspace:^",
     "cordis": "^4.0.0-rc.7"
   }
 }

packages/sandbox/sandbox-policy/src/index.ts

@@ -3,13 +3,17 @@
  * deployment's sandbox fallbacks plus per-session resolution: the file-effect
  * {@link SandboxMode}, the `workspace-write` root, and the override kit (the
  * `sandbox/mode` event, its fold, and its write path, from `./session-mode.ts`).
+ * Before each agent request, the owner also contributes the resolved policy to
+ * the cache-safe runtime-context snapshot. The agent loop logs that snapshot as
+ * model history, so replay reconstructs the same mode and root the enforcing
+ * consumers resolve without rewriting the stable system prompt.
  *
- * Both enforcing capability families read the SAME policy here: the sandboxed
- * bash executor (`@deepseek-ai/dsh-bash-sandbox`) and the sandboxed filesystem
- * provider (`@deepseek-ai/dsh-fs-sandbox`) consume the SAME resolved per-call
- * policy, so bash and fs can never confine to different roots — the split
- * world the sandbox RFC warns about. The service reads session state once at
- * the tool boundary; executors and providers remain session-free.
+ * Enforcing filesystem, one-shot bash, and terminal backends read the SAME
+ * resolved policy here. The context describes that policy without inventorying
+ * capabilities, while each backend retains its own enforcement dialect and each
+ * tool owns its operation-specific denial and escalation guidance. The service
+ * reads session state once at each operation boundary; executors and providers
+ * remain session-free.
  *
  * @module @deepseek-ai/dsh-sandbox-policy
  */
@@ -17,8 +21,10 @@
 import { resolve as resolvePath } from 'node:path'
 import { Context, Service } from 'cordis'
 import z from 'schemastery'
+import type {} from '@deepseek-ai/dsh-agent'
 import { canonicalPath, type SandboxExecutionPolicy, type SandboxMode } from '@deepseek-ai/dsh-sandbox'
 import type { Session } from '@deepseek-ai/dsh-session'
+import type {} from '@deepseek-ai/dsh-system-prompt'
 import { effectiveSandboxMode } from './session-mode.ts'
 
 export { SANDBOX_MODES, effectiveSandboxMode, setSandboxMode } from './session-mode.ts'
@@ -28,6 +34,23 @@ function resolveWorkspaceRoot(path: string): string {
   return resolvePath(canonicalPath(path))
 }
 
+/** Render the policy without claiming which capabilities are mounted. */
+function renderPolicyContext(policy: SandboxExecutionPolicy): string {
+  switch (policy.mode) {
+    case 'read-only':
+      return 'Current DSH file policy: read-only. Any available operation enforced by the DSH file sandbox cannot modify files in the standing mode. Do not refuse a required modification from this policy alone: try an available tool normally and follow any denial and escalation guidance it returns.'
+    case 'workspace-write':
+      return `Current DSH file policy: workspace-write. Any available operation enforced by the DSH file sandbox may modify files under the session workspace: ${JSON.stringify(policy.workspaceRoot)}. Some platform temporary areas may also be writable.`
+    case 'danger-full-access':
+      return 'Current DSH file policy: danger-full-access. The DSH file sandbox does not restrict file modifications by available operations.'
+    /* v8 ignore next 4 -- SandboxMode is a typed same-process closed union; this branch is only the static exhaustiveness guard. */
+    default: {
+      const mode: never = policy.mode
+      throw new Error(`unreachable sandbox mode: ${String(mode)}`)
+    }
+  }
+}
+
 declare module 'cordis' {
   interface Context {
     sandboxPolicy: SandboxPolicyService
@@ -61,9 +84,9 @@ export interface SandboxPolicyRequest {
 
 /**
  * The sandbox-policy service (`ctx.sandboxPolicy`). Owns the deployment
- * default mode and fallback workspace root. Tool layers call {@link resolve}
- * for each execution so a session's mode log and immutable cwd travel together
- * to every enforcing capability.
+ * default mode, fallback workspace root, and current request-time policy
+ * section. Tool layers call {@link resolve} for each execution so a session's
+ * mode log and immutable cwd travel together to every enforcing capability.
  */
 export class SandboxPolicyService extends Service {
   // Inline schema call: the config catalog walks `static Config` statically.
@@ -78,7 +101,6 @@ export class SandboxPolicyService extends Service {
   readonly defaultMode: SandboxMode
   /** The absolute `workspace-write` fallback root for calls without a session cwd. */
   readonly workspaceRoot: string
-
   constructor(ctx: Context, config: Config) {
     super(ctx, 'sandboxPolicy')
     // schemastery (static Config) already filled `mode`; the cast records that
@@ -86,6 +108,19 @@ export class SandboxPolicyService extends Service {
     // the process cwd is real branching, resolved absolute either way.
     this.defaultMode = config.mode as SandboxMode
     this.workspaceRoot = resolveWorkspaceRoot(config.workspaceRoot ?? process.cwd())
+
+    ctx.inject(['systemPrompt'], (scope: Context) => {
+      scope.systemPrompt.context({
+        name: 'sandbox:policy',
+        order: 110,
+        text: (context) => {
+          const session = context.agent?.session
+          return session === undefined
+            ? ''
+            : renderPolicyContext(this.resolve({ session }))
+        },
+      })
+    })
   }
 
