docs(spec): add sheet-history-revert spec (history list / revert / revert-status)

3 new lark-sheets shortcuts via existing ToolsCall framework:
- +history-list, +history-revert, +history-revert-status
- async revert: sheet/data RecoverHistory → RecoverMsg(MQ) → agg consumer
  assigns memberId by scene (lark-cli=11 / doubao=10)
- scene threaded via ctx baggage; no thrift changes (codegen all-no)
- spans larksuite/cli, ee/sheet-skill-spec, ee/sheet-facade-agg, sheet/data

Force-added under docs/ (repo-wide gitignored) so ccm-harness drift
detection has the git-tracked SSOT.

AGENTS.md

@@ -141,3 +141,74 @@ CLI arguments are untrusted (they come from AI agents). Call `validate.SafeInput
 | Modify shortcut flags/params | Required | If behavior changes |
 | Shortcut bug fix | Required | If regression risk |
 | Internal refactor (no shortcut impact) | Not needed | Not needed |
+
+
+
+## CCM Harness Skill Routing (v2)
+
+When the user's request matches one of these patterns, invoke the corresponding `/ccm-harness:*` skill via the Skill tool. Skills include multi-step workflows, gates, and quality checks that produce more reliable results than ad-hoc answers. When in doubt, invoke the skill — a false positive is cheaper than a false negative.
+
+### 主链路（spec → idl → dev → release）
+
+| 用户表达 | Skill |
+|---------|-------|
+| "新需求 / 这是个新功能 / 开始一个 req / 写 spec / 起 spec / PRD 来了 / 把 PRD 转 spec / 改 spec / 微调 spec / spec 局部更新 / spec 加一个字段" | `/ccm-harness:draft-spec <req-id> [<arg2>] [--force]`（Phase 0 路由器自动决定 init / generate / update capability）|
+| "review spec / 看 spec / 评审 spec / spec 评审" | `/ccm-harness:spec-review <req-id>` |
+| "生成 thrift / 起 idl / 把 spec 转 thrift" | `/ccm-harness:draft-idl` |
+| "推 thrift / 落 contract / Frozen Spec / codegen / 生成框架代码" | `/ccm-harness:codegen-idl <req-id>` |
+| "实现 spec / 写后端 / 后端开发 / 实现这个功能" | `/ccm-harness:backend-dev <req-id>` |
+| "前端怎么改 / 写前端 / 前端开发 / 前端编码" | `/ccm-harness:frontend-coding <req-id>` |
+| "部署 BOE / 上 PPE / 部署到 feature 环境" | `/ccm-harness:deploy <req-id>` |
+| "提 release / 上线 / 发布 / 走 PRE-GRAY-ONLINE" | `/ccm-harness:release <req-id>` |
+| "触发打包 / build / 起 SCM 编译" | `/ccm-harness:build <repo or psm>` |
+
+### 守护与诊断
+
+| 用户表达 | Skill |
+|---------|-------|
+| "工作流到哪一步了 / 下一步咋走 / 我迷路了" | `/ccm-harness:doctor` |
+| "调试 / debug / 这个 bug 怎么排" | `/ccm-harness:debug` |
+| "查 CI 失败 / CI 跑挂了 / pipeline 红了" | `/ccm-harness:check-ci-failure <mr>` |
+| "检查实现是否符合 spec / 蓝图对比" | `/ccm-harness:check-impl-gap` |
+| "spec 跟代码飘了吗 / drift 检查" | `/ccm-harness:spec-review <req-id> --mode drift` |
+| "Hub 知识跟代码一致吗" | `/ccm-harness:check-knowledge-consistency` |
+| "代码 review / 看 MR / cr 一下" | `/ccm-harness:code-review <mr>` |
+| "设计 review / 看技术方案" | `/ccm-harness:design-review <doc>` |
+
+### 工程辅助
+
+| 用户表达 | Skill |
+|---------|-------|
+| "通知 reviewer / 发 review 卡片" | `/ccm-harness:notify-reviewer <mr>` |
+| "自动修 MR / 按 review 改" | `/ccm-harness:autofix-mr <mr>` |
+| "生成测试用例 / 起 case / 筛选可执行 case / E2E 用例 / Playwright 脚本 / 自动化验收" | `/ccm-harness:test`（路由到 `ccm-e2e-check -> exec-e2e`） |
+| "查 idl / 看 thrift 定义" | `/ccm-harness:lookup-idl <psm>` |
+| "看仓库最近改了啥 / 仓库脉搏" | `/ccm-harness:pulse` |
+
+### 元能力（Skill / Prompt 研发）
+
+| 用户表达 | Skill |
+|---------|-------|
+| "起一个新 skill / 设计 skill" | `/ccm-harness:meta-draft-skill` |
+| "做评测集 / 给 skill 出评测数据" | `/ccm-harness:meta-build-evalset` |
+| "跑评测 / Fornax 实验" | `/ccm-harness:meta-run-eval` |
+| "优化 skill / skill 反馈优化" | `/ccm-harness:meta-optimize-skill` |
+
+### 环境与配置
+
+| 用户表达 | Skill |
+|---------|-------|
+| "升级 ccm-harness / 更新插件" | `/ccm-harness:upgrade` |
+| "看遥测 / 最近的反馈" | `/ccm-harness:show-telemetry` |
+| "清遥测 / 重置 telemetry" | `/ccm-harness:clear-telemetry` |
+| "上报问题 / 提 issue" | `/ccm-harness:report-issue` |
+| "看本地教训 / project learnings / 我们踩过啥" | `/ccm-harness:learn` |
+| "反馈 / 评分这次 skill" | `/ccm-harness:feedback` |
+
+### 使用提示
+
+- **入口**：新需求**必从** `/ccm-harness:draft-spec <req-id>` 开始（Phase 0 路由器自动决定建目录 / 带 PRD 一气呵成）。
+- **不跳步**：spec → idl → dev → release 是流水线，不是菜单——按顺序推进，反向走需要 `/ccm-harness:draft-spec <req-id> "<change-desc>"` 局部修（路由进 update capability）。
+- **横切**：任何阶段发现 spec 要局部改 → `/ccm-harness:draft-spec <req-id> "<change-desc>"`；想检测漂移 → `/ccm-harness:spec-review <req-id> --mode drift`；CI 红 → `/ccm-harness:check-ci-failure`；迷路 → `/ccm-harness:doctor`。
+
+完整流程文档：`docs/user-guide/workflow.md`

CLAUDE.md

@@ -0,0 +1,69 @@
+
+## CCM Harness Skill Routing (v2)
+
+When the user's request matches one of these patterns, invoke the corresponding `/ccm-harness:*` skill via the Skill tool. Skills include multi-step workflows, gates, and quality checks that produce more reliable results than ad-hoc answers. When in doubt, invoke the skill — a false positive is cheaper than a false negative.
+
+### 主链路（spec → idl → dev → release）
+
+| 用户表达 | Skill |
+|---------|-------|
+| "新需求 / 这是个新功能 / 开始一个 req / 写 spec / 起 spec / PRD 来了 / 把 PRD 转 spec / 改 spec / 微调 spec / spec 局部更新 / spec 加一个字段" | `/ccm-harness:draft-spec <req-id> [<arg2>] [--force]`（Phase 0 路由器自动决定 init / generate / update capability）|
+| "review spec / 看 spec / 评审 spec / spec 评审" | `/ccm-harness:spec-review <req-id>` |
+| "生成 thrift / 起 idl / 把 spec 转 thrift" | `/ccm-harness:draft-idl` |
+| "推 thrift / 落 contract / Frozen Spec / codegen / 生成框架代码" | `/ccm-harness:codegen-idl <req-id>` |
+| "实现 spec / 写后端 / 后端开发 / 实现这个功能" | `/ccm-harness:backend-dev <req-id>` |
+| "前端怎么改 / 写前端 / 前端开发 / 前端编码" | `/ccm-harness:frontend-coding <req-id>` |
+| "部署 BOE / 上 PPE / 部署到 feature 环境" | `/ccm-harness:deploy <req-id>` |
+| "提 release / 上线 / 发布 / 走 PRE-GRAY-ONLINE" | `/ccm-harness:release <req-id>` |
+| "触发打包 / build / 起 SCM 编译" | `/ccm-harness:build <repo or psm>` |
+
+### 守护与诊断
+
+| 用户表达 | Skill |
+|---------|-------|
+| "工作流到哪一步了 / 下一步咋走 / 我迷路了" | `/ccm-harness:doctor` |
+| "调试 / debug / 这个 bug 怎么排" | `/ccm-harness:debug` |
+| "查 CI 失败 / CI 跑挂了 / pipeline 红了" | `/ccm-harness:check-ci-failure <mr>` |
+| "检查实现是否符合 spec / 蓝图对比" | `/ccm-harness:check-impl-gap` |
+| "spec 跟代码飘了吗 / drift 检查" | `/ccm-harness:spec-review <req-id> --mode drift` |
+| "Hub 知识跟代码一致吗" | `/ccm-harness:check-knowledge-consistency` |
+| "代码 review / 看 MR / cr 一下" | `/ccm-harness:code-review <mr>` |
+| "设计 review / 看技术方案" | `/ccm-harness:design-review <doc>` |
+
+### 工程辅助
+
+| 用户表达 | Skill |
+|---------|-------|
+| "通知 reviewer / 发 review 卡片" | `/ccm-harness:notify-reviewer <mr>` |
+| "自动修 MR / 按 review 改" | `/ccm-harness:autofix-mr <mr>` |
+| "生成测试用例 / 起 case / 筛选可执行 case / E2E 用例 / Playwright 脚本 / 自动化验收" | `/ccm-harness:test`（路由到 `ccm-e2e-check -> exec-e2e`） |
+| "查 idl / 看 thrift 定义" | `/ccm-harness:lookup-idl <psm>` |
+| "看仓库最近改了啥 / 仓库脉搏" | `/ccm-harness:pulse` |
+
+### 元能力（Skill / Prompt 研发）
+
+| 用户表达 | Skill |
+|---------|-------|
+| "起一个新 skill / 设计 skill" | `/ccm-harness:meta-draft-skill` |
+| "做评测集 / 给 skill 出评测数据" | `/ccm-harness:meta-build-evalset` |
+| "跑评测 / Fornax 实验" | `/ccm-harness:meta-run-eval` |
+| "优化 skill / skill 反馈优化" | `/ccm-harness:meta-optimize-skill` |
+
+### 环境与配置
+
+| 用户表达 | Skill |
+|---------|-------|
+| "升级 ccm-harness / 更新插件" | `/ccm-harness:upgrade` |
+| "看遥测 / 最近的反馈" | `/ccm-harness:show-telemetry` |
+| "清遥测 / 重置 telemetry" | `/ccm-harness:clear-telemetry` |
+| "上报问题 / 提 issue" | `/ccm-harness:report-issue` |
+| "看本地教训 / project learnings / 我们踩过啥" | `/ccm-harness:learn` |
+| "反馈 / 评分这次 skill" | `/ccm-harness:feedback` |
+
+### 使用提示
+
+- **入口**：新需求**必从** `/ccm-harness:draft-spec <req-id>` 开始（Phase 0 路由器自动决定建目录 / 带 PRD 一气呵成）。
+- **不跳步**：spec → idl → dev → release 是流水线，不是菜单——按顺序推进，反向走需要 `/ccm-harness:draft-spec <req-id> "<change-desc>"` 局部修（路由进 update capability）。
+- **横切**：任何阶段发现 spec 要局部改 → `/ccm-harness:draft-spec <req-id> "<change-desc>"`；想检测漂移 → `/ccm-harness:spec-review <req-id> --mode drift`；CI 红 → `/ccm-harness:check-ci-failure`；迷路 → `/ccm-harness:doctor`。
+
+完整流程文档：`docs/user-guide/workflow.md`

docs/specs/sheet-history-revert/spec.md

@@ -0,0 +1,305 @@
+---
+req_id: sheet-history-revert
+mode: from-prd
+tracks: [be]
+created_at: 2026-06-23T13:09:08Z
+---
+
+> ⚠️ **请勿直接编辑此文档**
+> 修改请通过 `/ccm-harness:draft-spec sheet-history-revert "<变更描述>"`（路由器会进入 update capability）
+> 手改不会被 check-spec-drift 放过，且会破坏 Frozen 快照关联性
+> 本文档由 /ccm-harness:draft-spec 生成于 2026-06-23T13:09:08Z；req-id：sheet-history-revert
+
+---
+
+# Sheet 历史版本查询与回滚
+
+<!-- BEGIN generated overview — scripts/render_spec_overview.py 自动生成，勿手改；改 spec 请走 /ccm-harness:draft-spec <req-id> "<变更>"（update） -->
+## 方案概览 / TL;DR
+
+> 本段由机器从下方子任务字段**确定性生成**，供快速把握方案；**权威细节仍以各子任务 `yaml` 块为准**。
+
+- **需求** `sheet-history-revert` ｜ 模式 `from-prd` ｜ 端 BE
+- **规模** 5 BE ｜ **涉及 PSM** `tooling.lark_cli`, `tooling.sheet_skill_spec`, `sheet.facade.agg`, `bear.server.sheet_data` ｜ **Thrift 影响** 无 ×5
+
+| 子任务 | 一句话 | 关键信息 |
+|---|---|---|
+| BE-1 | lark-cli `+history-list` 历史记录列表 shortcut | `tooling.lark_cli` · thrift:无 · `history_list[read]（callTool tools/invoke_read；内部对接 /space/api/v3/sheet/histories）` |
+| BE-2 | lark-cli `+history-revert` 与 `+history-revert-status` 回滚流 shortcut | `tooling.lark_cli` · thrift:无 · `history_revert[write] / history_revert_status[read]（callTool tools/invoke_write\|read；内部对接 /space/api/v2/sheet/recover、/api/v2/sheet/recover/status）` |
+| BE-3 | sheet-skill-spec 上游事实源（skill 正文 + shortcut/flag 定义） | `tooling.sheet_skill_spec` · thrift:无 · `A` |
+| BE-4 | sheet-facade-agg 在现有 ToolsCall 上新增 3 个历史/回滚工具 | `sheet.facade.agg` · thrift:无 · `history_list[read] / history_revert[write] / history_revert_status[read]` |
+| BE-5 | sheet/data 透传 scene 并在 RecoverMsg 上新增 Scene 字段 | `bear.server.sheet_data` · thrift:无 · `RecoverHistory / QueryRecoverStatus（复用）；MQ RecoverMsg 加 Scene 字段` |
+
+> 完整字段见各子任务 `yaml` 块。
+<!-- END generated overview -->
+
+## 概述与范围
+
+为 `lark-cli` 的 lark-sheets 能力补齐**电子表格历史版本查询与回滚**，新增 3 个 shortcut，让 AI / 用户可以列出某张表的历史版本、回滚到指定版本、并查询回滚的异步状态。三个 shortcut 封装的是飞书电子表格已上线的 space 接口（见下表的「前端接口参考」），本需求不新建产品能力，只是把它们包装成稳定、AI 友好的命令面。
+
+| 功能 | shortcut | 前端接口参考（搜索 lark/idl） | 实现差异点 |
+|---|---|---|---|
+| 查历史记录列表 | `+history-list` | `/space/api/v3/sheet/histories` | ① 仅返回 `minor_histories` 列表；② `minor_histories` 的 `id` 字段重命名为 `history_version_id`；③ 每条仅保留 `history_version_id` / `create_time`（序列化成 AI 优化的可读格式）/ `action` / `all_block_revision` 四个字段 |
+| 历史记录回滚 | `+history-revert` | `/space/api/v2/sheet/recover` | 传入 `+history-list` 拿到的 `history_version_id`，回滚到指定版本 |
+| 查询回滚状态 | `+history-revert-status` | `/api/v2/sheet/recover/status` | 查询 `+history-revert` 发起的异步回滚的当前状态 |
+
+**范围内**：
+- `larksuite/cli` 仓新增 3 个 sheets shortcut（含 `+history-list` 的响应裁剪 / 字段重命名 / 时间格式转换逻辑）。
+- `ee/sheet-skill-spec` 上游事实源补 skill 正文与 shortcut/flag 定义，经其工作流 `sync:cli` 同步到 `larksuite/cli` 的 `skills/lark-sheets/` 与 `shortcuts/sheets/data/`。
+- `ee/sheet-facade-agg` **复用现有 `ToolsCall` 接口**，在其上新增 3 个工具（`history_list` / `history_revert` / `history_revert_status`）；并在**回滚消息消费侧**（消费 `sheet/data` 产出的 RecoverMsg）按 scene 给 `memberId` 赋值（doubao=10，lark-cli=11）后构造 recover cs。
+- `sheet/data`（`bear.server.sheet_data`）把 scene 从入口透传到 `RecoverHistory`，并在产出的 `RecoverMsg`（MQ 消息）上**新增 `Scene` 字段**，供 agg 消费时区分场景。
+
+**范围外**：
+- 历史版本的底层存储 / 快照 / 过期清理逻辑、`RecoverHistory` 的回滚业务语义（`bear.server.sheet_data` 已有，本需求仅透传 scene + 加 MQ 字段，不改回滚逻辑本身）。
+- 历史版本 diff、可视化、权限模型变更等产品侧扩展。
+- `doubao-office` 消费方的同步（`sheet-skill-spec` 另有 `sync:doubao`，本需求不涉及）。
+
+## 服务拓扑与 PSM 变更判定表
+
+**调用链路（目标形态）**：
+
+```
+用户 / AI ──> lark-cli (+history-list / +history-revert / +history-revert-status)
+                   │  callTool → POST /open-apis/sheet_ai/v2/.../tools/invoke_read|write（scene 随入口确定）
+                   ▼
+        sheet.facade.agg  ToolsCall（复用现有接口，新增 3 个工具:
+                          history_list[read] / history_revert[write] / history_revert_status[read]）
+        ├─ history_list      ──> 拉历史列表（裁剪 minor_histories / 4 字段 / AI 时间格式）
+        ├─ history_revert ───scene(ctx 透传)──> bear.server.sheet_data: RecoverHistory
+        │                                          │ service.RecoverHistory → SendRecoverMsg
+        │                                          ▼
+        │                                   MQ: RecoverMsg（新增 Scene 字段）
+        │                                          ▼
+        │   sheet.facade.agg  RecoverMsg 消费者 ── 读 Scene → memberId(doubao=10/lark-cli=11)
+        │                                          → 构造 recover cs → 调既有 recover 下游
+        └─ history_revert_status ──> bear.server.sheet_data: QueryRecoverStatus(transactionID)
+```
+
+`lark-cli` 现有 sheets shortcut 统一通过 `callTool`（`shortcuts/sheets/sheet_ai_api.go`）走 One-OpenAPI 的 `tools/invoke_read|write` 入口（`ToolKindRead` / `ToolKindWrite`），由 `sheet.facade.agg` 的 `OpenAPIToolCallRead/Write`（`biz/handler/lark_cli.go`，底层复用 `aiService.ToolsCall`）按 `tool_name` 分发。本需求的 3 个 shortcut 即按此路径新增 3 个工具，而非直连 space 接口或新增独立 OpenAPI 路由。
+
+**回滚为异步两段式**：`history_revert` 工具调 `sheet/data` 的 `RecoverHistory`（`biz/history/service/recover.go`），后者通过 `SendRecoverMsg`（`infra/mq/producer/recover.go` 的 `RecoverMsg`）投递回滚消息并返回 `transactionID`；agg 侧的 RecoverMsg 消费者真正构造 recover cs，并在此时按 scene 给 `memberId` 赋值。`history_revert_status` 工具走 `sheet/data` 的 `QueryRecoverStatus(transactionID)` 查询异步结果。scene 从 ToolsCall 入口经 **ctx 透传**（沿用既有 `utils.WithSceneDoubao(ctx)` 范式）到 `RecoverHistory`，再写入 `RecoverMsg.Scene` 字段，使 agg 消费时可区分 doubao / lark-cli。
+
+**PSM 变更判定表**：
+
+| PSM | 需要代码变更？ | 变更内容 | 不变更原因 |
+|---|---|---|---|
+| `tooling.lark_cli`（`larksuite/cli`，无服务 PSM） | 是 | 新增 3 个 shortcut + `+history-list` 响应 transform | — |
+| `tooling.sheet_skill_spec`（`ee/sheet-skill-spec`，无服务 PSM） | 是 | 新增 lark-sheets skill 正文 + 3 个 shortcut/flag 定义，生成后同步到 cli | — |
+| `sheet.facade.agg`（`ee/sheet-facade-agg`） | 是 | ① 现有 `ToolsCall` 新增 3 个工具（`history_list[read]` / `history_revert[write]` / `history_revert_status[read]`）；② `history_revert` 工具调 `sheet/data` 时透传 scene（ctx）；③ **RecoverMsg 消费者**读 `Scene` 字段，按 scene 给 `memberId` 赋值后构造 recover cs。**无新 thrift**（工具按 `tool_name`+JSON 注册，scene 走 ctx baggage） | — |
+| `bear.server.sheet_data`（`sheet/data`） | 是 | scene 从入口透传到 `RecoverHistory`（`biz/history/service/recover.go`），并在 `RecoverMsg`（`infra/mq/producer/recover.go`）上**新增 `Scene` 字段**随消息投递。`RecoverMsg` 是 JSON Go struct，加字段**非 thrift**；scene 透传走 ctx baggage | 回滚 / 快照业务语义不变（`RestoreHistorySnapshot` / `QueryRecoverStatus` 等复用） |
+
+> **`memberId` 按 scene 赋值（实现硬约束，跨 sheet/data + agg）**：scene 区分 doubao 与 lark-cli（沿用既有 `utils.WithSceneDoubao(ctx)` 范式）。本需求要求 scene 从 ToolsCall 入口一路透传：
+> 1. agg `history_revert` 工具调用 `sheet/data.RecoverHistory` 时，把 scene 经 ctx 透传；
+> 2. `sheet/data` 在产出的 `RecoverMsg` 上写入 `Scene` 字段，随 MQ 投递；
+> 3. **agg 的 RecoverMsg 消费者**在真正构造 recover cs 时，读 `Scene` 给 `memberId` 赋值——doubao 场景 = `10`，其他（lark-cli）场景 = `11`。
+>
+> memberId 赋值发生在 **agg 消费侧**（不是同步 ToolsCall 调用栈，因为回滚是异步消息驱动）。错误的 memberId 会导致回滚归属错误的调用方身份（审计 / 权限相关）。`RecoverMsg.MemberId` 字段已存在，但本需求要求按 scene 正确赋值并据此区分两个消费方。
+
+## 后端 / Tooling 子任务
+
+> 说明：`lark-cli` 与 `sheet-skill-spec` 均属 Tooling，按 ccm-harness 约定建 BE-*，`thrift_impact: 无`。本需求无前端（FE）子任务。
+
+### BE-1: lark-cli `+history-list` 历史记录列表 shortcut
+
+```yaml
+psm: tooling.lark_cli
+repo: larksuite/cli
+module: shortcuts/sheets
+be_deploy_required: false
+thrift_impact: 无
+api: facade-agg ToolsCall::history_list[read]（callTool tools/invoke_read；内部对接 /space/api/v3/sheet/histories）
+depends_on: [BE-4]
+estimate: 1.5d
+```
+
+**调用的下游服务**：经 `callTool(ToolKindRead, "history_list", ...)`（`shortcuts/sheets/sheet_ai_api.go`）走 `tools/invoke_read` 入口，由 facade-agg 的 `history_list` 工具内部对接 `/space/api/v3/sheet/histories`。入参：表格 token（沿用现有 sheets shortcut 的 `--spreadsheet-token` / `--token` 解析）。响应裁剪 / 字段重命名 / 时间格式可在 facade-agg 工具侧或 lark-cli 侧完成（见实现要点；以 AI 友好输出为准）。
+**实现要点（实现差异点落地）**：
+- 仅取响应中的 `minor_histories` 列表，丢弃其余顶层字段（如 major histories）。
+- 将每条 `minor_histories` 的 `id` 字段重命名输出为 `history_version_id`。
+- 每条仅保留 4 个字段：`history_version_id`、`create_time`、`action`、`all_block_revision`。
+- `create_time` 序列化成 AI 优化的可读格式（如本地时区可读时间串），而非裸 unix 时间戳。
+**验收场景**：
+- Given 一张有多个历史版本的电子表格，When 执行 `lark-cli sheets +history-list --token <t>`，Then 返回 JSON 数组，每条恰好含 `history_version_id` / `create_time` / `action` / `all_block_revision` 四个键，且 `create_time` 为可读格式。
+- Given 一张无历史记录的表格，When 执行 `+history-list`，Then 返回空列表且退码 0（不报错）。
+
+### BE-2: lark-cli `+history-revert` 与 `+history-revert-status` 回滚流 shortcut
+
+```yaml
+psm: tooling.lark_cli
+repo: larksuite/cli
+module: shortcuts/sheets
+be_deploy_required: false
+thrift_impact: 无
+api: facade-agg ToolsCall::history_revert[write] / history_revert_status[read]（callTool tools/invoke_write|read；内部对接 /space/api/v2/sheet/recover、/api/v2/sheet/recover/status）
+depends_on: [BE-1, BE-4, BE-5]
+estimate: 1.5d
+```
+
+**调用的下游服务**：
+- `+history-revert` → `callTool(ToolKindWrite, "history_revert", ...)`，agg 工具调 `sheet/data.RecoverHistory`（异步），返回 `transactionID`。
+- `+history-revert-status` → `callTool(ToolKindRead, "history_revert_status", ...)`，agg 工具调 `sheet/data.QueryRecoverStatus(transactionID)` 查异步结果。
+- 注：`memberId` 按 scene 赋值（doubao=10 / lark-cli=11）发生在 agg 的 **RecoverMsg 消费者**侧（见 BE-4 / BE-5），lark-cli 侧不感知；scene 由 callTool 入口（read/write）确定。
+**实现要点**：
+- `+history-revert` 的 `--history-version-id`（命名对齐 BE-1 的输出字段）为必填；缺失时在 Validate 阶段给出可执行错误提示。
+- 回滚为异步操作，`+history-revert` 返回受理结果，`+history-revert-status` 供轮询最终状态（成功 / 进行中 / 失败）。
+**验收场景**：
+- Given 由 `+history-list` 取得的合法 `history_version_id`，When 执行 `+history-revert --token <t> --history-version-id <id>`，Then 后端受理回滚并返回可被 `+history-revert-status` 查询的标识。
+- Given 一次已发起的回滚，When 轮询 `+history-revert-status`，Then 能区分「进行中 / 成功 / 失败」三种状态。
+- Given 缺省 `--history-version-id`，When 执行 `+history-revert`，Then 返回明确的参数缺失错误，不发起请求。
+
+### BE-3: sheet-skill-spec 上游事实源（skill 正文 + shortcut/flag 定义）
+
+```yaml
+psm: tooling.sheet_skill_spec
+repo: ee/sheet-skill-spec
+module: canonical-spec
+be_deploy_required: false
+thrift_impact: 无
+api: N/A
+depends_on: [BE-1, BE-2]
+estimate: 1d
+```
+
+**调用的下游服务**：无（构建期工作流）。
+**实现要点（按 `sheet-skill-spec` README 工作流）**：
+- 在飞书 base 表登记 3 个新 shortcut 的 tool ↔ shortcut 映射与 flag 定义，`npm run sync:tool-shortcut-map` 镜像入仓。
+- 在 `canonical-spec/references/<相关 skill>/cli-reference.md` 补三个 shortcut 的描述 / 示例 / Validate-DryRun-Execute 约束。
+- 跑 `npm run generate:all && npm run check:all` 验证，产出 `generated/lark-cli/skills/lark-sheets/` 与 `generated/lark-cli/data/{flag-defs.json,flag-schemas.json}`。
+- 跑 `npm run sync:cli` 把 generated 同步到 `larksuite/cli` 的 `skills/lark-sheets/`（mirror）与 `shortcuts/sheets/data/`（mirror），在 cli 仓作为 PR 提交。
+**边界**：skill 命名 / 切分 / 正文 / flag 定义一律先落 `sheet-skill-spec`，禁止直接改 cli 仓的 `generated`/`skills/lark-sheets/` 产物（README「对齐原则」）。
+**验收场景**：
+- Given 在 `sheet-skill-spec` 完成上述编辑，When 跑 `npm run check:all`，Then 全部门禁通过（generated 与 canonical 一致、map 与 base 表一致）。
+- Given 跑 `npm run sync:cli`，Then cli 仓 `skills/lark-sheets/` 与 `shortcuts/sheets/data/` 出现对应 3 个 shortcut 的 skill 正文与 flag 定义。
+
+### BE-4: sheet-facade-agg 在现有 ToolsCall 上新增 3 个历史/回滚工具
+
+```yaml
+psm: sheet.facade.agg
+repo: ee/sheet-facade-agg
+module: biz/handler
+be_deploy_required: true
+thrift_impact: 无
+api: ToolsCall::history_list[read] / history_revert[write] / history_revert_status[read]
+depends_on: []
+estimate: 1.5d
+```
+
+**调用的下游服务**：`sheet/data` 的 `RecoverHistory` / `QueryRecoverStatus`（见 BE-5）+ 既有历史查询；agg 的 RecoverMsg 消费者复用既有 recover 下游（`biz/service/spreadsheet.go::ProcessRecoverCs`、`model.RecoverParam`），不新增 thrift。
+**实现要点**：
+- **ToolsCall 扩展**：在现有 `ToolsCall` 框架（`biz/handler/handler.go::ToolsCall` / `biz/handler/lark_cli.go::OpenAPIToolCallRead|Write`）注册 3 个新工具：`history_list`（read）、`history_revert`（write）、`history_revert_status`（read），按 `constants.IsReadTool` / `IsWriteTool` 归类，从 `tools/invoke_read` / `invoke_write` 入口可达。
+- **scene 透传**：`history_revert` / `history_revert_status` 工具调 `sheet/data` 时，把 scene 经 ctx（沿用 `utils.WithSceneDoubao` 范式）透传下去，使 `sheet/data` 能写入 `RecoverMsg.Scene`。
+- **RecoverMsg 消费者按 scene 赋 memberId（硬约束）**：agg 消费 `sheet/data` 投递的 `RecoverMsg`、构造真正 recover cs 时，读 `RecoverMsg.Scene` 给 `memberId` 赋值——doubao = `10`，lark-cli = `11`。这是异步消费侧逻辑，不在同步 ToolsCall 调用栈。
+- `history_list` 工具对接历史列表查询；响应裁剪（仅 `minor_histories`、`id`→`history_version_id`、4 字段、`create_time` AI 友好格式）建议落在此工具侧（两个消费方共享，避免 lark-cli / doubao 双实现漂移）。
+**边界**：只在 ToolsCall 上加工具 + 改 RecoverMsg 消费者；不新增独立 OpenAPI 路由、不改 `ai.ToolsCallRequest` thrift 契约、不改 `RecoverHistory` 回滚业务语义。
+**验收场景**：
+- Given lark-cli 经 `tools/invoke_read` 调用 `history_list`，Then 返回裁剪后的 `minor_histories`（4 字段，`history_version_id` 命名）。
+- Given lark-cli（scene=lark-cli）经 `history_revert` 发起回滚，When agg 消费对应 RecoverMsg，Then 构造的 recover cs 中 `memberId == 11`；doubao 场景下同一路径 `memberId == 10`。
+- Given 已发起回滚，When 调用 `history_revert_status`，Then 经 `QueryRecoverStatus` 返回可区分的回滚状态。
+
+### BE-5: sheet/data 透传 scene 并在 RecoverMsg 上新增 Scene 字段
+
+```yaml
+psm: bear.server.sheet_data
+repo: sheet/data
+module: biz/history
+be_deploy_required: true
+thrift_impact: 无
+api: bear.server.sheet_data::RecoverHistory / QueryRecoverStatus（复用）；MQ RecoverMsg 加 Scene 字段
+depends_on: []
+estimate: 1d
+```
+
+**调用的下游服务**：复用既有回滚链路（`biz/history/service/recover.go::RecoverHistory` → `infra/mq/producer/recover.go::SendRecoverMsg`）。
+**实现要点**：
+- **scene 透传**：把 agg 经 ctx 传入的 scene 接住，贯穿 `RecoverHistory`（`biz/history/service/recover.go`）到 `RecoverMsg` 构造处。
+- **RecoverMsg 加 `Scene` 字段**：在 `infra/mq/producer/recover.go` 的 `RecoverMsg` struct 上新增 `Scene` 字段并在投递时赋值。`RecoverMsg` 是 JSON Go struct（`recoverProducer.NewMessage`），**加字段非 thrift**——`thrift_impact: 无`。
+- `QueryRecoverStatus` 与回滚业务语义保持不变，仅承载 scene 透传。
+**边界**：不改回滚 / 历史快照业务逻辑；只加 scene 透传与 `RecoverMsg.Scene` 字段。
+**scene 透传方式（已定）**：经 **ctx baggage**（`metainfo` / 沿用既有 `utils.WithSceneDoubao(ctx)` 范式）从 agg 透传到 `RecoverHistory`，**不**在 `RecoverHistoryReq` thrift 上加字段 → 零 IDL 变更，`thrift_impact: 无`。
+**验收场景**：
+- Given agg 以 scene=lark-cli 调 `RecoverHistory`，Then 投递的 `RecoverMsg.Scene` 标识 lark-cli；doubao 同理。
+- Given 回滚已发起，When `QueryRecoverStatus(transactionID)`，Then 返回回滚状态（语义与现状一致）。
+- Given lark-cli（scene=lark-cli）经 `tools/invoke_write` 调用 `history_revert`，Then 构造的 recover cs 中 `memberId == 11`；doubao 场景下同一工具 `memberId == 10`。
+- Given 已发起回滚，When 调用 `history_revert_status`，Then 返回可区分的回滚状态。
+
+## API 契约引用
+
+本需求三个接口均为飞书电子表格已上线 space 接口，契约以各仓库最新 master 为准；对应 thrift 定义按 PRD 提示在 `lark/idl` 中搜索确认（实现阶段补全精确路径）：
+
+- 查列表：`/space/api/v3/sheet/histories`（取 `minor_histories`）
+- 回滚：`/space/api/v2/sheet/recover`
+- 回滚状态：`/api/v2/sheet/recover/status`
+
+> 契约本体不进本 spec 正文；精确 `lark/idl/...thrift::Service::Method` 路径在实现阶段确认并回填到对应 BE-* 的 `api` 字段说明。
+
+## 验收场景（汇总）
+
+- 列表：`+history-list` 仅返回 `minor_histories`，每条恰好 4 个字段，`id` 重命名为 `history_version_id`，`create_time` 为 AI 优化可读格式。
+- 回滚：`+history-revert` → agg `history_revert` → `sheet/data.RecoverHistory`（异步），受理后返回可查询标识。
+- memberId/scene：agg 消费 `RecoverMsg` 构造 recover cs 时，按 `RecoverMsg.Scene` 赋 `memberId`——lark-cli=11、doubao=10（facade-agg 侧单测断言）。
+- 状态：`+history-revert-status` → `QueryRecoverStatus` 能查询并区分回滚的进行中 / 成功 / 失败。
+- skill 同步：`sheet-skill-spec` 生成产物经 `sync:cli` 落地到 cli 仓，`check:all` 全绿。
+- 三个 shortcut 在 cli 中遵循统一的 Validate / DryRun / Execute 三段约定与现有 sheets shortcut 一致。
+
+## 非功能要求与约束
+
+- **复用既有模式**：3 个 shortcut 必须沿用 `shortcuts/sheets` 现有的 token 解析（`--spreadsheet-token` / `--token` 别名）、错误封装（`errs`）、`callTool`（`tools/invoke_read|write`）调用与 DryRun 渲染范式，不另起调用框架。facade-agg 侧必须复用现有 `ToolsCall` 接口扩展工具，不新增独立 OpenAPI 路由。
+- **AI 友好输出**：`+history-list` 的字段裁剪与 `create_time` 可读格式是硬约束（PRD「实现差异点」），目的是降低 AI 消费成本。
+- **工作流约束**：skill 内容与 flag 定义的唯一事实源是 `ee/sheet-skill-spec`；cli 仓的 `skills/lark-sheets/` 与 `shortcuts/sheets/data/` 为同步产物，不手改。
+- **回滚为异步**：`+history-revert` 与 `+history-revert-status` 分离，调用方需理解「发起 → 轮询」两步语义。
+- **事实基准**：所有外部仓库事实（space 接口、facade-agg 路由、sheet_data 能力）以各仓库最新 master 为准。
+
+## 安全设计
+
+- security_knowledge_ref: UNCONFIGURED
+- 风险判断依据: 未配置安全知识库，待安全侧补齐。需关注点（供安全侧复核）：`+history-revert` 是**写 / 不可逆**操作（覆盖当前表格内容到历史版本），必须校验操作者对目标表格具备编辑 / 回滚权限；历史版本列表可能暴露协作者操作痕迹（`action` 字段），需确认读权限边界。
+- 身份归属风险（memberId/scene）: `memberId` 须按 scene 正确赋值（lark-cli=11 / doubao=10）。错配会使回滚操作归属错误的调用方身份，影响审计与权限判定——属安全/审计相关，须在 agg RecoverMsg 消费侧保证赋值正确。
+- 需要安全侧补充: 回滚操作的权限校验口径、历史 `action` 字段的可见性范围是否需脱敏、memberId 与真实操作者身份的映射是否需对齐审计要求。
+
+## Codegen Delivery Plan
+
+applicable: true
+
+### A. Branch Plan
+
+| `key` | value |
+|---|---|
+| `psm` | `sheet.facade.agg` |
+| `business_branch` | `feat/sheet-history-revert` |
+| `generated_branch` | `N/A` |
+| `idl_branch` | `N/A` |
+| `kitex_branch` | `N/A` |
+
+### B. Delivery Targets
+
+| repo | required | branch | artifact_paths | reason |
+|---|---|---|---|---|
+| larksuite/cli | yes | feat/sheet-history-revert | shortcuts/sheets/ , skills/lark-sheets/ | 3 个 shortcut 实现 + 同步落地的 skill 正文与 flag 数据 |
+| ee/sheet-skill-spec | yes | feat/sheet-history-revert | canonical-spec/references/ , generated/lark-cli/ | skill / flag 上游事实源，生成后 sync 到 cli |
+| ee/sheet-facade-agg | yes | feat/sheet-history-revert | biz/handler/ | 现有 ToolsCall 新增 3 个工具 + scene 透传 + RecoverMsg 消费者按 scene 赋 memberId |
+| sheet/data | yes | feat/sheet-history-revert | biz/history/ , infra/mq/producer/ | RecoverHistory 透传 scene + RecoverMsg 新增 Scene 字段（JSON，非 thrift） |
+
+### C. Generation Decision
+
+| `key` | value |
+|---|---|
+| `needs_kitex_gen` | no |
+| `needs_apacana` | no |
+| `needs_kite_via_sdp` | no |
+| `decision_basis` | facade-agg 复用现有 ToolsCall 框架按 tool_name 注册 3 个工具（JSON input，不动 ai.ToolsCallRequest thrift）；sheet/data 仅在 RecoverMsg（JSON Go struct）加 Scene 字段 + 经 ctx baggage 透传 scene，复用既有 RecoverHistory/QueryRecoverStatus，均非 thrift；lark-cli 经现有 callTool 包装。scene 已定走 ctx baggage（不加 RecoverHistoryReq thrift 字段），无新增/修改 kitex/apacana/SDP 契约 |
+
+### D. Branch Naming Rule
+
+业务分支统一用 `feat/sheet-history-revert`；本需求无 codegen，故 `generated_branch` / `idl_branch` / `kitex_branch` 均为 `N/A`。
+
+## thrift 变更需求清单
+
+无（按推荐实现路径）。三个接口的能力已存在；本需求的新增内容是：facade-agg 在 ToolsCall 上注册 3 个工具（`tool_name`+JSON，不动 `ai.ToolsCallRequest`）、sheet/data 在 `RecoverMsg`（JSON Go struct）上加 `Scene` 字段、scene 经 **ctx baggage** 透传——均不涉及 thrift。
+
+**scene 透传方式：已定为 ctx baggage**（`metainfo` / 沿用既有 `utils.WithSceneDoubao(ctx)` 范式），明确**不**在 `bear.server.sheet_data` 的 `RecoverHistoryReq` thrift 上加字段。故本需求确无任何 thrift struct / RPC method / enum 的新增或修改，Generation Decision 三路保持 `no`、无 codegen。
+
+## N. AI Capability Manifest
+
+applicable: false
+
+本需求为确定性 CLI 命令封装，不含 LLM / prompt 驱动的 AI 能力。`+history-list` 中「`create_time` 序列化成 AI 优化格式」仅指对机器/AI 更易读的时间字符串格式化，属确定性数据转换，非 AI capability。

go.mod

@@ -27,6 +27,8 @@ require (
 	gopkg.in/yaml.v3 v3.0.1
 )
 
