dsh-better-display 0.1.1 → 0.3.0

package/CHANGELOG.md

@@ -1,10 +1,194 @@
 # Changelog
 
-## Unreleased
+## 0.3.0 — 2026-09-21
+
+- 修复关闭自动折叠、查看过程详情后，再开启仍无法折叠的问题；重新开启会恢复自动折叠规则，仍保留文本选择保护和之后的手动展开。
+- 在现有阅读布局中接入官方反馈、工具详情、显式产物卡片、命令和未知节点渲染；官方渲染器负责注入、翻译、store 和子插槽。
+- 移除直接调用工具组件及模拟点击展开的逻辑，保留现有摘要、折叠、动效和文件快捷操作。
+- 接入官方文件链接服务，传递工具文件行号，补齐 RC2 本地 Markdown 图片及加载失败降级。
+- 修复滚动跟随拉回底部的问题，统一阅读列轨道和吸顶区域的位置。
+- 代码解释器结果按终端输出展示，上下文行兼容 `provenance` 和 `producer`。
+- 增加可重放的官方组件集成测试、自动折叠交互回归和 Harness 升级兼容检查，依赖对齐 Harness `0.1.5-rc.2`。
+- 设计范围、接口边界和待收敛模块见 `docs/official-rendering-bridge.md`。
+
+## 0.2.1 — 2026-09-20
+
+### Fixed
+
+- Pinned reader lanes no longer paint over the composer. The live status row
+  (`执行过程` / `正在使用工具`) pinned at the scrollport top with `z-index: 8`,
+  while the host's sticky composer seat owns `z-index: 7` and nothing between
+  the reader and `<body>` creates a stacking context. A scrolling reader clamps
+  every turn-level sticky lane down into the footer band — the lane stops at its
+  containing block's bottom, which passes behind the input card — so the reader
+  painted its own opaque plate over the composer.
+  - The live status lane now owns the measured lane under the toolbar, exactly
+    like the closed summary rows, instead of sharing the toolbar's lane.
+  - Lanes a scroll can clamp into the footer band stay below the composer seat
+    (`z-index` 6); only the top toolbar lane keeps 7, which also keeps it above
+    the host's code-block banners.
+  - Skin mode no longer carries a second lane ladder of its own.
+  - `tests/footer-precedence.test.ts` guards the ladder in the source sheet and
+    in the committed bundle.
+
+## 0.2.0 — 2026-09-18
+
+### Diff review and tool presentation
+
+- Re-engineered diff review panel: line addition/deletion statistics (`+A -R`) are pushed
+  flush to the right margin, perfectly aligned on the same 32px baseline as the summary text.
+- Diff overlay expands cleanly within the reading column (0px horizontal overflow) with
+  independent vertical scrolling (`max-height: 420px; overscroll-behavior: contain`),
+  preventing mouse-wheel events from leaking into the conversation stream.
+- Multi-file tabs scroll horizontally with adaptive content height on file switches.
+- PR #12 custom tool cards: delegate tool view rendering to `tool.call.toolview` slot with
+  error boundary isolation and auto-expand capabilities.
+
+### Translucent frosted glass and skin mode
+
+- Translucent frosted glass mode is an opt-in toggle in Settings.
+- When enabled, all card backgrounds, tool detail frames, and code blocks become translucent
+  with backdrop blur, eliminating opaque white patches and seamlessly adapting to third-party
+  wallpaper and window skins.
+
+### Refined folding and liveness
+
+- Reverted to a single, intuitive auto-fold toggle (**自动折叠开 / 关**) on the reading toolbar
+  and in Settings, removing complex multi-level rules while keeping reasoning and tools
+  completely unfolded when turned off.
+- Fixed line-wrapping bug so digest rows stay strictly single-line when space permits.
+- Real-time wait clocks (`WaitClock`), execution time tracking, and token throughput
+  metrics (`TurnMetrics`) fully preserved and active.
+- Animated disclosures include fallback timeouts to prevent rendering stalls.
+
+### Better Display settings
+
+- Glass, auto-fold, and deliverable open-mode share the root `dsh.reader.v1` store so
+  Settings and the reading view stay in sync across refresh.
+- Deliverable open-mode: default system app, optional Sidebar. Folder/reveal stay OS.
+- Generative-mcpapps skill installation detection and status reporting.
+
+### Compatibility and liveness
+
+- A wait now restarts when a tool returns. The handover set named kinds from the