   /**

packages/sandbox/sandbox-policy/src/session-mode.ts

@@ -5,11 +5,11 @@
  * `effective = fold(events) ?? the deployment default`, so an override
  * survives restart by replay, two sessions can never see each other's state,
  * and there is no external config store. The event is log-only (the
- * `approval/*` precedent): the model learns the mode from the boundary
- * markers in the enforcing tools, never from the event itself. EXECUTION
- * honors the fold through `ctx.sandboxPolicy.resolve()` — it stamps the mode
- * together with the calling session's workspace root onto each capability
- * call, weakest-precedence beneath an escalation grant.
+ * `approval/*` precedent): the policy owner projects the fold into each model
+ * request, while enforcing tools report operation-specific boundary markers.
+ * EXECUTION honors the same fold through `ctx.sandboxPolicy.resolve()` — it
+ * stamps the mode together with the calling session's workspace root onto each
+ * capability call, weakest-precedence beneath an escalation grant.
  *
  * The override is policy state shared by every enforcing family (bash and
  * filesystem alike), so it lives here in the policy package rather than in any

packages/sandbox/sandbox-policy/tests/policy.spec.ts

@@ -1,7 +1,7 @@
 /**
  * Tests for the sandbox-policy home: the deployment default (mode +
  * workspaceRoot) the service exposes, and the per-session `sandbox/mode`
- * override kit (fold + write path) both enforcing families read.
+ * override kit (fold + write path) every enforcing capability reads.
  */
 
 import { mkdirSync, mkdtempSync, realpathSync, rmSync, symlinkSync } from 'node:fs'
@@ -9,8 +9,10 @@ import { tmpdir } from 'node:os'
 import { join, resolve, sep } from 'node:path'
 import { describe, expect, it } from 'vitest'
 import { Context } from 'cordis'
+import type { Agent } from '@deepseek-ai/dsh-agent'
 import { Session, SessionId } from '@deepseek-ai/dsh-session'
 import SandboxPolicyService, { SANDBOX_MODES, effectiveSandboxMode, setSandboxMode } from '@deepseek-ai/dsh-sandbox-policy'
+import SystemPrompt, { renderContextSnapshot, renderPrompt } from '@deepseek-ai/dsh-system-prompt'
 
 async function mounted(config: { mode?: 'read-only' | 'workspace-write' | 'danger-full-access'; workspaceRoot?: string } = {}) {
   const ctx = new Context()
@@ -28,6 +30,15 @@ function session(id: string, cwd?: string): Session {
   })
 }
 
+function agentFor(activeSession: Session): Agent {
+  return { session: activeSession } as unknown as Agent
+}
+
+async function policyContext(ctx: Context, activeSession: Session): Promise<string | undefined> {
+  return (await ctx.systemPrompt.assemble({ agent: agentFor(activeSession) }))
+    .contexts.find(context => context.name === 'sandbox:policy')?.text
+}
+
 describe('SandboxPolicyService', () => {
   it('defaults to read-only under the process cwd', async () => {
     const ctx = await mounted()
@@ -114,12 +125,81 @@ describe('SandboxPolicyService', () => {
     await expect(ctx.plugin(SandboxPolicyService, { mode: 'yolo' as never })).rejects.toThrow()
   })
 