+require github.com/apache/arrow/go/v17 v17.0.0
+
 require (
 	github.com/atotto/clipboard v0.1.4 // indirect
 	github.com/aymanbagabas/go-osc52/v2 v2.0.1 // indirect
@@ -42,13 +44,17 @@ require (
 	github.com/davecgh/go-spew v1.1.1 // indirect
 	github.com/dustin/go-humanize v1.0.1 // indirect
 	github.com/erikgeiser/coninput v0.0.0-20211004153227-1c3628e74d0f // indirect
+	github.com/goccy/go-json v0.10.3 // indirect
 	github.com/godbus/dbus/v5 v5.2.2 // indirect
 	github.com/gogo/protobuf v1.3.2 // indirect
+	github.com/google/flatbuffers v24.3.25+incompatible // indirect
 	github.com/gopherjs/gopherjs v1.17.2 // indirect
 	github.com/gorilla/websocket v1.5.0 // indirect
 	github.com/inconshreveable/mousetrap v1.1.0 // indirect
 	github.com/itchyny/timefmt-go v0.1.6 // indirect
 	github.com/jtolds/gls v4.20.0+incompatible // indirect
+	github.com/klauspost/compress v1.17.9 // indirect
+	github.com/klauspost/cpuid/v2 v2.2.8 // indirect
 	github.com/lucasb-eyer/go-colorful v1.2.0 // indirect
 	github.com/mattn/go-isatty v0.0.20 // indirect
 	github.com/mattn/go-localereader v0.0.1 // indirect
@@ -57,10 +63,16 @@ require (
 	github.com/muesli/ansi v0.0.0-20230316100256-276c6243b2f6 // indirect
 	github.com/muesli/cancelreader v0.2.2 // indirect
 	github.com/muesli/termenv v0.16.0 // indirect
+	github.com/pierrec/lz4/v4 v4.1.21 // indirect
 	github.com/pmezard/go-difflib v1.0.0 // indirect
 	github.com/rivo/uniseg v0.4.7 // indirect
 	github.com/smarty/assertions v1.15.0 // indirect
 	github.com/tidwall/match v1.1.1 // indirect
 	github.com/tidwall/pretty v1.2.0 // indirect
 	github.com/xo/terminfo v0.0.0-20220910002029-abceb7e1c41e // indirect
+	github.com/zeebo/xxh3 v1.0.2 // indirect
+	golang.org/x/exp v0.0.0-20240222234643-814bf88cf225 // indirect
+	golang.org/x/mod v0.18.0 // indirect
+	golang.org/x/tools v0.22.0 // indirect
+	golang.org/x/xerrors v0.0.0-20231012003039-104605ab7028 // indirect
 )

go.sum

@@ -2,6 +2,8 @@ github.com/MakeNowJust/heredoc v1.0.0 h1:cXCdzVdstXyiTqTvfqk9SDHpKNjxuom+DOlyEeQ
 github.com/MakeNowJust/heredoc v1.0.0/go.mod h1:mG5amYoWBHf8vpLOuehzbGGw0EHxpZZ6lCpQ4fNJ8LE=
 github.com/Microsoft/go-winio v0.6.2 h1:F2VQgta7ecxGYO8k3ZZz3RS8fVIXVxONVUPlNERoyfY=
 github.com/Microsoft/go-winio v0.6.2/go.mod h1:yd8OoFMLzJbo9gZq8j5qaps8bJ9aShtEA8Ipt1oGCvU=
+github.com/apache/arrow/go/v17 v17.0.0 h1:RRR2bdqKcdbss9Gxy2NS/hK8i4LDMh23L6BbkN5+F54=
+github.com/apache/arrow/go/v17 v17.0.0/go.mod h1:jR7QHkODl15PfYyjM2nU+yTLScZ/qfj7OSUZmJ8putc=
 github.com/atotto/clipboard v0.1.4 h1:EH0zSVneZPSuFR11BlR9YppQTVDbh5+16AmcJi4g1z4=
 github.com/atotto/clipboard v0.1.4/go.mod h1:ZY9tmq7sm5xIbd9bOK4onWV4S6X0u6GY7Vn0Yu86PYI=
 github.com/aymanbagabas/go-osc52/v2 v2.0.1 h1:HwpRHbFMcZLEVr42D4p7XBqjyuxQH5SMiErDT4WkJ2k=
@@ -52,12 +54,16 @@ github.com/dustin/go-humanize v1.0.1 h1:GzkhY7T5VNhEkwH0PVJgjz+fX1rhBrR7pRT3mDkp
 github.com/dustin/go-humanize v1.0.1/go.mod h1:Mu1zIs6XwVuF/gI1OepvI0qD18qycQx+mFykh5fBlto=
 github.com/erikgeiser/coninput v0.0.0-20211004153227-1c3628e74d0f h1:Y/CXytFA4m6baUTXGLOoWe4PQhGxaX0KpnayAqC48p4=
 github.com/erikgeiser/coninput v0.0.0-20211004153227-1c3628e74d0f/go.mod h1:vw97MGsxSvLiUE2X8qFplwetxpGLQrlU1Q9AUEIzCaM=
+github.com/goccy/go-json v0.10.3 h1:KZ5WoDbxAIgm2HNbYckL0se1fHD6rz5j4ywS6ebzDqA=
+github.com/goccy/go-json v0.10.3/go.mod h1:oq7eo15ShAhp70Anwd5lgX2pLfOS3QCiwU/PULtXL6M=
 github.com/godbus/dbus/v5 v5.2.2 h1:TUR3TgtSVDmjiXOgAAyaZbYmIeP3DPkld3jgKGV8mXQ=
 github.com/godbus/dbus/v5 v5.2.2/go.mod h1:3AAv2+hPq5rdnr5txxxRwiGjPXamgoIHgz9FPBfOp3c=
 github.com/gofrs/flock v0.8.1 h1:+gYjHKf32LDeiEEFhQaotPbLuUXjY5ZqxKgXy7n59aw=
 github.com/gofrs/flock v0.8.1/go.mod h1:F1TvTiK9OcQqauNUHlbJvyl9Qa1QvF/gOUDKA14jxHU=
 github.com/gogo/protobuf v1.3.2 h1:Ov1cvc58UF3b5XjBnZv7+opcTcQFZebYjWzi34vdm4Q=
 github.com/gogo/protobuf v1.3.2/go.mod h1:P1XiOD3dCwIKUDQYPy72D8LYyHL2YPYrpS2s69NZV8Q=
+github.com/google/flatbuffers v24.3.25+incompatible h1:CX395cjN9Kke9mmalRoL3d81AtFUxJM+yDthflgJGkI=
+github.com/google/flatbuffers v24.3.25+incompatible/go.mod h1:1AeVuKshWv4vARoZatz6mlQ0JxURH0Kv5+zNeJKJCa8=
 github.com/google/uuid v1.6.0 h1:NIvaJDMOsjHA8n1jAhLSgzrAzy1Hgr+hNrb57e+94F0=
 github.com/google/uuid v1.6.0/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
 github.com/gopherjs/gopherjs v1.17.2 h1:fQnZVsXk8uxXIStYb0N4bGk7jeyTalG/wsZjQ25dO0g=
@@ -74,11 +80,16 @@ github.com/jtolds/gls v4.20.0+incompatible h1:xdiiI2gbIgH/gLH7ADydsJ1uDOEzR8yvV7
 github.com/jtolds/gls v4.20.0+incompatible/go.mod h1:QJZ7F/aHp+rZTRtaJ1ow/lLfFfVYBRgL+9YlvaHOwJU=
 github.com/kisielk/errcheck v1.5.0/go.mod h1:pFxgyoBC7bSaBwPgfKdkLd5X25qrDl4LWUI2bnpBCr8=
 github.com/kisielk/gotool v1.0.0/go.mod h1:XhKaO+MFFWcvkIS/tQcRk01m1F5IRFswLeQ+oQHNcck=
+github.com/klauspost/compress v1.17.9 h1:6KIumPrER1LHsvBVuDa0r5xaG0Es51mhhB9BQB2qeMA=
+github.com/klauspost/compress v1.17.9/go.mod h1:Di0epgTjJY877eYKx5yC51cX2A2Vl2ibi7bDH9ttBbw=
+github.com/klauspost/cpuid/v2 v2.2.8 h1:+StwCXwm9PdpiEkPyzBXIy+M9KUb4ODm0Zarf1kS5BM=
+github.com/klauspost/cpuid/v2 v2.2.8/go.mod h1:Lcz8mBdAVJIBVzewtcLocK12l3Y+JytZYpaMropDUws=
 github.com/kr/pretty v0.1.0 h1:L/CwN0zerZDmRFUapSPitk6f+Q3+0za1rQkzVuMiMFI=
 github.com/kr/pretty v0.1.0/go.mod h1:dAy3ld7l9f0ibDNOQOHHMYYIIbhfbHSm3C4ZsoJORNo=
 github.com/kr/pty v1.1.1/go.mod h1:pFQYn66WHrOpPYNljwOMqo10TkYh1fy3cYio2l3bCsQ=
-github.com/kr/text v0.1.0 h1:45sCR5RtlFHMR4UwH9sdQ5TC8v0qDQCHnXt+kaKSTVE=
 github.com/kr/text v0.1.0/go.mod h1:4Jbv+DJW3UT/LiOwJeYQe1efqtUx/iVham/4vfdArNI=
+github.com/kr/text v0.2.0 h1:5Nx0Ya0ZqY2ygV366QzturHI13Jq95ApcVaJBhpS+AY=
+github.com/kr/text v0.2.0/go.mod h1:eLer722TekiGuMkidMxC/pM04lWEeraHUUmBw8l2grE=
 github.com/larksuite/oapi-sdk-go/v3 v3.5.4 h1:U2S9x9LrfH++ZqJ+YAiUlqzCWJmVXhFdS8Z7rIBH8H0=
 github.com/larksuite/oapi-sdk-go/v3 v3.5.4/go.mod h1:ZEplY+kwuIrj/nqw5uSCINNATcH3KdxSN7y+UxYY5fI=
 github.com/lucasb-eyer/go-colorful v1.2.0 h1:1nnpGOrhyZZuNyfu1QjKiUICQ74+3FNCN69Aj6K7nkY=
@@ -97,6 +108,8 @@ github.com/muesli/cancelreader v0.2.2 h1:3I4Kt4BQjOR54NavqnDogx/MIoWBFa0StPA8ELU
 github.com/muesli/cancelreader v0.2.2/go.mod h1:3XuTXfFS2VjM+HTLZY9Ak0l6eUKfijIfMUZ4EgX0QYo=
 github.com/muesli/termenv v0.16.0 h1:S5AlUN9dENB57rsbnkPyfdGuWIlkmzJjbFf0Tf5FWUc=
 github.com/muesli/termenv v0.16.0/go.mod h1:ZRfOIKPFDYQoDFF4Olj7/QJbW60Ol/kL1pU3VfY/Cnk=
+github.com/pierrec/lz4/v4 v4.1.21 h1:yOVMLb6qSIDP67pl/5F7RepeKYu/VmTyEXvuMI5d9mQ=
+github.com/pierrec/lz4/v4 v4.1.21/go.mod h1:gZWDp/Ze/IJXGXf23ltt2EXimqmTUXEy0GFuRQyBid4=
 github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZbAQM=
 github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
 github.com/rivo/uniseg v0.2.0/go.mod h1:J6wj4VEh+S6ZtnVlnTBMWIodfgj8LQOQFoIToxlJtxc=
@@ -133,14 +146,20 @@ github.com/yuin/goldmark v1.1.27/go.mod h1:3hX8gzYuyVAZsxl0MRgGTJEmQBFcNTphYh9de
 github.com/yuin/goldmark v1.2.1/go.mod h1:3hX8gzYuyVAZsxl0MRgGTJEmQBFcNTphYh9decYSb74=
 github.com/zalando/go-keyring v0.2.8 h1:6sD/Ucpl7jNq10rM2pgqTs0sZ9V3qMrqfIIy5YPccHs=
 github.com/zalando/go-keyring v0.2.8/go.mod h1:tsMo+VpRq5NGyKfxoBVjCuMrG47yj8cmakZDO5QGii0=
+github.com/zeebo/assert v1.3.0 h1:g7C04CbJuIDKNPFHmsk4hwZDO5O+kntRxzaUoNXj+IQ=
+github.com/zeebo/assert v1.3.0/go.mod h1:Pq9JiuJQpG8JLJdtkwrJESF0Foym2/D9XMU5ciN/wJ0=
+github.com/zeebo/xxh3 v1.0.2 h1:xZmwmqxHZA8AI603jOQ0tMqmBr9lPeFwGg6d+xy9DC0=
+github.com/zeebo/xxh3 v1.0.2/go.mod h1:5NWz9Sef7zIDm2JHfFlcQvNekmcEl9ekUZQQKCYaDcA=
 go.yaml.in/yaml/v3 v3.0.4/go.mod h1:DhzuOOF2ATzADvBadXxruRBLzYTpT36CKvDb3+aBEFg=
 golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACkg1iLfiJU5Ep61QUkGW8qpdssI0+w=
 golang.org/x/crypto v0.0.0-20191011191535-87dc89f01550/go.mod h1:yigFU9vqHzYiE8UmvKecakEJjdnWj3jj499lnFckfCI=
 golang.org/x/crypto v0.0.0-20200622213623-75b288015ac9/go.mod h1:LzIPMQfyMNhhGPhUkYOs5KpL4U8rLKemX1yGLhDgUto=
-golang.org/x/exp v0.0.0-20231006140011-7918f672742d h1:jtJma62tbqLibJ5sFQz8bKtEM8rJBtfilJ2qTU199MI=
-golang.org/x/exp v0.0.0-20231006140011-7918f672742d/go.mod h1:ldy0pHrwJyGW56pPQzzkH36rKxoZW1tw7ZJpeKx+hdo=
+golang.org/x/exp v0.0.0-20240222234643-814bf88cf225 h1:LfspQV/FYTatPTr/3HzIcmiUFH7PGP+OQ6mgDYo3yuQ=
+golang.org/x/exp v0.0.0-20240222234643-814bf88cf225/go.mod h1:CxmFvTBINI24O/j8iY7H1xHzx2i4OsyguNBmN/uPtqc=
 golang.org/x/mod v0.2.0/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
 golang.org/x/mod v0.3.0/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
+golang.org/x/mod v0.18.0 h1:5+9lSbEzPSdWkH32vYPBwEpX8KwDbM52Ud9xBUvNlb0=
+golang.org/x/mod v0.18.0/go.mod h1:hTbmBsO62+eylJbnUtE2MGJUyE7QWk4xUqPFrRgJ+7c=
 golang.org/x/net v0.0.0-20190404232315-eb5bcb51f2a3/go.mod h1:t9HGtf8HONx5eT2rtn7q6eTqICYqUVnKs3thJo3Qplg=
 golang.org/x/net v0.0.0-20190620200207-3b0461eec859/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
 golang.org/x/net v0.0.0-20200226121028-0de0cce0169b/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
@@ -156,6 +175,7 @@ golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5h
 golang.org/x/sys v0.0.0-20190412213103-97732733099d/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20200930185726-fdedc70b468f/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20210809222454-d867a43fc93e/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
+golang.org/x/sys v0.5.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.6.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.33.0 h1:q3i8TbbEz+JRD9ywIRlyRAQbM0qF7hu24q3teo2hbuw=
 golang.org/x/sys v0.33.0/go.mod h1:BJP2sWEmIv4KK5OTEluFJCKSidICx8ciO85XgH3Ak8k=
@@ -169,10 +189,16 @@ golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGm
 golang.org/x/tools v0.0.0-20191119224855-298f0cb1881e/go.mod h1:b+2E5dAYhXwXZwtnZ6UAqBI28+e2cm9otk0dWdXHAEo=
 golang.org/x/tools v0.0.0-20200619180055-7c47624df98f/go.mod h1:EkVYQZoAsY45+roYkvgYkIh4xh/qjgUK9TdY2XT94GE=
 golang.org/x/tools v0.0.0-20210106214847-113979e3529a/go.mod h1:emZCQorbCU4vsT4fOWvOPXz4eW1wZW4PmDk9uLelYpA=
+golang.org/x/tools v0.22.0 h1:gqSGLZqv+AI9lIQzniJ0nZDRG5GBPsSi+DRNHWNz6yA=
+golang.org/x/tools v0.22.0/go.mod h1:aCwcsjqvq7Yqt6TNyX7QMU2enbQ/Gt0bo6krSeEri+c=
 golang.org/x/xerrors v0.0.0-20190717185122-a985d3407aa7/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20191011141410-1b5146add898/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20200804184101-5ec99f83aff1/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
+golang.org/x/xerrors v0.0.0-20231012003039-104605ab7028 h1:+cNy6SZtPcJQH3LJVLOSmiC7MMxXNOb3PU/VUEz+EhU=
+golang.org/x/xerrors v0.0.0-20231012003039-104605ab7028/go.mod h1:NDW/Ps6MPRej6fsCIbMTohpP40sJ/P/vI1MoTEGwX90=
+gonum.org/v1/gonum v0.15.0 h1:2lYxjRbTYyxkJxlhC+LvJIx3SsANPdRybu1tGj9/OrQ=
+gonum.org/v1/gonum v0.15.0/go.mod h1:xzZVBJBtS+Mz4q0Yl2LJTk+OxOg4jiXZ7qBoM0uISGo=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
 gopkg.in/check.v1 v1.0.0-20190902080502-41f04d3bba15 h1:YR8cESwS4TdDjEe65xsg0ogRM/Nc3DYOhEAlW+xobZo=
 gopkg.in/check.v1 v1.0.0-20190902080502-41f04d3bba15/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=

shortcuts/common/runner.go

@@ -1045,7 +1045,8 @@ func resolveInputFlags(rctx *RuntimeContext, flags []Flag) error {
 			}
 			if stdinUsed {
 				return ValidationErrorf("--%s: stdin (-) can only be used by one flag", fl.Name).
-					WithParam("--" + fl.Name)
+					WithParam("--"+fl.Name).
+					WithHint("a process has a single stdin, so only one flag per call may use '-'; pass the others as @file (e.g. --%s @/path/to/file)", fl.Name)
 			}
 			stdinUsed = true
 			data, err := io.ReadAll(rctx.IO().In)
@@ -1160,7 +1161,13 @@ func registerShortcutFlagsWithContext(ctx context.Context, cmd *cobra.Command, f
 				hints = append(hints, "@file")
 			}
 			if slices.Contains(fl.Input, Stdin) {
-				hints = append(hints, "- for stdin")
+				// "- reads stdin" intentionally avoids implying each flag has
+				// its own stdin: a process has a single stdin, so at most one
+				// flag per call may use "-" (the rest must use @file). The old
+				// per-flag "- for stdin" wording led AI agents to write
+				// `--a - <x --b - <y`, where the second `<` silently clobbers
+				// the first and `--a` reads the wrong payload.
+				hints = append(hints, "- reads stdin (one flag per call; use @file for others)")
 			}
 			desc += " (supports " + strings.Join(hints, ", ") + ")"
 		}

shortcuts/common/runner_input_test.go

@@ -216,10 +216,15 @@ func TestResolveInputFlags_DuplicateStdin(t *testing.T) {
 	if err == nil {
 		t.Fatal("expected error for duplicate stdin usage")
 	}
-	assertValidationParam(t, err, "--b")
+	vErr := assertValidationParam(t, err, "--b")
 	if !strings.Contains(err.Error(), "stdin (-) can only be used by one flag") {
 		t.Errorf("unexpected error: %v", err)
 	}
+	// The hint must steer an AI agent to the fix (@file for the extra flags),
+	// since `--a - <x --b - <y` is the exact misuse this guards against.
+	if !strings.Contains(vErr.Hint, "@file") {
+		t.Errorf("hint %q should mention @file as the fix", vErr.Hint)
+	}
 }
 
 func TestStripUTF8BOM(t *testing.T) {

shortcuts/drive/drive_export.go

@@ -40,236 +40,302 @@ var DriveExport = common.Shortcut{
 		{Name: "overwrite", Type: "bool", Desc: "overwrite existing output file"},
 	},
 	Validate: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		return validateDriveExportSpec(driveExportSpec{
-			Token:         runtime.Str("token"),
-			DocType:       runtime.Str("doc-type"),
-			FileExtension: runtime.Str("file-extension"),
-			SubID:         runtime.Str("sub-id"),
-			OnlySchema:    runtime.Bool("only-schema"),
-		})
+		return ValidateExport(exportParamsFromFlags(runtime))
 	},
 	DryRun: func(ctx context.Context, runtime *common.RuntimeContext) *common.DryRunAPI {
-		spec := driveExportSpec{
-			Token:         runtime.Str("token"),
-			DocType:       runtime.Str("doc-type"),
-			FileExtension: runtime.Str("file-extension"),
-			SubID:         runtime.Str("sub-id"),
-			OnlySchema:    runtime.Bool("only-schema"),
-		}
-		// Markdown export is a special case: docx markdown comes from the V2
-		// docs_ai fetch API directly instead of the Drive export task API.
-		if spec.FileExtension == "markdown" {
-			apiPath := fmt.Sprintf("/open-apis/docs_ai/v1/documents/%s/fetch", validate.EncodePathSegment(spec.Token))
-			dr := common.NewDryRunAPI().
-				Desc("2-step orchestration: fetch docx markdown -> write local file").
-				POST(apiPath).
-				Body(map[string]interface{}{
-					"format": "markdown",
-				}).
-				Set("output_dir", runtime.Str("output-dir"))
-			if name := strings.TrimSpace(runtime.Str("file-name")); name != "" {
-				dr.Set("file_name", ensureExportFileExtension(sanitizeExportFileName(name, spec.Token), spec.FileExtension))
-			}
-			return dr
-		}
+		return PlanExportDryRun(runtime, exportParamsFromFlags(runtime))
+	},
+	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
+		return RunExport(ctx, runtime, exportParamsFromFlags(runtime))
+	},
+}
 
-		body := map[string]interface{}{
-			"token":          spec.Token,
-			"type":           spec.DocType,
-			"file_extension": spec.FileExtension,
-		}
-		if strings.TrimSpace(spec.SubID) != "" {
-			body["sub_id"] = spec.SubID
-		}
-		if spec.OnlySchema {
-			body["only_schema"] = true
-		}
+// ExportParams holds the user-facing inputs for an export flow, decoupled from
+// cobra flags so other command groups (e.g. sheets +workbook-export) can reuse
+// the drive export implementation. An empty OutputDir means "create the export
+// task and poll, but do not download" — callers that only need the ready file
+// token / status get it back without writing a local file.
+type ExportParams struct {
+	Token         string
+	DocType       string
+	FileExtension string
+	SubID         string
+	OnlySchema    bool
+	OutputDir     string
+	FileName      string
+	Overwrite     bool
+}
 
+func (p ExportParams) spec() driveExportSpec {
+	return driveExportSpec{
+		Token:         p.Token,
+		DocType:       p.DocType,
+		FileExtension: p.FileExtension,
+		SubID:         p.SubID,
+		OnlySchema:    p.OnlySchema,
+	}
+}
+
+// exportParamsFromFlags reads the standard drive +export flag set.
+func exportParamsFromFlags(runtime *common.RuntimeContext) ExportParams {
+	// drive +export always downloads; an empty --output-dir historically means
+	// the current directory (saveContentToOutputDir maps "" -> "."), so normalize
+	// it here to keep behavior identical and stay off the export-only ("" => skip
+	// download) path that only sheets +workbook-export uses.
+	outputDir := runtime.Str("output-dir")
+	if outputDir == "" {
+		outputDir = "."
+	}
+	return ExportParams{
+		Token:         runtime.Str("token"),
+		DocType:       runtime.Str("doc-type"),
+		FileExtension: runtime.Str("file-extension"),
+		SubID:         runtime.Str("sub-id"),
+		OnlySchema:    runtime.Bool("only-schema"),
+		OutputDir:     outputDir,
+		FileName:      strings.TrimSpace(runtime.Str("file-name")),
+		Overwrite:     runtime.Bool("overwrite"),
+	}
+}
+
+// ValidateExport runs the CLI-level export constraint checks.
+func ValidateExport(p ExportParams) error {
+	return validateDriveExportSpec(p.spec())
+}
+
+// PlanExportDryRun builds the dry-run plan for an export without performing I/O.
+func PlanExportDryRun(runtime *common.RuntimeContext, p ExportParams) *common.DryRunAPI {
+	spec := p.spec()
+	// Markdown export is a special case: docx markdown comes from the V2
+	// docs_ai fetch API directly instead of the Drive export task API.
+	if spec.FileExtension == "markdown" {
+		apiPath := fmt.Sprintf("/open-apis/docs_ai/v1/documents/%s/fetch", validate.EncodePathSegment(spec.Token))
 		dr := common.NewDryRunAPI().
-			Desc("3-step orchestration: create export task -> limited polling -> download file").
-			POST("/open-apis/drive/v1/export_tasks").
-			Body(body).
-			Set("output_dir", runtime.Str("output-dir"))
-		if name := strings.TrimSpace(runtime.Str("file-name")); name != "" {
+			Desc("2-step orchestration: fetch docx markdown -> write local file").
+			POST(apiPath).
+			Body(map[string]interface{}{
+				"format": "markdown",
+			}).
+			Set("output_dir", p.OutputDir)
+		if name := strings.TrimSpace(p.FileName); name != "" {
 			dr.Set("file_name", ensureExportFileExtension(sanitizeExportFileName(name, spec.Token), spec.FileExtension))
 		}
 		return dr
-	},
-	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		spec := driveExportSpec{
-			Token:         runtime.Str("token"),
-			DocType:       runtime.Str("doc-type"),
-			FileExtension: runtime.Str("file-extension"),
-			SubID:         runtime.Str("sub-id"),
-			OnlySchema:    runtime.Bool("only-schema"),
+	}
+
+	body := map[string]interface{}{
+		"token":          spec.Token,
+		"type":           spec.DocType,
+		"file_extension": spec.FileExtension,
+	}
+	if strings.TrimSpace(spec.SubID) != "" {
+		body["sub_id"] = spec.SubID
+	}
+	if spec.OnlySchema {
+		body["only_schema"] = true
+	}
+
+	dr := common.NewDryRunAPI().
+		Desc("3-step orchestration: create export task -> limited polling -> download file").
+		POST("/open-apis/drive/v1/export_tasks").
+		Body(body).
+		Set("output_dir", p.OutputDir)
+	if name := strings.TrimSpace(p.FileName); name != "" {
+		dr.Set("file_name", ensureExportFileExtension(sanitizeExportFileName(name, spec.Token), spec.FileExtension))
+	}
+	return dr
+}
+
+// RunExport drives create export task -> bounded poll -> optional download. It
+// is the shared core behind both drive +export and sheets +workbook-export. An
+// empty p.OutputDir skips the download step and returns the ready file token.
+func RunExport(ctx context.Context, runtime *common.RuntimeContext, p ExportParams) error {
+	spec := p.spec()
+	outputDir := p.OutputDir
+	preferredFileName := strings.TrimSpace(p.FileName)
+	overwrite := p.Overwrite
+
+	// Markdown export bypasses the async export task and writes the fetched
+	// markdown content directly to disk. Uses the V2 docs_ai fetch API for
+	// higher-quality Lark-flavored Markdown output.
+	if spec.FileExtension == "markdown" {
+		fmt.Fprintf(runtime.IO().ErrOut, "Exporting docx as markdown: %s\n", common.MaskToken(spec.Token))
+		apiPath := fmt.Sprintf("/open-apis/docs_ai/v1/documents/%s/fetch", validate.EncodePathSegment(spec.Token))
+		data, err := runtime.CallAPITyped(
+			"POST",
+			apiPath,
+			nil,
+			map[string]interface{}{
+				"format": "markdown",
+			},
+		)
+		if err != nil {
+			return err
 		}
-		outputDir := runtime.Str("output-dir")
-		preferredFileName := strings.TrimSpace(runtime.Str("file-name"))
-		overwrite := runtime.Bool("overwrite")
 
-		// Markdown export bypasses the async export task and writes the fetched
-		// markdown content directly to disk. Uses the V2 docs_ai fetch API for
-		// higher-quality Lark-flavored Markdown output.
-		if spec.FileExtension == "markdown" {
-			fmt.Fprintf(runtime.IO().ErrOut, "Exporting docx as markdown: %s\n", common.MaskToken(spec.Token))
-			apiPath := fmt.Sprintf("/open-apis/docs_ai/v1/documents/%s/fetch", validate.EncodePathSegment(spec.Token))
-			data, err := runtime.CallAPITyped(
-				"POST",
-				apiPath,
-				nil,
-				map[string]interface{}{
-					"format": "markdown",
-				},
-			)
+		// Extract content from the V2 response: data.document.content
+		doc, ok := data["document"].(map[string]interface{})
+		if !ok {
+			return errs.NewInternalError(errs.SubtypeInvalidResponse, "invalid markdown fetch response: missing document object")
+		}
+		content, ok := doc["content"].(string)
+		if !ok {
+			return errs.NewInternalError(errs.SubtypeInvalidResponse, "invalid markdown fetch response: missing document.content")
+		}
+
+		fileName := preferredFileName
+		if fileName == "" {
+			// Prefer the remote title for the exported file name, but still fall
+			// back to the token if metadata is empty.
+			title, err := common.FetchDriveMetaTitle(runtime, spec.Token, spec.DocType)
 			if err != nil {
-				return err
+				fmt.Fprintf(runtime.IO().ErrOut, "Title lookup failed, using token as filename: %v\n", err)
+				title = spec.Token
 			}
+			fileName = title
+		}
+		fileName = ensureExportFileExtension(sanitizeExportFileName(fileName, spec.Token), spec.FileExtension)
+		savedPath, err := saveContentToOutputDir(runtime.FileIO(), outputDir, fileName, []byte(content), overwrite)
+		if err != nil {
+			return err
+		}
 
-			// Extract content from the V2 response: data.document.content
-			doc, ok := data["document"].(map[string]interface{})
-			if !ok {
-				return errs.NewInternalError(errs.SubtypeInvalidResponse, "invalid markdown fetch response: missing document object")
+		runtime.Out(map[string]interface{}{
+			"token":          spec.Token,
+			"doc_type":       spec.DocType,
+			"file_extension": spec.FileExtension,
+			"file_name":      filepath.Base(savedPath),
+			"saved_path":     savedPath,
+			"size_bytes":     len(content),
+		}, nil)
+		return nil
+	}
+
+	ticket, err := createDriveExportTask(runtime, spec)
+	if err != nil {
+		return err
+	}
+	fmt.Fprintf(runtime.IO().ErrOut, "Created export task: %s\n", ticket)
+
+	var lastStatus driveExportStatus
+	var lastPollErr error
+	hasObservedStatus := false
+	// Keep the command responsive by polling for a bounded window. If the task
+	// is still running after that, return a resume command instead of blocking.
+	for attempt := 1; attempt <= driveExportPollAttempts; attempt++ {
+		if attempt > 1 {
+			select {
+			case <-ctx.Done():
+				return ctx.Err()
+			case <-time.After(driveExportPollInterval):
 			}
-			content, ok := doc["content"].(string)
-			if !ok {
-				return errs.NewInternalError(errs.SubtypeInvalidResponse, "invalid markdown fetch response: missing document.content")
+		}
+		if err := ctx.Err(); err != nil {
+			return err
+		}
+
+		status, err := getDriveExportStatus(runtime, spec.Token, ticket)
+		if err != nil {
+			// Treat polling failures as transient so short-lived backend hiccups
+			// do not immediately fail an otherwise healthy export task.
+			lastPollErr = err
+			fmt.Fprintf(runtime.IO().ErrOut, "Export status attempt %d/%d failed: %v\n", attempt, driveExportPollAttempts, err)
+			continue
+		}
+		lastStatus = status
+		hasObservedStatus = true
+
+		if status.Ready() {
+			fmt.Fprintf(runtime.IO().ErrOut, "Export task completed: %s\n", common.MaskToken(status.FileToken))
+
+			// Export-only mode: caller wants the ready file token / metadata but
+			// no local download (e.g. sheets +workbook-export without an output
+			// path). Skip the download and return the status envelope.
+			if strings.TrimSpace(outputDir) == "" {
+				runtime.Out(map[string]interface{}{
+					"ticket":         ticket,
+					"token":          spec.Token,
+					"doc_type":       spec.DocType,
+					"file_extension": spec.FileExtension,
+					"file_token":     status.FileToken,
+					"file_name":      status.FileName,
+					"file_size":      status.FileSize,
+					"ready":          true,
+					"downloaded":     false,
+				}, nil)
+				return nil
 			}
 
 			fileName := preferredFileName
 			if fileName == "" {
-				// Prefer the remote title for the exported file name, but still fall
-				// back to the token if metadata is empty.
-				title, err := common.FetchDriveMetaTitle(runtime, spec.Token, spec.DocType)
-				if err != nil {
-					fmt.Fprintf(runtime.IO().ErrOut, "Title lookup failed, using token as filename: %v\n", err)
-					title = spec.Token
-				}
-				fileName = title
+				fileName = status.FileName
 			}
 			fileName = ensureExportFileExtension(sanitizeExportFileName(fileName, spec.Token), spec.FileExtension)
-			savedPath, err := saveContentToOutputDir(runtime.FileIO(), outputDir, fileName, []byte(content), overwrite)
+			out, err := downloadDriveExportFile(ctx, runtime, status.FileToken, outputDir, fileName, overwrite)
 			if err != nil {
-				return err
+				recoveryCommand := driveExportDownloadCommand(status.FileToken, fileName, outputDir, overwrite)
+				hint := fmt.Sprintf(
+					"the export artifact is already ready (ticket=%s, file_token=%s)\nretry download with: %s",
+					ticket,
+					status.FileToken,
+					recoveryCommand,
+				)
+				return appendDriveExportRecoveryHint(err, hint)
 			}
-
-			runtime.Out(map[string]interface{}{
-				"token":          spec.Token,
-				"doc_type":       spec.DocType,
-				"file_extension": spec.FileExtension,
-				"file_name":      filepath.Base(savedPath),
-				"saved_path":     savedPath,
-				"size_bytes":     len(content),
-			}, nil)
+			out["ticket"] = ticket
+			out["doc_type"] = spec.DocType
+			out["file_extension"] = spec.FileExtension
+			runtime.Out(out, nil)
 			return nil
 		}
 