+  conversation contract (`tool-result`) while the reader matches against the chat
+  layer's kinds, where a returned tool is the `tool-call` row whose `data.root`
+  carries a result. No node ever had the old name, so the branch was dead and a
+  returned tool never restarted the clock. The set and every kind it reasons about
+  are now typed against the host's own kind union, so a wrong name fails `tsc`
+  instead of failing silently at runtime, and a tool that is still running is
+  explicitly not a wait.
+- The fold choreography can no longer stall. Each phase advances on animation
+  promises and rendered frames, and a cancelled animation rejects while a hidden
+  tab delivers no frames — either one used to leave the machine in a non-idle
+  phase forever, which holds the frame source at the shown snapshot and pauses the
+  text reveal, so the turn looked stuck until the view remounted. Every phase now
+  also has a wall-clock deadline that forces the next phase.
+- While the choreography holds the reveal, the stream buffer keeps absorbing the
+  source instead of returning early, so text never resumes from a stale target
+  once the fold settles.
+- Status shows how many sub-agents the turn dispatched, read from the host's own
+  turn-process row instead of being counted again here.
+- Drop the `command-input` render branch: that string is not a chat node kind in
+  any released host, so it only made the file look like it handled a case that
+  cannot occur.
+- Give every animated disclosure a wall-clock deadline as well. `fill: 'both'`
+  pins the opening keyframe, so a Web Animation that never reaches `onfinish`
+  (cancelled, unmounted, or skipped by the compositor) left the row in the DOM at
+  zero height and zero opacity — clicking it looked like nothing happened.
+- Add a Better Display settings section: deliverables still open in the system app
+  by default, with an optional right-Sidebar preview, plus generative-mcpapps
+  skill-root detection and install guidance.
+- Guard pending-submission image echoes so a text-only send cannot crash
+  `conversation.view` (`images` / `attachments` may be missing).
+- Install guidance names only conventional relative roots (`.dsh/skills`,
+  `.agents/skills`) and never prints host home or plugin pack absolutes.
+
+### Reading and folding
+
+- A later reasoning step folds only the finished process run it follows. The run
+  that is still streaming stays fully expanded, so a long turn reads as
+  alternating digests and prose instead of collapsing mid-thought.
+- Fold intensity 2 (`过程摘要`) is PR #14's process-only mode: prose is never
+  folded; each finished run of reasoning / tool / record steps collapses into
+  one digest that names what the tools actually did (`读取 Reader.tsx`,
+  `运行 pnpm test`), read from the same tool identity the tool cards use. It is
+  not the default — standard fold (level 1) keeps current-main choreography.
+- A finished process-only turn folds its trailing run too, and skips the
+  duplicate closed-turn counter row it no longer needs.
+
+### Reading surface
+
+- When frosted glass is on, sticky lanes use the same liquid glass as the
+  dsh-auto-memory pane: translucent `bg-layer-2` wash, `blur(28px)`, hairline
+  border, 16px radius and a soft lift. Detail chips stay transparent until
+  hovered or focused. When the toggle is off, chrome matches current main.
+- The toolbar reserves the width of its whole control group. Measuring only the
+  first button let the status lane paint over every later control.
+- A compaction divider now arrives instead of simply appearing: the rule sweeps
+  out from its centre, the pill settles in, and the dial turns once. Gated by