-  it('unregisters cleanly from a child fiber (HMR safety)', async () => {
+  it('disposes the service and context contribution from a child fiber (HMR safety)', async () => {
     const ctx = new Context()
+    await ctx.plugin(SystemPrompt)
     const fiber = await ctx.plugin(SandboxPolicyService, {})
     expect(ctx.sandboxPolicy).toBeDefined()
+    expect(await policyContext(ctx, session('sess-hmr'))).toContain('read-only')
     await fiber.dispose()
     expect(ctx.get('sandboxPolicy')).toBeUndefined()
+    expect((await ctx.systemPrompt.assemble()).contexts.find(context => context.name === 'sandbox:policy')).toBeUndefined()
+  })
+})
+
+describe('sandbox:policy request context', () => {
+  async function promptMounted(config: { mode?: 'read-only' | 'workspace-write' | 'danger-full-access'; workspaceRoot?: string } = {}): Promise<Context> {
+    const ctx = new Context()
+    await ctx.plugin(SystemPrompt)
+    await ctx.plugin(SandboxPolicyService, config)
+    return ctx
+  }
+
+  it.each(['read-only', 'workspace-write', 'danger-full-access'] as const)('renders the exact %s policy without a capability inventory', async (mode) => {
+    const ctx = await promptMounted({ mode, workspaceRoot: '/fallback' })
+    const workspaceRoot = resolve('/projects/current')
+    const expected = {
+      'read-only': 'Current DSH file policy: read-only. Any available operation enforced by the DSH file sandbox cannot modify files in the standing mode. Do not refuse a required modification from this policy alone: try an available tool normally and follow any denial and escalation guidance it returns.',
+      'workspace-write': `Current DSH file policy: workspace-write. Any available operation enforced by the DSH file sandbox may modify files under the session workspace: ${JSON.stringify(workspaceRoot)}. Some platform temporary areas may also be writable.`,
+      'danger-full-access': 'Current DSH file policy: danger-full-access. The DSH file sandbox does not restrict file modifications by available operations.',
+    } as const
+
+    expect(await policyContext(ctx, session(`sess-${mode}`, '/projects/../projects/current'))).toBe(expected[mode])
+  })
+
+  it('keeps the complete rendered prompt byte-stable across TMPDIR changes', async () => {
+    const ctx = await promptMounted({ mode: 'workspace-write' })
+    const active = session('sess-tmpdir-stability', '/projects/current')
+    const previous = process.env.TMPDIR
+    try {
+      process.env.TMPDIR = '/tmp/first-host-temp'
+      const firstAssembly = await ctx.systemPrompt.assemble({ agent: agentFor(active) })
+      const firstPrompt = renderPrompt(firstAssembly)
+      const firstContext = renderContextSnapshot(firstAssembly)
+      process.env.TMPDIR = '/tmp/second-host-temp'
+      const secondAssembly = await ctx.systemPrompt.assemble({ agent: agentFor(active) })
+      expect(renderPrompt(secondAssembly)).toBe(firstPrompt)
+      expect(renderContextSnapshot(secondAssembly)).toBe(firstContext)
+      expect(firstContext).not.toContain('host-temp')
+    } finally {
+      if (previous === undefined) delete process.env.TMPDIR
+      else process.env.TMPDIR = previous
+    }
+  })
+
+  it('reflects the latest durable switch on the next assembly and stays byte-stable otherwise', async () => {
+    const ctx = await promptMounted()
+    const active = session('sess-switch', '/projects/current')
+    const first = await policyContext(ctx, active)
+    expect(await policyContext(ctx, active)).toBe(first)
+
+    setSandboxMode(active, 'danger-full-access')
+    const danger = await policyContext(ctx, active)
+    expect(danger).toBe('Current DSH file policy: danger-full-access. The DSH file sandbox does not restrict file modifications by available operations.')
+    expect(await policyContext(ctx, active)).toBe(danger)
+
+    setSandboxMode(active, 'workspace-write')
+    expect(await policyContext(ctx, active)).toBe(`Current DSH file policy: workspace-write. Any available operation enforced by the DSH file sandbox may modify files under the session workspace: ${JSON.stringify(resolve('/projects/current'))}. Some platform temporary areas may also be writable.`)
+  })
+
+  it('reconstructs resumed policy from the session log and omits diagnostics without an agent', async () => {
+    const active = session('sess-resume', '/projects/current')
+    setSandboxMode(active, 'workspace-write')
+    const resumed = new Session(active.id, active.events, active.header)
+    const ctx = await promptMounted({ mode: 'read-only' })
+
+    expect(await policyContext(ctx, resumed)).toContain('workspace-write')
+    expect((await ctx.systemPrompt.assemble()).contexts.find(context => context.name === 'sandbox:policy')?.text).toBe('')
   })
 })
 

packages/sandbox/sandbox-policy/tsconfig.json

@@ -17,12 +17,18 @@
     {
       "path": "../../../vendor/schemastery"
     },
+    {
+      "path": "../../core/agent"
+    },
     {
       "path": "../sandbox"
     },
     {
       "path": "../../core/session"
     },
+    {
+      "path": "../../core/system-prompt"
+    },
     {
       "path": "../../support/invariants"
     }

packages/subagent/subagent-inprocess/tests/inheritance.spec.ts

@@ -100,7 +100,20 @@ describe('in-process policy inheritance', () => {
       const request = child.session.events.find(
         (event): event is SessionEvent<'request/header'> => event.type === 'request/header',
       )
-      expect(request?.data.header.system).toContain('Approval prompts are disabled')
+      const runtimeContext = child.session.events.find(
+        (event): event is SessionEvent<'user/message'> => event.type === 'user/message'
+          && event.data.source.kind === 'plugin'
+          && event.data.source.plugin === '@deepseek-ai/dsh-system-prompt',
+      )
+      if (request === undefined || runtimeContext === undefined) throw new Error('child request lacks its runtime policy context')
+      expect(runtimeContext.seq).toBeLessThan(request.seq)
+      const contextText = runtimeContext.data.content
+        .filter((block): block is Extract<ContentBlock, { type: 'text' }> => block.type === 'text')
+        .map(block => block.text)
+        .join('\n')
+      expect(contextText).toContain('Current DSH file policy: read-only')
+      expect(contextText).toContain('Approval prompts are disabled')
+      expect(request.data.header.system).not.toContain('Approval prompts are disabled')
       expect(parent.session.events).toHaveLength(parentLogLength)
     } finally {
       await run.dispose()