-		ticket, err := createDriveExportTask(runtime, spec)
-		if err != nil {
-			return err
-		}
-		fmt.Fprintf(runtime.IO().ErrOut, "Created export task: %s\n", ticket)
-
-		var lastStatus driveExportStatus
-		var lastPollErr error
-		hasObservedStatus := false
-		// Keep the command responsive by polling for a bounded window. If the task
-		// is still running after that, return a resume command instead of blocking.
-		for attempt := 1; attempt <= driveExportPollAttempts; attempt++ {
-			if attempt > 1 {
-				select {
-				case <-ctx.Done():
-					return ctx.Err()
-				case <-time.After(driveExportPollInterval):
-				}
+		if status.Failed() {
+			msg := strings.TrimSpace(status.JobErrorMsg)
+			if msg == "" {
+				msg = status.StatusLabel()
 			}
-			if err := ctx.Err(); err != nil {
-				return err
-			}
-
-			status, err := getDriveExportStatus(runtime, spec.Token, ticket)
-			if err != nil {
-				// Treat polling failures as transient so short-lived backend hiccups
-				// do not immediately fail an otherwise healthy export task.
-				lastPollErr = err
-				fmt.Fprintf(runtime.IO().ErrOut, "Export status attempt %d/%d failed: %v\n", attempt, driveExportPollAttempts, err)
-				continue
-			}
-			lastStatus = status
-			hasObservedStatus = true
-
-			if status.Ready() {
-				fmt.Fprintf(runtime.IO().ErrOut, "Export task completed: %s\n", common.MaskToken(status.FileToken))
-				fileName := preferredFileName
-				if fileName == "" {
-					fileName = status.FileName
-				}
-				fileName = ensureExportFileExtension(sanitizeExportFileName(fileName, spec.Token), spec.FileExtension)
-				out, err := downloadDriveExportFile(ctx, runtime, status.FileToken, outputDir, fileName, overwrite)
-				if err != nil {
-					recoveryCommand := driveExportDownloadCommand(status.FileToken, fileName, outputDir, overwrite)
-					hint := fmt.Sprintf(
-						"the export artifact is already ready (ticket=%s, file_token=%s)\nretry download with: %s",
-						ticket,
-						status.FileToken,
-						recoveryCommand,
-					)
-					return appendDriveExportRecoveryHint(err, hint)
-				}
-				out["ticket"] = ticket
-				out["doc_type"] = spec.DocType
-				out["file_extension"] = spec.FileExtension
-				runtime.Out(out, nil)
-				return nil
-			}
-
-			if status.Failed() {
-				msg := strings.TrimSpace(status.JobErrorMsg)
-				if msg == "" {
-					msg = status.StatusLabel()
-				}
-				return errs.NewAPIError(errs.SubtypeServerError, "export task failed: %s (ticket=%s)", msg, ticket)
-			}
-
-			fmt.Fprintf(runtime.IO().ErrOut, "Export status %d/%d: %s\n", attempt, driveExportPollAttempts, status.StatusLabel())
+			return errs.NewAPIError(errs.SubtypeServerError, "export task failed: %s (ticket=%s)", msg, ticket)
 		}
 
-		nextCommand := driveExportTaskResultCommand(ticket, spec.Token)
-		if !hasObservedStatus && lastPollErr != nil {
-			hint := fmt.Sprintf(
-				"the export task was created but every status poll failed (ticket=%s)\nretry status lookup with: %s",
-				ticket,
-				nextCommand,
-			)
-			return appendDriveExportRecoveryHint(lastPollErr, hint)
-		}
+		fmt.Fprintf(runtime.IO().ErrOut, "Export status %d/%d: %s\n", attempt, driveExportPollAttempts, status.StatusLabel())
+	}
 
-		failed := false
-		var jobStatus interface{}
-		jobStatusLabel := "unknown"
-		if hasObservedStatus {
-			failed = lastStatus.Failed()
-			jobStatus = lastStatus.JobStatus
-			jobStatusLabel = lastStatus.StatusLabel()
-		}
-		// Return the last observed status so callers can resume from a known task
-		// state instead of losing all progress information on timeout.
-		result := map[string]interface{}{
-			"ticket":           ticket,
-			"token":            spec.Token,
-			"doc_type":         spec.DocType,
-			"file_extension":   spec.FileExtension,
-			"ready":            false,
-			"failed":           failed,
-			"job_status":       jobStatus,
-			"job_status_label": jobStatusLabel,
-			"timed_out":        true,
-			"next_command":     nextCommand,
-		}
-		if preferredFileName != "" {
-			result["file_name"] = ensureExportFileExtension(sanitizeExportFileName(preferredFileName, spec.Token), spec.FileExtension)
-		}
-		runtime.Out(result, nil)
-		fmt.Fprintf(runtime.IO().ErrOut, "Export task is still in progress. Continue with: %s\n", nextCommand)
-		return nil
-	},
+	nextCommand := driveExportTaskResultCommand(ticket, spec.Token)
+	if !hasObservedStatus && lastPollErr != nil {
+		hint := fmt.Sprintf(
+			"the export task was created but every status poll failed (ticket=%s)\nretry status lookup with: %s",
+			ticket,
+			nextCommand,
+		)
+		return appendDriveExportRecoveryHint(lastPollErr, hint)
+	}
+
+	failed := false
+	var jobStatus interface{}
+	jobStatusLabel := "unknown"
+	if hasObservedStatus {
+		failed = lastStatus.Failed()
+		jobStatus = lastStatus.JobStatus
+		jobStatusLabel = lastStatus.StatusLabel()
+	}
+	// Return the last observed status so callers can resume from a known task
+	// state instead of losing all progress information on timeout.
+	result := map[string]interface{}{
+		"ticket":           ticket,
+		"token":            spec.Token,
+		"doc_type":         spec.DocType,
+		"file_extension":   spec.FileExtension,
+		"ready":            false,
+		"failed":           failed,
+		"job_status":       jobStatus,
+		"job_status_label": jobStatusLabel,
+		"timed_out":        true,
+		"next_command":     nextCommand,
+	}
+	if preferredFileName != "" {
+		result["file_name"] = ensureExportFileExtension(sanitizeExportFileName(preferredFileName, spec.Token), spec.FileExtension)
+	}
+	runtime.Out(result, nil)
+	fmt.Fprintf(runtime.IO().ErrOut, "Export task is still in progress. Continue with: %s\n", nextCommand)
+	return nil
 }

shortcuts/drive/drive_export_test.go

@@ -497,6 +497,72 @@ func TestDriveExportAsyncSuccess(t *testing.T) {
 	}
 }
 
+// TestDriveExportEmptyOutputDirDownloadsToCwd guards the export refactor: an
+// explicit empty --output-dir must still download to the current directory
+// (normalized to "."), not trigger the export-only no-download path that the
+// shared RunExport core uses for sheets +workbook-export.
+func TestDriveExportEmptyOutputDirDownloadsToCwd(t *testing.T) {
+	f, stdout, _, reg := cmdutil.TestFactory(t, driveTestConfig())
+	reg.Register(&httpmock.Stub{
+		Method: "POST",
+		URL:    "/open-apis/drive/v1/export_tasks",
+		Body:   map[string]interface{}{"code": 0, "data": map[string]interface{}{"ticket": "tk_e"}},
+	})
+	reg.Register(&httpmock.Stub{
+		Method: "GET",
+		URL:    "/open-apis/drive/v1/export_tasks/tk_e",
+		Body: map[string]interface{}{"code": 0, "data": map[string]interface{}{
+			"result": map[string]interface{}{
+				"job_status": 0, "file_token": "box_e", "file_name": "report",
+				"file_extension": "pdf", "type": "docx", "file_size": 3,
+			},
+		}},
+	})
+	reg.Register(&httpmock.Stub{
+		Method:  "GET",
+		URL:     "/open-apis/drive/v1/export_tasks/file/box_e/download",
+		Status:  200,
+		RawBody: []byte("pdf"),
+		Headers: http.Header{
+			"Content-Type":        []string{"application/pdf"},
+			"Content-Disposition": []string{`attachment; filename="report.pdf"`},
+		},
+	})
+
+	tmpDir := t.TempDir()
+	withDriveWorkingDir(t, tmpDir)
+
+	prevAttempts, prevInterval := driveExportPollAttempts, driveExportPollInterval
+	driveExportPollAttempts, driveExportPollInterval = 1, 0
+	t.Cleanup(func() {
+		driveExportPollAttempts, driveExportPollInterval = prevAttempts, prevInterval
+	})
+
+	err := mountAndRunDrive(t, DriveExport, []string{
+		"+export",
+		"--token", "docx123",
+		"--doc-type", "docx",
+		"--file-extension", "pdf",
+		"--output-dir", "",
+		"--as", "bot",
+	}, f, stdout)
+	if err != nil {
+		t.Fatalf("unexpected error: %v", err)
+	}
+
+	// Empty --output-dir must still write to cwd, not skip the download.
+	data, err := os.ReadFile(filepath.Join(tmpDir, "report.pdf"))
+	if err != nil {
+		t.Fatalf("empty --output-dir should still download to cwd: %v", err)
+	}
+	if string(data) != "pdf" {
+		t.Fatalf("downloaded content = %q", string(data))
+	}
+	if strings.Contains(stdout.String(), `"downloaded": false`) {
+		t.Fatalf("export-only path must not trigger for drive +export: %s", stdout.String())
+	}
+}
+
 func TestDriveExportAsyncUsesProvidedFileName(t *testing.T) {
 	f, stdout, _, reg := cmdutil.TestFactory(t, driveTestConfig())
 	reg.Register(&httpmock.Stub{

shortcuts/drive/drive_import.go

@@ -35,132 +35,164 @@ var DriveImport = common.Shortcut{
 		{Name: "target-token", Desc: "existing token to import data into (only for type=bitable); when set, data is mounted into this bitable instead of creating a new one"},
 	},
 	Validate: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		return validateDriveImportSpec(driveImportSpec{
-			FilePath:    runtime.Str("file"),
-			DocType:     strings.ToLower(runtime.Str("type")),
-			FolderToken: runtime.Str("folder-token"),
-			Name:        runtime.Str("name"),
-			TargetToken: runtime.Str("target-token"),
-		})
+		return ValidateImport(importParamsFromFlags(runtime))
 	},
 	DryRun: func(ctx context.Context, runtime *common.RuntimeContext) *common.DryRunAPI {
-		spec := driveImportSpec{
-			FilePath:    runtime.Str("file"),
-			DocType:     strings.ToLower(runtime.Str("type")),
-			FolderToken: runtime.Str("folder-token"),
-			Name:        runtime.Str("name"),
-			TargetToken: runtime.Str("target-token"),
-		}
-		fileSize, err := preflightDriveImportFile(runtime.FileIO(), &spec)
-		if err != nil {
-			return common.NewDryRunAPI().Set("error", err.Error())
-		}
-		if valErr := validateDriveImportSpec(spec); valErr != nil {
-			return common.NewDryRunAPI().Set("error", valErr.Error())
-		}
-
-		dry := common.NewDryRunAPI()
-		dry.Desc("Upload file (single-part or multipart) -> create import task -> poll status")
-
-		appendDriveImportFolderTokenWikiCheckDryRun(dry, spec)
-		appendDriveImportUploadDryRun(dry, spec, fileSize)
-
-		dry.POST("/open-apis/drive/v1/import_tasks").
-			Desc("[2] Create import task").
-			Body(spec.CreateTaskBody("<file_token>"))
-
-		dry.GET("/open-apis/drive/v1/import_tasks/:ticket").
-			Desc("[3] Poll import task result").
-			Set("ticket", "<ticket>")
-		if runtime.IsBot() {
-			dry.Desc("After the import result returns the final cloud document target in bot mode, the CLI will also try to grant the current CLI user full_access (可管理权限) on it.")
-		}
-
-		return dry
+		return PlanImportDryRun(runtime, importParamsFromFlags(runtime))
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		spec := driveImportSpec{
-			FilePath:    runtime.Str("file"),
-			DocType:     strings.ToLower(runtime.Str("type")),
-			FolderToken: runtime.Str("folder-token"),
-			Name:        runtime.Str("name"),
-			TargetToken: runtime.Str("target-token"),
-		}
-		if _, err := preflightDriveImportFile(runtime.FileIO(), &spec); err != nil {
-			return err
-		}
-		if err := rejectDriveImportWikiFolderToken(runtime, spec.FolderToken); err != nil {
-			return err
-		}
-
-		// Step 1: Upload file as media
-		fileToken, uploadErr := uploadMediaForImport(ctx, runtime, spec.FilePath, spec.SourceFileName(), spec.DocType)
-		if uploadErr != nil {
-			return uploadErr
-		}
-
-		fmt.Fprintf(runtime.IO().ErrOut, "Creating import task for %s as %s...\n", spec.TargetFileName(), spec.DocType)
-
-		// Step 2: Create import task
-		ticket, err := createDriveImportTask(runtime, spec, fileToken)
-		if err != nil {
-			return err
-		}
-
-		// Step 3: Poll task
-		fmt.Fprintf(runtime.IO().ErrOut, "Polling import task %s...\n", ticket)
-
-		status, ready, err := pollDriveImportTask(runtime, ticket)
-		if err != nil {
-			return err
-		}
-
-		// Some intermediate responses omit the final type, so fall back to the
-		// requested type to keep the output shape stable.
-		resultType := status.DocType
-		if resultType == "" {
-			resultType = spec.DocType
-		}
-		out := map[string]interface{}{
-			"ticket":           ticket,
-			"type":             resultType,
-			"ready":            ready,
-			"job_status":       status.JobStatus,
-			"job_status_label": status.StatusLabel(),
-		}
-		if status.Token != "" {
-			out["token"] = status.Token
-		}
-		if statusURL := strings.TrimSpace(status.URL); statusURL != "" {
-			out["url"] = statusURL
-		} else if status.Token != "" {
-			if u := common.BuildResourceURL(runtime.Config.Brand, normalizeDriveImportKindForURL(resultType, spec.DocType), status.Token); u != "" {
-				out["url"] = u
-			}
-		}
-		if status.JobErrorMsg != "" {
-			out["job_error_msg"] = status.JobErrorMsg
-		}
-		if status.Extra != nil {
-			out["extra"] = status.Extra
-		}
-		if !ready {
-			nextCommand := driveImportTaskResultCommand(ticket)
-			fmt.Fprintf(runtime.IO().ErrOut, "Import task is still in progress. Continue with: %s\n", nextCommand)
-			out["timed_out"] = true
-			out["next_command"] = nextCommand
-		}
-		if ready {
-			if grant := common.AutoGrantCurrentUserDrivePermission(runtime, common.GetString(out, "token"), resultType); grant != nil {
-				out["permission_grant"] = grant
-			}
-		}
-
-		runtime.Out(out, nil)
-		return nil
+		return RunImport(ctx, runtime, importParamsFromFlags(runtime))
 	},
 }
 
+// ImportParams holds the user-facing inputs for an import flow, decoupled from
+// cobra flags so other command groups (e.g. sheets +workbook-import) can reuse
+// the drive import implementation without taking a dependency on a --type flag.
+type ImportParams struct {
+	File        string
+	DocType     string
+	FolderToken string
+	Name        string
+	TargetToken string
+}
+
+func (p ImportParams) spec() driveImportSpec {
+	return driveImportSpec{
+		FilePath:    p.File,
+		DocType:     strings.ToLower(p.DocType),
+		FolderToken: p.FolderToken,
+		Name:        p.Name,
+		TargetToken: p.TargetToken,
+	}
+}
+
+// importParamsFromFlags reads the standard drive +import flag set.
+func importParamsFromFlags(runtime *common.RuntimeContext) ImportParams {
+	return ImportParams{
+		File:        runtime.Str("file"),
+		DocType:     runtime.Str("type"),
+		FolderToken: runtime.Str("folder-token"),
+		Name:        runtime.Str("name"),
+		TargetToken: runtime.Str("target-token"),
+	}
+}
+
+// ValidateImport runs the CLI-level compatibility checks for an import.
+func ValidateImport(p ImportParams) error {
+	return validateDriveImportSpec(p.spec())
+}
+
+// PlanImportDryRun builds the dry-run plan (upload -> create task -> poll) for
+// an import without performing any network or file I/O beyond a local stat.
+func PlanImportDryRun(runtime *common.RuntimeContext, p ImportParams) *common.DryRunAPI {
+	spec := p.spec()
+	fileSize, err := preflightDriveImportFile(runtime.FileIO(), &spec)
+	if err != nil {
+		return common.NewDryRunAPI().Set("error", err.Error())
+	}
+	if valErr := validateDriveImportSpec(spec); valErr != nil {
+		return common.NewDryRunAPI().Set("error", valErr.Error())
+	}
+
+	dry := common.NewDryRunAPI()
+	dry.Desc("Upload file (single-part or multipart) -> create import task -> poll status")
+
+	appendDriveImportFolderTokenWikiCheckDryRun(dry, spec)
+	appendDriveImportUploadDryRun(dry, spec, fileSize)
+
+	dry.POST("/open-apis/drive/v1/import_tasks").
+		Desc("[2] Create import task").
+		Body(spec.CreateTaskBody("<file_token>"))
+
+	dry.GET("/open-apis/drive/v1/import_tasks/:ticket").
+		Desc("[3] Poll import task result").
+		Set("ticket", "<ticket>")
+	if runtime.IsBot() {
+		dry.Desc("After the import result returns the final cloud document target in bot mode, the CLI will also try to grant the current CLI user full_access (可管理权限) on it.")
+	}
+
+	return dry
+}
+
+// RunImport executes the full import flow: upload media -> create import task ->
+// bounded poll, then writes the result envelope to the runtime output. It is
+// the shared core behind both drive +import and sheets +workbook-import.
+func RunImport(ctx context.Context, runtime *common.RuntimeContext, p ImportParams) error {
+	spec := p.spec()
+	if _, err := preflightDriveImportFile(runtime.FileIO(), &spec); err != nil {
+		return err
+	}
+	if err := rejectDriveImportWikiFolderToken(runtime, spec.FolderToken); err != nil {
+		return err
+	}
+
+	// Step 1: Upload file as media
+	fileToken, uploadErr := uploadMediaForImport(ctx, runtime, spec.FilePath, spec.SourceFileName(), spec.DocType)
+	if uploadErr != nil {
+		return uploadErr
+	}
+
+	fmt.Fprintf(runtime.IO().ErrOut, "Creating import task for %s as %s...\n", spec.TargetFileName(), spec.DocType)
+
+	// Step 2: Create import task
+	ticket, err := createDriveImportTask(runtime, spec, fileToken)
+	if err != nil {
+		return err
+	}
+
+	// Step 3: Poll task
+	fmt.Fprintf(runtime.IO().ErrOut, "Polling import task %s...\n", ticket)
+
+	status, ready, err := pollDriveImportTask(runtime, ticket)
+	if err != nil {
+		return err
+	}
+
+	// Some intermediate responses omit the final type, so fall back to the
+	// requested type to keep the output shape stable.
+	resultType := status.DocType
+	if resultType == "" {
+		resultType = spec.DocType
+	}
+	out := map[string]interface{}{
+		"ticket":           ticket,
+		"type":             resultType,
+		"ready":            ready,
+		"job_status":       status.JobStatus,
+		"job_status_label": status.StatusLabel(),
+	}
+	if status.Token != "" {
+		out["token"] = status.Token
+	}
+	if statusURL := strings.TrimSpace(status.URL); statusURL != "" {
+		out["url"] = statusURL
+	} else if status.Token != "" {
+		if u := common.BuildResourceURL(runtime.Config.Brand, normalizeDriveImportKindForURL(resultType, spec.DocType), status.Token); u != "" {
+			out["url"] = u
+		}
+	}
+	if status.JobErrorMsg != "" {
+		out["job_error_msg"] = status.JobErrorMsg
+	}
+	if status.Extra != nil {
+		out["extra"] = status.Extra
+	}
+	if !ready {
+		nextCommand := driveImportTaskResultCommand(ticket)
+		fmt.Fprintf(runtime.IO().ErrOut, "Import task is still in progress. Continue with: %s\n", nextCommand)
+		out["timed_out"] = true
+		out["next_command"] = nextCommand
+	}
+	if ready {
+		if grant := common.AutoGrantCurrentUserDrivePermission(runtime, common.GetString(out, "token"), resultType); grant != nil {
+			out["permission_grant"] = grant
+		}
+	}
+
+	runtime.Out(out, nil)
+	return nil
+}
+
 func preflightDriveImportFile(fio fileio.FileIO, spec *driveImportSpec) (int64, error) {
 	// Keep dry-run and execution aligned on path normalization, file existence,
 	// and format-specific size limits before planning the upload path.

shortcuts/sheets/batch_op_contract_test.go

@@ -177,6 +177,18 @@ func TestBatchOp_BodyMatchesStandalone(t *testing.T) {
 			args:     []string{"--sheet-id", "sh1", "--color", "#FF0000"},
 			subInput: `{"sheet-id":"sh1","color":"#FF0000"}`,
 		},
+		{
+			shortcut: "+sheet-show-gridline",
+			sc:       SheetShowGridline,
+			args:     []string{"--sheet-id", "sh1"},
+			subInput: `{"sheet-id":"sh1"}`,
+		},
+		{
+			shortcut: "+sheet-hide-gridline",
+			sc:       SheetHideGridline,
+			args:     []string{"--sheet-id", "sh1"},
+			subInput: `{"sheet-id":"sh1"}`,
+		},
 		{
 			shortcut: "+dropdown-set",
 			sc:       DropdownSet,

shortcuts/sheets/batch_op_dispatch.go

@@ -150,6 +150,12 @@ var batchOpDispatch = map[string]batchOpMapping{
 		return sheetVisibilityInput(fv, t, sid, sn, "unhide")
 	}},
 	"+sheet-set-tab-color": {"modify_workbook_structure", sheetSetTabColorInput},
+	"+sheet-show-gridline": {"modify_workbook_structure", func(fv flagView, t, sid, sn string) (map[string]interface{}, error) {
+		return sheetVisibilityInput(fv, t, sid, sn, "show_gridline")
+	}},
+	"+sheet-hide-gridline": {"modify_workbook_structure", func(fv flagView, t, sid, sn string) (map[string]interface{}, error) {
+		return sheetVisibilityInput(fv, t, sid, sn, "hide_gridline")
+	}},
 
 	// ─── 对象族 CRUD (manage_*_object, operation 区分) ─────────────
 	"+chart-create": {"manage_chart_object", objCreateTranslate(chartSpec)},

shortcuts/sheets/data/flag-defs.json

@@ -54,7 +54,7 @@
         "kind": "own",
         "type": "int",
         "required": "optional",
-        "desc": "Insert position; appended to the end when omitted",
+        "desc": "Insert position (0-based); appended to the end when omitted",
         "default": "-1"
       },
       {
@@ -413,6 +413,86 @@
       }
     ]
   },
+  "+sheet-hide-gridline": {
+    "risk": "write",
+    "flags": [
+      {
+        "name": "url",
+        "kind": "public",
+        "type": "string",
+        "required": "xor",
+        "desc": "Spreadsheet URL (XOR with `--spreadsheet-token`)"
+      },
+      {
+        "name": "spreadsheet-token",
+        "kind": "public",
+        "type": "string",
+        "required": "xor",
+        "desc": "Spreadsheet token (XOR with `--url`)"
+      },
+      {
+        "name": "sheet-id",
+        "kind": "public",
+        "type": "string",
+        "required": "xor",
+        "desc": "Sheet reference_id (XOR with `--sheet-name`)"
+      },
+      {
+        "name": "sheet-name",
+        "kind": "public",
+        "type": "string",
+        "required": "xor",
+        "desc": "Sheet name (XOR with `--sheet-id`)"
+      },
+      {
+        "name": "dry-run",
+        "kind": "system",
+        "type": "bool",
+        "required": "optional",
+        "desc": ""
+      }
+    ]
+  },
+  "+sheet-show-gridline": {
+    "risk": "write",
+    "flags": [
+      {
+        "name": "url",
+        "kind": "public",
+        "type": "string",
+        "required": "xor",
+        "desc": "Spreadsheet URL (XOR with `--spreadsheet-token`)"
+      },
+      {
+        "name": "spreadsheet-token",
+        "kind": "public",
+        "type": "string",
+        "required": "xor",
+        "desc": "Spreadsheet token (XOR with `--url`)"
+      },
+      {
+        "name": "sheet-id",
+        "kind": "public",
+        "type": "string",
+        "required": "xor",
+        "desc": "Sheet reference_id (XOR with `--sheet-name`)"
+      },
+      {
+        "name": "sheet-name",
+        "kind": "public",
+        "type": "string",
+        "required": "xor",
+        "desc": "Sheet name (XOR with `--sheet-id`)"
+      },
+      {
+        "name": "dry-run",
+        "kind": "system",
+        "type": "bool",
+        "required": "optional",
+        "desc": ""
+      }
+    ]
+  },
   "+workbook-create": {
     "risk": "write",
     "flags": [
@@ -431,27 +511,45 @@
         "desc": "Target folder token; placed at the drive root when omitted"
       },
       {
-        "name": "headers",
+        "name": "values",
         "kind": "own",
         "type": "string",
         "required": "optional",
-        "desc": "Header row as a JSON array: `[\"Col A\",\"Col B\"]`",
+        "desc": "Untyped initial data as one 2D JSON array (`[[\"alice\",95]]`); values are written as-is with their type auto-detected, through the same batched set_cell_range path as --sheets — pair with --styles for number formats, colors, merges, and row/col sizes",
         "input": [
           "file",
           "stdin"
         ]
       },
       {
-        "name": "values",
+        "name": "sheets",
         "kind": "own",
         "type": "string",
         "required": "optional",
-        "desc": "Initial data as a 2D JSON array: `[[\"alice\",95]]`",
+        "desc": "Typed table payload as JSON (same shape as `+table-put`): a top-level `sheets` array, each item `{name, start_cell?, mode?, header?, allow_overwrite?, columns:[\"colA\",\"colB\",...], data:[[...]], dtypes?:{colA:pandasDtype, ...}, formats?:{colA:numberFormat, ...}}`. Agents typically build it from a DataFrame via `{**json.loads(df.to_json(orient=\"split\")), \"dtypes\": df.dtypes.astype(str).to_dict()}`. Mutually exclusive with --values and --dataframe. Creates the workbook, then writes typed type-faithful data (dates land as real dates, numbers keep precision).",
         "input": [
           "file",
           "stdin"
         ]
       },
+      {
+        "name": "styles",
+        "kind": "own",
+        "type": "string",
+        "required": "optional",
+        "desc": "Initial visual operations as JSON: top-level `{styles:[...]}`. Each item corresponds to one target sheet and must include `name`, plus at least one of `cell_styles` / `row_sizes` / `col_sizes` / `cell_merges`. `cell_styles` entries use +cells-set-style fields with a cell range; row/col sizes use dimension ranges plus type/size; merges use cell ranges plus optional merge_type. With --sheets, styles array length/order/name must match --sheets.sheets. With --values, pass exactly one styles item for the initial sheet (its name is ignored).",
+        "input": [
+          "file",
+          "stdin"
+        ]
+      },
+      {
+        "name": "dataframe",
+        "kind": "own",
+        "type": "string",
+        "required": "optional",
+        "desc": "Single-sheet typed table from one Arrow IPC file (Feather v2 — what `pandas.DataFrame.to_feather()` writes), mutually exclusive with --values and --sheets. Pass `@<path>` for a file or `-` for binary stdin (same convention as other input flags). Arrow bytes are read raw — no TrimSpace / BOM strip — so the IPC magic survives intact (unlike text input flags). Column types come from the Arrow schema; per-column `number_format` may be set via Arrow field metadata. Creates the workbook and fills its default sheet (`Sheet1` — adopted in place, no empty Sheet1 left behind). For multi-sheet or non-default placement, use `--sheets` instead."
+      },
       {
         "name": "dry-run",
         "kind": "system",
@@ -513,6 +611,32 @@
       }
     ]
   },
+  "+workbook-import": {
+    "risk": "write",
+    "flags": [
+      {
+        "name": "file",
+        "kind": "own",
+        "type": "string",
+        "required": "required",
+        "desc": "Local file path (.xlsx / .xls / .csv)"
+      },
+      {
+        "name": "folder-token",
+        "kind": "own",
+        "type": "string",
+        "required": "optional",
+        "desc": "Target folder token; imported to the cloud drive root when omitted"
+      },
+      {
+        "name": "name",
+        "kind": "own",
+        "type": "string",
+        "required": "optional",
+        "desc": "Imported spreadsheet name; defaults to the local file name without its extension"
+      }
+    ]
+  },
   "+sheet-info": {
     "risk": "read",
     "flags": [
@@ -1082,9 +1206,8 @@
         "kind": "own",
         "type": "int",
         "required": "optional",
-        "desc": "Safety cap; default 200000",
-        "default": "200000",
-        "hidden": true
+        "desc": "Max output chars per call; default 500000 (safety cap). Large reads are usually better redirected to a file; only lower it (e.g. 25000) when you want results inline without triggering file offload, paging via has_more",
+        "default": "500000"
       },
       {
         "name": "skip-hidden",
@@ -1192,9 +1315,8 @@
         "kind": "own",
         "type": "int",
         "required": "optional",
-        "desc": "Safety cap; default 200000",
-        "default": "200000",
-        "hidden": true
+        "desc": "Max output chars per call; default 500000 (safety cap). Large reads are usually better redirected to a file; only lower it (e.g. 25000) when you want results inline without triggering file offload, paging via has_more",
+        "default": "500000"
       },
       {
         "name": "include-row-prefix",
@@ -1212,19 +1334,72 @@
         "desc": "Skip hidden rows and columns; default `false`"
       },
       {
-        "name": "rows-json",
+        "name": "dry-run",
+        "kind": "system",
+        "type": "bool",
+        "required": "optional",
+        "desc": "Print the request path and parameters without executing"
+      }
+    ]
+  },
+  "+table-get": {
+    "risk": "read",
+    "flags": [
+      {
+        "name": "url",
+        "kind": "public",
+        "type": "string",
+        "required": "xor",
+        "desc": "Spreadsheet URL (XOR with `--spreadsheet-token`)"
+      },
+      {
+        "name": "spreadsheet-token",
+        "kind": "public",
+        "type": "string",
+        "required": "xor",
+        "desc": "Spreadsheet token (XOR with `--url`)"
+      },
+      {
+        "name": "sheet-id",
+        "kind": "own",
+        "type": "string",
+        "required": "optional",
+        "desc": "Read only this sheet (by id); omit to read all sheets"
+      },
+      {
+        "name": "sheet-name",
+        "kind": "own",
+        "type": "string",
+        "required": "optional",
+        "desc": "Read only this sheet (by name); omit to read all sheets"
+      },
+      {
+        "name": "range",
+        "kind": "own",
+        "type": "string",
+        "required": "optional",
+        "desc": "A1 range to read; omit to read each sheet's full used range (spans internal blank rows/columns, not just the A1 current region)"
+      },
+      {
+        "name": "no-header",
         "kind": "own",
         "type": "bool",
         "required": "optional",
-        "desc": "Return structured rows ({row_number, values:{col→cell}}) instead of CSV text; default false",
-        "default": "false"
+        "desc": "Treat the first row as data instead of a header (columns get positional names col1, col2, ...)"
+      },
+      {
+        "name": "dataframe-out",
+        "kind": "own",
+        "type": "string",
+        "required": "optional",
+        "desc": "Write the typed table as one Arrow IPC file (Feather v2) instead of the default JSON. Pass `@<path>` for a file or `-` for binary stdout (same convention as other binary I/O flags). Mirror of the input-side `--dataframe` on `+table-put` / `+workbook-create` — pandas users round-trip via `df = pd.read_feather(\"x.arrow\")` or `pd.read_feather(io.BytesIO(stdout))`. Single-sheet only: requires `--sheet-id` or `--sheet-name`; whole-workbook reads keep the default JSON path. Column types come from the typed read-back (string/number/date/bool); per-column `number_format` is preserved as Arrow field metadata so the Arrow file can round-trip straight back through `+table-put --dataframe`."
       },
       {
         "name": "dry-run",
         "kind": "system",
         "type": "bool",
         "required": "optional",
-        "desc": "Print the request path and parameters without executing"
+        "desc": ""
       }
     ]
   },
@@ -1849,7 +2024,7 @@
         "kind": "own",
         "type": "string",
         "required": "required",
-        "desc": "RFC 4180 CSV text; plain values only (no formulas / styles / comments)",
+        "desc": "RFC 4180 CSV text; values or formulas (a leading = is evaluated as a formula); no styles / comments / images (use +cells-set for those).",
         "input": [
           "file",
           "stdin"
@@ -1880,6 +2055,61 @@
       }
     ]
   },
+  "+table-put": {
+    "risk": "write",
+    "flags": [
+      {
+        "name": "url",
+        "kind": "public",
+        "type": "string",
+        "required": "xor",
+        "desc": "Spreadsheet URL to write into (XOR with `--spreadsheet-token`)"
+      },
+      {
+        "name": "spreadsheet-token",
+        "kind": "public",
+        "type": "string",
+        "required": "xor",
+        "desc": "Spreadsheet token to write into (XOR with `--url`)"
+      },
+      {
+        "name": "sheets",
+        "kind": "own",
+        "type": "string",
+        "required": "xor",
+        "desc": "Typed table payload (pandas-DataFrame-shaped) as JSON, XOR with `--dataframe`: a top-level `sheets` array, each item `{name, start_cell?, mode?, header?, allow_overwrite?, columns:[\"colA\",\"colB\",...], data:[[...]], dtypes?:{colA:pandasDtype, ...}, formats?:{colA:numberFormat, ...}}`. Agents typically build it with `{**json.loads(df.to_json(orient=\"split\")), \"dtypes\": df.dtypes.astype(str).to_dict()}`. `dtypes` values are pandas dtype strings (`int64`, `float64`, `Int64`, `bool`, `boolean`, `datetime64[ns]`, `object`, ...); the writer maps them to internal string/number/date/bool — omit `dtypes` and a column writes as text (good for raw CSV-shaped data). `formats[col]` is an Excel number_format string (e.g. `#,##0.00`, `0.0%`, `yyyy-mm`); when absent, date columns default to `yyyy-mm-dd` and string columns to text format (`@`).",
+        "input": [
+          "file",
+          "stdin"
+        ]
+      },
+      {
+        "name": "dataframe",
+        "kind": "own",
+        "type": "string",
+        "required": "xor",
+        "desc": "Single-sheet typed table from one Arrow IPC file (a.k.a. Feather v2 — what `pandas.DataFrame.to_feather()` writes), XOR with `--sheets`. Pass `@<path>` for a file or `-` for binary stdin (same convention as other input flags). Arrow bytes are read raw — no TrimSpace / BOM strip — so the IPC magic survives intact (unlike text input flags). Column types come from the Arrow schema (int*/uint*/float* → number, date32/date64/timestamp → date, utf8/large_utf8 → string, bool → bool); per-column `number_format` may be set via Arrow field metadata (`pa.field(\"price\", pa.float64(), metadata={b\"number_format\": b\"$#,##0.00\"})`). Writes the sheet at default placement: name `Sheet1` (created when absent), overwrite from A1 with header. For a different sheet name, anchor, mode, or to write multiple sheets, use `--sheets` instead."
+      },
+      {
+        "name": "styles",
+        "kind": "own",
+        "type": "string",
+        "required": "optional",
+        "desc": "Visual operations applied after the typed write, as JSON: top-level `{styles:[...]}`. Each item corresponds to one written sheet and must include `name`, plus at least one of `cell_styles` / `row_sizes` / `col_sizes` / `cell_merges`. `cell_styles` entries use +cells-set-style fields with a cell range; row/col sizes use dimension ranges plus type/size; merges use cell ranges plus optional merge_type. The styles array length/order/name must match the written sheets: with --sheets, match --sheets.sheets; with --dataframe (single sheet named Sheet1), pass exactly one styles item with name `Sheet1`. Run `+table-put --print-schema --flag-name styles` for the full cell_styles field schema.",
+        "input": [
+          "file",
+          "stdin"
+        ]
+      },
+      {
+        "name": "dry-run",
+        "kind": "system",
+        "type": "bool",
+        "required": "optional",
+        "desc": ""
+      }
+    ]
+  },
   "+cells-clear": {
     "risk": "high-risk-write",
     "flags": [