+  `data-motion=off` and `prefers-reduced-motion` like every other transition.
+
+### Diff review
+
+- Read the changed line counts from the tool result the host already attaches
+  (`meta.diffs`, `{ path, oldText, newText }`), and fall back to the call's own
+  arguments for `write` / `edit` / `str_replace_editor` when a host build does
+  not send them. The name whitelist matters: several unrelated tools take a
+  field called `content`, and counting those invented additions for calls that
+  changed no file.
+- Fold a parent call's child calls into its own counts. A `run_code` that writes
+  files reports what its children changed rather than what its own arguments
+  contain, so a run whose edits live one level down no longer loses its counts.
+  Each child contributes through its own arguments, read from either the landed
+  result or the pending call.
+- `+N -M` sits at the end of the row in green/red and opens a panel listing one
+  entry per changed file — the file tabs, the removed lines on a red wash and
+  the added lines on a green one, using the same `DiffBlock` primitive the
+  official tool row renders. The panel is in flow, never a floating popup, so
+  the flow cell's `overflow: clip` cannot cut it off.
+- Opening and closing the panel animate real height in the plugin's own motion
+  language, and the card is brought back into view when it opens below the
+  fold; an open panel keeps its caret lit.
+
+### Waiting and timing
+
+- Show how long the model has been given the turn, next to 「深度求索中」and the
+  in-flow wait indicator. The clock is anchored to the last event that handed
+  control to the model — a returned tool, an injected context, a run command,
+  the user's own message — never to the start of the turn, so a wait that has
+  just begun does not inherit the minutes the tools already spent.
+- Past ten seconds the readout adds a「暂未响应」badge, and the moment the model
+  produces its first block the whole indicator is withdrawn. A tool that is
+  still running is not a wait: the tool is the one working.
+- A command that ran long keeps its duration after it returns (warning colour
+  past ten seconds); a fast one shows nothing once it is done. Running steps
+  show a live seconds readout and a shimmering summary while they work.
+
+### Scroll
+
+- Tail-follow only detaches on a real upward move by the reader. Content growth
+  and the follow easing itself also move `scrollTop`, and reading that as "the
+  user left the bottom" froze following mid-turn.
+- Disable browser scroll anchoring on the conversation scroller while the reader
+  is mounted: it moved the viewport on its own as the transcript grew.
+- Only a focused text field suspends following, and only inside the reader.
+  Focusing the composer used to stop the transcript from advancing.
+- Stream at the source's own rate: add a feed-forward term estimated from the
+  arrival rate on top of the proportional controller. A purely proportional
+  reveal settles at a constant `catchUpMs` of lag regardless of model speed,
+  which read as 「慢」 while the backlog grew with faster models; the feed-forward
+  term drains the backlog and the proportional term absorbs transport jitter.
+- Reveal a batch of words on one clock instead of one word per fixed gap.
+  Per-word spacing pinned the rate near 16 words/second, so a fast model still
+  arrived one word at a time; a batch now lands inside one short window while a
+  lone new word keeps the original typing cadence.
+- The reasoning pane's own follow steps with the backlog rather than at a fixed
+  step-and-hold: it used to advance two lines every ~1.3 seconds no matter how
+  fast the text arrived, which is what made the pane look like it was lagging
+  behind a fast model.
+### Host DOM hooks
 