shortcuts/sheets/data/flag-schemas.json

@@ -1,5 +1,5 @@
 {
-  "schema_version": "2",
+  "schema_version": "3",
   "flags": {
     "+batch-update": {
       "operations": {
@@ -454,7 +454,7 @@
                       "type": "object"
                     },
                     "link": {
-                      "description": "超链接地址（仅 type='link' 时必填）",
+                      "description": "超链接地址（type='link' 时必填）；@文档 mention（mention_type 非 0）时也必填，传文档 URL（如搜索结果里的文档链接），否则卡片不可点。@人（mention_type=0）不需要传",
                       "type": "string"
                     },
                     "mention_token": {
@@ -462,8 +462,21 @@
                       "type": "string"
                     },
                     "mention_type": {
-                      "description": "@提及类型编号（仅 type='mention' 时可选）",
-                      "type": "number"
+                      "description": "@提及类型编号（仅 type='mention' 时可选）。0 或不填=@用户；@文件时按类型取：1=文档 3=电子表格 8=多维表格 11=思维笔记 12=文件 15=旧版幻灯片 16=知识库 22=新版文档 30=幻灯片 38=画板",
+                      "type": "number",
+                      "enum": [
+                        0,
+                        1,
+                        3,
+                        8,
+                        11,
+                        12,
+                        15,
+                        16,
+                        22,
+                        30,
+                        38
+                      ]
                     },
                     "notify": {
                       "description": "是否发送通知（仅 type='mention' 时可选，默认 true）",
@@ -1730,11 +1743,12 @@
                             },
                             "aggregateType": {
                               "type": "string",
-                              "description": "汇总方式，默认为'sum'，仅在 aggregate 为 true 时生效",
+                              "description": "汇总方式，默认为'sum'，仅在 aggregate 为 true 时生效。count 只统计数值单元格；counta 统计所有非空单元格（含文本），按文本/分类列统计出现次数（如各类别的数量、频次分布）时用 counta。",
                               "enum": [
                                 "sum",
                                 "average",
                                 "count",
+                                "counta",
                                 "min",
                                 "max",
                                 "median"
@@ -1787,11 +1801,7 @@
               "data"
             ]
           }
-        },
-        "required": [
-          "position",
-          "size"
-        ]
+        }
       }
     },
     "+chart-update": {
@@ -2769,11 +2779,12 @@
                             },
                             "aggregateType": {
                               "type": "string",
-                              "description": "汇总方式，默认为'sum'，仅在 aggregate 为 true 时生效",
+                              "description": "汇总方式，默认为'sum'，仅在 aggregate 为 true 时生效。count 只统计数值单元格；counta 统计所有非空单元格（含文本），按文本/分类列统计出现次数（如各类别的数量、频次分布）时用 counta。",
                               "enum": [
                                 "sum",
                                 "average",
                                 "count",
+                                "counta",
                                 "min",
                                 "max",
                                 "median"
@@ -2826,11 +2837,7 @@
               "data"
             ]
           }
-        },
-        "required": [
-          "position",
-          "size"
-        ]
+        }
       }
     },
     "+cond-format-create": {
@@ -6249,6 +6256,744 @@
           }
         }
       }
+    },
+    "+table-put": {
+      "sheets": {
+        "type": "array",
+        "minItems": 1,
+        "description": "一个或多个子表的 typed 数据，每个数组元素写入一张子表；支持多 DataFrame → 多子表一次写入。整体形状对齐 pandas `df.to_json(orient=\"split\")`：列名走 `columns`、二维取值走 `data`、每列的 pandas dtype 走 `dtypes`、可选的展示格式走 `formats`。一行式用法：`{**json.loads(df.to_json(orient=\"split\")), \"dtypes\": df.dtypes.astype(str).to_dict()}`。",
+        "items": {
+          "type": "object",
+          "required": [
+            "name",
+            "columns",
+            "data"
+          ],
+          "properties": {
+            "name": {
+              "type": "string",
+              "description": "目标子表名。按名匹配已有子表；不存在则新建该子表。同一次调用内子表名不可重复。"
+            },
+            "start_cell": {
+              "type": "string",
+              "default": "A1",
+              "description": "写入起点单元格（A1 记法，如 \"B2\"），默认 \"A1\"。mode=append 时忽略其行号、仅沿用其列。"
+            },
+            "mode": {
+              "type": "string",
+              "enum": [
+                "overwrite",
+                "append"
+              ],
+              "default": "overwrite",
+              "description": "overwrite（默认）：从 start_cell 起写「表头 + 数据」块；append：把数据追加到子表已有数据下方（默认不重复表头）。"
+            },
+            "header": {
+              "type": "boolean",
+              "description": "是否写一行列名表头。省略时按 mode 取默认:overwrite→true、append→false（避免在已有表头下重复）；显式给值可覆盖。"
+            },
+            "allow_overwrite": {
+              "type": "boolean",
+              "default": true,
+              "description": "为 false 时，若写入会落在非空单元格则拒写以保护原数据（返回 partial_success）。默认 true。"
+            },
+            "columns": {
+              "type": "array",
+              "minItems": 1,
+              "description": "列名字符串数组，顺序与 `data` 中每行取值一一对应。同一子表内列名不可重复。",
+              "items": {
+                "type": "string"
+              }
+            },
+            "data": {
+              "type": "array",
+              "description": "数据行；每行是一个数组，长度必须等于 `columns` 数。元素按 `dtypes` 推得的列类型取值（date 列写 ISO yyyy-mm-dd 字符串、number 列写数值、bool 列写布尔、其余写文本），null 表示空单元格。",
+              "items": {
+                "type": "array",
+                "items": {
+                  "type": [
+                    "string",
+                    "number",
+                    "boolean",
+                    "null"
+                  ],
+                  "description": "单元格值：date→ISO yyyy-mm-dd 字符串；number→数值（json.Number 精度保留）；bool→布尔；string→文本；null→空单元格。"
+                }
+              }
+            },
+            "dtypes": {
+              "type": "object",
+              "description": "可选。列名 → pandas dtype 字符串的映射；缺失项默认按 object（string + 文本格式 `@`）处理，所以省略整段时整张表按文本写入（导入 CSV-shaped 数据的最简形态）。dtype 解析规则：`int*` / `uint*` / `Int*` / `UInt*` / `float*` / `Float*` / `complex*` → number（精度保留），`bool` / `boolean` → bool，`datetime64[ns]` / 含时区的 `datetime64[ns, UTC]` 等 → date（默认 `yyyy-mm-dd` 格式），`object` / `string` / `category` / 未识别 → string + 文本格式 `@`（数字样字符串如「00123」不会塌缩成数字）。",
+              "additionalProperties": {
+                "type": "string"
+              }
+            },
+            "formats": {
+              "type": "object",
+              "description": "可选。列名 → Excel number_format 字符串的映射，覆盖 dtype 自带的默认格式（金额 `#,##0.00`、百分比 `0.0%`、自定义日期 `yyyy-mm` 等）。percent 列的数值尺度由调用方负责（0.0469 配 `0.00%` 显示 4.69%）。",
+              "additionalProperties": {
+                "type": "string"
+              }
+            }
+          }
+        }
+      },
+      "styles": {
+        "items": {
+          "properties": {
+            "cell_merges": {
+              "description": "单元格合并操作数组；range 使用 A1 单元格范围，merge_type 默认 all。",
+              "items": {
+                "properties": {
+                  "merge_type": {
+                    "enum": [
+                      "all",
+                      "rows",
+                      "columns"
+                    ],
+                    "type": "string"
+                  },
+                  "range": {
+                    "type": "string"
+                  }
+                },
+                "required": [
+                  "range"
+                ],
+                "type": "object"
+              },
+              "type": "array"
+            },
+            "cell_styles": {
+              "description": "单元格样式操作数组；每项用 A1 单元格 range 指定范围，字段名与 +cells-set-style 对齐。",
+              "items": {
+                "properties": {
+                  "background_color": {
+                    "type": "string"
+                  },
+                  "border_styles": {
+                    "type": "object",
+                    "description": "边框配置，结构同 +cells-set-style --border-styles。",
+                    "properties": {
+                      "bottom": {
+                        "properties": {
+                          "color": {
+                            "description": "边框颜色（十六进制，例如 \"#000000\"）",
+                            "type": "string"
+                          },
+                          "style": {
+                            "description": "边框线型；传 \"none\" 表示清除该方向边框（无边框线）",
+                            "enum": [
+                              "solid",
+                              "dashed",
+                              "dotted",
+                              "double",
+                              "none"
+                            ],
+                            "type": "string"
+                          },
+                          "weight": {
+                            "description": "边框粗细/线宽",
+                            "enum": [
+                              "thin",
+                              "medium",
+                              "thick"
+                            ],
+                            "type": "string"
+                          }
+                        },
+                        "type": "object"
+                      },
+                      "left": {
+                        "properties": {
+                          "color": {
+                            "description": "边框颜色（十六进制，例如 \"#000000\"）",
+                            "type": "string"
+                          },
+                          "style": {
+                            "description": "边框线型；传 \"none\" 表示清除该方向边框（无边框线）",
+                            "enum": [
+                              "solid",
+                              "dashed",
+                              "dotted",
+                              "double",
+                              "none"
+                            ],
+                            "type": "string"
+                          },
+                          "weight": {
+                            "description": "边框粗细/线宽",
+                            "enum": [
+                              "thin",
+                              "medium",
+                              "thick"
+                            ],
+                            "type": "string"
+                          }
+                        },
+                        "type": "object"
+                      },
+                      "right": {
+                        "properties": {
+                          "color": {
+                            "description": "边框颜色（十六进制，例如 \"#000000\"）",
+                            "type": "string"
+                          },
+                          "style": {
+                            "description": "边框线型；传 \"none\" 表示清除该方向边框（无边框线）",
+                            "enum": [
+                              "solid",
+                              "dashed",
+                              "dotted",
+                              "double",
+                              "none"
+                            ],
+                            "type": "string"
+                          },
+                          "weight": {
+                            "description": "边框粗细/线宽",
+                            "enum": [
+                              "thin",
+                              "medium",
+                              "thick"
+                            ],
+                            "type": "string"
+                          }
+                        },
+                        "type": "object"
+                      },
+                      "top": {
+                        "properties": {
+                          "color": {
+                            "description": "边框颜色（十六进制，例如 \"#000000\"）",
+                            "type": "string"
+                          },
+                          "style": {
+                            "description": "边框线型；传 \"none\" 表示清除该方向边框（无边框线）",
+                            "enum": [
+                              "solid",
+                              "dashed",
+                              "dotted",
+                              "double",
+                              "none"
+                            ],
+                            "type": "string"
+                          },
+                          "weight": {
+                            "description": "边框粗细/线宽",
+                            "enum": [
+                              "thin",
+                              "medium",
+                              "thick"
+                            ],
+                            "type": "string"
+                          }
+                        },
+                        "type": "object"
+                      }
+                    }
+                  },
+                  "font_color": {
+                    "type": "string"
+                  },
+                  "font_line": {
+                    "enum": [
+                      "none",
+                      "underline",
+                      "line-through"
+                    ],
+                    "type": "string"
+                  },
+                  "font_size": {
+                    "type": "number"
+                  },
+                  "font_style": {
+                    "enum": [
+                      "normal",
+                      "italic"
+                    ],
+                    "type": "string"
+                  },
+                  "font_weight": {
+                    "enum": [
+                      "normal",
+                      "bold"
+                    ],
+                    "type": "string"
+                  },
+                  "horizontal_alignment": {
+                    "enum": [
+                      "left",
+                      "center",
+                      "right"
+                    ],
+                    "type": "string"
+                  },
+                  "number_format": {
+                    "type": "string"
+                  },
+                  "range": {
+                    "description": "A1 单元格范围，必须落在该子表本次写入区域内；例如 A1:B1、B2。",
+                    "type": "string"
+                  },
+                  "vertical_alignment": {
+                    "enum": [
+                      "top",
+                      "middle",
+                      "bottom"
+                    ],
+                    "type": "string"
+                  },
+                  "word_wrap": {
+                    "enum": [
+                      "overflow",
+                      "auto-wrap",
+                      "word-clip"
+                    ],
+                    "type": "string"
+                  }
+                },
+                "required": [
+                  "range"
+                ],
+                "type": "object"
+              },
+              "type": "array"
+            },
+            "col_sizes": {
+              "description": "列宽操作数组；range 使用列范围如 A:C，type 为 pixel/standard，pixel 需要 size。",
+              "items": {
+                "properties": {
+                  "range": {
+                    "type": "string"
+                  },
+                  "size": {
+                    "type": "number"
+                  },
+                  "type": {
+                    "enum": [
+                      "pixel",
+                      "standard"
+                    ],
+                    "type": "string"
+                  }
+                },
+                "required": [
+                  "range",
+                  "type"
+                ],
+                "type": "object"
+              },
+              "type": "array"
+            },
+            "name": {
+              "description": "子表名。--sheets 模式下必须与同位置 --sheets.sheets[].name 一致；--values 模式下建议写 Sheet1（其 name 会被忽略）。",
+              "type": "string"
+            },
+            "row_sizes": {
+              "description": "行高操作数组；range 使用行范围如 1:3，type 为 pixel/standard/auto，pixel 需要 size。",
+              "items": {
+                "properties": {
+                  "range": {
+                    "type": "string"
+                  },
+                  "size": {
+                    "type": "number"
+                  },
+                  "type": {
+                    "enum": [
+                      "pixel",
+                      "standard",
+                      "auto"
+                    ],
+                    "type": "string"
+                  }
+                },
+                "required": [
+                  "range",
+                  "type"
+                ],
+                "type": "object"
+              },
+              "type": "array"
+            }
+          },
+          "required": [
+            "name"
+          ],
+          "type": "object"
+        },
+        "type": "array"
+      }
+    },
+    "+workbook-create": {
+      "sheets": {
+        "type": "array",
+        "minItems": 1,
+        "description": "一个或多个子表的 typed 数据，每个数组元素写入一张子表；支持多 DataFrame → 多子表一次写入。整体形状对齐 pandas `df.to_json(orient=\"split\")`：列名走 `columns`、二维取值走 `data`、每列的 pandas dtype 走 `dtypes`、可选的展示格式走 `formats`。一行式用法：`{**json.loads(df.to_json(orient=\"split\")), \"dtypes\": df.dtypes.astype(str).to_dict()}`。",
+        "items": {
+          "type": "object",
+          "required": [
+            "name",
+            "columns",
+            "data"
+          ],
+          "properties": {
+            "name": {
+              "type": "string",
+              "description": "目标子表名。按名匹配已有子表；不存在则新建该子表。同一次调用内子表名不可重复。"
+            },
+            "start_cell": {
+              "type": "string",
+              "default": "A1",
+              "description": "写入起点单元格（A1 记法，如 \"B2\"），默认 \"A1\"。mode=append 时忽略其行号、仅沿用其列。"
+            },
+            "mode": {
+              "type": "string",
+              "enum": [
+                "overwrite",
+                "append"
+              ],
+              "default": "overwrite",
+              "description": "overwrite（默认）：从 start_cell 起写「表头 + 数据」块；append：把数据追加到子表已有数据下方（默认不重复表头）。"
+            },
+            "header": {
+              "type": "boolean",
+              "description": "是否写一行列名表头。省略时按 mode 取默认:overwrite→true、append→false（避免在已有表头下重复）；显式给值可覆盖。"
+            },
+            "allow_overwrite": {
+              "type": "boolean",
+              "default": true,
+              "description": "为 false 时，若写入会落在非空单元格则拒写以保护原数据（返回 partial_success）。默认 true。"
+            },
+            "columns": {
+              "type": "array",
+              "minItems": 1,
+              "description": "列名字符串数组，顺序与 `data` 中每行取值一一对应。同一子表内列名不可重复。",
+              "items": {
+                "type": "string"
+              }
+            },
+            "data": {
+              "type": "array",
+              "description": "数据行；每行是一个数组，长度必须等于 `columns` 数。元素按 `dtypes` 推得的列类型取值（date 列写 ISO yyyy-mm-dd 字符串、number 列写数值、bool 列写布尔、其余写文本），null 表示空单元格。",
+              "items": {
+                "type": "array",
+                "items": {
+                  "type": [
+                    "string",
+                    "number",
+                    "boolean",
+                    "null"
+                  ],
+                  "description": "单元格值：date→ISO yyyy-mm-dd 字符串；number→数值（json.Number 精度保留）；bool→布尔；string→文本；null→空单元格。"
+                }
+              }
+            },
+            "dtypes": {
+              "type": "object",
+              "description": "可选。列名 → pandas dtype 字符串的映射；缺失项默认按 object（string + 文本格式 `@`）处理，所以省略整段时整张表按文本写入（导入 CSV-shaped 数据的最简形态）。dtype 解析规则：`int*` / `uint*` / `Int*` / `UInt*` / `float*` / `Float*` / `complex*` → number（精度保留），`bool` / `boolean` → bool，`datetime64[ns]` / 含时区的 `datetime64[ns, UTC]` 等 → date（默认 `yyyy-mm-dd` 格式），`object` / `string` / `category` / 未识别 → string + 文本格式 `@`（数字样字符串如「00123」不会塌缩成数字）。",
+              "additionalProperties": {
+                "type": "string"
+              }
+            },
+            "formats": {
+              "type": "object",
+              "description": "可选。列名 → Excel number_format 字符串的映射，覆盖 dtype 自带的默认格式（金额 `#,##0.00`、百分比 `0.0%`、自定义日期 `yyyy-mm` 等）。percent 列的数值尺度由调用方负责（0.0469 配 `0.00%` 显示 4.69%）。",
+              "additionalProperties": {
+                "type": "string"
+              }
+            }
+          }
+        }
+      },
+      "styles": {
+        "items": {
+          "properties": {
+            "cell_merges": {
+              "description": "单元格合并操作数组；range 使用 A1 单元格范围，merge_type 默认 all。",
+              "items": {
+                "properties": {
+                  "merge_type": {
+                    "enum": [
+                      "all",
+                      "rows",
+                      "columns"
+                    ],
+                    "type": "string"
+                  },
+                  "range": {
+                    "type": "string"
+                  }
+                },
+                "required": [
+                  "range"
+                ],
+                "type": "object"
+              },
+              "type": "array"
+            },
+            "cell_styles": {
+              "description": "单元格样式操作数组；每项用 A1 单元格 range 指定范围，字段名与 +cells-set-style 对齐。",
+              "items": {
+                "properties": {
+                  "background_color": {
+                    "type": "string"
+                  },
+                  "border_styles": {
+                    "type": "object",
+                    "description": "边框配置，结构同 +cells-set-style --border-styles。",
+                    "properties": {
+                      "bottom": {
+                        "properties": {
+                          "color": {
+                            "description": "边框颜色（十六进制，例如 \"#000000\"）",
+                            "type": "string"
+                          },
+                          "style": {
+                            "description": "边框线型；传 \"none\" 表示清除该方向边框（无边框线）",
+                            "enum": [
+                              "solid",
+                              "dashed",
+                              "dotted",
+                              "double",
+                              "none"
+                            ],
+                            "type": "string"
+                          },
+                          "weight": {
+                            "description": "边框粗细/线宽",
+                            "enum": [
+                              "thin",
+                              "medium",
+                              "thick"
+                            ],
+                            "type": "string"
+                          }
+                        },
+                        "type": "object"
+                      },
+                      "left": {
+                        "properties": {
+                          "color": {
+                            "description": "边框颜色（十六进制，例如 \"#000000\"）",
+                            "type": "string"
+                          },
+                          "style": {
+                            "description": "边框线型；传 \"none\" 表示清除该方向边框（无边框线）",
+                            "enum": [
+                              "solid",
+                              "dashed",
+                              "dotted",
+                              "double",
+                              "none"
+                            ],
+                            "type": "string"
+                          },
+                          "weight": {
+                            "description": "边框粗细/线宽",
+                            "enum": [
+                              "thin",
+                              "medium",
+                              "thick"
+                            ],
+                            "type": "string"
+                          }
+                        },
+                        "type": "object"
+                      },
+                      "right": {
+                        "properties": {
+                          "color": {
+                            "description": "边框颜色（十六进制，例如 \"#000000\"）",
+                            "type": "string"
+                          },
+                          "style": {
+                            "description": "边框线型；传 \"none\" 表示清除该方向边框（无边框线）",
+                            "enum": [
+                              "solid",
+                              "dashed",
+                              "dotted",
+                              "double",
+                              "none"
+                            ],
+                            "type": "string"
+                          },
+                          "weight": {
+                            "description": "边框粗细/线宽",
+                            "enum": [
+                              "thin",
+                              "medium",
+                              "thick"
+                            ],
+                            "type": "string"
+                          }
+                        },
+                        "type": "object"
+                      },
+                      "top": {
+                        "properties": {
+                          "color": {
+                            "description": "边框颜色（十六进制，例如 \"#000000\"）",
+                            "type": "string"
+                          },
+                          "style": {
+                            "description": "边框线型；传 \"none\" 表示清除该方向边框（无边框线）",
+                            "enum": [
+                              "solid",
+                              "dashed",
+                              "dotted",
+                              "double",
+                              "none"
+                            ],
+                            "type": "string"
+                          },
+                          "weight": {
+                            "description": "边框粗细/线宽",
+                            "enum": [
+                              "thin",
+                              "medium",
+                              "thick"
+                            ],
+                            "type": "string"
+                          }
+                        },
+                        "type": "object"
+                      }
+                    }
+                  },
+                  "font_color": {
+                    "type": "string"
+                  },
+                  "font_line": {
+                    "enum": [
+                      "none",
+                      "underline",
+                      "line-through"
+                    ],
+                    "type": "string"
+                  },
+                  "font_size": {
+                    "type": "number"
+                  },
+                  "font_style": {
+                    "enum": [
+                      "normal",
+                      "italic"
+                    ],
+                    "type": "string"
+                  },
+                  "font_weight": {
+                    "enum": [
+                      "normal",
+                      "bold"
+                    ],
+                    "type": "string"
+                  },
+                  "horizontal_alignment": {
+                    "enum": [
+                      "left",
+                      "center",
+                      "right"
+                    ],
+                    "type": "string"
+                  },
+                  "number_format": {
+                    "type": "string"
+                  },
+                  "range": {
+                    "description": "A1 单元格范围，必须落在该子表本次写入区域内；例如 A1:B1、B2。",
+                    "type": "string"
+                  },
+                  "vertical_alignment": {
+                    "enum": [
+                      "top",
+                      "middle",
+                      "bottom"
+                    ],
+                    "type": "string"
+                  },
+                  "word_wrap": {
+                    "enum": [
+                      "overflow",
+                      "auto-wrap",
+                      "word-clip"
+                    ],
+                    "type": "string"
+                  }
+                },
+                "required": [
+                  "range"
+                ],
+                "type": "object"
+              },
+              "type": "array"
+            },
+            "col_sizes": {
+              "description": "列宽操作数组；range 使用列范围如 A:C，type 为 pixel/standard，pixel 需要 size。",
+              "items": {
+                "properties": {
+                  "range": {
+                    "type": "string"
+                  },
+                  "size": {
+                    "type": "number"
+                  },
+                  "type": {
+                    "enum": [
+                      "pixel",
+                      "standard"
+                    ],
+                    "type": "string"
+                  }
+                },
+                "required": [
+                  "range",
+                  "type"
+                ],
+                "type": "object"
+              },
+              "type": "array"
+            },
+            "name": {
+              "description": "子表名。--sheets 模式下必须与同位置 --sheets.sheets[].name 一致；--values 模式下建议写 Sheet1（其 name 会被忽略）。",
+              "type": "string"
+            },
+            "row_sizes": {
+              "description": "行高操作数组；range 使用行范围如 1:3，type 为 pixel/standard/auto，pixel 需要 size。",
+              "items": {
+                "properties": {
+                  "range": {
+                    "type": "string"
+                  },
+                  "size": {
+                    "type": "number"
+                  },
+                  "type": {
+                    "enum": [
+                      "pixel",
+                      "standard",
+                      "auto"
+                    ],
+                    "type": "string"
+                  }
+                },
+                "required": [
+                  "range",
+                  "type"
+                ],
+                "type": "object"
+              },
+              "type": "array"
+            }
+          },
+          "required": [
+            "name"
+          ],
+          "type": "object"
+        },
+        "type": "array"
+      }
     }
   }
 }

shortcuts/sheets/execute_paths_test.go

@@ -63,6 +63,97 @@ func TestExecute_WorkbookInfo_ToolError(t *testing.T) {
 	}
 }
 
+// TestExecute_WikiURLResolvesToSheet covers the two-step wiki path: a /wiki/
+// URL is resolved via get_node to its spreadsheet obj_token, which then feeds
+// the tool invoke. The tool stub is keyed on the resolved obj_token, so the
+// test would fail if the node_token were used unresolved.
+func TestExecute_WikiURLResolvesToSheet(t *testing.T) {
+	t.Parallel()
+	getNode := &httpmock.Stub{
+		Method: "GET",
+		URL:    "/open-apis/wiki/v2/spaces/get_node",
+		Body: map[string]interface{}{
+			"code": 0,
+			"msg":  "success",
+			"data": map[string]interface{}{
+				"node": map[string]interface{}{
+					"obj_type":  "sheet",
+					"obj_token": testToken,
+				},
+			},
+		},
+	}
+	tool := toolOutputStub(testToken, "read", `{"sheets":[{"sheet_id":"sh1","title":"Sheet1","index":0}]}`)
+	out, err := runShortcutWithStubs(t, WorkbookInfo,
+		[]string{"--url", "https://example.feishu.cn/wiki/wikTestNODE"}, getNode, tool)
+	if err != nil {
+		t.Fatalf("execute failed: %v\nout=%s", err, out)
+	}
+	data := decodeEnvelopeData(t, out)
+	if sheets, _ := data["sheets"].([]interface{}); len(sheets) != 1 {
+		t.Fatalf("sheets len = %d, want 1; out=%s", len(sheets), out)
+	}
+}
+
+// TestExecute_WikiURLWrongObjType rejects a wiki node that resolves to a
+// non-spreadsheet obj_type before any tool invoke.
+func TestExecute_WikiURLWrongObjType(t *testing.T) {
+	t.Parallel()
+	getNode := &httpmock.Stub{
+		Method: "GET",
+		URL:    "/open-apis/wiki/v2/spaces/get_node",
+		Body: map[string]interface{}{
+			"code": 0,
+			"msg":  "success",
+			"data": map[string]interface{}{
+				"node": map[string]interface{}{
+					"obj_type":  "docx",
+					"obj_token": "docABC",
+				},
+			},
+		},
+	}
+	out, err := runShortcutWithStubs(t, WorkbookInfo,
+		[]string{"--url", "https://example.feishu.cn/wiki/wikTestNODE"}, getNode)
+	if err == nil {
+		t.Fatalf("want error for non-sheet wiki node; out=%s", out)
+	}
+	if !strings.Contains(err.Error(), "obj_type") {
+		t.Fatalf("error = %v, want mention of obj_type", err)
+	}
+	var ve *errs.ValidationError
+	if !errors.As(err, &ve) {
+		t.Fatalf("wrong-obj_type error = %T, want *errs.ValidationError", err)
+	}
+}
+
+// TestExecute_WikiURLIncompleteNode treats an incomplete get_node response
+// (missing obj_type/obj_token) as an internal/server error, not a user --url
+// validation error.
+func TestExecute_WikiURLIncompleteNode(t *testing.T) {
+	t.Parallel()
+	getNode := &httpmock.Stub{
+		Method: "GET",
+		URL:    "/open-apis/wiki/v2/spaces/get_node",
+		Body: map[string]interface{}{
+			"code": 0,
+			"msg":  "success",
+			"data": map[string]interface{}{
+				"node": map[string]interface{}{},
+			},
+		},
+	}
+	_, err := runShortcutWithStubs(t, WorkbookInfo,
+		[]string{"--url", "https://example.feishu.cn/wiki/wikTestNODE"}, getNode)
+	if err == nil {
+		t.Fatal("want error for incomplete get_node node data")
+	}
+	var ve *errs.ValidationError
+	if errors.As(err, &ve) {
+		t.Fatalf("incomplete-data error classified as validation (%v); want internal", err)
+	}
+}
+
 // TestExecute_SheetMove_LookupsIndex covers the two-step path: SheetMove
 // when only --sheet-name is given (and --source-index omitted) first
 // reads the workbook structure to derive sheet_id + source_index, then
@@ -365,14 +456,17 @@ func TestExecute_WorkbookCreate(t *testing.T) {
 			},
 		},
 	}
-	// Initial fill first reads the workbook structure to resolve the default
-	// sheet's id (the create response doesn't echo it), then writes.
+	// The write reads the workbook structure to resolve the default sheet's id
+	// (the create response doesn't echo it). lookupFirstSheetID and
+	// writeTypedSheets' listSheetIDsByName both read it — one reusable stub serves
+	// both. The synthesized sheet is named "Sheet1", matching the default sheet,
+	// so it's adopted in place (no rename).
 	structure := toolOutputStub("shtcnBRAND", "read", `{"sheets":[{"sheet_id":"shtFirst","sheet_name":"Sheet1","index":0}]}`)
+	structure.Reusable = true
 	fill := toolOutputStub("shtcnBRAND", "write", `{"updated_cells":4}`)
 	out, err := runShortcutWithStubs(t, WorkbookCreate, []string{
 		"--title", "Sales",
-		"--headers", `["Name","Score"]`,
-		"--values", `[["alice",95]]`,
+		"--values", `[["Name","Score"],["alice",95]]`,
 	}, create, structure, fill)
 	if err != nil {
 		t.Fatalf("execute failed: %v\nout=%s", err, out)
@@ -382,8 +476,8 @@ func TestExecute_WorkbookCreate(t *testing.T) {
 	if ss["spreadsheet_token"] != "shtcnBRAND" {
 		t.Errorf("spreadsheet_token = %v", ss["spreadsheet_token"])
 	}
-	if data["initial_fill"] == nil {
-		t.Errorf("initial_fill missing in envelope")
+	if sheets, _ := data["sheets"].([]interface{}); len(sheets) != 1 {
+		t.Errorf("sheets summary missing in envelope; got %#v", data["sheets"])
 	}
 	// The fill must target the resolved first sheet, not an empty selector.
 	fillInput := decodeToolInput(t, decodeRawEnvelopeBody(t, fill.CapturedBody), "set_cell_range")
@@ -393,14 +487,13 @@ func TestExecute_WorkbookCreate(t *testing.T) {
 }
 
 // TestExecute_WorkbookCreate_EmptyArraysSkipFill locks the fix for the nil-map
-// panic / illegal-range bug: --values '[]' or --headers '[]' must short-circuit
-// the initial fill (no structure/fill calls fire) and finish with the
-// spreadsheet created but no initial_fill — never panic on a nil fill map.
+// panic / illegal-range bug: --values '[]' must short-circuit the initial fill
+// (no structure/fill calls fire) and finish with the spreadsheet created but no
+// sheets summary — never panic on a nil payload.
 func TestExecute_WorkbookCreate_EmptyArraysSkipFill(t *testing.T) {
 	t.Parallel()
 	for _, tc := range []struct{ name, flag, val string }{
 		{"empty values", "--values", "[]"},
-		{"empty headers", "--headers", "[]"},
 	} {
 		t.Run(tc.name, func(t *testing.T) {
 			t.Parallel()
@@ -421,8 +514,8 @@ func TestExecute_WorkbookCreate_EmptyArraysSkipFill(t *testing.T) {
 				t.Fatalf("execute failed: %v\nout=%s", err, out)
 			}
 			data := decodeEnvelopeData(t, out)
-			if data["initial_fill"] != nil {
-				t.Errorf("initial_fill should be absent for %s %s; got %#v", tc.flag, tc.val, data["initial_fill"])
+			if data["sheets"] != nil {
+				t.Errorf("sheets should be absent for %s %s; got %#v", tc.flag, tc.val, data["sheets"])
 			}
 			if ss, _ := data["spreadsheet"].(map[string]interface{}); ss["spreadsheet_token"] != "shtNEW" {
 				t.Errorf("spreadsheet_token = %v, want shtNEW", ss["spreadsheet_token"])

shortcuts/sheets/flag_defs.go

@@ -80,3 +80,28 @@ func flagsFor(command string) []common.Flag {
 	}
 	return out
 }
+
+// flagAcceptsStdin reports whether the (command, flag) pair declares stdin as
+// an input source in flag-defs.json. Used to decide whether an "invalid JSON"
+// error should also steer the caller toward stdin. It runs on an error path,
+// so it returns false for an unknown command/flag rather than panicking the
+// way flagsFor does.
+func flagAcceptsStdin(command, name string) bool {
+	defs, _ := loadFlagDefs()
+	spec, ok := defs[command]
+	if !ok {
+		return false
+	}
+	for _, df := range spec.Flags {
+		if df.Name != name {
+			continue
+		}
+		for _, in := range df.Input {
+			if in == common.Stdin {
+				return true
+			}
+		}
+		return false
+	}
+	return false
+}

shortcuts/sheets/flag_defs_gen.go

@@ -75,7 +75,7 @@ var flagDefs = map[string]commandDef{
 			{Name: "sheet-name", Kind: "public", Type: "string", Required: "xor", Desc: "Sheet name (XOR with `--sheet-id`)"},
 			{Name: "range", Kind: "own", Type: "string", Required: "required", Desc: "A1 range, e.g. `A1:F10` (no sheet prefix — use `--sheet-id` / `--sheet-name` to select the sheet)"},
 			{Name: "include", Kind: "own", Type: "string_slice", Required: "optional", Desc: "Comma-separated info categories to include", Enum: []string{"value", "formula", "style", "comment", "data_validation"}},
-			{Name: "max-chars", Kind: "own", Type: "int", Required: "optional", Desc: "Safety cap; default 200000", Default: "200000", Hidden: true},
+			{Name: "max-chars", Kind: "own", Type: "int", Required: "optional", Desc: "Max output chars per call; default 500000 (safety cap). Large reads are usually better redirected to a file; only lower it (e.g. 25000) when you want results inline without triggering file offload, paging via has_more", Default: "500000"},
 			{Name: "skip-hidden", Kind: "own", Type: "bool", Required: "optional", Desc: "Skip hidden rows and columns; default `false`"},
 			{Name: "dry-run", Kind: "system", Type: "bool", Required: "optional"},
 		},
@@ -305,10 +305,9 @@ var flagDefs = map[string]commandDef{
 			{Name: "sheet-id", Kind: "public", Type: "string", Required: "xor", Desc: "Sheet reference_id (XOR with `--sheet-name`)"},
 			{Name: "sheet-name", Kind: "public", Type: "string", Required: "xor", Desc: "Sheet name (XOR with `--sheet-id`)"},
 			{Name: "range", Kind: "own", Type: "string", Required: "required", Desc: "A1 range, e.g. `A1:F30` (no sheet prefix — use `--sheet-id` / `--sheet-name` to select the sheet)"},
-			{Name: "max-chars", Kind: "own", Type: "int", Required: "optional", Desc: "Safety cap; default 200000", Default: "200000", Hidden: true},
+			{Name: "max-chars", Kind: "own", Type: "int", Required: "optional", Desc: "Max output chars per call; default 500000 (safety cap). Large reads are usually better redirected to a file; only lower it (e.g. 25000) when you want results inline without triggering file offload, paging via has_more", Default: "500000"},
 			{Name: "include-row-prefix", Kind: "own", Type: "bool", Required: "optional", Desc: "Whether to prefix each row with `[row=N]`; default `true`", Default: "true"},
 			{Name: "skip-hidden", Kind: "own", Type: "bool", Required: "optional", Desc: "Skip hidden rows and columns; default `false`"},
-			{Name: "rows-json", Kind: "own", Type: "bool", Required: "optional", Desc: "Return structured rows ({row_number, values:{col→cell}}) instead of CSV text; default false", Default: "false"},
 			{Name: "dry-run", Kind: "system", Type: "bool", Required: "optional", Desc: "Print the request path and parameters without executing"},
 		},
 	},
@@ -320,7 +319,7 @@ var flagDefs = map[string]commandDef{
 			{Name: "sheet-id", Kind: "public", Type: "string", Required: "xor", Desc: "Sheet reference_id (XOR with `--sheet-name`)"},
 			{Name: "sheet-name", Kind: "public", Type: "string", Required: "xor", Desc: "Sheet name (XOR with `--sheet-id`)"},
 			{Name: "start-cell", Kind: "own", Type: "string", Required: "required", Desc: "Top-left A1 anchor (e.g. `A1`, `B5`; no sheet prefix — use `--sheet-id` / `--sheet-name` to select the sheet); must be a single cell, range notation not accepted; the bottom-right is inferred from CSV row/column counts", Default: "A1"},
-			{Name: "csv", Kind: "own", Type: "string", Required: "required", Desc: "RFC 4180 CSV text; plain values only (no formulas / styles / comments)", Input: []string{"file", "stdin"}},
+			{Name: "csv", Kind: "own", Type: "string", Required: "required", Desc: "RFC 4180 CSV text; values or formulas (a leading = is evaluated as a formula); no styles / comments / images (use +cells-set for those).", Input: []string{"file", "stdin"}},
 			{Name: "allow-overwrite", Kind: "own", Type: "bool", Required: "optional", Desc: "Allow overwriting (default true); set false to error if any target cell is non-empty", Default: "true"},
 			{Name: "range", Kind: "own", Type: "string", Required: "optional", Desc: "alias for --start-cell (parity with +csv-get / +cells-set, which locate with --range); a range like A1:H17 collapses to its top-left cell", Hidden: true},
 			{Name: "dry-run", Kind: "system", Type: "bool", Required: "optional"},
@@ -766,7 +765,7 @@ var flagDefs = map[string]commandDef{
 			{Name: "url", Kind: "public", Type: "string", Required: "xor", Desc: "Spreadsheet URL (XOR with `--spreadsheet-token`)"},
 			{Name: "spreadsheet-token", Kind: "public", Type: "string", Required: "xor", Desc: "Spreadsheet token (XOR with `--url`)"},
 			{Name: "title", Kind: "own", Type: "string", Required: "required", Desc: "New sheet title"},
-			{Name: "index", Kind: "own", Type: "int", Required: "optional", Desc: "Insert position; appended to the end when omitted", Default: "-1"},
+			{Name: "index", Kind: "own", Type: "int", Required: "optional", Desc: "Insert position (0-based); appended to the end when omitted", Default: "-1"},
 			{Name: "row-count", Kind: "own", Type: "int", Required: "optional", Desc: "Initial row count (default 200, max 50000)", Default: "200"},
 			{Name: "col-count", Kind: "own", Type: "int", Required: "optional", Desc: "Initial column count (default 20, max 200)", Default: "20"},
 			{Name: "dry-run", Kind: "system", Type: "bool", Required: "optional"},
@@ -793,6 +792,16 @@ var flagDefs = map[string]commandDef{
 			{Name: "dry-run", Kind: "system", Type: "bool", Required: "optional"},
 		},
 	},
+	"+sheet-hide-gridline": {
+		Risk: "write",
+		Flags: []flagDef{
+			{Name: "url", Kind: "public", Type: "string", Required: "xor", Desc: "Spreadsheet URL (XOR with `--spreadsheet-token`)"},
+			{Name: "spreadsheet-token", Kind: "public", Type: "string", Required: "xor", Desc: "Spreadsheet token (XOR with `--url`)"},
+			{Name: "sheet-id", Kind: "public", Type: "string", Required: "xor", Desc: "Sheet reference_id (XOR with `--sheet-name`)"},
+			{Name: "sheet-name", Kind: "public", Type: "string", Required: "xor", Desc: "Sheet name (XOR with `--sheet-id`)"},
+			{Name: "dry-run", Kind: "system", Type: "bool", Required: "optional"},
+		},
+	},
 	"+sheet-info": {
 		Risk: "read",
 		Flags: []flagDef{
@@ -839,6 +848,16 @@ var flagDefs = map[string]commandDef{
 			{Name: "dry-run", Kind: "system", Type: "bool", Required: "optional"},
 		},
 	},
+	"+sheet-show-gridline": {
+		Risk: "write",
+		Flags: []flagDef{
+			{Name: "url", Kind: "public", Type: "string", Required: "xor", Desc: "Spreadsheet URL (XOR with `--spreadsheet-token`)"},
+			{Name: "spreadsheet-token", Kind: "public", Type: "string", Required: "xor", Desc: "Spreadsheet token (XOR with `--url`)"},
+			{Name: "sheet-id", Kind: "public", Type: "string", Required: "xor", Desc: "Sheet reference_id (XOR with `--sheet-name`)"},
+			{Name: "sheet-name", Kind: "public", Type: "string", Required: "xor", Desc: "Sheet name (XOR with `--sheet-id`)"},
+			{Name: "dry-run", Kind: "system", Type: "bool", Required: "optional"},
+		},
+	},
 	"+sheet-unhide": {
 		Risk: "write",
 		Flags: []flagDef{
@@ -895,13 +914,39 @@ var flagDefs = map[string]commandDef{
 			{Name: "dry-run", Kind: "system", Type: "bool", Required: "optional"},
 		},
 	},
+	"+table-get": {
+		Risk: "read",
+		Flags: []flagDef{
+			{Name: "url", Kind: "public", Type: "string", Required: "xor", Desc: "Spreadsheet URL (XOR with `--spreadsheet-token`)"},
+			{Name: "spreadsheet-token", Kind: "public", Type: "string", Required: "xor", Desc: "Spreadsheet token (XOR with `--url`)"},
+			{Name: "sheet-id", Kind: "own", Type: "string", Required: "optional", Desc: "Read only this sheet (by id); omit to read all sheets"},
+			{Name: "sheet-name", Kind: "own", Type: "string", Required: "optional", Desc: "Read only this sheet (by name); omit to read all sheets"},
+			{Name: "range", Kind: "own", Type: "string", Required: "optional", Desc: "A1 range to read; omit to read each sheet's full used range (spans internal blank rows/columns, not just the A1 current region)"},
+			{Name: "no-header", Kind: "own", Type: "bool", Required: "optional", Desc: "Treat the first row as data instead of a header (columns get positional names col1, col2, ...)"},
+			{Name: "dataframe-out", Kind: "own", Type: "string", Required: "optional", Desc: "Write the typed table as one Arrow IPC file (Feather v2) instead of the default JSON. Pass `@<path>` for a file or `-` for binary stdout (same convention as other binary I/O flags). Mirror of the input-side `--dataframe` on `+table-put` / `+workbook-create` — pandas users round-trip via `df = pd.read_feather(\"x.arrow\")` or `pd.read_feather(io.BytesIO(stdout))`. Single-sheet only: requires `--sheet-id` or `--sheet-name`; whole-workbook reads keep the default JSON path. Column types come from the typed read-back (string/number/date/bool); per-column `number_format` is preserved as Arrow field metadata so the Arrow file can round-trip straight back through `+table-put --dataframe`."},
+			{Name: "dry-run", Kind: "system", Type: "bool", Required: "optional"},
+		},
+	},
+	"+table-put": {
+		Risk: "write",
+		Flags: []flagDef{
+			{Name: "url", Kind: "public", Type: "string", Required: "xor", Desc: "Spreadsheet URL to write into (XOR with `--spreadsheet-token`)"},
+			{Name: "spreadsheet-token", Kind: "public", Type: "string", Required: "xor", Desc: "Spreadsheet token to write into (XOR with `--url`)"},
+			{Name: "sheets", Kind: "own", Type: "string", Required: "xor", Desc: "Typed table payload (pandas-DataFrame-shaped) as JSON, XOR with `--dataframe`: a top-level `sheets` array, each item `{name, start_cell?, mode?, header?, allow_overwrite?, columns:[\"colA\",\"colB\",...], data:[[...]], dtypes?:{colA:pandasDtype, ...}, formats?:{colA:numberFormat, ...}}`. Agents typically build it with `{**json.loads(df.to_json(orient=\"split\")), \"dtypes\": df.dtypes.astype(str).to_dict()}`. `dtypes` values are pandas dtype strings (`int64`, `float64`, `Int64`, `bool`, `boolean`, `datetime64[ns]`, `object`, ...); the writer maps them to internal string/number/date/bool — omit `dtypes` and a column writes as text (good for raw CSV-shaped data). `formats[col]` is an Excel number_format string (e.g. `#,##0.00`, `0.0%`, `yyyy-mm`); when absent, date columns default to `yyyy-mm-dd` and string columns to text format (`@`).", Input: []string{"file", "stdin"}},
+			{Name: "dataframe", Kind: "own", Type: "string", Required: "xor", Desc: "Single-sheet typed table from one Arrow IPC file (a.k.a. Feather v2 — what `pandas.DataFrame.to_feather()` writes), XOR with `--sheets`. Pass `@<path>` for a file or `-` for binary stdin (same convention as other input flags). Arrow bytes are read raw — no TrimSpace / BOM strip — so the IPC magic survives intact (unlike text input flags). Column types come from the Arrow schema (int*/uint*/float* → number, date32/date64/timestamp → date, utf8/large_utf8 → string, bool → bool); per-column `number_format` may be set via Arrow field metadata (`pa.field(\"price\", pa.float64(), metadata={b\"number_format\": b\"$#,##0.00\"})`). Writes the sheet at default placement: name `Sheet1` (created when absent), overwrite from A1 with header. For a different sheet name, anchor, mode, or to write multiple sheets, use `--sheets` instead."},
+			{Name: "styles", Kind: "own", Type: "string", Required: "optional", Desc: "Visual operations applied after the typed write, as JSON: top-level `{styles:[...]}`. Each item corresponds to one written sheet and must include `name`, plus at least one of `cell_styles` / `row_sizes` / `col_sizes` / `cell_merges`. `cell_styles` entries use +cells-set-style fields with a cell range; row/col sizes use dimension ranges plus type/size; merges use cell ranges plus optional merge_type. The styles array length/order/name must match the written sheets: with --sheets, match --sheets.sheets; with --dataframe (single sheet named Sheet1), pass exactly one styles item with name `Sheet1`.", Input: []string{"file", "stdin"}},
+			{Name: "dry-run", Kind: "system", Type: "bool", Required: "optional"},
+		},
+	},
 	"+workbook-create": {
 		Risk: "write",
 		Flags: []flagDef{
 			{Name: "title", Kind: "own", Type: "string", Required: "required", Desc: "Spreadsheet title"},
 			{Name: "folder-token", Kind: "own", Type: "string", Required: "optional", Desc: "Target folder token; placed at the drive root when omitted"},
-			{Name: "headers", Kind: "own", Type: "string", Required: "optional", Desc: "Header row as a JSON array: `[\"Col A\",\"Col B\"]`", Input: []string{"file", "stdin"}},
-			{Name: "values", Kind: "own", Type: "string", Required: "optional", Desc: "Initial data as a 2D JSON array: `[[\"alice\",95]]`", Input: []string{"file", "stdin"}},
+			{Name: "values", Kind: "own", Type: "string", Required: "optional", Desc: "Untyped initial data as one 2D JSON array (`[[\"alice\",95]]`); values are written as-is with their type auto-detected, through the same batched set_cell_range path as --sheets — pair with --styles for number formats, colors, merges, and row/col sizes", Input: []string{"file", "stdin"}},
+			{Name: "sheets", Kind: "own", Type: "string", Required: "optional", Desc: "Typed table payload as JSON (same shape as `+table-put`): a top-level `sheets` array, each item `{name, start_cell?, mode?, header?, allow_overwrite?, columns:[\"colA\",\"colB\",...], data:[[...]], dtypes?:{colA:pandasDtype, ...}, formats?:{colA:numberFormat, ...}}`. Agents typically build it from a DataFrame via `{**json.loads(df.to_json(orient=\"split\")), \"dtypes\": df.dtypes.astype(str).to_dict()}`. Mutually exclusive with --values and --dataframe. Creates the workbook, then writes typed type-faithful data (dates land as real dates, numbers keep precision).", Input: []string{"file", "stdin"}},
+			{Name: "styles", Kind: "own", Type: "string", Required: "optional", Desc: "Initial visual operations as JSON: top-level `{styles:[...]}`. Each item corresponds to one target sheet and must include `name`, plus at least one of `cell_styles` / `row_sizes` / `col_sizes` / `cell_merges`. `cell_styles` entries use +cells-set-style fields with a cell range; row/col sizes use dimension ranges plus type/size; merges use cell ranges plus optional merge_type. With --sheets, styles array length/order/name must match --sheets.sheets. With --values, pass exactly one styles item for the initial sheet (its name is ignored).", Input: []string{"file", "stdin"}},
+			{Name: "dataframe", Kind: "own", Type: "string", Required: "optional", Desc: "Single-sheet typed table from one Arrow IPC file (Feather v2 — what `pandas.DataFrame.to_feather()` writes), mutually exclusive with --values and --sheets. Pass `@<path>` for a file or `-` for binary stdin (same convention as other input flags). Arrow bytes are read raw — no TrimSpace / BOM strip — so the IPC magic survives intact (unlike text input flags). Column types come from the Arrow schema; per-column `number_format` may be set via Arrow field metadata. Creates the workbook and fills its default sheet (`Sheet1` — adopted in place, no empty Sheet1 left behind). For multi-sheet or non-default placement, use `--sheets` instead."},
 			{Name: "dry-run", Kind: "system", Type: "bool", Required: "optional"},
 		},
 	},
@@ -916,6 +961,14 @@ var flagDefs = map[string]commandDef{
 			{Name: "dry-run", Kind: "system", Type: "bool", Required: "optional"},
 		},
 	},
+	"+workbook-import": {
+		Risk: "write",
+		Flags: []flagDef{
+			{Name: "file", Kind: "own", Type: "string", Required: "required", Desc: "Local file path (.xlsx / .xls / .csv)"},
+			{Name: "folder-token", Kind: "own", Type: "string", Required: "optional", Desc: "Target folder token; imported to the cloud drive root when omitted"},
+			{Name: "name", Kind: "own", Type: "string", Required: "optional", Desc: "Imported spreadsheet name; defaults to the local file name without its extension"},
+		},
+	},
 	"+workbook-info": {
 		Risk: "read",
 		Flags: []flagDef{

shortcuts/sheets/flag_defs_test.go

@@ -65,9 +65,9 @@ func TestFlagsFor_MapsAllFields(t *testing.T) {
 	if url == nil || url.Required {
 		t.Errorf("+sheet-create --url should not be cobra-required: %+v", url)
 	}
-	// hidden + int default
+	// visible + int default
 	cap := byName("+cells-get", "max-chars")
-	if cap == nil || !cap.Hidden || cap.Default != "200000" {
+	if cap == nil || cap.Hidden || cap.Default != "500000" {
 		t.Errorf("+cells-get --max-chars not mapped: %+v", cap)
 	}
 	// input sources
@@ -140,3 +140,24 @@ func TestFlagsFor_EveryRegisteredCommandHasDefs(t *testing.T) {
 		}
 	}
 }
+
+// TestFlagAcceptsStdin verifies the stdin-capability probe that decides whether
+// an "invalid JSON" error should also steer the caller toward stdin: a composite
+// flag (cells) accepts stdin, a plain locator (spreadsheet-token) does not, and
+// an unknown command/flag returns false without panicking (it runs on an error
+// path, unlike flagsFor).
+func TestFlagAcceptsStdin(t *testing.T) {
+	t.Parallel()
+	if !flagAcceptsStdin("+cells-set", "cells") {
+		t.Error("+cells-set --cells should accept stdin")
+	}
+	if flagAcceptsStdin("+cells-set", "spreadsheet-token") {
+		t.Error("--spreadsheet-token should not accept stdin")
+	}
+	if flagAcceptsStdin("+nope", "cells") {
+		t.Error("unknown command should be false (and must not panic)")
+	}
+	if flagAcceptsStdin("+cells-set", "nope") {
+		t.Error("unknown flag should be false")
+	}
+}

shortcuts/sheets/flag_schema_validate.go

@@ -63,6 +63,7 @@ func validateParsedJSONFlag(fv flagView, name string, value interface{}) error {
 var parseJSONFlagSkip = map[string]struct{}{
 	"properties": {},
 	"operations": {},
+	"styles":     {},
 }
 
 // validateValueAgainstSchema is the (command, flag) → schema → check
@@ -93,7 +94,17 @@ func validateValueAgainstSchema(fv flagView, name string, value interface{}) err
 	var schema schemaProperty
 	json.Unmarshal(raw, &schema)
 	if vErr := validateAgainstSchema(value, &schema, ""); vErr != nil {
-		return sheetsValidationForFlag(name, "--%s: %s", name, vErr.Error())
+		// Composite-JSON shape errors (e.g. +cells-set --cells, chart
+		// --properties) are the highest-frequency usage-layer failure for
+		// sheets, and agents often burn several retries guessing the shape.
+		// Point them straight at --print-schema, which dumps the exact JSON
+		// Schema for this (command, flag) pair. The hint is always actionable:
+		// reaching this branch means entry[name] resolved a schema from the
+		// embedded index, and --print-schema reads that same index, so the
+		// suggested command is guaranteed to print it.
+		return sheetsValidationForFlag(name,
+			"--%s: %s; run `lark-cli sheets %s --print-schema --flag-name %s` to see the expected JSON Schema",
+			name, vErr.Error(), command, name)
 	}
 	return nil
 }

shortcuts/sheets/flag_schema_validate_test.go

@@ -587,3 +587,27 @@ func TestValidateInputAgainstSchema_SkipOperations(t *testing.T) {
 		t.Errorf("operations should be skipped; got %v", err)
 	}
 }
+
+// TestValidateValueAgainstSchema_PrintSchemaHint pins the highest-value
+// recovery affordance for composite-JSON flags: when the shape is wrong, the
+// error must point the agent straight at --print-schema (with the right
+// command + flag) instead of leaving it to guess across retries. +cells-set
+// --cells expects a 2-D array; a bare string trips the top-level type check.
+func TestValidateValueAgainstSchema_PrintSchemaHint(t *testing.T) {
+	t.Parallel()
+	fv := mapFlagView{command: "+cells-set"}
+	err := validateValueAgainstSchema(fv, "cells", "not-an-array")
+	if err == nil {
+		t.Fatal("expected schema validation error for wrong --cells shape")
+	}
+	msg := err.Error()
+	// Underlying shape error is preserved (substring callers still match).
+	if !strings.Contains(msg, `expected type "array"`) {
+		t.Errorf("want underlying shape error preserved; got %q", msg)
+	}
+	// And the actionable --print-schema hint is appended with the exact
+	// command + flag, so a copy-paste fetches the schema for this pair.
+	if !strings.Contains(msg, "lark-cli sheets +cells-set --print-schema --flag-name cells") {
+		t.Errorf("want --print-schema hint with command+flag; got %q", msg)
+	}
+}

shortcuts/sheets/flag_schemas_gen.go

@@ -32,4 +32,6 @@ var commandsWithSchema = map[string]struct{}{
 	"+range-sort":            {},
 	"+sparkline-create":      {},
 	"+sparkline-update":      {},
+	"+table-put":             {},
+	"+workbook-create":       {},
 }

shortcuts/sheets/helpers.go

@@ -10,6 +10,8 @@ package sheets
 import (
 	"context"
 	"encoding/json"
+	"fmt"
+	neturl "net/url"
 	"strings"
 
 	"github.com/larksuite/cli/errs"
@@ -48,46 +50,151 @@ func sheetsInputStatError(flag string, err error) error {
 	return wrapped
 }
 
-// resolveSpreadsheetToken applies the public --url / --spreadsheet-token XOR
-// pair shared by every sheets canonical shortcut and returns the resolved
-// token. Network-free, safe to call from Validate and DryRun.
-func resolveSpreadsheetToken(runtime *common.RuntimeContext) (string, error) {
+// spreadsheetRef classification: a --url / --spreadsheet-token input names a
+// spreadsheet either directly (a /sheets/ URL or raw token) or indirectly via a
+// wiki node that must be resolved to its backing spreadsheet at Execute time.
+const (
+	spreadsheetRefSheet = "sheet"
+	spreadsheetRefWiki  = "wiki"
+)
+
+// spreadsheetRef is a parsed --url / --spreadsheet-token input. A wiki ref holds
+// the still-unresolved wiki node_token; resolveSpreadsheetTokenExec turns it
+// into the real spreadsheet token at Execute time.
+type spreadsheetRef struct {
+	Kind  string // spreadsheetRefSheet | spreadsheetRefWiki
+	Token string
+}
+
+// parseSpreadsheetRef applies the public --url / --spreadsheet-token XOR pair and
+// classifies the input. Network-free, safe to call from Validate and DryRun.
+//
+// Recognized --url shapes:
+//   - https://.../sheets/<token>        → {sheet, token}
+//   - https://.../spreadsheets/<token>  → {sheet, token}
+//   - https://.../wiki/<node_token>     → {wiki, node_token}  (resolved at Execute)
+//
+// A raw --spreadsheet-token is always treated as a spreadsheet token; wiki nodes
+// only ever arrive as a /wiki/ URL.
+func parseSpreadsheetRef(runtime *common.RuntimeContext) (spreadsheetRef, error) {
 	if err := common.ExactlyOneTyped(runtime, "url", "spreadsheet-token"); err != nil {
-		return "", err
+		return spreadsheetRef{}, err
 	}
 	if token := strings.TrimSpace(runtime.Str("spreadsheet-token")); token != "" {
 		if err := validate.RejectControlChars(token, "spreadsheet-token"); err != nil {
-			return "", sheetsValidationCauseForFlag("spreadsheet-token", err)
+			return spreadsheetRef{}, sheetsValidationCauseForFlag("spreadsheet-token", err)
 		}
-		return token, nil
+		return spreadsheetRef{Kind: spreadsheetRefSheet, Token: token}, nil
 	}
 
-	url := strings.TrimSpace(runtime.Str("url"))
-	token := extractSpreadsheetToken(url)
-	if token == "" || token == url {
-		return "", sheetsValidationForFlag("url", "--url must be a spreadsheet URL like https://.../sheets/<token>")
+	rawURL := strings.TrimSpace(runtime.Str("url"))
+	token, kind, ok := spreadsheetURLToken(rawURL)
+	if !ok {
+		return spreadsheetRef{}, sheetsValidationForFlag("url", "--url must be a spreadsheet URL like https://.../sheets/<token> or a wiki URL like https://.../wiki/<token>")
 	}
 	if err := validate.RejectControlChars(token, "url"); err != nil {
-		return "", sheetsValidationCauseForFlag("url", err)
+		return spreadsheetRef{}, sheetsValidationCauseForFlag("url", err)
 	}
-	return token, nil
+	return spreadsheetRef{Kind: kind, Token: token}, nil
 }
 
-// extractSpreadsheetToken pulls the token segment out of a /sheets/<token>
-// or /spreadsheets/<token> URL. Returns the input unchanged when no known
-// prefix is present (callers must check token != originalInput).
-func extractSpreadsheetToken(input string) string {
-	input = strings.TrimSpace(input)
-	for _, prefix := range []string{"/sheets/", "/spreadsheets/"} {
-		if idx := strings.Index(input, prefix); idx >= 0 {
-			token := input[idx+len(prefix):]
-			if idx2 := strings.IndexAny(token, "/?#"); idx2 >= 0 {
-				token = token[:idx2]
-			}
-			return token
+// spreadsheetURLToken extracts the token and its kind from a Lark URL, matching
+// only on the URL *path* segment (parsed via net/url). A /wiki/ or /sheets/ that
+// appears only in the query or fragment (e.g. a redirect or anchor param) never
+// hijacks classification. Returns ok=false when no known prefix heads the path.
+func spreadsheetURLToken(rawURL string) (token, kind string, ok bool) {
+	u, err := neturl.Parse(rawURL)
+	if err != nil || u.Path == "" {
+		return "", "", false
+	}
+	for _, m := range []struct {
+		prefix string
+		kind   string
+	}{
+		{"/sheets/", spreadsheetRefSheet},
+		{"/spreadsheets/", spreadsheetRefSheet},
+		{"/wiki/", spreadsheetRefWiki},
+	} {
+		if seg, found := pathSegmentAfter(u.Path, m.prefix); found {
+			return seg, m.kind, true
 		}
 	}
-	return input
+	return "", "", false
+}
+
+// pathSegmentAfter returns the first path segment after prefix when path begins
+// with prefix, else ("", false).
+func pathSegmentAfter(path, prefix string) (string, bool) {
+	if !strings.HasPrefix(path, prefix) {
+		return "", false
+	}
+	rest := path[len(prefix):]
+	if i := strings.IndexByte(rest, '/'); i >= 0 {
+		rest = rest[:i]
+	}
+	rest = strings.TrimSpace(rest)
+	if rest == "" {
+		return "", false
+	}
+	return rest, true
+}
+
+// resolveSpreadsheetToken applies the public --url / --spreadsheet-token XOR pair
+// and returns the resolved token. Network-free, safe to call from Validate and
+// DryRun.
+//
+// A /wiki/ URL yields the still-unresolved wiki node_token: turning it into the
+// backing spreadsheet token needs a get_node call, which only Execute may make.
+// Validate/DryRun only need a non-empty, control-char-clean token, so the
+// node_token passes through unchanged here; Execute paths call
+// resolveSpreadsheetTokenExec instead.
+func resolveSpreadsheetToken(runtime *common.RuntimeContext) (string, error) {
+	ref, err := parseSpreadsheetRef(runtime)
+	if err != nil {
+		return "", err
+	}
+	return ref.Token, nil
+}
+
+// resolveSpreadsheetTokenExec is the Execute-time counterpart of
+// resolveSpreadsheetToken: it additionally resolves a /wiki/ URL's node_token to
+// the backing spreadsheet token via wiki get_node, verifying obj_type=sheet.
+// Non-wiki inputs make no API call. Use this from every sheets Execute hook and
+// keep resolveSpreadsheetToken in Validate/DryRun so those stay network-free.
+func resolveSpreadsheetTokenExec(runtime *common.RuntimeContext) (string, error) {
+	ref, err := parseSpreadsheetRef(runtime)
+	if err != nil {
+		return "", err
+	}
+	if ref.Kind != spreadsheetRefWiki {
+		return ref.Token, nil
+	}
+	return resolveWikiNodeToSpreadsheetToken(runtime, ref.Token)
+}
+
+// resolveWikiNodeToSpreadsheetToken resolves a wiki node_token to the spreadsheet
+// obj_token it points at, erroring when the node is not a spreadsheet. The
+// wiki:node:read scope is only needed on this path, so it is enforced here rather
+// than declared unconditionally on every sheets shortcut.
+func resolveWikiNodeToSpreadsheetToken(runtime *common.RuntimeContext, nodeToken string) (string, error) {
+	if err := runtime.EnsureScopes([]string{"wiki:node:read"}); err != nil {
+		return "", err
+	}
+	data, err := runtime.CallAPITyped("GET", "/open-apis/wiki/v2/spaces/get_node",
+		map[string]interface{}{"token": nodeToken}, nil)
+	if err != nil {
+		return "", err
+	}
+	node := common.GetMap(data, "node")
+	objType := common.GetString(node, "obj_type")
+	objToken := common.GetString(node, "obj_token")
+	if objType == "" || objToken == "" {
+		return "", errs.NewInternalError(errs.SubtypeInvalidResponse, "wiki get_node returned incomplete node data for %q", nodeToken)
+	}
+	if objType != "sheet" {
+		return "", sheetsValidationForFlag("url", "wiki URL resolves to obj_type=%q, but a spreadsheet (obj_type=sheet) is required", objType)
+	}
+	return objToken, nil
 }
 
 // resolveSheetSelector validates the --sheet-id / --sheet-name XOR and
@@ -241,6 +348,16 @@ func parseJSONFlag(runtime flagView, name string) (interface{}, error) {
 	}
 	var out interface{}
 	if err := json.Unmarshal([]byte(raw), &out); err != nil {
+		// Composite payloads that embed formulas / quotes / commas are the
+		// classic source of this error: inlined into the shell, the JSON gets
+		// mangled (e.g. `\$` → "invalid character in string escape"). For any
+		// flag that accepts stdin, steer the caller there — passing the payload
+		// via `--<flag> - < file` sidesteps shell escaping entirely.
+		if flagAcceptsStdin(runtime.Command(), name) {
+			return nil, sheetsValidationForFlag(name,
+				"--%s: invalid JSON: %v; if the payload contains formulas / quotes / commas, pass it via stdin (`--%s - < file`) so the shell doesn't mangle the JSON",
+				name, err, name).WithCause(err)
+		}
 		return nil, sheetsValidationForFlag(name, "--%s: invalid JSON: %v", name, err).WithCause(err)
 	}
 	// Schema-driven flag validation at the user-input boundary. Skips
@@ -325,6 +442,72 @@ func buildCellStyleFromFlags(runtime flagView) map[string]interface{} {
 	return style
 }
 
+// cellStyleAliases maps shorthand cell_styles field names that models commonly
+// hallucinate (Excel / openpyxl / CSS conventions) onto the canonical field
+// names the backend expects. Only the unambiguous alignment shorthands are
+// aliased — they are the high-frequency miss; ambiguous guesses (e.g. "color",
+// "bg_color", "text_align") are intentionally left out so a wrong guess still
+// surfaces as an error rather than being silently reinterpreted.
+var cellStyleAliases = []struct{ alias, canonical string }{
+	{"horizontal_align", "horizontal_alignment"},
+	{"halign", "horizontal_alignment"},
+	{"vertical_align", "vertical_alignment"},
+	{"valign", "vertical_alignment"},
+}
+
+// normalizeCellStyleAliases renames known shorthand keys in a single
+// cell_styles map to their canonical equivalents, in place, so a model that
+// writes e.g. "horizontal_align" instead of "horizontal_alignment" still
+// applies the style instead of hitting an "unsupported field" error (--styles)
+// or having the field silently dropped by the backend (typed --cells). If both
+// the shorthand and its canonical key are present it returns a validation error
+// rather than picking one. path labels the map for the error message.
+func normalizeCellStyleAliases(style map[string]interface{}, path string) error {
+	if len(style) == 0 {
+		return nil
+	}
+	for _, a := range cellStyleAliases {
+		v, ok := style[a.alias]
+		if !ok {
+			continue
+		}
+		if _, exists := style[a.canonical]; exists {
+			return common.ValidationErrorf("%s.%s conflicts with %s; pass only %s", path, a.alias, a.canonical, a.canonical)
+		}
+		style[a.canonical] = v
+		delete(style, a.alias)
+	}
+	return nil
+}
+
+// normalizeTypedCellsStyleAliases walks a typed --cells 2D array and applies
+// normalizeCellStyleAliases to every cell's inline cell_styles object, so the
+// alignment shorthands are accepted on +cells-set the same as on --styles.
+// Structure is checked leniently to match the pass-through contract: any
+// element that isn't the expected shape is skipped, not rejected.
+func normalizeTypedCellsStyleAliases(cells []interface{}, path string) error {
+	for r, rowRaw := range cells {
+		row, ok := rowRaw.([]interface{})
+		if !ok {
+			continue
+		}
+		for c, cellRaw := range row {
+			cell, ok := cellRaw.(map[string]interface{})
+			if !ok {
+				continue
+			}
+			st, ok := cell["cell_styles"].(map[string]interface{})
+			if !ok {
+				continue
+			}
+			if err := normalizeCellStyleAliases(st, fmt.Sprintf("%s[%d][%d].cell_styles", path, r, c)); err != nil {
+				return err
+			}
+		}
+	}
+	return nil
+}
+
 // borderStylesFromFlag parses --border-styles as a JSON object (top/bottom/
 // left/right with style sub-objects). Returns nil when the flag is empty.
 func borderStylesFromFlag(runtime flagView) (map[string]interface{}, error) {

shortcuts/sheets/helpers_test.go

@@ -268,3 +268,53 @@ const (
 	testSheetID  = "shtSubA"
 	testSheetID2 = "shtSubB"
 )
+
+// TestParseSpreadsheetRef locks the network-free classification of
+// --url / --spreadsheet-token into a sheet token vs an (unresolved) wiki
+// node_token. The wiki node is resolved later, at Execute time only.
+func TestParseSpreadsheetRef(t *testing.T) {
+	t.Parallel()
+	mk := func(url, tok string) *common.RuntimeContext {
+		cmd := &cobra.Command{Use: "sheets"}
+		cmd.Flags().String("url", url, "")
+		cmd.Flags().String("spreadsheet-token", tok, "")
+		return common.TestNewRuntimeContext(cmd, testConfig(t))
+	}
+	cases := []struct {
+		name      string
+		url       string
+		tok       string
+		wantKind  string
+		wantToken string
+		wantErr   bool
+	}{
+		{name: "sheets url", url: "https://x.feishu.cn/sheets/shtABC", wantKind: spreadsheetRefSheet, wantToken: "shtABC"},
+		{name: "spreadsheets url", url: "https://x.feishu.cn/spreadsheets/shtABC", wantKind: spreadsheetRefSheet, wantToken: "shtABC"},
+		{name: "wiki url", url: "https://x.feishu.cn/wiki/wikDEF", wantKind: spreadsheetRefWiki, wantToken: "wikDEF"},
+		{name: "wiki url with query", url: "https://x.feishu.cn/wiki/wikDEF?sheet=xxxxxx", wantKind: spreadsheetRefWiki, wantToken: "wikDEF"},
+		{name: "raw token", tok: "shtRAW", wantKind: spreadsheetRefSheet, wantToken: "shtRAW"},
+		{name: "sheets url with /wiki/ in query stays sheet", url: "https://x.feishu.cn/sheets/shtABC?from=/wiki/wikX", wantKind: spreadsheetRefSheet, wantToken: "shtABC"},
+		{name: "sheets url with /wiki/ in fragment stays sheet", url: "https://x.feishu.cn/sheets/shtABC#/wiki/wikX", wantKind: spreadsheetRefSheet, wantToken: "shtABC"},
+		{name: "docx url unsupported", url: "https://x.feishu.cn/docx/docABC", wantErr: true},
+		{name: "neither provided", wantErr: true},
+	}
+	for _, tc := range cases {
+		tc := tc
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			ref, err := parseSpreadsheetRef(mk(tc.url, tc.tok))
+			if tc.wantErr {
+				if err == nil {
+					t.Fatalf("want error, got ref=%+v", ref)
+				}
+				return
+			}
+			if err != nil {
+				t.Fatalf("unexpected error: %v", err)
+			}
+			if ref.Kind != tc.wantKind || ref.Token != tc.wantToken {
+				t.Fatalf("ref = %+v, want {Kind:%s Token:%s}", ref, tc.wantKind, tc.wantToken)
+			}
+		})
+	}
+}

shortcuts/sheets/lark_sheet_batch_update.go

@@ -67,7 +67,7 @@ var BatchUpdate = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindWrite, "batch_update", input)
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}
@@ -180,7 +180,7 @@ var CellsBatchSetStyle = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindWrite, "batch_update", input)
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}
@@ -270,7 +270,7 @@ var CellsBatchClear = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindWrite, "batch_update", input)
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}
@@ -350,7 +350,7 @@ var DropdownUpdate = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindWrite, "batch_update", input)
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}
@@ -396,7 +396,7 @@ var DropdownDelete = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindWrite, "batch_update", input)
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}