-- Add a Better Display settings section: deliverables still open in the system app by default, with an optional right-Sidebar preview, plus generative-mcpapps skill-root detection and install guidance.
-- Guard pending-submission image echoes so a text-only send cannot crash `conversation.view` (`images` / `attachments` may be missing).
-- Install guidance names only conventional relative roots (`.dsh/skills`, `.agents/skills`) and never prints host home or plugin pack absolutes.
 - Keep ChatView's `data-chat-flow=""` hook on the Reader column so skins that hide `[data-composer-seat]` when the scrollport has no chat-flow (maid-atelier, phoebe-atelier, and others) still show the composer in reading view.
 
 ## 0.1.1 — 2026-09-16

package/README.en.md

@@ -2,25 +2,41 @@
 
 [中文](./README.md)
 
+Prefer npm (version-pinable):
+
+```sh
+dsh plugin --profile web add dsh-better-display@0.3.0
+```
+
+Or latest:
+
+```sh
+dsh plugin --profile web add dsh-better-display
+```
+
+Fallback: install from GitHub (tracks the default branch tip):
+
 ```sh
 dsh plugin --profile web add github:aa2246740/dsh-better-display
 ```
 
-You need official `dsh` (or `npx @deepseek-ai/dsh`) and **pnpm** on PATH. `dsh plugin add` runs pnpm in `$DSH_HOME/profiles/web`. This repo commits built `lib/`, so a git install does not need `prepare` or a profile `allowBuilds` entry.
+You need official `dsh` (or `npx @deepseek-ai/dsh`) and **pnpm** on PATH. `dsh plugin add` runs pnpm in `$DSH_HOME/profiles/web`. This repo commits built `lib/`, so git / npm installs do not need `prepare` or a profile `allowBuilds` entry.
 
 Then restart that Host and reload the page. `dsh plugin add` writes the profile. It does not hot-load a running process.
 
 Adds a **阅读** tab to DeepSeek Harness. While a turn runs you see steps, thinking, and progress. After a successful turn those collapse and the final answer stays. Native Chat / Trajectory, the composer, model picker, tools, and approvals stay. The reading column keeps ChatView's `data-chat-flow` hook so third-party skins that gate the composer on that mark still treat Reader as an interactive conversation.
 
-A ````mcp-app` fence in the final answer mounts as an interactive card in the reading view, inside `<iframe sandbox="allow-scripts allow-forms">` without `allow-same-origin`. The card can fill the next prompt via JSON-RPC. The skill pack is [`skills/generative-mcpapps/`](skills/generative-mcpapps/). Settings → **Better Display** can preview deliverables in the right Sidebar (system app remains the default) and reports whether that skill is installed in a harness skill root.
+A ````mcp-app` fence in the final answer mounts as an interactive card in the reading view, inside `<iframe sandbox="allow-scripts allow-forms">` without `allow-same-origin`. The card can fill the next prompt via JSON-RPC. The skill pack is [`skills/generative-mcpapps/`](skills/generative-mcpapps/). Settings → **Better Display** can preview deliverables in the right Sidebar (system app remains the default), turn on translucent frosted glass (off by default), toggle process auto-folding (On is the default), and reports whether that skill is installed in a harness skill root.
 
 Targets DeepSeek Harness **0.1.5-rc.2**. Display only. It does not change Agent execution, the SDK, or credentials. Node.js `^22.19.0 || >=24`. New sessions default to reading.
 
+**0.3.0** keeps the existing reading layout, folding, and motion while using official feedback, tool details, file cards, and file links. It also fixes process content staying expanded after auto-folding is re-enabled. See the [official integration notes](docs/official-rendering-bridge.md) for coverage and upgrade checks.
+
 From a local checkout or tarball:
 
 ```sh
 dsh plugin --profile web add ./dsh-better-display
-dsh plugin --profile web add ./dsh-better-display-0.1.1.tgz
+dsh plugin --profile web add ./dsh-better-display-0.3.0.tgz
 ```
 
 `dsh.bundle` is captured at Host boot. Do not also insert the same row by hand in the profile `cordis.patch.yml`, or it will mount twice.

package/README.md

@@ -2,25 +2,41 @@
 
 [English](./README.en.md)
 
+推荐用 npm（可钉版本）：
+
+```sh
+dsh plugin --profile web add dsh-better-display@0.3.0
+```
+
+也可以装 latest：
+
+```sh
+dsh plugin --profile web add dsh-better-display
+```
+
+备选：从 GitHub 直装（跟默认分支最新提交）：
+
 ```sh
 dsh plugin --profile web add github:aa2246740/dsh-better-display
 ```
 
-PATH 上要有官方 `dsh`（没有就用 `npx @deepseek-ai/dsh`）和 **pnpm**。`dsh plugin add` 会在 `$DSH_HOME/profiles/web` 里跑 pnpm。仓库已提交编译好的 `lib/`，git 安装不用 `prepare`，也不用改 profile 的 `allowBuilds`。
+PATH 上要有官方 `dsh`（没有就用 `npx @deepseek-ai/dsh`）和 **pnpm**。`dsh plugin add` 会在 `$DSH_HOME/profiles/web` 里跑 pnpm。仓库已提交编译好的 `lib/`，git / npm 安装都不用 `prepare`，也不用改 profile 的 `allowBuilds`。
 