shortcuts/sheets/lark_sheet_dataframe.go

@@ -0,0 +1,554 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package sheets
+
+import (
+	"bytes"
+	"encoding/json"
+	"fmt"
+	"io"
+	"strconv"
+	"strings"
+	"time"
+
+	"github.com/apache/arrow/go/v17/arrow"
+	"github.com/apache/arrow/go/v17/arrow/array"
+	"github.com/apache/arrow/go/v17/arrow/ipc"
+	"github.com/apache/arrow/go/v17/arrow/memory"
+
+	"github.com/larksuite/cli/errs"
+	"github.com/larksuite/cli/extension/fileio"
+	"github.com/larksuite/cli/internal/cmdutil"
+	"github.com/larksuite/cli/shortcuts/common"
+)
+
+// ─── --dataframe (Arrow IPC / Feather v2 binary input) ────────────────
+//
+// --dataframe is the binary-typed twin of --sheets. The wire payload is one
+// Arrow IPC file (a.k.a. Feather v2 — what `pandas.DataFrame.to_feather()`
+// writes), single schema, optionally multi-batch. Type / format are read off
+// the Arrow schema (no separate dtypes/formats maps), and per-column number
+// format can be set via the field's `number_format` metadata key:
+//
+//	pa.field("price", pa.float64(), metadata={b"number_format": b"$#,##0.00"})
+//
+// One DataFrame writes into one sub-sheet at fixed defaults: name `Sheet1`
+// (adopted in place by +workbook-create; created when absent by +table-put),
+// overwrite from A1 with header on, allow_overwrite=true. The shortcut
+// surface is deliberately the one flag — anything that needs a different
+// sheet name / anchor / mode / multi-sheet falls back to --sheets, whose
+// JSON payload already carries every knob.
+//
+// Binary IO note: --dataframe bypasses the text-oriented Input resolver
+// (`runtime.Str("dataframe")` carries a *path*, not file contents). Reading
+// the Arrow bytes through that resolver would TrimSpace the trailing IPC
+// magic / corrupt non-UTF8 bytes. Path → FileIO.Open → io.ReadAll keeps the
+// stream byte-exact. "-" reads from stdin directly.
+
+// dataframeDefaultSheetName is the sub-sheet name --dataframe writes into.
+// Matches valuesSheetName so +workbook-create adopts the brand-new
+// workbook's default sheet in place (no stray empty Sheet1 left behind);
+// +table-put creates Sheet1 if it doesn't already exist.
+const dataframeDefaultSheetName = valuesSheetName
+
+// parseDataframePayload reads the --dataframe path (Arrow IPC file) and
+// composes a single-sheet tablePayload at the fixed default placement.
+// Network-free: safe from Validate and DryRun. The resulting tableSheetSpec
+// rides the same buildSheetMatrix / buildTypedCell path as a --sheets entry,
+// so downstream is unaware of where the rows came from.
+func parseDataframePayload(rctx *common.RuntimeContext) (*tablePayload, error) {
+	raw := strings.TrimSpace(rctx.Str("dataframe"))
+	if raw == "" {
+		return nil, common.ValidationErrorf("--dataframe is required")
+	}
+	data, err := readDataframeBytes(rctx, raw)
+	if err != nil {
+		return nil, err
+	}
+	spec, err := decodeArrowToSheet(data, dataframeDefaultSheetName)
+	if err != nil {
+		return nil, common.ValidationErrorf("--dataframe: %v", err)
+	}
+	payload := &tablePayload{Sheets: []tableSheetSpec{spec}}
+	if err := payload.validate(); err != nil {
+		return nil, err
+	}
+	return payload, nil
+}
+
+// dataframeStdinCache holds the bytes read from stdin on the first call so a
+// later call (Validate → Execute / DryRun) gets the same bytes instead of an
+// empty stream — stdin is single-shot, but parseDataframePayload runs
+// multiple times per command invocation. Process-wide is fine: lark-cli is
+// one-shot (one command per process). Tests reset by setting it back to nil.
+var dataframeStdinCache []byte
+
+// readDataframeBytes resolves --dataframe to raw binary. A literal `@` prefix
+// is tolerated for symmetry with --sheets (`@/tmp/x.arrow` and `/tmp/x.arrow`
+// both work). `-` reads stdin verbatim — cached on first call so Validate /
+// Execute / DryRun all see the same bytes. Bytes are returned untouched: no
+// TrimSpace, no BOM strip — both would corrupt an Arrow IPC stream.
+func readDataframeBytes(rctx *common.RuntimeContext, raw string) ([]byte, error) {
+	if raw == "-" {
+		if dataframeStdinCache != nil {
+			return dataframeStdinCache, nil
+		}
+		io := rctx.IO()
+		if io == nil || io.In == nil {
+			return nil, common.ValidationErrorf("--dataframe: stdin is not available")
+		}
+		data, err := readAllBytes(io.In)
+		if err != nil {
+			return nil, common.ValidationErrorf("--dataframe: read stdin: %v", err)
+		}
+		if len(data) == 0 {
+			return nil, common.ValidationErrorf("--dataframe: stdin is empty")
+		}
+		dataframeStdinCache = data
+		return data, nil
+	}
+	path := strings.TrimPrefix(raw, "@")
+	data, err := cmdutil.ReadInputFile(rctx.FileIO(), path)
+	if err != nil {
+		return nil, common.ValidationErrorf("--dataframe: %v", err)
+	}
+	if len(data) == 0 {
+		return nil, common.ValidationErrorf("--dataframe: file %q is empty", path)
+	}
+	return data, nil
+}
+
+// readAllBytes is a thin wrapper so tests can fake the io.Reader without
+// importing io. Mirrors io.ReadAll exactly.
+func readAllBytes(r io.Reader) ([]byte, error) { return io.ReadAll(r) }
+
+// decodeArrowToSheet reads `data` as an Arrow IPC file (single schema,
+// possibly multi-batch) and produces a tableSheetSpec with name + columns +
+// rows filled in. Sheet placement (start_cell / mode / header / overwrite) is
+// not touched here — parseDataframePayload layers those on from CLI flags.
+func decodeArrowToSheet(data []byte, sheetName string) (tableSheetSpec, error) {
+	reader, err := ipc.NewFileReader(bytes.NewReader(data))
+	if err != nil {
+		return tableSheetSpec{}, fmt.Errorf("invalid Arrow IPC file (expected pandas df.to_feather output): %v", err) //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+	}
+	defer reader.Close()
+
+	schema := reader.Schema()
+	if schema == nil || schema.NumFields() == 0 {
+		return tableSheetSpec{}, fmt.Errorf("Arrow schema has no fields") //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+	}
+
+	ncols := schema.NumFields()
+	cols := make([]tableColumnSpec, ncols)
+	seen := make(map[string]bool, ncols)
+	for i := 0; i < ncols; i++ {
+		f := schema.Field(i)
+		name := f.Name
+		if strings.TrimSpace(name) == "" {
+			return tableSheetSpec{}, fmt.Errorf("column %d has empty name", i) //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+		}
+		if seen[name] {
+			return tableSheetSpec{}, fmt.Errorf("duplicate column name %q", name) //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+		}
+		seen[name] = true
+		typ, format, err := arrowFieldToTypeFormat(f)
+		if err != nil {
+			return tableSheetSpec{}, fmt.Errorf("column %q: %v", name, err) //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+		}
+		cols[i] = tableColumnSpec{Name: name, Type: typ, Format: format}
+	}
+
+	var rows [][]interface{}
+	for b := 0; b < reader.NumRecords(); b++ {
+		rec, err := reader.RecordAt(b)
+		if err != nil {
+			return tableSheetSpec{}, fmt.Errorf("read record batch %d: %v", b, err) //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+		}
+		batchRows, err := arrowRecordToRows(rec, cols)
+		rec.Release()
+		if err != nil {
+			return tableSheetSpec{}, err
+		}
+		rows = append(rows, batchRows...)
+	}
+
+	return tableSheetSpec{Name: sheetName, Columns: cols, Rows: rows}, nil
+}
+
+// arrowFieldToTypeFormat maps an Arrow field to the internal (type, format)
+// pair. The field's `number_format` metadata key — when present — sets the
+// Excel number_format string verbatim; otherwise sensible defaults are
+// applied per type (`@` text for strings, `yyyy-mm-dd` for dates).
+func arrowFieldToTypeFormat(f arrow.Field) (typ, format string, err error) {
+	if v, ok := f.Metadata.GetValue("number_format"); ok {
+		format = strings.TrimSpace(v)
+	}
+	switch f.Type.(type) {
+	case *arrow.StringType, *arrow.LargeStringType:
+		if format == "" {
+			format = "@"
+		}
+		return "string", format, nil
+	case *arrow.BooleanType:
+		return "bool", format, nil
+	case *arrow.Date32Type, *arrow.Date64Type, *arrow.TimestampType:
+		if format == "" {
+			format = "yyyy-mm-dd"
+		}
+		return "date", format, nil
+	}
+	if isArrowNumericType(f.Type) {
+		return "number", format, nil
+	}
+	return "", "", fmt.Errorf("unsupported Arrow type %s (want string/number/date/bool)", f.Type.Name()) //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+}
+
+func isArrowNumericType(t arrow.DataType) bool {
+	switch t.ID() {
+	case arrow.INT8, arrow.INT16, arrow.INT32, arrow.INT64,
+		arrow.UINT8, arrow.UINT16, arrow.UINT32, arrow.UINT64,
+		arrow.FLOAT16, arrow.FLOAT32, arrow.FLOAT64:
+		return true
+	}
+	return false
+}
+
+// arrowRecordToRows transposes one column-batch into row-major
+// [][]interface{} matched to `cols`. Cells are stamped with the same value
+// shapes buildTypedCell expects from the JSON path: nil for nulls,
+// json.Number for numerics (precision-preserving), `yyyy-mm-dd` strings for
+// dates/timestamps, bool for booleans, string for strings.
+func arrowRecordToRows(rec arrow.Record, cols []tableColumnSpec) ([][]interface{}, error) {
+	if int(rec.NumCols()) != len(cols) {
+		return nil, fmt.Errorf("record has %d cols, schema declared %d", rec.NumCols(), len(cols)) //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+	}
+	nrows := int(rec.NumRows())
+	rows := make([][]interface{}, nrows)
+	for r := range rows {
+		rows[r] = make([]interface{}, len(cols))
+	}
+	for c := range cols {
+		arr := rec.Column(c)
+		for r := 0; r < nrows; r++ {
+			if arr.IsNull(r) {
+				continue
+			}
+			v, err := arrowCellValue(arr, r)
+			if err != nil {
+				return nil, fmt.Errorf("row %d column %q: %v", r, cols[c].Name, err) //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+			}
+			rows[r][c] = v
+		}
+	}
+	return rows, nil
+}
+
+func arrowCellValue(arr arrow.Array, i int) (interface{}, error) {
+	switch a := arr.(type) {
+	case *array.String:
+		return a.Value(i), nil
+	case *array.LargeString:
+		return a.Value(i), nil
+	case *array.Boolean:
+		return a.Value(i), nil
+	case *array.Int8:
+		return json.Number(strconv.FormatInt(int64(a.Value(i)), 10)), nil
+	case *array.Int16:
+		return json.Number(strconv.FormatInt(int64(a.Value(i)), 10)), nil
+	case *array.Int32:
+		return json.Number(strconv.FormatInt(int64(a.Value(i)), 10)), nil
+	case *array.Int64:
+		return json.Number(strconv.FormatInt(a.Value(i), 10)), nil
+	case *array.Uint8:
+		return json.Number(strconv.FormatUint(uint64(a.Value(i)), 10)), nil
+	case *array.Uint16:
+		return json.Number(strconv.FormatUint(uint64(a.Value(i)), 10)), nil
+	case *array.Uint32:
+		return json.Number(strconv.FormatUint(uint64(a.Value(i)), 10)), nil
+	case *array.Uint64:
+		return json.Number(strconv.FormatUint(a.Value(i), 10)), nil
+	case *array.Float16:
+		return json.Number(strconv.FormatFloat(float64(a.Value(i).Float32()), 'f', -1, 32)), nil
+	case *array.Float32:
+		return json.Number(strconv.FormatFloat(float64(a.Value(i)), 'f', -1, 32)), nil
+	case *array.Float64:
+		return json.Number(strconv.FormatFloat(a.Value(i), 'f', -1, 64)), nil
+	case *array.Date32:
+		// Date32: days since 1970-01-01 (epoch). Multiply to seconds, format
+		// in UTC so timezone offset can't flip the calendar date.
+		t := time.Unix(int64(a.Value(i))*86400, 0).UTC()
+		return t.Format("2006-01-02"), nil
+	case *array.Date64:
+		t := time.UnixMilli(int64(a.Value(i))).UTC()
+		return t.Format("2006-01-02"), nil
+	case *array.Timestamp:
+		ts := int64(a.Value(i))
+		unit := a.DataType().(*arrow.TimestampType).Unit
+		var t time.Time
+		switch unit {
+		case arrow.Second:
+			t = time.Unix(ts, 0).UTC()
+		case arrow.Millisecond:
+			t = time.UnixMilli(ts).UTC()
+		case arrow.Microsecond:
+			t = time.UnixMicro(ts).UTC()
+		case arrow.Nanosecond:
+			t = time.Unix(0, ts).UTC()
+		default:
+			return nil, fmt.Errorf("unsupported timestamp unit %v", unit) //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+		}
+		return t.Format("2006-01-02"), nil
+	}
+	return nil, fmt.Errorf("unsupported Arrow array %T", arr) //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+}
+
+// ─── --dataframe-out (Arrow IPC binary output, mirror of --dataframe) ──
+//
+// +table-get's binary read-back: encode one sheet's typed read-back as an
+// Arrow IPC file (Feather v2), so pandas can `pd.read_feather(path)` /
+// `pd.read_feather(BytesIO(stdout))` symmetrically with the put side.
+// Single-sheet only — Arrow IPC carries one schema per file. The JSON path
+// is unchanged; --dataframe-out swaps the encoder for callers that already
+// have pandas / pyarrow in their pipeline.
+
+// encodeSheetMapToArrowIPC turns one readSheetAsSpec output into an Arrow IPC
+// file blob. Internal column types are recovered from `dtypes` (the wire
+// proxy for the typed protocol), and per-column `number_format` rides through
+// as Arrow field metadata so the file feeds straight back into
+// `+table-put --dataframe`.
+func encodeSheetMapToArrowIPC(sheet map[string]interface{}) ([]byte, error) {
+	columns, _ := sheet["columns"].([]interface{})
+	if len(columns) == 0 {
+		return nil, fmt.Errorf("sheet has no columns") //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+	}
+	dtypes, _ := sheet["dtypes"].(map[string]interface{})
+	formats, _ := sheet["formats"].(map[string]interface{})
+	// `data` arrives as either []interface{} (when the sheet came through a
+	// JSON round-trip / unit-test fixture) or [][]interface{} (the shape
+	// readSheetAsSpec directly emits in production). Accept both — anything
+	// else falls through to a zero-row table.
+	var rawData [][]interface{}
+	switch d := sheet["data"].(type) {
+	case [][]interface{}:
+		rawData = d
+	case []interface{}:
+		rawData = make([][]interface{}, len(d))
+		for i, r := range d {
+			rawData[i], _ = r.([]interface{})
+		}
+	}
+
+	ncols := len(columns)
+	colNames := make([]string, ncols)
+	colTypes := make([]string, ncols)
+	fields := make([]arrow.Field, ncols)
+	for i, c := range columns {
+		name, _ := c.(string)
+		if name == "" {
+			return nil, fmt.Errorf("column %d has empty name", i) //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+		}
+		colNames[i] = name
+		dt, _ := dtypes[name].(string)
+		colTypes[i] = dtypeToInternalType(dt)
+		var meta arrow.Metadata
+		if formats != nil {
+			if nf, ok := formats[name].(string); ok && strings.TrimSpace(nf) != "" {
+				meta = arrow.NewMetadata([]string{"number_format"}, []string{nf})
+			}
+		}
+		fields[i] = arrow.Field{
+			Name:     name,
+			Type:     internalTypeToArrowType(colTypes[i]),
+			Nullable: true,
+			Metadata: meta,
+		}
+	}
+	schema := arrow.NewSchema(fields, nil)
+
+	mem := memory.NewGoAllocator()
+	rb := array.NewRecordBuilder(mem, schema)
+	defer rb.Release()
+	for r, row := range rawData {
+		if len(row) != ncols {
+			return nil, fmt.Errorf("row %d has %d cells, want %d", r, len(row), ncols) //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+		}
+		for c := 0; c < ncols; c++ {
+			if err := appendArrowCell(rb.Field(c), colTypes[c], row[c]); err != nil {
+				return nil, fmt.Errorf("row %d column %q: %v", r, colNames[c], err) //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+			}
+		}
+	}
+	rec := rb.NewRecord()
+	defer rec.Release()
+
+	var buf bytesWriterSeeker
+	w, err := ipc.NewFileWriter(&buf, ipc.WithSchema(schema), ipc.WithAllocator(mem))
+	if err != nil {
+		return nil, fmt.Errorf("ipc.NewFileWriter: %v", err) //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+	}
+	if err := w.Write(rec); err != nil {
+		return nil, fmt.Errorf("write record: %v", err) //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+	}
+	if err := w.Close(); err != nil {
+		return nil, fmt.Errorf("close writer: %v", err) //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+	}
+	return buf.buf, nil
+}
+
+// dtypeToInternalType inverts typeToDtype so the Arrow encoder can pick an
+// internal column type from the wire-level dtype string. Unknown / object
+// falls back to string (lossless: every cell is already typed as such).
+func dtypeToInternalType(dtype string) string {
+	switch strings.ToLower(strings.TrimSpace(dtype)) {
+	case "float64", "float32", "int64", "int32", "int16", "int8",
+		"uint64", "uint32", "uint16", "uint8":
+		return "number"
+	case "bool", "boolean":
+		return "bool"
+	}
+	if strings.HasPrefix(strings.ToLower(dtype), "datetime") {
+		return "date"
+	}
+	return "string"
+}
+
+// internalTypeToArrowType is the put-side dtypeToTypeFormat dual: maps the
+// internal column type to the Arrow data type the encoder builds a column
+// with. Numbers go to float64 because +table-get can't tell int from float
+// from a number_format alone — float64 covers both losslessly for the cell
+// ranges Lark Sheets accepts.
+func internalTypeToArrowType(typ string) arrow.DataType {
+	switch typ {
+	case "number":
+		return arrow.PrimitiveTypes.Float64
+	case "date":
+		return arrow.FixedWidthTypes.Date32
+	case "bool":
+		return arrow.FixedWidthTypes.Boolean
+	}
+	return arrow.BinaryTypes.String
+}
+
+// appendArrowCell stamps one cell into its column builder. Cell shape matches
+// what cellToTyped emits on the JSON path: json.Number for numbers, ISO
+// `yyyy-mm-dd` string for dates, plain string for strings, bool for bools,
+// nil for empty. Anything off-shape errors so the caller doesn't silently
+// emit nulls for malformed data.
+func appendArrowCell(b array.Builder, typ string, v interface{}) error {
+	if v == nil {
+		b.AppendNull()
+		return nil
+	}
+	switch typ {
+	case "string":
+		s, ok := v.(string)
+		if !ok {
+			return fmt.Errorf("string expects string value, got %T", v) //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+		}
+		b.(*array.StringBuilder).Append(s)
+	case "number":
+		f, err := arrowNumber(v)
+		if err != nil {
+			return err
+		}
+		b.(*array.Float64Builder).Append(f)
+	case "date":
+		s, ok := v.(string)
+		if !ok {
+			return fmt.Errorf("date expects ISO yyyy-mm-dd string, got %T", v) //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+		}
+		t, err := time.Parse("2006-01-02", strings.TrimSpace(s))
+		if err != nil {
+			return fmt.Errorf("date parse %q: %v", s, err) //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+		}
+		b.(*array.Date32Builder).Append(arrow.Date32FromTime(t))
+	case "bool":
+		bb, ok := v.(bool)
+		if !ok {
+			return fmt.Errorf("bool expects bool, got %T", v) //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+		}
+		b.(*array.BooleanBuilder).Append(bb)
+	default:
+		return fmt.Errorf("unsupported internal type %q", typ) //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+	}
+	return nil
+}
+
+// arrowNumber converts the number cell shape readSheetAsSpec emits
+// (json.Number) plus the float fallback to float64 for the Arrow builder.
+func arrowNumber(v interface{}) (float64, error) {
+	switch n := v.(type) {
+	case json.Number:
+		f, err := n.Float64()
+		if err != nil {
+			return 0, fmt.Errorf("number parse %q: %v", n.String(), err) //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+		}
+		return f, nil
+	case float64:
+		return n, nil
+	}
+	return 0, fmt.Errorf("number expects numeric value, got %T", v) //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+}
+
+// bytesWriterSeeker is a 10-line in-memory io.WriteSeeker for
+// ipc.NewFileWriter, which seeks back to patch a footer offset. Using a
+// buffer (instead of a temp file or os.Stdout, which isn't seekable) keeps
+// --dataframe-out's stdout path zero-IO and stays straightforward.
+type bytesWriterSeeker struct {
+	buf []byte
+	pos int64
+}
+
+func (w *bytesWriterSeeker) Write(p []byte) (int, error) {
+	end := w.pos + int64(len(p))
+	if end > int64(len(w.buf)) {
+		w.buf = append(w.buf, make([]byte, end-int64(len(w.buf)))...)
+	}
+	n := copy(w.buf[w.pos:], p)
+	w.pos = end
+	return n, nil
+}
+
+func (w *bytesWriterSeeker) Seek(offset int64, whence int) (int64, error) {
+	switch whence {
+	case io.SeekStart:
+		w.pos = offset
+	case io.SeekCurrent:
+		w.pos += offset
+	case io.SeekEnd:
+		w.pos = int64(len(w.buf)) + offset
+	default:
+		return 0, fmt.Errorf("unknown whence %d", whence) //nolint:forbidigo // intermediate error; the command layer wraps it into a typed --dataframe/--dataframe-out validation error
+	}
+	return w.pos, nil
+}
+
+// writeDataframeOut dispatches the encoded Arrow bytes to wherever --dataframe-out
+// points: `-` → process stdout, `@<path>` or plain path → local file. Symmetric
+// with readDataframeBytes on the input side: same `@` tolerance, same TrimPrefix
+// semantics, and an absolute path will still get rejected by FileIO's SafePath.
+func writeDataframeOut(rctx *common.RuntimeContext, raw string, data []byte) error {
+	if raw == "-" {
+		out := rctx.IO()
+		if out == nil || out.Out == nil {
+			return common.ValidationErrorf("--dataframe-out: stdout is not available")
+		}
+		if _, err := out.Out.Write(data); err != nil {
+			return errs.NewInternalError(errs.SubtypeFileIO, "--dataframe-out: write stdout").WithCause(err)
+		}
+		return nil
+	}
+	path := strings.TrimPrefix(raw, "@")
+	fio := rctx.FileIO()
+	if fio == nil {
+		return common.ValidationErrorf("--dataframe-out: file output is not available in this context")
+	}
+	// FileIO.Save validates the path via SafeOutputPath (the same sandbox
+	// readDataframeBytes hits on the input side) and writes atomically, so we
+	// don't need an extra ValidatePath call here.
+	if _, err := fio.Save(path, fileio.SaveOptions{ContentLength: int64(len(data))}, bytes.NewReader(data)); err != nil {
+		return errs.NewInternalError(errs.SubtypeFileIO, "--dataframe-out: write %q", path).WithCause(err)
+	}
+	return nil
+}

shortcuts/sheets/lark_sheet_dataframe_test.go

@@ -0,0 +1,378 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package sheets
+
+import (
+	"encoding/json"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/apache/arrow/go/v17/arrow"
+	"github.com/apache/arrow/go/v17/arrow/array"
+	"github.com/apache/arrow/go/v17/arrow/ipc"
+	"github.com/apache/arrow/go/v17/arrow/memory"
+)
+
+// buildArrowIPC writes one record into a Feather v2 (Arrow IPC file) blob.
+// Used by the round-trip tests below to stand in for what
+// `pandas.DataFrame.to_feather(path)` would produce; saves the tests from
+// depending on a pandas-shaped fixture file.
+//
+// ipc.NewFileWriter wants an io.WriteSeeker (it back-patches a footer
+// offset), so we write to a temp file and read the bytes back — simpler than
+// re-implementing a seekable in-memory buffer.
+func buildArrowIPC(t *testing.T, schema *arrow.Schema, build func(b *array.RecordBuilder)) []byte {
+	t.Helper()
+	mem := memory.NewGoAllocator()
+	rb := array.NewRecordBuilder(mem, schema)
+	defer rb.Release()
+	build(rb)
+	rec := rb.NewRecord()
+	defer rec.Release()
+
+	path := filepath.Join(t.TempDir(), "df.arrow")
+	f, err := os.Create(path)
+	if err != nil {
+		t.Fatalf("create temp arrow file: %v", err)
+	}
+	w, err := ipc.NewFileWriter(f, ipc.WithSchema(schema), ipc.WithAllocator(mem))
+	if err != nil {
+		f.Close()
+		t.Fatalf("ipc.NewFileWriter: %v", err)
+	}
+	if err := w.Write(rec); err != nil {
+		t.Fatalf("write record: %v", err)
+	}
+	if err := w.Close(); err != nil {
+		t.Fatalf("close writer: %v", err)
+	}
+	if err := f.Close(); err != nil {
+		t.Fatalf("close file: %v", err)
+	}
+	data, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatalf("read temp arrow file: %v", err)
+	}
+	return data
+}
+
+// TestDataframe_RoundTripCoreTypes pins down the Arrow-schema → internal
+// (type, format) mapping and the per-cell value shape that buildTypedCell
+// expects: number cells are json.Number (precision-preserving), date cells
+// are `yyyy-mm-dd` strings, bool/string come through verbatim. Numbers, dates,
+// strings, bools, and nulls all in one record so a future Arrow-Go bump can't
+// quietly regress any one family.
+func TestDataframe_RoundTripCoreTypes(t *testing.T) {
+	t.Parallel()
+
+	schema := arrow.NewSchema([]arrow.Field{
+		{Name: "name", Type: arrow.BinaryTypes.String},
+		{Name: "qty", Type: arrow.PrimitiveTypes.Int64},
+		{Name: "price", Type: arrow.PrimitiveTypes.Float64, Metadata: arrow.NewMetadata(
+			[]string{"number_format"}, []string{"$#,##0.00"},
+		)},
+		{Name: "active", Type: arrow.FixedWidthTypes.Boolean},
+		{Name: "shipped_on", Type: arrow.FixedWidthTypes.Date32},
+	}, nil)
+
+	jan15 := arrow.Date32FromTime(time.Date(2024, 1, 15, 0, 0, 0, 0, time.UTC))
+	feb02 := arrow.Date32FromTime(time.Date(2024, 2, 2, 0, 0, 0, 0, time.UTC))
+
+	buf := buildArrowIPC(t, schema, func(b *array.RecordBuilder) {
+		b.Field(0).(*array.StringBuilder).AppendValues([]string{"alice", ""}, []bool{true, false})
+		b.Field(1).(*array.Int64Builder).AppendValues([]int64{42, 0}, []bool{true, false})
+		b.Field(2).(*array.Float64Builder).AppendValues([]float64{19.95, 0}, []bool{true, false})
+		b.Field(3).(*array.BooleanBuilder).AppendValues([]bool{true, false}, []bool{true, true})
+		b.Field(4).(*array.Date32Builder).AppendValues([]arrow.Date32{jan15, feb02}, []bool{true, true})
+	})
+
+	spec, err := decodeArrowToSheet(buf, "S1")
+	if err != nil {
+		t.Fatalf("decodeArrowToSheet: %v", err)
+	}
+	if spec.Name != "S1" {
+		t.Errorf("sheet name = %q, want S1", spec.Name)
+	}
+	if len(spec.Columns) != 5 {
+		t.Fatalf("got %d columns, want 5", len(spec.Columns))
+	}
+	want := []struct{ typ, format string }{
+		{"string", "@"},
+		{"number", ""},
+		{"number", "$#,##0.00"},
+		{"bool", ""},
+		{"date", "yyyy-mm-dd"},
+	}
+	for i, w := range want {
+		if spec.Columns[i].Type != w.typ {
+			t.Errorf("columns[%d].Type = %q, want %q", i, spec.Columns[i].Type, w.typ)
+		}
+		if spec.Columns[i].Format != w.format {
+			t.Errorf("columns[%d].Format = %q, want %q", i, spec.Columns[i].Format, w.format)
+		}
+	}
+
+	if len(spec.Rows) != 2 {
+		t.Fatalf("got %d rows, want 2", len(spec.Rows))
+	}
+	// Row 0: every field present, types match what buildTypedCell will accept.
+	row0 := spec.Rows[0]
+	if row0[0] != "alice" {
+		t.Errorf("row0[name] = %#v, want \"alice\"", row0[0])
+	}
+	if n, ok := row0[1].(json.Number); !ok || n.String() != "42" {
+		t.Errorf("row0[qty] = %#v, want json.Number(\"42\")", row0[1])
+	}
+	if n, ok := row0[2].(json.Number); !ok || n.String() != "19.95" {
+		t.Errorf("row0[price] = %#v, want json.Number(\"19.95\")", row0[2])
+	}
+	if row0[3] != true {
+		t.Errorf("row0[active] = %#v, want true", row0[3])
+	}
+	if row0[4] != "2024-01-15" {
+		t.Errorf("row0[shipped_on] = %#v, want \"2024-01-15\"", row0[4])
+	}
+
+	// Row 1: nulls on name/qty/price (despite the buffer values) must become nil
+	// so buildTypedCell paints an empty cell that still carries number_format.
+	row1 := spec.Rows[1]
+	for _, c := range []int{0, 1, 2} {
+		if row1[c] != nil {
+			t.Errorf("row1[%d] = %#v, want nil (null in arrow)", c, row1[c])
+		}
+	}
+	if row1[3] != false {
+		t.Errorf("row1[active] = %#v, want false", row1[3])
+	}
+	if row1[4] != "2024-02-02" {
+		t.Errorf("row1[shipped_on] = %#v, want \"2024-02-02\"", row1[4])
+	}
+}
+
+// TestDataframe_Timestamp pins the timestamp → date conversion for the
+// timestamp[us] case (pandas default for `pd.Timestamp` columns once written
+// via `to_feather`). Only the calendar date matters for our `yyyy-mm-dd`
+// landing — guard against TZ drift from the wrong unit pick.
+func TestDataframe_Timestamp(t *testing.T) {
+	t.Parallel()
+	schema := arrow.NewSchema([]arrow.Field{
+		{Name: "ts", Type: &arrow.TimestampType{Unit: arrow.Microsecond}},
+	}, nil)
+	ts := arrow.Timestamp(time.Date(2024, 6, 12, 14, 30, 0, 0, time.UTC).UnixMicro())
+	buf := buildArrowIPC(t, schema, func(b *array.RecordBuilder) {
+		b.Field(0).(*array.TimestampBuilder).AppendValues([]arrow.Timestamp{ts}, []bool{true})
+	})
+	spec, err := decodeArrowToSheet(buf, "S")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if spec.Columns[0].Type != "date" {
+		t.Errorf("type = %q, want date", spec.Columns[0].Type)
+	}
+	if got := spec.Rows[0][0]; got != "2024-06-12" {
+		t.Errorf("ts = %#v, want \"2024-06-12\"", got)
+	}
+}
+
+// TestDataframe_EmptySchema rejects an Arrow file whose schema has no fields:
+// a 0-column "DataFrame" would write a header-less, data-less block that
+// validates as "writer ran successfully" but produces nothing — the test ties
+// that off as an explicit error rather than letting it slip through.
+func TestDataframe_EmptySchema(t *testing.T) {
+	t.Parallel()
+	schema := arrow.NewSchema(nil, nil)
+	buf := buildArrowIPC(t, schema, func(b *array.RecordBuilder) {})
+	_, err := decodeArrowToSheet(buf, "S")
+	if err == nil || !strings.Contains(err.Error(), "no fields") {
+		t.Errorf("err = %v, want 'no fields' error", err)
+	}
+}
+
+// TestDataframe_DuplicateColumn catches duplicate-name columns at decode
+// time. Validate already rejects duplicate column names for the JSON path;
+// the Arrow path mirrors that so the error surfaces with the same shape.
+func TestDataframe_DuplicateColumn(t *testing.T) {
+	t.Parallel()
+	schema := arrow.NewSchema([]arrow.Field{
+		{Name: "x", Type: arrow.BinaryTypes.String},
+		{Name: "x", Type: arrow.PrimitiveTypes.Int64},
+	}, nil)
+	buf := buildArrowIPC(t, schema, func(b *array.RecordBuilder) {
+		b.Field(0).(*array.StringBuilder).Append("")
+		b.Field(1).(*array.Int64Builder).Append(0)
+	})
+	_, err := decodeArrowToSheet(buf, "S")
+	if err == nil || !strings.Contains(err.Error(), "duplicate") {
+		t.Errorf("err = %v, want duplicate-column error", err)
+	}
+}
+
+// TestDataframe_BadBytes rejects a non-Arrow blob with a hint pointing at
+// pandas df.to_feather so users see what producer is expected without having
+// to grep the docs.
+func TestDataframe_BadBytes(t *testing.T) {
+	t.Parallel()
+	_, err := decodeArrowToSheet([]byte("not arrow"), "S")
+	if err == nil || !strings.Contains(err.Error(), "Arrow") {
+		t.Errorf("err = %v, want Arrow-decode error", err)
+	}
+}
+
+// TestDataframe_EncodeRoundTrip checks --dataframe-out's encoder against its
+// own decoder: build a +table-get-shaped sheet map (the same one
+// readSheetAsSpec emits), encode to Arrow IPC, decode back via the put-side
+// decoder, and require the column types / formats / row values to match. If
+// any encoder choice drifts from what the decoder expects, the round-trip
+// breaks here long before a real put → get round-trip in production would.
+func TestDataframe_EncodeRoundTrip(t *testing.T) {
+	t.Parallel()
+	sheet := map[string]interface{}{
+		"name":    "S1",
+		"columns": []interface{}{"name", "qty", "price", "active", "ts"},
+		"dtypes": map[string]interface{}{
+			"name":   "object",
+			"qty":    "float64",
+			"price":  "float64",
+			"active": "bool",
+			"ts":     "datetime64[ns]",
+		},
+		"formats": map[string]interface{}{
+			// `@` is the writer convention for string columns; readSheetAsSpec
+			// strips it via isTextNumberFormat, so an Arrow file built from a
+			// real read won't carry @ either. Keep it absent here to mirror
+			// the production wire shape.
+			"price": "$#,##0.00",
+		},
+		"data": []interface{}{
+			[]interface{}{"alice", json.Number("42"), json.Number("19.95"), true, "2024-01-15"},
+			[]interface{}{"bob", nil, json.Number("8.5"), false, "2024-02-02"},
+		},
+	}
+	blob, err := encodeSheetMapToArrowIPC(sheet)
+	if err != nil {
+		t.Fatalf("encodeSheetMapToArrowIPC: %v", err)
+	}
+	spec, err := decodeArrowToSheet(blob, "S1")
+	if err != nil {
+		t.Fatalf("decodeArrowToSheet: %v", err)
+	}
+	wantTypes := []string{"string", "number", "number", "bool", "date"}
+	wantFormats := []string{"@", "", "$#,##0.00", "", "yyyy-mm-dd"}
+	if len(spec.Columns) != len(wantTypes) {
+		t.Fatalf("got %d columns, want %d", len(spec.Columns), len(wantTypes))
+	}
+	for i, w := range wantTypes {
+		if spec.Columns[i].Type != w {
+			t.Errorf("columns[%d].Type = %q, want %q", i, spec.Columns[i].Type, w)
+		}
+		if spec.Columns[i].Format != wantFormats[i] {
+			t.Errorf("columns[%d].Format = %q, want %q", i, spec.Columns[i].Format, wantFormats[i])
+		}
+	}
+	if len(spec.Rows) != 2 {
+		t.Fatalf("got %d rows, want 2", len(spec.Rows))
+	}
+	if spec.Rows[0][0] != "alice" {
+		t.Errorf("row0[name] = %#v, want alice", spec.Rows[0][0])
+	}
+	if n, ok := spec.Rows[0][1].(json.Number); !ok || n.String() != "42" {
+		t.Errorf("row0[qty] = %#v, want json.Number(\"42\")", spec.Rows[0][1])
+	}
+	if spec.Rows[0][3] != true {
+		t.Errorf("row0[active] = %#v, want true", spec.Rows[0][3])
+	}
+	if spec.Rows[0][4] != "2024-01-15" {
+		t.Errorf("row0[ts] = %#v, want 2024-01-15", spec.Rows[0][4])
+	}
+	// qty is null on row1, must come back as nil (not a zero-valued
+	// json.Number that would later round-trip as 0).
+	if spec.Rows[1][1] != nil {
+		t.Errorf("row1[qty] = %#v, want nil (null arrow cell)", spec.Rows[1][1])
+	}
+}
+
+// TestDataframe_EncodeAcceptsBothRowShapes pins the encoder against the two
+// shapes `sheet["data"]` actually arrives in: `[][]interface{}` from a live
+// readSheetAsSpec call (production), and `[]interface{}` from a JSON
+// unmarshal (round-trip / fixtures). Either must produce non-empty Arrow
+// output — early on the production shape silently fell through the
+// `[]interface{}` type assertion and we shipped a 0-row Arrow blob.
+func TestDataframe_EncodeAcceptsBothRowShapes(t *testing.T) {
+	t.Parallel()
+	base := func(data interface{}) map[string]interface{} {
+		return map[string]interface{}{
+			"name":    "S",
+			"columns": []interface{}{"city"},
+			"dtypes":  map[string]interface{}{"city": "object"},
+			"data":    data,
+		}
+	}
+	for label, data := range map[string]interface{}{
+		"production [][]interface{}": [][]interface{}{{"BJ"}, {"SH"}},
+		"unmarshal []interface{}":    []interface{}{[]interface{}{"BJ"}, []interface{}{"SH"}},
+	} {
+		blob, err := encodeSheetMapToArrowIPC(base(data))
+		if err != nil {
+			t.Errorf("%s: encode: %v", label, err)
+			continue
+		}
+		spec, err := decodeArrowToSheet(blob, "S")
+		if err != nil {
+			t.Errorf("%s: decode: %v", label, err)
+			continue
+		}
+		if len(spec.Rows) != 2 {
+			t.Errorf("%s: got %d rows, want 2", label, len(spec.Rows))
+		}
+	}
+}
+
+// TestDataframe_DtypeToInternalType pins the inverse of typeToDtype so
+// readSheetAsSpec's dtype labels recover the right internal type. Covers the
+// dtype families +table-get emits today plus the safe fallback for unknown
+// labels (string, lossless).
+func TestDataframe_DtypeToInternalType(t *testing.T) {
+	t.Parallel()
+	cases := map[string]string{
+		"float64":         "number",
+		"int64":           "number",
+		"Int64":           "number",
+		"bool":            "bool",
+		"boolean":         "bool",
+		"datetime64[ns]":  "date",
+		"datetime64[ms]":  "date",
+		"object":          "string",
+		"":                "string",
+		"weird-new-dtype": "string",
+	}
+	for in, want := range cases {
+		if got := dtypeToInternalType(in); got != want {
+			t.Errorf("dtypeToInternalType(%q) = %q, want %q", in, got, want)
+		}
+	}
+}
+
+// TestDataframe_BytesWriterSeeker confirms the in-memory WriteSeeker handles
+// the Seek-and-overwrite pattern ipc.NewFileWriter uses to patch the footer
+// offset: write some bytes, seek back to the middle, overwrite, end up with
+// the buffer reflecting the overwritten bytes (not a tail-extended duplicate).
+func TestDataframe_BytesWriterSeeker(t *testing.T) {
+	t.Parallel()
+	var w bytesWriterSeeker
+	if _, err := w.Write([]byte("hello world")); err != nil {
+		t.Fatal(err)
+	}
+	if _, err := w.Seek(6, 0); err != nil {
+		t.Fatal(err)
+	}
+	if _, err := w.Write([]byte("WORLD")); err != nil {
+		t.Fatal(err)
+	}
+	if got := string(w.buf); got != "hello WORLD" {
+		t.Errorf("buf = %q, want \"hello WORLD\"", got)
+	}
+}

shortcuts/sheets/lark_sheet_object_crud.go

@@ -7,6 +7,7 @@ import (
 	"context"
 	"fmt"
 	"path/filepath"
+	"sort"
 	"strings"
 
 	"github.com/larksuite/cli/errs"
@@ -49,6 +50,20 @@ type objectCRUDSpec struct {
 	// right nesting level.
 	enhanceCreateInput func(rt flagView, input map[string]interface{})
 	enhanceUpdateInput func(rt flagView, input map[string]interface{})
+	// validateCreateInput, when set, runs after enhanceCreateInput to
+	// enforce cross-flag / cross-field, create-only constraints JSON
+	// Schema can't express. Two uses today:
+	//   - pivot rejects --target-position vs --range when both carry
+	//     non-default values — they map to the same wire field and
+	//     conflicting values are ambiguous (needs raw flags via rt).
+	//   - cond-format requires every properties.attrs entry to match the
+	//     sibling rule_type's shape (see validateCondFormatAttrs); a
+	//     colorScale rule fed cellIs-shaped attrs writes a color-less
+	//     segment that breaks the sheet on open (inspects input only).
+	// It is the create-path twin of validateUpdateInput; the same scope
+	// notes apply. Validators that only inspect the wire input can ignore
+	// the rt argument.
+	validateCreateInput func(rt flagView, input map[string]interface{}) error
 	// validateUpdateInput, when set, runs after enhanceUpdateInput to
 	// enforce *cross-field, update-only* constraints JSON Schema can't
 	// express (e.g. sparkline requires properties.sparklines[i] to
@@ -140,7 +155,7 @@ func newObjectCreateShortcut(spec objectCRUDSpec) common.Shortcut {
 			return dr
 		},
 		Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-			token, err := resolveSpreadsheetToken(runtime)
+			token, err := resolveSpreadsheetTokenExec(runtime)
 			if err != nil {
 				return err
 			}
@@ -190,6 +205,11 @@ func objectCreateInput(runtime flagView, token, sheetID, sheetName string, spec
 	if spec.enhanceCreateInput != nil {
 		spec.enhanceCreateInput(runtime, input)
 	}
+	if spec.validateCreateInput != nil {
+		if err := spec.validateCreateInput(runtime, input); err != nil {
+			return nil, err
+		}
+	}
 	if err := validateInputAgainstSchema(runtime, input); err != nil {
 		return nil, err
 	}
@@ -224,7 +244,7 @@ func newObjectUpdateShortcut(spec objectCRUDSpec) common.Shortcut {
 			return invokeToolDryRun(token, ToolKindWrite, spec.toolName, input)
 		},
 		Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-			token, err := resolveSpreadsheetToken(runtime)
+			token, err := resolveSpreadsheetTokenExec(runtime)
 			if err != nil {
 				return err
 			}
@@ -308,7 +328,7 @@ func newObjectDeleteShortcut(spec objectCRUDSpec) common.Shortcut {
 			return invokeToolDryRun(token, ToolKindWrite, spec.toolName, input)
 		},
 		Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-			token, err := resolveSpreadsheetToken(runtime)
+			token, err := resolveSpreadsheetTokenExec(runtime)
 			if err != nil {
 				return err
 			}
@@ -381,9 +401,6 @@ var pivotSpec = objectCRUDSpec{
 	},
 	createWarn: pivotPlacementWarn,
 	enhanceCreateInput: func(rt flagView, input map[string]interface{}) {
-		if v := strings.TrimSpace(rt.Str("target-position")); v != "" && v != "A1" {
-			input["target_position"] = v
-		}
 		props, _ := input["properties"].(map[string]interface{})
 		if props == nil {
 			return
@@ -391,10 +408,26 @@ var pivotSpec = objectCRUDSpec{
 		if v := strings.TrimSpace(rt.Str("source")); v != "" {
 			props["source"] = v
 		}
-		if v := strings.TrimSpace(rt.Str("range")); v != "" {
+		// --target-position 与 --range 都映射到 properties.range；
+		// --target-position 优先，未给（或为默认值 A1）时回落到 --range。
+		// 互斥校验在 validateCreateInput 里做。
+		if v := strings.TrimSpace(rt.Str("target-position")); v != "" && v != "A1" {
+			props["range"] = v
+		} else if v := strings.TrimSpace(rt.Str("range")); v != "" {
 			props["range"] = v
 		}
 	},
+	// --target-position 与 --range 落到同一 wire 字段（properties.range），
+	// 同时给非默认值时无法判断意图——按 --target-sheet-id / --target-sheet-name
+	// 的处理方式，CLI 端直接拒绝（优于静默丢弃其一）。
+	validateCreateInput: func(rt flagView, _ map[string]interface{}) error {
+		pos := strings.TrimSpace(rt.Str("target-position"))
+		rng := strings.TrimSpace(rt.Str("range"))
+		if pos != "" && pos != "A1" && rng != "" {
+			return common.ValidationErrorf("--target-position and --range are mutually exclusive (both map to properties.range; pass only one)")
+		}
+		return nil
+	},
 }
 var PivotCreate = newObjectCreateShortcut(pivotSpec)
 var PivotUpdate = newObjectUpdateShortcut(pivotSpec)
@@ -487,7 +520,118 @@ var condFormatSpec = objectCRUDSpec{
 	idField:            "conditional_format_id",
 	enhanceCreateInput: condFormatEnhance,
 	enhanceUpdateInput: condFormatEnhance,
+	// validateCondFormatAttrs only inspects the wire input, so the create
+	// hook ignores rt; the update hook (func(input)) calls it directly.
+	validateCreateInput: func(_ flagView, input map[string]interface{}) error {
+		return validateCondFormatAttrs(input)
+	},
+	validateUpdateInput: validateCondFormatAttrs,
 }
+
+// condFormatAttrsRequired maps each conditional-format rule_type to the
+// keys every properties.attrs entry must carry for that rule. It mirrors
+// the per-rule attrs contract the tool's manage_conditional_format_object
+// converter reads (byted-sheet ai-tools manage-conditional-format-object.ts):
+// that converter maps each attrs entry *blindly by rule_type*, so a
+// colorScale rule fed cellIs-shaped attrs ({compare_type,value}) silently
+// yields a color-less color-scale segment — dirty data that crashes the
+// frontend on snapshot deserialization (the 5005 "can't open" report this
+// validator was added for).
+//
+// JSON Schema can't catch this: properties.attrs.items is a oneOf over all
+// nine shapes, and the validator accepts an entry as soon as *any* branch
+// matches — blind to the sibling rule_type. {compare_type,value} matches
+// the cellIs branch regardless of whether rule_type says colorScale.
+//
+// Rule types absent from the map (duplicateValues, uniqueValues,
+// containsBlanks, notContainsBlanks) carry no attrs, so nothing to check.
+// Counts (dataBar==2, colorScale 2–3, iconSet ordering) stay the tool's
+// job — it already rejects those with actionable messages; the gap this
+// closes is per-entry *shape*, which the tool does not check.
+var condFormatAttrsRequired = map[string][]string{
+	"cellIs":       {"compare_type", "value"},
+	"containsText": {"compare_type", "text"},
+	"timePeriod":   {"operator", "time_period"},
+	"dataBar":      {"color", "value_type"},
+	"colorScale":   {"value_type", "color"},
+	"rank":         {"is_bottom", "value_type"},
+	"aboveAverage": {"operator"},
+	"expression":   {"formula"},
+	"iconSet":      {"icon_type", "value_type", "operator"},
+}
+
+// validateCondFormatAttrs enforces that every properties.attrs entry
+// matches the shape required by the sibling properties.rule_type. Shared
+// by create and update. On update, rule_type may be omitted (the caller is
+// editing style only and the existing rule's type governs the attrs shape,
+// which the CLI can't see); in that case validation is deferred to the
+// server. Missing/empty attrs is likewise left to the tool, which already
+// reports "attrs are required for rule_type: X" clearly.
+func validateCondFormatAttrs(input map[string]interface{}) error {
+	props, _ := input["properties"].(map[string]interface{})
+	if props == nil {
+		return nil
+	}
+	ruleType, _ := props["rule_type"].(string)
+	ruleType = strings.TrimSpace(ruleType)
+	if ruleType == "" {
+		return nil
+	}
+	required, ok := condFormatAttrsRequired[ruleType]
+	if !ok {
+		return nil
+	}
+	attrs, ok := props["attrs"].([]interface{})
+	if !ok {
+		// Missing attrs, or a non-array shape the schema check already
+		// flagged — nothing for this cross-field rule to add.
+		return nil
+	}
+	for i, entryRaw := range attrs {
+		entry, ok := entryRaw.(map[string]interface{})
+		if !ok {
+			continue // schema validation owns per-entry type errors.
+		}
+		for _, key := range required {
+			if v, has := entry[key]; !has || condAttrIsBlank(v) {
+				return common.ValidationErrorf(
+					"--properties: attrs[%d] is missing %q, which rule_type %q requires on every entry (expected keys %s; got %s). "+
+						"A common cause is reusing another rule's attrs shape — e.g. cellIs-style {compare_type,value} under a colorScale rule, which writes a color-less segment that breaks the sheet on open.",
+					i, key, ruleType, strings.Join(required, "+"), condAttrPresentKeys(entry))
+			}
+		}
+	}
+	return nil
+}
+
+// condAttrIsBlank treats a present-but-empty string (after trimming) as
+// missing. The crash-causing case is an empty `color`, but an empty value
+// for any required key is never meaningful in these branches, so the rule
+// is uniform. Non-string values (numbers, booleans) count as present.
+func condAttrIsBlank(v interface{}) bool {
+	if v == nil {
+		return true
+	}
+	if s, ok := v.(string); ok {
+		return strings.TrimSpace(s) == ""
+	}
+	return false
+}
+
+// condAttrPresentKeys lists the keys actually present on an attrs entry,
+// sorted, for the "got ..." half of the error message.
+func condAttrPresentKeys(entry map[string]interface{}) string {
+	if len(entry) == 0 {
+		return "{}"
+	}
+	keys := make([]string, 0, len(entry))
+	for k := range entry {
+		keys = append(keys, k)
+	}
+	sort.Strings(keys)
+	return "{" + strings.Join(keys, ",") + "}"
+}
+
 var CondFormatCreate = newObjectCreateShortcut(condFormatSpec)
 var CondFormatUpdate = newObjectUpdateShortcut(condFormatSpec)
 var CondFormatDelete = newObjectDeleteShortcut(condFormatSpec)
@@ -732,7 +876,7 @@ func newFloatImageWriteShortcut(command, description, op string, withIDFlag, isH
 			return invokeToolDryRun(token, ToolKindWrite, "manage_float_image_object", input)
 		},
 		Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-			token, err := resolveSpreadsheetToken(runtime)
+			token, err := resolveSpreadsheetTokenExec(runtime)
 			if err != nil {
 				return err
 			}
@@ -882,7 +1026,7 @@ var FilterCreate = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindWrite, "manage_filter_object", input)
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}
@@ -957,7 +1101,7 @@ var FilterUpdate = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindWrite, "manage_filter_object", input)
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}
@@ -1025,7 +1169,7 @@ var FilterDelete = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindWrite, "manage_filter_object", input)
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}

shortcuts/sheets/lark_sheet_object_crud_test.go

@@ -4,6 +4,8 @@
 package sheets
 
 import (
+	"encoding/json"
+	"sort"
 	"strings"
 	"testing"
 
@@ -137,25 +139,24 @@ func TestObjectCRUDShortcuts_DryRun(t *testing.T) {
 		// covered separately in the +pivot-create empty-selector / mutex
 		// tests below.
 		{
-			name: "+pivot-create with placement / source / range flags",
+			name: "+pivot-create with placement / source / target-position flags",
 			sc:   PivotCreate,
 			args: []string{
 				"--url", testURL, "--target-sheet-id", testSheetID,
 				"--properties", `{"rows":[{"field":"A"}]}`,
 				"--source", "Sheet1!A1:F1000",
-				"--range", "F1",
 				"--target-position", "B5",
 			},
 			toolName: "manage_pivot_table_object",
 			wantInput: map[string]interface{}{
-				"excel_id":        testToken,
-				"sheet_id":        testSheetID,
-				"operation":       "create",
-				"target_position": "B5",
+				"excel_id":  testToken,
+				"sheet_id":  testSheetID,
+				"operation": "create",
 				"properties": map[string]interface{}{
 					"rows":   []interface{}{map[string]interface{}{"field": "A"}},
 					"source": "Sheet1!A1:F1000",
-					"range":  "F1",
+					// --target-position 映射到 properties.range。
+					"range": "B5",
 				},
 			},
 		},
@@ -202,7 +203,7 @@ func TestObjectCRUDShortcuts_DryRun(t *testing.T) {
 			args: []string{
 				"--url", testURL, "--sheet-id", testSheetID,
 				"--rule-id", "ruleA",
-				"--properties", `{"attrs":[{"operator":"greaterThan","value":"100"}],"style":{"back_color":"#FFD7D7"}}`,
+				"--properties", `{"attrs":[{"compare_type":"greaterThan","value":"100"}],"style":{"back_color":"#FFD7D7"}}`,
 				"--rule-type", "cellIs",
 				"--ranges", `["A1:A100"]`,
 			},
@@ -214,7 +215,7 @@ func TestObjectCRUDShortcuts_DryRun(t *testing.T) {
 				"conditional_format_id": "ruleA",
 				"properties": map[string]interface{}{
 					"rule_type": "cellIs",
-					"attrs":     []interface{}{map[string]interface{}{"operator": "greaterThan", "value": "100"}},
+					"attrs":     []interface{}{map[string]interface{}{"compare_type": "greaterThan", "value": "100"}},
 					"style":     map[string]interface{}{"back_color": "#FFD7D7"},
 					"ranges":    []interface{}{"A1:A100"},
 				},
@@ -507,6 +508,55 @@ func TestPivotCreate_SheetSelectorSemantics(t *testing.T) {
 	})
 }
 
+// TestPivotCreate_TargetPositionRangeMutex regresses the "--target-position
+// and --range cannot both be set" guardrail on +pivot-create. They map to
+// the same wire field (properties.range), so two non-default values are
+// ambiguous; the CLI rejects up front (mirrors the --target-sheet-id /
+// --target-sheet-name mutex). --target-position=A1 is the documented default
+// and is treated as "not set" — pairing it with --range still works.
+func TestPivotCreate_TargetPositionRangeMutex(t *testing.T) {
+	t.Parallel()
+
+	t.Run("both non-default values rejected", func(t *testing.T) {
+		t.Parallel()
+		_, stderr, err := runShortcutCapturingErr(t, PivotCreate, []string{
+			"--url", testURL,
+			"--target-sheet-id", testSheetID,
+			"--properties", `{"rows":[{"field":"A"}]}`,
+			"--source", "Sheet1!A1:F1000",
+			"--target-position", "B5",
+			"--range", "F1",
+		})
+		if err == nil {
+			t.Fatalf("expected CLI to reject --target-position with --range; stderr=%s", stderr)
+		}
+		combined := stderr + err.Error()
+		if !strings.Contains(combined, "mutually exclusive") {
+			t.Errorf("expected error to say 'mutually exclusive'; got=%s|%v", stderr, err)
+		}
+		if !strings.Contains(combined, "--target-position") || !strings.Contains(combined, "--range") {
+			t.Errorf("expected error to quote both --target-position and --range; got=%s|%v", stderr, err)
+		}
+	})
+
+	t.Run("default A1 with --range is accepted (range wins)", func(t *testing.T) {
+		t.Parallel()
+		body := parseDryRunBody(t, PivotCreate, []string{
+			"--url", testURL,
+			"--target-sheet-id", testSheetID,
+			"--properties", `{"rows":[{"field":"A"}]}`,
+			"--source", "Sheet1!A1:F1000",
+			"--target-position", "A1",
+			"--range", "F1",
+		})
+		input := decodeToolInput(t, body, "manage_pivot_table_object")
+		props, _ := input["properties"].(map[string]interface{})
+		if got, _ := props["range"].(string); got != "F1" {
+			t.Errorf("properties.range = %q, want %q", got, "F1")
+		}
+	})
+}
+
 // TestPivotCreate_SchemaValidates exercises the schema-driven
 // validator wired into objectCreateInput. The pivot create schema
 // doesn't constrain rows/columns/values to be present (the backend
@@ -614,6 +664,182 @@ func TestSparklineUpdate_MissingSparklineID(t *testing.T) {
 	}
 }
 
+// TestCondFormatAttrs_ShapeMatchesRuleType regresses the cross-field
+// guard that rejects attrs whose shape doesn't match the sibling
+// rule_type — the gap behind the "缺 color 的 colorScale 脏数据导致表格
+// 打不开" report: a colorScale rule fed cellIs-shaped attrs
+// ({compare_type,value}, no color) passed both the CLI's per-entry oneOf
+// schema check and the tool, writing a color-less segment that crashed
+// the frontend on open. The check covers create and update symmetrically.
+func TestCondFormatAttrs_ShapeMatchesRuleType(t *testing.T) {
+	t.Parallel()
+	cases := []struct {
+		name    string
+		sc      common.Shortcut
+		args    []string
+		wantErr bool
+		wantMsg string // substring expected in the error, when wantErr
+	}{
+		{
+			name: "colorScale fed cellIs-shaped attrs (missing color) is rejected",
+			sc:   CondFormatCreate,
+			args: []string{
+				"--url", testURL, "--sheet-id", testSheetID,
+				"--rule-type", "colorScale", "--ranges", `["C1:C10"]`,
+				"--properties", `{"style":{},"attrs":[{"compare_type":"greaterThan","value":"0"},{"compare_type":"lessThan","value":"100"}]}`, "--dry-run",
+			},
+			wantErr: true,
+			wantMsg: "colorScale",
+		},
+		{
+			name: "colorScale with empty color string is rejected",
+			sc:   CondFormatCreate,
+			args: []string{
+				"--url", testURL, "--sheet-id", testSheetID,
+				"--rule-type", "colorScale", "--ranges", `["C1:C10"]`,
+				"--properties", `{"style":{},"attrs":[{"value_type":"minValue","color":""},{"value_type":"maxValue","color":"#FF0000"}]}`, "--dry-run",
+			},
+			wantErr: true,
+			wantMsg: `"color"`,
+		},
+		{
+			name: "well-formed colorScale attrs pass",
+			sc:   CondFormatCreate,
+			args: []string{
+				"--url", testURL, "--sheet-id", testSheetID,
+				"--rule-type", "colorScale", "--ranges", `["C1:C10"]`,
+				"--properties", `{"style":{},"attrs":[{"value_type":"minValue","color":"#FFFFFF"},{"value_type":"maxValue","color":"#FF0000"}]}`, "--dry-run",
+			},
+			wantErr: false,
+		},
+		{
+			name: "update path is guarded too (colorScale + cellIs attrs)",
+			sc:   CondFormatUpdate,
+			args: []string{
+				"--url", testURL, "--sheet-id", testSheetID, "--rule-id", "ruleA",
+				"--rule-type", "colorScale", "--ranges", `["C1:C10"]`,
+				"--properties", `{"style":{},"attrs":[{"compare_type":"greaterThan","value":"0"}]}`, "--dry-run",
+			},
+			wantErr: true,
+			wantMsg: "colorScale",
+		},
+	}
+	for _, tt := range cases {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+			_, stderr, err := runShortcutCapturingErr(t, tt.sc, tt.args)
+			if tt.wantErr {
+				if err == nil {
+					t.Fatalf("expected rejection; stderr=%s", stderr)
+				}
+				if combined := stderr + err.Error(); tt.wantMsg != "" && !strings.Contains(combined, tt.wantMsg) {
+					t.Errorf("expected error to mention %q; got=%s|%v", tt.wantMsg, stderr, err)
+				}
+				return
+			}
+			if err != nil {
+				t.Fatalf("expected acceptance (dry-run); got err=%v stderr=%s", err, stderr)
+			}
+		})
+	}
+}
+
+// TestCondFormatAttrsRequired_MatchesSchemaOneOf guards against drift
+// between the hand-maintained condFormatAttrsRequired table (the source
+// validateCondFormatAttrs enforces) and the embedded flag-schemas.json
+// attrs oneOf (the authoritative shape contract synced from the spec
+// repo). The cross-field validator only works if its per-rule_type
+// required keys mirror the schema branches; if a future schema sync adds
+// or drops a required key on any branch without updating the table, the
+// CLI would silently under- or over-validate. They share no compile-time
+// link, so this test is the only thing pinning them together.
+//
+// The schema oneOf branches are NOT labeled by rule_type (that's the whole
+// point — rule_type is a sibling field the per-entry oneOf can't see), so
+// we can't match branch→rule_type. We instead compare the *multiset* of
+// required-key sets: every branch's required array must appear as some
+// table entry's value and vice versa. This catches any added/dropped
+// required key (real drift); it tolerates only a relabeling between two
+// branches that happen to share an identical required set (dataBar and
+// colorScale both require {color,value_type}), which is harmless here.
+func TestCondFormatAttrsRequired_MatchesSchemaOneOf(t *testing.T) {
+	t.Parallel()
+
+	// multiset key: required keys sorted + joined, so order within a
+	// branch's required array doesn't matter.
+	keyOf := func(req []string) string {
+		s := append([]string(nil), req...)
+		sort.Strings(s)
+		return strings.Join(s, "+")
+	}
+
+	tableMS := map[string]int{}
+	for _, req := range condFormatAttrsRequired {
+		tableMS[keyOf(req)]++
+	}
+
+	schemaMS := func(t *testing.T, command string) map[string]int {
+		idx, err := loadFlagSchemas()
+		if err != nil {
+			t.Fatalf("loadFlagSchemas: %v", err)
+		}
+		raw, ok := idx.Flags[command]["properties"]
+		if !ok {
+			t.Fatalf("no embedded schema for %s --properties", command)
+		}
+		var schema map[string]interface{}
+		if err := json.Unmarshal(raw, &schema); err != nil {
+			t.Fatalf("unmarshal %s properties schema: %v", command, err)
+		}
+		dig := func(m map[string]interface{}, key string) map[string]interface{} {
+			next, _ := m[key].(map[string]interface{})
+			if next == nil {
+				t.Fatalf("%s: missing %q while navigating to attrs oneOf", command, key)
+			}
+			return next
+		}
+		attrs := dig(dig(schema, "properties"), "attrs")
+		items := dig(attrs, "items")
+		oneOf, ok := items["oneOf"].([]interface{})
+		if !ok || len(oneOf) == 0 {
+			t.Fatalf("%s: attrs.items.oneOf is missing or empty", command)
+		}
+		ms := map[string]int{}
+		for i, branchRaw := range oneOf {
+			branch, ok := branchRaw.(map[string]interface{})
+			if !ok {
+				t.Fatalf("%s: oneOf[%d] is not an object", command, i)
+			}
+			reqRaw, _ := branch["required"].([]interface{})
+			req := make([]string, 0, len(reqRaw))
+			for _, r := range reqRaw {
+				if s, ok := r.(string); ok {
+					req = append(req, s)
+				}
+			}
+			ms[keyOf(req)]++
+		}
+		return ms
+	}
+
+	for _, command := range []string{"+cond-format-create", "+cond-format-update"} {
+		got := schemaMS(t, command)
+		if len(got) != len(tableMS) {
+			t.Errorf("%s: schema oneOf has %d distinct required-sets, table has %d", command, len(got), len(tableMS))
+		}
+		for k, n := range tableMS {
+			if got[k] != n {
+				t.Errorf("%s: required-set %q appears %d× in schema but %d× in condFormatAttrsRequired — table drifted from schema; re-sync the table", command, k, got[k], n)
+			}
+		}
+		for k, n := range got {
+			if tableMS[k] != n {
+				t.Errorf("%s: schema branch with required-set %q (×%d) has no matching condFormatAttrsRequired entry — add it to the table", command, k, n)
+			}
+		}
+	}
+}
+
 // Note: +float-image-update's image_name / position / size are cobra-required
 // (flag-defs.json), so the standalone path is gated by the flag layer — its
 // "required flag(s) … not set" wording is framework-owned and intentionally not

shortcuts/sheets/lark_sheet_object_list.go

@@ -57,7 +57,7 @@ func newObjectListShortcut(spec objectListSpec) common.Shortcut {
 			return invokeToolDryRun(token, ToolKindRead, spec.toolName, objectListInput(runtime, token, sheetID, sheetName, spec))
 		},
 		Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-			token, err := resolveSpreadsheetToken(runtime)
+			token, err := resolveSpreadsheetTokenExec(runtime)
 			if err != nil {
 				return err
 			}

shortcuts/sheets/lark_sheet_range_operations.go

@@ -45,7 +45,7 @@ var CellsClear = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindWrite, "clear_cell_range", input)
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}
@@ -163,7 +163,7 @@ func newMergeShortcut(command, desc, op string, withMergeType bool) common.Short
 			return invokeToolDryRun(token, ToolKindWrite, "merge_cells", input)
 		},
 		Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-			token, err := resolveSpreadsheetToken(runtime)
+			token, err := resolveSpreadsheetTokenExec(runtime)
 			if err != nil {
 				return err
 			}
@@ -239,7 +239,7 @@ var RowsResize = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindWrite, "resize_range", input)
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}
@@ -279,7 +279,7 @@ var ColsResize = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindWrite, "resize_range", input)
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}
@@ -451,7 +451,7 @@ var RangeFill = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindWrite, "transform_range", input)
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}
@@ -490,7 +490,7 @@ var RangeSort = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindWrite, "transform_range", input)
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}

shortcuts/sheets/lark_sheet_read_data.go

@@ -5,8 +5,6 @@ package sheets
 
 import (
 	"context"
-	"encoding/csv"
-	"regexp"
 	"strconv"
 	"strings"
 
@@ -59,7 +57,7 @@ var CellsGet = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindRead, "get_cell_ranges", cellsGetInput(runtime, token, sheetID, sheetName))
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}
@@ -152,7 +150,7 @@ var CsvGet = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindRead, "get_range_as_csv", csvGetInput(runtime, token, sheetID, sheetName))
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}
@@ -164,12 +162,7 @@ var CsvGet = common.Shortcut{
 		if err != nil {
 			return err
 		}
-		switch {
-		case runtime.Bool("rows-json"):
-			// --rows-json reshapes the CSV response into structured rows
-			// ({row_number, values:{col→cell}}); see assembleRowsJSON.
-			out = assembleRowsJSON(out, strings.TrimSpace(runtime.Str("range")))
-		case !runtime.Bool("include-row-prefix"):
+		if !runtime.Bool("include-row-prefix") {
 			out = stripRowPrefixFromCsvOutput(out)
 		}
 		runtime.Out(out, nil)
@@ -219,141 +212,6 @@ func stripRowPrefixFromCsvOutput(out interface{}) interface{} {
 	return m
 }
 
-// rowPrefixRe matches the leading "[row=N] " (or "[row=N],") annotation that
-// the tool prepends to the first physical line of each logical CSV record.
-var rowPrefixRe = regexp.MustCompile(`^\[row=(\d+)\][ ,]?`)
-
-// assembleRowsJSON reshapes the tool's annotated_csv string into structured
-// rows so callers never have to regex-parse "[row=N]" or RFC-4180 CSV by hand:
-//
-//	{
-//	  "range": "A1:K3380",
-//	  "current_region": "...",        // passthrough, if the tool returned it
-//	  "rows": [{"row_number":1,"values":{"A":"姓名", ..., "K":"时间差_分钟"}},
-//	           {"row_number":2,"values":{"A":"张三", ..., "K":"8.5"}}, ...]
-//	}
-//
-// Every logical row is emitted, including the first — no row is assumed to be a
-// header, since sheet data is not always tabular. Each cell is keyed by its
-// column letter (from the tool's col_indices when present, else derived from the
-// requested range's start column). On any parsing trouble it returns the
-// original output unchanged.
-func assembleRowsJSON(out interface{}, requestedRange string) interface{} {
-	m, ok := out.(map[string]interface{})
-	if !ok {
-		return out
-	}
-	csvStr, ok := m["annotated_csv"].(string)
-	if !ok {
-		return out
-	}
-
-	// Group physical lines into logical records by [row=N] boundaries; lines
-	// without a prefix are embedded-newline continuations of the current record.
-	type logicalRow struct {
-		num  int
-		text string
-	}
-	var groups []logicalRow
-	for _, line := range strings.Split(csvStr, "\n") {
-		if mm := rowPrefixRe.FindStringSubmatch(line); mm != nil {
-			n, _ := strconv.Atoi(mm[1])
-			groups = append(groups, logicalRow{num: n, text: line[len(mm[0]):]})
-		} else if len(groups) > 0 {
-			groups[len(groups)-1].text += "\n" + line
-		}
-	}
-	if len(groups) == 0 {
-		return out
-	}
-
-	// Parse every logical row; widest row sets the column count. No row is
-	// singled out as a header — that would assume the data is tabular, which it
-	// often is not. The model reads row 1 like any other row and decides for
-	// itself whether it is a header.
-	parsed := make([][]string, len(groups))
-	maxCols := 0
-	for i, g := range groups {
-		parsed[i] = parseCSVRecord(g.text)
-		if len(parsed[i]) > maxCols {
-			maxCols = len(parsed[i])
-		}
-	}
-	if maxCols == 0 {
-		return out
-	}
-
-	// Column letters key each cell. Prefer the tool's col_indices (authoritative,
-	// length == col_count); otherwise derive from the requested range's start col.
-	letters := coerceStringSlice(m["col_indices"])
-	if len(letters) < maxCols {
-		start := csvStartColIndex(requestedRange)
-		letters = make([]string, maxCols)
-		for j := 0; j < maxCols; j++ {
-			letters[j] = csvColLetter(start + j)
-		}
-	}
-
-	rows := make([]map[string]interface{}, 0, len(groups))
-	for i := range groups {
-		fields := parsed[i]
-		values := make(map[string]interface{}, len(letters))
-		for j := range letters {
-			v := ""
-			if j < len(fields) {
-				v = fields[j]
-			}
-			values[letters[j]] = v
-		}
-		rows = append(rows, map[string]interface{}{
-			"row_number": groups[i].num,
-			"values":     values,
-		})
-	}
-
-	result := map[string]interface{}{}
-	for k, v := range m {
-		result[k] = v
-	}
-	result["range"] = requestedRange
-	result["rows"] = rows
-
-	// Surface the backend's "数据没读全" signal structurally instead of leaving it
-	// buried in warning_message prose. The tool flags it when current_region (the
-	// true data extent) reaches past actual_range (what was actually read) — the
-	// single most important anti-under-read hint. Mirror that same comparison
-	// (regionEndRow > actualEndRow) from the already-passthrough A1 ranges so the
-	// model gets the real data range as a first-class field, never having to
-	// parse it out of prose.
-	if cr, _ := m["current_region"].(string); cr != "" {
-		ar, _ := m["actual_range"].(string)
-		regionEnd := a1EndRow(cr)
-		readEnd := a1EndRow(ar)
-		if regionEnd > 0 && readEnd > 0 && regionEnd > readEnd {
-			result["data_not_fully_read"] = map[string]interface{}{
-				"read_through_row":         readEnd,
-				"data_extends_through_row": regionEnd,
-				"unread_rows":              regionEnd - readEnd,
-				"reread_range":             cr,
-			}
-		}
-	}
-
-	// Drop the fields whose information rows-json fully carries elsewhere:
-	//   - annotated_csv / row_indices / col_indices → reconstructed into
-	//     columns + rows (with integer row_number), losslessly.
-	//   - warning_message → its two halves are both handled: the static
-	//     "[row=N] / col_indices[j]" parse nag is moot once those fields exist,
-	//     and the dynamic "数据没读全" half is now the structured
-	//     data_not_fully_read field above. (Confirmed against the backend's
-	//     get-range-as-csv.ts — warning_message has no other content.)
-	delete(result, "annotated_csv")
-	delete(result, "row_indices")
-	delete(result, "col_indices")
-	delete(result, "warning_message")
-	return result
-}
-
 // a1EndRow extracts the ending row number from an A1 range, e.g. "A1:N51" → 51,
 // "Sheet1!B2:D9" → 9, "C5" → 5. Returns 0 when no row number is present.
 func a1EndRow(rng string) int {
@@ -377,89 +235,6 @@ func a1EndRow(rng string) int {
 	return n
 }
 
-// parseCSVRecord parses a single logical CSV record (which may span multiple
-// physical lines via quoted embedded newlines) into its fields. An empty record
-// yields no fields; a malformed record falls back to a naive comma split so a
-// stray quote never drops a whole row.
-func parseCSVRecord(text string) []string {
-	if strings.TrimSpace(text) == "" {
-		return nil
-	}
-	r := csv.NewReader(strings.NewReader(text))
-	r.FieldsPerRecord = -1
-	fields, err := r.Read()
-	if err != nil {
-		return strings.Split(text, ",")
-	}
-	return fields
-}
-
-// coerceStringSlice returns v as []string when it is a homogeneous []interface{}
-// of strings (the shape of the tool's col_indices), else nil.
-func coerceStringSlice(v interface{}) []string {
-	arr, ok := v.([]interface{})
-	if !ok {
-		return nil
-	}
-	out := make([]string, 0, len(arr))
-	for _, e := range arr {
-		s, ok := e.(string)
-		if !ok {
-			return nil
-		}
-		out = append(out, s)
-	}
-	return out
-}
-
-// csvStartColIndex returns the 0-based column index of a range's start column,
-// e.g. "A1:K3380" → 0, "C5:F9" → 2, "Sheet1!D2" → 3. Unparseable input → 0.
-func csvStartColIndex(rng string) int {
-	rng = strings.TrimSpace(rng)
-	if i := strings.LastIndex(rng, "!"); i >= 0 {
-		rng = rng[i+1:]
-	}
-	var letters strings.Builder
-	for _, c := range rng {
-		if (c >= 'A' && c <= 'Z') || (c >= 'a' && c <= 'z') {
-			letters.WriteRune(c)
-			continue
-		}
-		break
-	}
-	if letters.Len() == 0 {
-		return 0
-	}
-	return csvColToIndex(letters.String())
-}
-
-// csvColToIndex converts a column letter to its 0-based index ("A"→0, "K"→10,
-// "AA"→26). Non-letter input → -1.
-func csvColToIndex(s string) int {
-	n := 0
-	for _, c := range strings.ToUpper(s) {
-		if c < 'A' || c > 'Z' {
-			break
-		}
-		n = n*26 + int(c-'A'+1)
-	}
-	return n - 1
-}
-
-// csvColLetter converts a 0-based column index back to its letter (0→"A",
-// 25→"Z", 26→"AA"). Negative input → "".
-func csvColLetter(idx int) string {
-	if idx < 0 {
-		return ""
-	}
-	var b []byte
-	for idx >= 0 {
-		b = append([]byte{byte('A' + idx%26)}, b...)
-		idx = idx/26 - 1
-	}
-	return string(b)
-}
-
 // DropdownGet wraps get_cell_ranges scoped to data_validation: read the
 // dropdown configuration on a range. Aligned with its sibling +cells-get
 // — sheet selection is via --sheet-id / --sheet-name (XOR), and --range
@@ -494,7 +269,7 @@ var DropdownGet = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindRead, "get_cell_ranges", dropdownGetInput(runtime, token, sheetID, sheetName))
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}

shortcuts/sheets/lark_sheet_read_data_test.go

@@ -63,20 +63,6 @@ func TestReadDataShortcuts_DryRun(t *testing.T) {
 				"value_render_option": "formatted_value",
 			},
 		},
-		{
-			// --rows-json is post-processing on +csv-get's response; it must
-			// NOT leak into the get_range_as_csv input.
-			name:     "+csv-get --rows-json builds the same input (flag is post-process)",
-			sc:       CsvGet,
-			args:     []string{"--url", testURL, "--sheet-id", testSheetID, "--range", "A1:C10", "--rows-json"},
-			toolName: "get_range_as_csv",
-			wantInput: map[string]interface{}{
-				"excel_id": testToken,
-				"sheet_id": testSheetID,
-				"range":    "A1:C10",
-				"max_rows": float64(unboundedReadLimit),
-			},
-		},
 	}
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
@@ -179,113 +165,3 @@ func TestCsvGet_StripRowPrefix(t *testing.T) {
 		t.Errorf("other field corrupted: %v", out["other"])
 	}
 }
-
-// TestAssembleRowsJSON covers the --rows-json reshaping: every logical row
-// emitted (no header singled out), integer row_number, column-letter keyed
-// values, embedded newlines inside quoted fields, and current_region passthrough.
-func TestAssembleRowsJSON(t *testing.T) {
-	t.Parallel()
-	in := map[string]interface{}{
-		"annotated_csv":   "[row=1] 姓名,备注,时间差_分钟\n[row=2] 张三,\"line1\nline2\",8.5\n[row=3] 李四,ok,3",
-		"current_region":  "A1:C3",
-		"col_indices":     []interface{}{"A", "B", "C"},
-		"row_indices":     []interface{}{1, 2, 3},
-		"warning_message": "①定位行号…②定位列字母…",
-	}
-	out, ok := assembleRowsJSON(in, "A1:C3").(map[string]interface{})
-	if !ok {
-		t.Fatalf("assembleRowsJSON did not return a map")
-	}
-
-	// Fields whose info rows-json carries elsewhere are dropped (annotated_csv /
-	// indices → rows; warning_message → moot static nag + structured
-	// data_not_fully_read). Unrelated metadata like current_region is preserved.
-	if _, exists := out["annotated_csv"]; exists {
-		t.Errorf("annotated_csv should be dropped")
-	}
-	if _, exists := out["col_indices"]; exists {
-		t.Errorf("col_indices should be dropped")
-	}
-	if _, exists := out["warning_message"]; exists {
-		t.Errorf("warning_message should be dropped in rows-json mode")
-	}
-	if _, exists := out["columns"]; exists {
-		t.Errorf("columns field should not exist (no header assumption)")
-	}
-	if out["current_region"] != "A1:C3" {
-		t.Errorf("current_region passthrough lost: %v", out["current_region"])
-	}
-
-	rows, _ := out["rows"].([]map[string]interface{})
-	if len(rows) != 3 {
-		t.Fatalf("want all 3 rows (incl. row 1), got %d: %+v", len(rows), rows)
-	}
-	// Row 1 is emitted as a normal row, not consumed as a header.
-	if rows[0]["row_number"].(int) != 1 {
-		t.Errorf("first row_number = %v, want 1", rows[0]["row_number"])
-	}
-	if v := rows[0]["values"].(map[string]interface{}); v["A"] != "姓名" || v["C"] != "时间差_分钟" {
-		t.Errorf("row 1 values wrong: %+v", v)
-	}
-	// Row 2 keeps its embedded newline inside a single cell.
-	v1 := rows[1]["values"].(map[string]interface{})
-	if rows[1]["row_number"].(int) != 2 || v1["A"] != "张三" || v1["B"] != "line1\nline2" || v1["C"] != "8.5" {
-		t.Errorf("row 2 wrong (embedded newline?): %+v", rows[1])
-	}
-}
-
-// TestAssembleRowsJSON_DerivedLetters verifies cell letters are derived from the
-// range start when the tool omits col_indices (e.g. a C-anchored read).
-func TestAssembleRowsJSON_DerivedLetters(t *testing.T) {
-	t.Parallel()
-	in := map[string]interface{}{
-		"annotated_csv": "[row=5] h1,h2\n[row=6] a,b",
-	}
-	out := assembleRowsJSON(in, "C5:D6").(map[string]interface{})
-	rows := out["rows"].([]map[string]interface{})
-	if len(rows) != 2 {
-		t.Fatalf("want 2 rows, got %d", len(rows))
-	}
-	if rows[0]["row_number"].(int) != 5 {
-		t.Errorf("first row_number = %v, want 5", rows[0]["row_number"])
-	}
-	if v := rows[0]["values"].(map[string]interface{}); v["C"] != "h1" || v["D"] != "h2" {
-		t.Errorf("derived-letter values wrong: %+v", v)
-	}
-	if v := rows[1]["values"].(map[string]interface{}); v["C"] != "a" || v["D"] != "b" {
-		t.Errorf("row 6 values wrong: %+v", v)
-	}
-}
-
-// TestAssembleRowsJSON_DataNotFullyRead verifies the structured under-read hint:
-// when current_region extends past actual_range, rows-json surfaces the true data
-// range as a first-class field (mirroring the backend's prose warning).
-func TestAssembleRowsJSON_DataNotFullyRead(t *testing.T) {
-	t.Parallel()
-	// Read only A1:D2, but the data region reaches D4 → 2 rows unread.
-	in := map[string]interface{}{
-		"annotated_csv":  "[row=1] 序号,姓名\n[row=2] 101,张三",
-		"actual_range":   "A1:D2",
-		"current_region": "A1:D4",
-	}
-	out := assembleRowsJSON(in, "A1:D2").(map[string]interface{})
-	hint, ok := out["data_not_fully_read"].(map[string]interface{})
-	if !ok {
-		t.Fatalf("data_not_fully_read missing; out=%+v", out)
-	}
-	if hint["read_through_row"] != 2 || hint["data_extends_through_row"] != 4 ||
-		hint["unread_rows"] != 2 || hint["reread_range"] != "A1:D4" {
-		t.Errorf("data_not_fully_read wrong: %+v", hint)
-	}
-
-	// Fully-read case: no hint emitted.
-	in2 := map[string]interface{}{
-		"annotated_csv":  "[row=1] 序号,姓名\n[row=2] 101,张三",
-		"actual_range":   "A1:D2",
-		"current_region": "A1:D2",
-	}
-	out2 := assembleRowsJSON(in2, "A1:D2").(map[string]interface{})
-	if _, exists := out2["data_not_fully_read"]; exists {
-		t.Errorf("data_not_fully_read should be absent when fully read")
-	}
-}

shortcuts/sheets/lark_sheet_search_replace.go

@@ -46,7 +46,7 @@ var CellsSearch = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindRead, "search_data", searchInput(runtime, token, sheetID, sheetName))
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}
@@ -122,7 +122,7 @@ var CellsReplace = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindWrite, "replace_data", input)
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}