 然后重启这个 Host，再刷新页面。`dsh plugin add` 只写 profile，不会热挂正在跑的进程。
 
 给 DeepSeek Harness 加一个 **阅读** 页签：执行时能看到步骤、思考和进度；整轮成功结束后把过程收起来，留下最终回答。原版「对话 / 轨迹」、输入框、模型选择、工具和审批都还在。阅读列保留宿主 ChatView 的 `data-chat-flow` 钩子，依赖该标记显示输入框的第三方皮肤不会把阅读页当成仅检视视图。
 
-最终回答里的 ````mcp-app` 代码块会在阅读视图里挂成交互卡片，跑在 `<iframe sandbox="allow-scripts allow-forms">` 里，没有 `allow-same-origin`。卡片可以通过 JSON-RPC 把下一轮 prompt 填进输入框。技能包在 [`skills/generative-mcpapps/`](skills/generative-mcpapps/)。设置里的 **Better Display** 可把产物改为右侧栏预览（默认仍用系统应用），并检测该技能是否已装进宿主技能目录。
+最终回答里的 ````mcp-app` 代码块会在阅读视图里挂成交互卡片，跑在 `<iframe sandbox="allow-scripts allow-forms">` 里，没有 `allow-same-origin`。卡片可以通过 JSON-RPC 把下一轮 prompt 填进输入框。技能包在 [`skills/generative-mcpapps/`](skills/generative-mcpapps/)。设置里的 **Better Display** 可把产物改为右侧栏预览（默认仍用系统应用），打开半透明毛玻璃（默认关），设置过程自动折叠开关（默认开），并检测该技能是否已装进宿主技能目录。
 
 面向 DeepSeek Harness **0.1.5-rc.2**。只改展示，不改 Agent 执行、SDK 或模型凭据。Node.js `^22.19.0 || >=24`。新会话默认进阅读。
 
+**0.3.0** 保留现有阅读布局、折叠和动效，接入官方反馈、工具详情、文件卡片及文件链接；并修复重新开启自动折叠后仍保持展开的问题。接入范围和升级检查见 [官方能力接入说明](docs/official-rendering-bridge.md)。
+
 本地目录或 tarball：
 
 ```sh
 dsh plugin --profile web add ./dsh-better-display
-dsh plugin --profile web add ./dsh-better-display-0.1.1.tgz
+dsh plugin --profile web add ./dsh-better-display-0.3.0.tgz
 ```
 
 `dsh.bundle` 是开机捕获的。不要再往 profile 的 `cordis.patch.yml` 手写同一条 insert，会重复挂载。

package/compat/harness-rc2.json

@@ -0,0 +1,59 @@
+{
+  "harnessVersion": "0.1.5-rc.2",
+  "hashes": {
+    "packages/client/ui-slots/src/index.ts": "07b7aec2248b6ee6b56626f7aaba5dbc97cf765f1967dc7e60d6cef07ba7c81e",
+    "packages/client/ui-slots/src/renderer.ts": "66201b2ded3ab3ea133881de76737b3a0253f94f59b84594462109f052c63b19",
+    "packages/client/ui-renderer/src/client/registry.ts": "46442dfe72c837296f9f827c681dfba55ebfe3549a3895975e945511a1fe9c25",
+    "packages/client/ui-renderer/src/client/scoped-slots.tsx": "bcdd61674a52fae843ef21b258a8a8568a49e310217cda187c73159e925fa8a0",
+    "packages/client/ui-chat/src/client/contract/slots.ts": "66deb5545891ab05cb8b7a3b43139d96a9b90271e64b7d32fadc229337ede589",
+    "packages/client/ui-tool/src/client/contract/slots.ts": "9d0bba9e047864e4c31842e8a0010de383075e12c74e7c962fc6cffdc54221bf",
+    "packages/client/ui-deliverables/src/client/Deliverables.tsx": "d599330309bd4973bdb14d4be953b5bf3113c5e4bde3ac854738fe1fce7e4991",
+    "packages/client/ui-chat/src/client/chat/AssistantMarkdown.tsx": "ad25664afa92be5436ce354b3bef37201187a1034898567cb120ca3bd34d51df",
+    "packages/client/ui-chat/src/client/chat/MessageItem.tsx": "75e0bea127342d2f4185b6d0f89d74d0c8bdd01858be85b4cc859f7596b073d6",
+    "packages/client/ui-chat/src/client/chat/AssistantNodeView.tsx": "315e5112b78a3da86417f55778cf26a204acd0eee984a762dcc87bd900f2e718",
+    "packages/client/ui-chat/src/client/chat/TurnTailNodeView.tsx": "6a0167ec521e64e58c3c5588078dd3a4e7ebe55309e276b04271b303f82ba96b",
+    "packages/client/ui-chat/src/client/chat/TurnProcessNodeView.tsx": "d397f1e92f76585fa49b924b3d43d829ad503a4ffa821660234d4b51e5cc743f",
+    "packages/client/ui-tool/src/client/tool/models/tool-call-model.ts": "dd5033d26a6ed147b70e310506d311d55ccac6a54d08a7e24834d232226a06a3",
+    "packages/client/ui-primitives/src/markdown/MarkdownText.tsx": "d8c6aa23232aa4a76339c49c02df5888f5530a7a1bbb417ef14a42a1e41d5d94",
+    "packages/client/ui-primitives/src/markdown/render.tsx": "3576c16a37846a55f215562d15df9394cf5761fc46eae8f89fdcea8420f09d3d",
+    "packages/client/ui-primitives/src/markdown/parse.ts": "6be57559995c4e717d6d704207eb2fec86db9714c2a4bcc5b0c3f6648c8990f0",
+    "packages/client/ui-primitives/src/markdown/incremental.ts": "eaa266bb77dfb646738d0fdd79be4f1c4bc5490f7bf3f4127f9ffb72ef741ff8",
+    "packages/client/ui-primitives/src/markdown/katex.tsx": "8227e04d29333d2568021ed012143237b457852bce25d82cf394d72b941e74a5"
+  },
+  "registrations": [
+    "conversation.chat.assistant-actions | feedback | packages/client/ui-message-feedback/src/client/index.ts",
+    "conversation.chat.node | assistant-step | packages/client/ui-chat/src/client/chat/register-node-renderers.ts",
+    "conversation.chat.node | command | packages/client/ui-chat/src/client/chat/register-node-renderers.ts",
+    "conversation.chat.node | command-input | packages/client/ui-goal/src/client/index.ts",
+    "conversation.chat.node | compaction | packages/client/ui-chat/src/client/chat/register-node-renderers.ts",
+    "conversation.chat.node | context | packages/client/ui-chat/src/client/chat/register-node-renderers.ts",
+    "conversation.chat.node | manual-compaction | packages/client/ui-chat/src/client/chat/register-node-renderers.ts",
+    "conversation.chat.node | model-retry | packages/client/ui-chat/src/client/chat/register-node-renderers.ts",
+    "conversation.chat.node | steering | packages/client/ui-chat/src/client/chat/register-node-renderers.ts",
+    "conversation.chat.node | system-prompt | packages/client/ui-chat/src/client/chat/register-node-renderers.ts",
+    "conversation.chat.node | tool-call | packages/client/ui-tool/src/client/apply.ts",
+    "conversation.chat.node | turn-error | packages/client/ui-chat/src/client/chat/register-node-renderers.ts",
+    "conversation.chat.node | turn-max-tokens | packages/client/ui-chat/src/client/chat/register-node-renderers.ts",
+    "conversation.chat.node | turn-process | packages/client/ui-chat/src/client/chat/register-node-renderers.ts",
+    "conversation.chat.node | turn-tail | packages/client/ui-chat/src/client/chat/register-node-renderers.ts",
+    "conversation.chat.node | unknown | packages/client/ui-chat/src/client/chat/register-node-renderers.ts",
+    "conversation.chat.node | user | packages/client/ui-chat/src/client/chat/register-node-renderers.ts",
+    "conversation.chat.node | workflow-run | packages/client/ui-workflow-run/src/client/index.ts",
+    "conversation.chat.turnTail | (chain/single) | packages/client/ui-deliverables/src/client/index.ts",
+    "conversation.message.images | (chain/single) | packages/client/ui-attachment/src/client/index.ts",
+    "tool.call.images | (chain/single) | packages/client/ui-attachment/src/client/index.ts",
+    "tool.call.toolview | ask_user_question | packages/client/ui-tool/src/client/tool/toolviews/ask-question-row.tsx",
+    "tool.call.toolview | bash | packages/client/ui-tool/src/client/tool/toolviews/bash-sample.tsx",
+    "tool.call.toolview | edit | packages/client/ui-tool/src/client/tool/toolviews/file-mutation-row.tsx",
+    "tool.call.toolview | glob | packages/client/ui-tool/src/client/tool/toolviews/search-row.tsx",
+    "tool.call.toolview | grep | packages/client/ui-tool/src/client/tool/toolviews/search-row.tsx",
+    "tool.call.toolview | present | packages/client/ui-deliverables/src/client/index.ts",
+    "tool.call.toolview | read | packages/client/ui-tool/src/client/tool/toolviews/read-row.tsx",
+    "tool.call.toolview | read_image | packages/client/ui-tool/src/client/tool/toolviews/read-image-row.tsx",
+    "tool.call.toolview | skill | packages/client/ui-skill/src/client/index.ts",
+    "tool.call.toolview | todo_write | packages/client/ui-tool/src/client/tool/toolviews/todo-row.tsx",
+    "tool.call.toolview | web_fetch | packages/client/ui-tool/src/client/tool/toolviews/web-row.tsx",
+    "tool.call.toolview | web_search | packages/client/ui-tool/src/client/tool/toolviews/web-row.tsx",
+    "tool.call.toolview | write | packages/client/ui-tool/src/client/tool/toolviews/file-mutation-row.tsx"
+  ]
+}

package/docs/official-rendering-bridge.md

@@ -0,0 +1,65 @@
+# 官方能力接入说明
+
+`0.3.0` 在 `v0.2.1` 的阅读布局上接入官方内容和控件，经过试用验收后发布。当前验收目标是 Harness `0.1.5-rc.2`。后续仍需维护的自有展示逻辑和升级边界见下文。
+
+## 一个具体例子
+
+一条回答的持久化 `messageId` 进入 `OfficialActions`，随后进入 Reader 自己的回答操作插槽。里面的反馈组件、状态注入和弹窗控制器来自当前已加载的官方反馈插件。点击按钮后，由官方控制器打开原来的官方弹窗；Reader 不实现反馈请求、类别或状态管理。
+
+工具详情同理：传递 `callId`、原始 `block`、工作目录、Host home、图片加载器、带行号的文件打开操作和轨迹入口，再由官方工具组件呈现。原来只取函数组件、手工传四个参数并模拟点击展开的路径已经移除。
+
+## 代码归属
+
+| 模块 | 负责什么 |
+| --- | --- |
+| `official-slots.tsx` | 用公开注册表 API 将官方注册信息映射到 Reader 的独立插槽；保留注入、locale、store、selector 和子插槽。监听贡献增删，逐条挂载和释放。 |
+| `OfficialContent.tsx` | 将回答身份、工具、节点和产物上下文传给官方控件。这里只组合，不处理业务请求。 |
+| `Reader.tsx`、`ToolActivity.tsx` | 原有阅读布局、过程组织、摘要和交互。工具详情里的控件采用官方形式。 |
+| `platform-media.ts` | RC2 的本地媒体 URL 约定；读取授权仍由 Host 文件 API 负责。 |
+| `compat/harness-rc2.json` | 已验收的官方接口、被自有展示覆盖的实现和注册清单基线。 |
+
+## 为什么有插槽映射
+
+RC2 的一个子插槽只能有一个声明者，只有声明它的组件能获得相应 `renderSlot` 权限。Reader 不能把官方插槽再声明一次，也不能从 `ctx.slots` 任意渲染子插槽。
+
+本适配用 `spec`、`entriesOfSlot`、`register`、`inject`、`subscribe` 建立 Reader 自己的名字空间，交给同一个官方渲染器运行。组件声明的子插槽名称在一层薄包装中转换，其余属性原样传递。没有私有源码导入，没有修改官方注册对象或官方 DOM，没有复制官方渲染器。现有控件保持组件和注入身份；增加别的控件不会重挂它们。
+
+这仍是 RC2 的受限适配方案。若官方将来提供可直接组合的共享渲染入口，应优先替换这层映射。
+
+## 这版覆盖的能力
+
+| 能力 | 处理方式 |
+| --- | --- |
+| 回答操作、反馈 | 官方 `conversation.chat.assistant-actions`，使用持久化 messageId；中断的临时片段不伪造身份。 |
+| 工具详情 | 官方 `tool.call.toolview`，包括 memo、注入状态、翻译和递归子插槽；未注册的工具保留原有结果预览。 |
+| 显式 `present` 产物 | 官方 turnTail 中的文件卡片和菜单，保留官方状态和打开操作。 |
+| 已修改文件快捷操作 | 保持原 Reader 的复制、定位和打开模式；仅从 Reader 的官方 tail 展示参数中排除重复的 produced chips，不改变官方源数据。未知匹配结构原样呈现。 |
+| 文件行号和正文文件链接 | 保留工具传来的行号；正文优先采用官方 file-mention 服务，现有路径匹配作为补充。 |
+| 本地 Markdown 图片 | 和 RC2 相同的同源文件 API，加载失败保留替代文本，危险协议不直接渲染。 |
+| 命令、新节点类型 | 接入官方节点渲染及其子插槽；无可用渲染器时保留记录并提供“在对话中查看”。 |
+
+阅读布局、折叠、逐字动效、阅读速度、默认设置保持现版。移除了未被使用的全局 `body.dataset.readerFolding` 写入。
+
+## 验证与升级
+
+在选定 Harness 运行时上执行：
+
+```sh
+node scripts/link-harness-dependencies.mjs "$DSHX_HARNESS"
+npm run build
+npm test
+npm run test:official
+npm run test:auto-fold
+```
+
+`test:official` 使用测试目标里的真实官方 SlotRegistry、React renderer、反馈插件、ReadRow 和 Deliverables，只有远程业务服务使用内存替身。测试私有源码仅用于验证，不进入交付 bundle。浏览器验证使用本机固定的 Codex Playwright runtime，不附着用户会话。
+
+`test:auto-fold` 使用真实 Reader 和独立的全局、会话 store，验证关闭后阅读再开启、此前步骤、设置同步、文本选择及连续切换；分别覆盖普通动画和减少动态效果。
+
+`build` 先检查兼容基线。Harness 版本、敏感接口、尚未收敛的官方实现或相关注册清单改变时，构建明确要求审查。审查差异并通过功能与视觉测试后，才可运行 `check-harness-compat.mjs --record` 记录新基线。记录本身不构成兼容证明。
+
+## 后续仍需完成
+
+Markdown 定制仍保留逐字动效和 MCP 展示钩子，官方公共组件还没有对应的注入接口。上下文行、部分过程展示和工具摘要也仍有自有实现。本分支先建立官方能力的完整运行路径，后续按视觉基线逐模块收敛这些副本，并拆分 Reader 的阅读策略与动画执行。不能据此承诺所有未来版本零维护，也不承诺任意第三方插件兼容。
+
+已有客户端通过 HMR 更新时，无需新增 Host 或改写会话数据；应先核对实际安装来源、构建产物和同页行为。正式版本保留 Git 标签及发布包，便于按版本安装和回退。