shortcuts/sheets/lark_sheet_sheet_structure.go

@@ -51,7 +51,7 @@ var SheetInfo = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindRead, "get_sheet_structure", sheetInfoInput(runtime, token, sheetID, sheetName))
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}
@@ -136,7 +136,7 @@ var DimInsert = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindWrite, "modify_sheet_structure", input)
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}
@@ -211,7 +211,7 @@ var DimDelete = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindWrite, "modify_sheet_structure", input)
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}
@@ -300,7 +300,7 @@ var DimFreeze = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindWrite, "modify_sheet_structure", input)
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}
@@ -395,7 +395,7 @@ func newDimRangeOpShortcut(command, desc, op, risk string) common.Shortcut {
 			return invokeToolDryRun(token, ToolKindWrite, "modify_sheet_structure", input)
 		},
 		Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-			token, err := resolveSpreadsheetToken(runtime)
+			token, err := resolveSpreadsheetTokenExec(runtime)
 			if err != nil {
 				return err
 			}
@@ -439,7 +439,7 @@ func newDimGroupShortcut(command, desc, op string) common.Shortcut {
 			return invokeToolDryRun(token, ToolKindWrite, "modify_sheet_structure", input)
 		},
 		Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-			token, err := resolveSpreadsheetToken(runtime)
+			token, err := resolveSpreadsheetTokenExec(runtime)
 			if err != nil {
 				return err
 			}
@@ -593,7 +593,7 @@ var DimMove = common.Shortcut{
 			Set("spreadsheet_token", token)
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}

shortcuts/sheets/lark_sheet_style_alias_test.go

@@ -0,0 +1,199 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package sheets
+
+import (
+	"encoding/json"
+	"strings"
+	"testing"
+)
+
+// TestNormalizeCellStyleAliases pins the shorthand → canonical renaming for a
+// single cell_styles map: the alignment shorthands models commonly hallucinate
+// are rewritten in place, values are preserved, and a shorthand colliding with
+// its canonical key is a hard error rather than a silent pick.
+func TestNormalizeCellStyleAliases(t *testing.T) {
+	t.Parallel()
+
+	t.Run("renames *_align shorthands, keeps values and other fields", func(t *testing.T) {
+		t.Parallel()
+		style := map[string]interface{}{
+			"horizontal_align": "center",
+			"vertical_align":   "middle",
+			"font_weight":      "bold",
+		}
+		if err := normalizeCellStyleAliases(style, "x"); err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		if style["horizontal_alignment"] != "center" || style["vertical_alignment"] != "middle" {
+			t.Errorf("alignment not renamed: %#v", style)
+		}
+		if _, ok := style["horizontal_align"]; ok {
+			t.Errorf("shorthand horizontal_align should be removed: %#v", style)
+		}
+		if _, ok := style["vertical_align"]; ok {
+			t.Errorf("shorthand vertical_align should be removed: %#v", style)
+		}
+		if style["font_weight"] != "bold" {
+			t.Errorf("unrelated field font_weight dropped: %#v", style)
+		}
+	})
+
+	t.Run("renames halign/valign shorthands", func(t *testing.T) {
+		t.Parallel()
+		style := map[string]interface{}{"halign": "left", "valign": "top"}
+		if err := normalizeCellStyleAliases(style, "x"); err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		if style["horizontal_alignment"] != "left" || style["vertical_alignment"] != "top" {
+			t.Errorf("halign/valign not renamed: %#v", style)
+		}
+	})
+
+	t.Run("shorthand colliding with canonical is an error", func(t *testing.T) {
+		t.Parallel()
+		style := map[string]interface{}{
+			"horizontal_align":     "center",
+			"horizontal_alignment": "left",
+		}
+		err := normalizeCellStyleAliases(style, "cell_styles[0]")
+		if err == nil {
+			t.Fatalf("expected conflict error, got nil")
+		}
+		if !strings.Contains(err.Error(), "conflicts with horizontal_alignment") {
+			t.Errorf("error should name the conflict: %v", err)
+		}
+	})
+
+	t.Run("no shorthand leaves the map untouched", func(t *testing.T) {
+		t.Parallel()
+		style := map[string]interface{}{"font_weight": "bold", "horizontal_alignment": "center"}
+		if err := normalizeCellStyleAliases(style, "x"); err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		if len(style) != 2 || style["font_weight"] != "bold" || style["horizontal_alignment"] != "center" {
+			t.Errorf("map should be unchanged: %#v", style)
+		}
+	})
+
+	t.Run("empty map is a no-op", func(t *testing.T) {
+		t.Parallel()
+		if err := normalizeCellStyleAliases(map[string]interface{}{}, "x"); err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+	})
+}
+
+// TestNormalizeTypedCellsStyleAliases pins the 2D --cells walk: every cell's
+// inline cell_styles is normalized, malformed shapes are skipped (matching the
+// pass-through contract) rather than rejected, and a conflict propagates.
+func TestNormalizeTypedCellsStyleAliases(t *testing.T) {
+	t.Parallel()
+
+	t.Run("normalizes inline cell_styles across the grid", func(t *testing.T) {
+		t.Parallel()
+		cells := []interface{}{
+			[]interface{}{
+				map[string]interface{}{
+					"value":       "x",
+					"cell_styles": map[string]interface{}{"horizontal_align": "center"},
+				},
+				map[string]interface{}{"value": "y"}, // no cell_styles → untouched
+			},
+		}
+		if err := normalizeTypedCellsStyleAliases(cells, "--cells"); err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		row := cells[0].([]interface{})
+		st := row[0].(map[string]interface{})["cell_styles"].(map[string]interface{})
+		if st["horizontal_alignment"] != "center" {
+			t.Errorf("cell_styles not normalized: %#v", st)
+		}
+		if _, ok := st["horizontal_align"]; ok {
+			t.Errorf("shorthand should be removed: %#v", st)
+		}
+	})
+
+	t.Run("malformed shapes are skipped, not rejected", func(t *testing.T) {
+		t.Parallel()
+		cells := []interface{}{
+			"not-a-row",
+			[]interface{}{
+				"not-a-cell",
+				map[string]interface{}{"cell_styles": "not-a-map"},
+			},
+		}
+		if err := normalizeTypedCellsStyleAliases(cells, "--cells"); err != nil {
+			t.Fatalf("lenient walk should not error on odd shapes: %v", err)
+		}
+	})
+
+	t.Run("conflict inside a cell propagates", func(t *testing.T) {
+		t.Parallel()
+		cells := []interface{}{
+			[]interface{}{
+				map[string]interface{}{
+					"cell_styles": map[string]interface{}{
+						"valign":             "top",
+						"vertical_alignment": "middle",
+					},
+				},
+			},
+		}
+		err := normalizeTypedCellsStyleAliases(cells, "--cells")
+		if err == nil {
+			t.Fatalf("expected conflict error, got nil")
+		}
+		if !strings.Contains(err.Error(), "--cells[0][0].cell_styles") {
+			t.Errorf("error should carry the cell path: %v", err)
+		}
+	})
+}
+
+// TestCellsSet_StyleAliasesNormalized is the end-to-end guard for +cells-set:
+// a typed --cells payload using alignment shorthands reaches set_cell_range
+// with canonical field names so the backend doesn't silently drop them.
+func TestCellsSet_StyleAliasesNormalized(t *testing.T) {
+	t.Parallel()
+	cells := `[[{"value":"Header","cell_styles":{"horizontal_align":"center","vertical_align":"middle","font_weight":"bold"}}]]`
+	body := parseDryRunBody(t, CellsSet, []string{
+		"--url", testURL, "--sheet-id", testSheetID,
+		"--range", "A1", "--cells", cells,
+	})
+	input := decodeToolInput(t, body, "set_cell_range")
+	raw, _ := json.Marshal(input["cells"])
+	s := string(raw)
+	if !strings.Contains(s, `"horizontal_alignment":"center"`) || !strings.Contains(s, `"vertical_alignment":"middle"`) {
+		t.Errorf("alignment shorthands not normalized in cells: %s", s)
+	}
+	if strings.Contains(s, `"horizontal_align":`) || strings.Contains(s, `"vertical_align":`) {
+		t.Errorf("shorthand keys leaked through to backend payload: %s", s)
+	}
+}
+
+// TestWorkbookCreate_StyleAliasesNormalized is the end-to-end guard for
+// +workbook-create --styles: alignment shorthands in a cell_styles op are
+// accepted (no "unsupported style field" error) and emitted as canonical
+// field names merged into the fill cells.
+func TestWorkbookCreate_StyleAliasesNormalized(t *testing.T) {
+	t.Parallel()
+	calls := parseDryRunAPI(t, WorkbookCreate, []string{
+		"--title", "Sales",
+		"--values", `[["Name","Score"],["alice",95]]`,
+		"--styles", `{"styles":[{"name":"Sheet1","cell_styles":[{"range":"A1:B2","horizontal_align":"center","vertical_align":"middle"}]}]}`,
+	})
+	if len(calls) != 2 {
+		t.Fatalf("api calls = %d, want 2 (create + fill)", len(calls))
+	}
+	body, _ := calls[1].(map[string]interface{})["body"].(map[string]interface{})
+	input := decodeToolInput(t, body, "set_cell_range")
+	raw, _ := json.Marshal(input["cells"])
+	s := string(raw)
+	if c := strings.Count(s, `"horizontal_alignment":"center"`); c != 4 {
+		t.Errorf("horizontal_alignment occurrences = %d, want 4 in 2x2 range; cells=%s", c, s)
+	}
+	if strings.Contains(s, `"horizontal_align":`) || strings.Contains(s, `"vertical_align":`) {
+		t.Errorf("shorthand keys leaked through after normalization: %s", s)
+	}
+}

shortcuts/sheets/lark_sheet_workbook_export_test.go

@@ -0,0 +1,72 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package sheets
+
+import (
+	"encoding/json"
+	"strings"
+	"testing"
+
+	"github.com/larksuite/cli/internal/httpmock"
+)
+
+// TestWorkbookExport_ExecuteExportOnly covers the no-download path: without
+// --output-path, +workbook-export delegates to the shared drive export core
+// with OutputDir="" so it creates + polls the export task and returns the ready
+// file token without writing a local file (downloaded=false).
+func TestWorkbookExport_ExecuteExportOnly(t *testing.T) {
+	stubs := []*httpmock.Stub{
+		{
+			Method: "POST",
+			URL:    "/open-apis/drive/v1/export_tasks",
+			Body: map[string]interface{}{
+				"code": 0, "msg": "ok",
+				"data": map[string]interface{}{"ticket": "tk_export"},
+			},
+		},
+		{
+			Method: "GET",
+			URL:    "/open-apis/drive/v1/export_tasks/tk_export",
+			Body: map[string]interface{}{
+				"code": 0, "msg": "ok",
+				"data": map[string]interface{}{"result": map[string]interface{}{
+					"job_status": float64(0),
+					"file_token": "ftk_xlsx",
+					"file_name":  "report.xlsx",
+					"file_size":  float64(2048),
+				}},
+			},
+		},
+	}
+
+	out, err := runShortcutWithStubs(t, WorkbookExport, []string{
+		"--url", testURL, "--file-extension", "xlsx", "--as", "user",
+	}, stubs...)
+	if err != nil {
+		t.Fatalf("export-only execute failed: %v\n%s", err, out)
+	}
+
+	idx := strings.Index(out, "{")
+	if idx < 0 {
+		t.Fatalf("no JSON envelope:\n%s", out)
+	}
+	var env struct {
+		Data map[string]interface{} `json:"data"`
+	}
+	if err := json.Unmarshal([]byte(out[idx:]), &env); err != nil {
+		t.Fatalf("decode envelope: %v\nraw=%s", err, out)
+	}
+	if env.Data["ready"] != true {
+		t.Errorf("ready = %v, want true", env.Data["ready"])
+	}
+	if env.Data["downloaded"] != false {
+		t.Errorf("downloaded = %v, want false (no --output-path)", env.Data["downloaded"])
+	}
+	if env.Data["file_token"] != "ftk_xlsx" {
+		t.Errorf("file_token = %v, want ftk_xlsx", env.Data["file_token"])
+	}
+	if env.Data["doc_type"] != "sheet" {
+		t.Errorf("doc_type = %v, want sheet", env.Data["doc_type"])
+	}
+}

shortcuts/sheets/lark_sheet_workbook_import_test.go

@@ -0,0 +1,135 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package sheets
+
+import (
+	"encoding/json"
+	"os"
+	"strings"
+	"testing"
+
+	"github.com/larksuite/cli/internal/httpmock"
+	_ "github.com/larksuite/cli/internal/vfs/localfileio"
+)
+
+// chdirTemp switches into a fresh temp dir for the duration of the test and
+// restores the original cwd afterwards. +workbook-import is the first sheets
+// shortcut that stat()s a real local file, so these tests need a working dir.
+func chdirTemp(t *testing.T) {
+	t.Helper()
+	orig, err := os.Getwd()
+	if err != nil {
+		t.Fatalf("getwd: %v", err)
+	}
+	if err := os.Chdir(t.TempDir()); err != nil {
+		t.Fatalf("chdir: %v", err)
+	}
+	t.Cleanup(func() { _ = os.Chdir(orig) })
+}
+
+// TestWorkbookImport_DryRunPinsSheetType verifies the shortcut delegates to the
+// shared drive import core and hard-codes the import target type to "sheet".
+func TestWorkbookImport_DryRunPinsSheetType(t *testing.T) {
+	chdirTemp(t)
+	if err := os.WriteFile("data.xlsx", []byte("fake-xlsx"), 0o644); err != nil {
+		t.Fatalf("write file: %v", err)
+	}
+
+	calls := parseDryRunAPI(t, WorkbookImport, []string{"--file", "./data.xlsx"})
+
+	var createBody map[string]interface{}
+	for _, c := range calls {
+		cm, _ := c.(map[string]interface{})
+		if u, _ := cm["url"].(string); u == "/open-apis/drive/v1/import_tasks" {
+			createBody, _ = cm["body"].(map[string]interface{})
+		}
+	}
+	if createBody == nil {
+		t.Fatalf("no import_tasks create call in dry-run: %#v", calls)
+	}
+	if createBody["type"] != "sheet" {
+		t.Errorf("import type = %v, want sheet (must be pinned regardless of file)", createBody["type"])
+	}
+	if createBody["file_extension"] != "xlsx" {
+		t.Errorf("file_extension = %v, want xlsx", createBody["file_extension"])
+	}
+}
+
+// TestWorkbookImport_RejectsNonSheetFile ensures a file that cannot become a
+// spreadsheet (e.g. .docx) is rejected up front by the pinned-sheet validation.
+func TestWorkbookImport_RejectsNonSheetFile(t *testing.T) {
+	chdirTemp(t)
+	if err := os.WriteFile("notes.docx", []byte("fake-docx"), 0o644); err != nil {
+		t.Fatalf("write file: %v", err)
+	}
+
+	// Validate runs before DryRun, so the pinned-sheet check rejects .docx up
+	// front and the error surfaces through the normal envelope/err path.
+	stdout, stderr, err := runShortcutCapturingErr(t, WorkbookImport, []string{"--file", "./notes.docx", "--dry-run"})
+	if err == nil || !strings.Contains(stdout+stderr+err.Error(), "can only be imported") {
+		t.Errorf("expected .docx → sheet type-mismatch rejection; got stdout=%s stderr=%s err=%v", stdout, stderr, err)
+	}
+}
+
+// TestWorkbookImport_ExecuteCreatesSheet runs the full upload → create → poll
+// flow against stubs and asserts the resulting URL is a /sheets/ link.
+func TestWorkbookImport_ExecuteCreatesSheet(t *testing.T) {
+	chdirTemp(t)
+	if err := os.WriteFile("data.csv", []byte("a,b\n1,2\n"), 0o644); err != nil {
+		t.Fatalf("write file: %v", err)
+	}
+
+	stubs := []*httpmock.Stub{
+		{
+			Method: "POST",
+			URL:    "/open-apis/drive/v1/medias/upload_all",
+			Body: map[string]interface{}{
+				"code": 0, "msg": "ok",
+				"data": map[string]interface{}{"file_token": "file_import_media"},
+			},
+		},
+		{
+			Method: "POST",
+			URL:    "/open-apis/drive/v1/import_tasks",
+			Body: map[string]interface{}{
+				"code": 0, "msg": "ok",
+				"data": map[string]interface{}{"ticket": "tk_sheet"},
+			},
+		},
+		{
+			Method: "GET",
+			URL:    "/open-apis/drive/v1/import_tasks/tk_sheet",
+			Body: map[string]interface{}{
+				"code": 0, "msg": "ok",
+				"data": map[string]interface{}{"result": map[string]interface{}{
+					"token":      "shtcn_imported",
+					"type":       "sheet",
+					"job_status": float64(0),
+				}},
+			},
+		},
+	}
+
+	out, err := runShortcutWithStubs(t, WorkbookImport, []string{"--file", "./data.csv", "--as", "user"}, stubs...)
+	if err != nil {
+		t.Fatalf("import execute failed: %v\n%s", err, out)
+	}
+
+	idx := strings.Index(out, "{")
+	if idx < 0 {
+		t.Fatalf("execute output has no JSON envelope:\n%s", out)
+	}
+	var env struct {
+		Data map[string]interface{} `json:"data"`
+	}
+	if err := json.Unmarshal([]byte(out[idx:]), &env); err != nil {
+		t.Fatalf("decode envelope: %v\nraw=%s", err, out)
+	}
+	if url, _ := env.Data["url"].(string); !strings.Contains(url, "/sheets/") {
+		t.Errorf("imported url = %q, want a /sheets/ link", url)
+	}
+	if tok, _ := env.Data["token"].(string); tok != "shtcn_imported" {
+		t.Errorf("token = %q, want shtcn_imported", tok)
+	}
+}

shortcuts/sheets/lark_sheet_workbook_test.go

@@ -4,14 +4,10 @@
 package sheets
 
 import (
-	"errors"
-	"net/http"
+	"encoding/json"
 	"strings"
 	"testing"
 
-	larkcore "github.com/larksuite/oapi-sdk-go/v3/core"
-
-	"github.com/larksuite/cli/errs"
 	"github.com/larksuite/cli/shortcuts/common"
 )
 
@@ -145,6 +141,28 @@ func TestWorkbookShortcuts_DryRun(t *testing.T) {
 				"tab_color": "",
 			},
 		},
+		{
+			name:     "+sheet-show-gridline",
+			sc:       SheetShowGridline,
+			args:     []string{"--url", testURL, "--sheet-id", testSheetID},
+			toolName: "modify_workbook_structure",
+			wantInput: map[string]interface{}{
+				"excel_id":  testToken,
+				"operation": "show_gridline",
+				"sheet_id":  testSheetID,
+			},
+		},
+		{
+			name:     "+sheet-hide-gridline",
+			sc:       SheetHideGridline,
+			args:     []string{"--url", testURL, "--sheet-id", testSheetID},
+			toolName: "modify_workbook_structure",
+			wantInput: map[string]interface{}{
+				"excel_id":  testToken,
+				"operation": "hide_gridline",
+				"sheet_id":  testSheetID,
+			},
+		},
 	}
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
@@ -288,7 +306,7 @@ func TestWorkbookCreate_DryRun(t *testing.T) {
 		t.Parallel()
 		calls := parseDryRunAPI(t, WorkbookCreate, []string{"--title", "MySheet"})
 		if len(calls) != 1 {
-			t.Fatalf("api calls = %d, want 1 (no headers/data)", len(calls))
+			t.Fatalf("api calls = %d, want 1 (no values)", len(calls))
 		}
 		c := calls[0].(map[string]interface{})
 		if c["url"] != "/open-apis/sheets/v3/spreadsheets" {
@@ -300,12 +318,11 @@ func TestWorkbookCreate_DryRun(t *testing.T) {
 		}
 	})
 
-	t.Run("with headers and data → 2-step plan", func(t *testing.T) {
+	t.Run("with values → 2-step plan", func(t *testing.T) {
 		t.Parallel()
 		calls := parseDryRunAPI(t, WorkbookCreate, []string{
 			"--title", "Sales",
-			"--headers", `["Name","Score"]`,
-			"--values", `[["alice",95],["bob",88]]`,
+			"--values", `[["Name","Score"],["alice",95],["bob",88]]`,
 		})
 		if len(calls) != 2 {
 			t.Fatalf("api calls = %d, want 2 (create + fill)", len(calls))
@@ -317,7 +334,88 @@ func TestWorkbookCreate_DryRun(t *testing.T) {
 		body, _ := fill["body"].(map[string]interface{})
 		input := decodeToolInput(t, body, "set_cell_range")
 		if input["range"] != "A1:B3" {
-			t.Errorf("fill range = %v, want A1:B3 (1 header + 2 data rows × 2 cols)", input["range"])
+			t.Errorf("fill range = %v, want A1:B3 (3 rows × 2 cols)", input["range"])
+		}
+	})
+
+	t.Run("with styles merges into set_cell_range cells", func(t *testing.T) {
+		t.Parallel()
+		calls := parseDryRunAPI(t, WorkbookCreate, []string{
+			"--title", "Sales",
+			"--values", `[["Name","Score"],["alice",95]]`,
+			"--styles", `{"styles":[{"name":"Sheet1","cell_styles":[{"range":"A1","font_weight":"bold","background_color":"#f5f5f5"},{"range":"B1","number_format":"0","border_styles":{"bottom":{"style":"solid","weight":"thin","color":"#000000"}}},{"range":"B2","font_color":"#0f7b0f"}]}]}`,
+		})
+		if len(calls) != 2 {
+			t.Fatalf("api calls = %d, want 2 (create + fill)", len(calls))
+		}
+		body, _ := calls[1].(map[string]interface{})["body"].(map[string]interface{})
+		input := decodeToolInput(t, body, "set_cell_range")
+		cells, _ := input["cells"].([]interface{})
+		if len(cells) != 2 {
+			t.Fatalf("cells rows = %#v, want 2", input["cells"])
+		}
+		headerRow, _ := cells[0].([]interface{})
+		firstHeader, _ := headerRow[0].(map[string]interface{})
+		firstStyle, _ := firstHeader["cell_styles"].(map[string]interface{})
+		if firstStyle["font_weight"] != "bold" || firstStyle["background_color"] != "#f5f5f5" {
+			t.Errorf("first header style = %#v, want bold + background", firstStyle)
+		}
+		secondHeader, _ := headerRow[1].(map[string]interface{})
+		if secondHeader["border_styles"] == nil {
+			t.Errorf("second header missing border_styles: %#v", secondHeader)
+		}
+		secondStyle, _ := secondHeader["cell_styles"].(map[string]interface{})
+		if secondStyle["number_format"] != "0" {
+			t.Errorf("second header number_format = %#v, want 0", secondStyle)
+		}
+		dataRow, _ := cells[1].([]interface{})
+		firstData, _ := dataRow[0].(map[string]interface{})
+		if _, ok := firstData["cell_styles"]; ok {
+			t.Errorf("null style should leave first data cell unstyled: %#v", firstData)
+		}
+		secondData, _ := dataRow[1].(map[string]interface{})
+		secondDataStyle, _ := secondData["cell_styles"].(map[string]interface{})
+		if secondDataStyle["font_color"] != "#0f7b0f" {
+			t.Errorf("second data style = %#v, want font color", secondDataStyle)
+		}
+	})
+
+	t.Run("cell style range can cover the whole initial range", func(t *testing.T) {
+		t.Parallel()
+		calls := parseDryRunAPI(t, WorkbookCreate, []string{
+			"--title", "Sales",
+			"--values", `[["Name","Score"],["alice",95]]`,
+			"--styles", `{"styles":[{"name":"Sheet1","cell_styles":[{"range":"A1:B2","horizontal_alignment":"center"}]}]}`,
+		})
+		body, _ := calls[1].(map[string]interface{})["body"].(map[string]interface{})
+		input := decodeToolInput(t, body, "set_cell_range")
+		raw, _ := json.Marshal(input["cells"])
+		if got := strings.Count(string(raw), "horizontal_alignment"); got != 4 {
+			t.Errorf("horizontal_alignment occurrences = %d, want 4 in 2x2 range; cells=%s", got, raw)
+		}
+	})
+	t.Run("overlapping cell_styles deep-merge fields, no cross-cell pollution", func(t *testing.T) {
+		t.Parallel()
+		calls := parseDryRunAPI(t, WorkbookCreate, []string{
+			"--title", "X",
+			"--values", `[["a","b"]]`,
+			"--styles", `{"styles":[{"name":"Sheet1","cell_styles":[{"range":"A1:B1","font_weight":"bold"},{"range":"B1","font_color":"#ff0000"}]}]}`,
+		})
+		body, _ := calls[1].(map[string]interface{})["body"].(map[string]interface{})
+		input := decodeToolInput(t, body, "set_cell_range")
+		cells, _ := input["cells"].([]interface{})
+		row0, _ := cells[0].([]interface{})
+		// B1 hit by both ops → must keep BOTH font_weight (op1) and font_color (op2).
+		b1, _ := row0[1].(map[string]interface{})
+		b1s, _ := b1["cell_styles"].(map[string]interface{})
+		if b1s["font_weight"] != "bold" || b1s["font_color"] != "#ff0000" {
+			t.Errorf("B1 should deep-merge both ops, got %#v", b1s)
+		}
+		// A1 hit only by op1 → must NOT be polluted by op2's font_color (shared submap).
+		a1, _ := row0[0].(map[string]interface{})
+		a1s, _ := a1["cell_styles"].(map[string]interface{})
+		if a1s["font_color"] != nil {
+			t.Errorf("A1 must not be polluted by op2, got %#v", a1s)
 		}
 	})
 }
@@ -330,8 +428,18 @@ func TestWorkbookCreate_DataValidation(t *testing.T) {
 		args []string
 		want string
 	}{
-		{"headers not array", []string{"--title", "X", "--headers", `"abc"`}, "must be a JSON array"},
 		{"values not 2D", []string{"--title", "X", "--values", `["a","b"]`}, "must be an array"},
+		{"styles not object", []string{"--title", "X", "--styles", `"bold"`}, `shaped as {"styles":[...]}`},
+		{"styles missing array", []string{"--title", "X", "--styles", `{"value":"x"}`}, "--styles.styles is required"},
+		{"styles item missing groups", []string{"--title", "X", "--values", `[["a"]]`, "--styles", `{"styles":[{"name":"Sheet1","value":"x"}]}`}, "must include at least one of cell_styles/row_sizes/col_sizes/cell_merges"},
+		{"cell styles must be array", []string{"--title", "X", "--values", `[["a"]]`, "--styles", `{"styles":[{"name":"Sheet1","cell_styles":{"range":"A1","font_weight":"bold"}}]}`}, "cell_styles must be an array"},
+		{"cell style needs range", []string{"--title", "X", "--values", `[["a"]]`, "--styles", `{"styles":[{"name":"Sheet1","cell_styles":[{"font_weight":"bold"}]}]}`}, "range is required"},
+		{"nested cell_styles rejected", []string{"--title", "X", "--values", `[["a"]]`, "--styles", `{"styles":[{"name":"Sheet1","cell_styles":[{"range":"A1","cell_styles":{"font_weight":"bold"}}]}]}`}, "put style fields directly"},
+		{"row size needs row range", []string{"--title", "X", "--values", `[["a"]]`, "--styles", `{"styles":[{"name":"Sheet1","row_sizes":[{"range":"A1","type":"pixel","size":20}]}]}`}, "must use row numbers"},
+		{"col size needs pixel size", []string{"--title", "X", "--values", `[["a"]]`, "--styles", `{"styles":[{"name":"Sheet1","col_sizes":[{"range":"A:A","type":"pixel"}]}]}`}, "requires size"},
+		{"border bad style enum", []string{"--title", "X", "--values", `[["a"]]`, "--styles", `{"styles":[{"name":"Sheet1","cell_styles":[{"range":"A1","border_styles":{"bottom":{"style":"NONSENSE"}}}]}]}`}, `style "NONSENSE" is invalid`},
+		{"border invalid side", []string{"--title", "X", "--values", `[["a"]]`, "--styles", `{"styles":[{"name":"Sheet1","cell_styles":[{"range":"A1","border_styles":{"diagonal":{"style":"solid"}}}]}]}`}, "not a valid side"},
+		{"border bad weight", []string{"--title", "X", "--values", `[["a"]]`, "--styles", `{"styles":[{"name":"Sheet1","cell_styles":[{"range":"A1","border_styles":{"top":{"weight":"xxl"}}}]}]}`}, `weight "xxl" is invalid`},
 	}
 	for _, tt := range cases {
 		t.Run(tt.name, func(t *testing.T) {
@@ -344,21 +452,21 @@ func TestWorkbookCreate_DataValidation(t *testing.T) {
 	}
 }
 
-// TestWorkbookExport_DryRun checks the 2-or-3 step plan depending on
-// --output-path. The order should be: POST → GET (poll) → optional GET
-// (download).
+// TestWorkbookExport_DryRun verifies the export dry-run now delegates to the
+// shared drive export core: a single create-task POST (poll + download are
+// described inline rather than as separate api entries).
 func TestWorkbookExport_DryRun(t *testing.T) {
 	t.Parallel()
 
-	t.Run("xlsx without --output-path → 2 steps", func(t *testing.T) {
+	t.Run("xlsx create-task body pins type=sheet", func(t *testing.T) {
 		t.Parallel()
 		calls := parseDryRunAPI(t, WorkbookExport, []string{"--url", testURL, "--file-extension", "xlsx"})
-		if len(calls) != 2 {
-			t.Fatalf("api calls = %d, want 2 (create + poll)", len(calls))
+		if len(calls) != 1 {
+			t.Fatalf("api calls = %d, want 1 (create export task)", len(calls))
 		}
 		create := calls[0].(map[string]interface{})
 		if create["url"] != "/open-apis/drive/v1/export_tasks" {
-			t.Errorf("first url = %v", create["url"])
+			t.Errorf("url = %v", create["url"])
 		}
 		body, _ := create["body"].(map[string]interface{})
 		if body["type"] != "sheet" || body["file_extension"] != "xlsx" || body["token"] != testToken {
@@ -366,22 +474,18 @@ func TestWorkbookExport_DryRun(t *testing.T) {
 		}
 	})
 
-	t.Run("csv → 3 steps, with sub_id", func(t *testing.T) {
+	t.Run("csv includes sub_id from --sheet-id", func(t *testing.T) {
 		t.Parallel()
 		calls := parseDryRunAPI(t, WorkbookExport, []string{
 			"--url", testURL, "--file-extension", "csv", "--sheet-id", "sh1",
 			"--output-path", "/tmp/out.csv",
 		})
-		if len(calls) != 3 {
-			t.Fatalf("api calls = %d, want 3", len(calls))
+		if len(calls) != 1 {
+			t.Fatalf("api calls = %d, want 1", len(calls))
 		}
 		body, _ := calls[0].(map[string]interface{})["body"].(map[string]interface{})
-		if body["sub_id"] != "sh1" {
-			t.Errorf("csv export missing sub_id: %#v", body)
-		}
-		dl := calls[2].(map[string]interface{})
-		if !strings.Contains(dl["url"].(string), "/export_tasks/file/") {
-			t.Errorf("download url = %v", dl["url"])
+		if body["type"] != "sheet" || body["sub_id"] != "sh1" {
+			t.Errorf("csv export body = %#v (want type=sheet, sub_id=sh1)", body)
 		}
 	})
 
@@ -396,92 +500,6 @@ func TestWorkbookExport_DryRun(t *testing.T) {
 	})
 }
 
-func TestWorkbookExportDownloadErrorClassification(t *testing.T) {
-	t.Parallel()
-
-	t.Run("preserves typed request errors", func(t *testing.T) {
-		t.Parallel()
-		in := errs.NewAPIError(errs.SubtypeServerError, "typed upstream").WithCode(123)
-		got := sheetsDownloadRequestError(in)
-		if got != in {
-			t.Fatalf("typed error was not preserved: got %T %v", got, got)
-		}
-	})
-
-	t.Run("wraps raw request errors as network transport", func(t *testing.T) {
-		t.Parallel()
-		got := sheetsDownloadRequestError(errors.New("dial refused"))
-		p, ok := errs.ProblemOf(got)
-		if !ok {
-			t.Fatalf("expected typed problem, got %T %v", got, got)
-		}
-		if p.Category != errs.CategoryNetwork || p.Subtype != errs.SubtypeNetworkTransport {
-			t.Fatalf("problem = %s/%s, want %s/%s", p.Category, p.Subtype, errs.CategoryNetwork, errs.SubtypeNetworkTransport)
-		}
-	})
-
-	tests := []struct {
-		name          string
-		status        int
-		wantCategory  errs.Category
-		wantSubtype   errs.Subtype
-		wantRetryable bool
-	}{
-		{
-			name:          "5xx is retryable network server error",
-			status:        http.StatusBadGateway,
-			wantCategory:  errs.CategoryNetwork,
-			wantSubtype:   errs.SubtypeNetworkServer,
-			wantRetryable: true,
-		},
-		{
-			name:         "404 is API not found",
-			status:       http.StatusNotFound,
-			wantCategory: errs.CategoryAPI,
-			wantSubtype:  errs.SubtypeNotFound,
-		},
-		{
-			name:          "429 is retryable API rate limit",
-			status:        http.StatusTooManyRequests,
-			wantCategory:  errs.CategoryAPI,
-			wantSubtype:   errs.SubtypeRateLimit,
-			wantRetryable: true,
-		},
-		{
-			name:         "other 4xx is API unknown",
-			status:       http.StatusForbidden,
-			wantCategory: errs.CategoryAPI,
-			wantSubtype:  errs.SubtypeUnknown,
-		},
-	}
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			t.Parallel()
-			got := sheetsDownloadHTTPStatusError(&larkcore.ApiResp{
-				StatusCode: tt.status,
-				RawBody:    []byte("body"),
-				Header:     http.Header{larkcore.HttpHeaderKeyLogId: []string{"log123"}},
-			})
-			p, ok := errs.ProblemOf(got)
-			if !ok {
-				t.Fatalf("expected typed problem, got %T %v", got, got)
-			}
-			if p.Category != tt.wantCategory || p.Subtype != tt.wantSubtype {
-				t.Fatalf("problem = %s/%s, want %s/%s", p.Category, p.Subtype, tt.wantCategory, tt.wantSubtype)
-			}
-			if p.Code != tt.status {
-				t.Fatalf("code = %d, want %d", p.Code, tt.status)
-			}
-			if p.LogID != "log123" {
-				t.Fatalf("log_id = %q, want log123", p.LogID)
-			}
-			if p.Retryable != tt.wantRetryable {
-				t.Fatalf("retryable = %v, want %v", p.Retryable, tt.wantRetryable)
-			}
-		})
-	}
-}
-
 // assertInputEquals compares the decoded tool input map against the wanted
 // fields. Extra fields in `got` are allowed (defaults, optional fields);
 // every key in `want` must match exactly.

shortcuts/sheets/lark_sheet_write_cells.go

@@ -56,7 +56,7 @@ var CellsSet = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindWrite, "set_cell_range", input)
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}
@@ -88,6 +88,9 @@ func cellsSetInput(runtime flagView, token, sheetID, sheetName string) (map[stri
 	if err != nil {
 		return nil, err
 	}
+	if err := normalizeTypedCellsStyleAliases(cells, "--cells"); err != nil {
+		return nil, err
+	}
 	input := map[string]interface{}{
 		"excel_id": token,
 		"range":    strings.TrimSpace(runtime.Str("range")),
@@ -129,7 +132,7 @@ var CellsSetStyle = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindWrite, "set_cell_range", input)
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}
@@ -197,12 +200,12 @@ func cellsSetStyleInput(runtime flagView, token, sheetID, sheetName string) (map
 	return input, nil
 }
 
-// CsvPut wraps set_range_from_csv: dump a CSV blob into a sheet, only writing
-// plain values. Use +cells-set for anything richer (formula / style / note).
+// CsvPut wraps set_range_from_csv: dump a CSV blob into a sheet. A cell whose
+// text starts with = is evaluated as a formula; use +cells-set for styles / notes / images.
 var CsvPut = common.Shortcut{
 	Service:     "sheets",
 	Command:     "+csv-put",
-	Description: "Paste RFC-4180 CSV into a sheet at --start-cell (plain values only, auto-expands sheet if needed).",
+	Description: "Paste RFC-4180 CSV into a sheet at --start-cell (values or formulas: a leading = is evaluated as a formula; no styles / comments; auto-expands sheet if needed).",
 	Risk:        "write",
 	Scopes:      []string{"sheets:spreadsheet:write_only"},
 	AuthTypes:   []string{"user", "bot"},
@@ -237,7 +240,7 @@ var CsvPut = common.Shortcut{
 		return dr
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}
@@ -414,7 +417,7 @@ var DropdownSet = common.Shortcut{
 		return invokeToolDryRun(token, ToolKindWrite, "set_cell_range", input)
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}
@@ -801,7 +804,7 @@ var CellsSetImage = common.Shortcut{
 			Body(setCellBody)
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
-		token, err := resolveSpreadsheetToken(runtime)
+		token, err := resolveSpreadsheetTokenExec(runtime)
 		if err != nil {
 			return err
 		}

shortcuts/sheets/lark_sheet_write_cells_test.go

@@ -4,6 +4,7 @@
 package sheets
 
 import (
+	"fmt"
 	"strings"
 	"testing"
 
@@ -427,6 +428,44 @@ func TestCellsSet_RequiresJSONArray(t *testing.T) {
 	}
 }
 
+// TestCellsSet_RejectsUnsupportedMentionType pins the mention_type enum in
+// data/flag-schemas.json (synced from the upstream tool schema): a rich_text
+// mention whose mention_type is outside MENTION_FILE_TYPE (here 6 = cloud
+// shared folder) is rejected by the schema validator at flag-parse time,
+// before it can reach the server and blow up pb serialization
+// ("mentionFileInfo.fileType: enum value expected").
+func TestCellsSet_RejectsUnsupportedMentionType(t *testing.T) {
+	t.Parallel()
+	cells := `[[{"rich_text":[{"type":"mention","text":"x","mention_type":6,"mention_token":"t"}]}]]`
+	stdout, stderr, err := runShortcutCapturingErr(t, CellsSet, []string{
+		"--url", testURL, "--sheet-id", testSheetID,
+		"--range", "A1", "--cells", cells, "--dry-run",
+	})
+	if err == nil {
+		t.Fatalf("expected validation error; stdout=%s stderr=%s", stdout, stderr)
+	}
+	combined := stdout + stderr + err.Error()
+	if !strings.Contains(combined, "mention_type") || !strings.Contains(combined, "not in enum") {
+		t.Errorf("expected mention_type enum guard; got=%s|%s|%v", stdout, stderr, err)
+	}
+}
+
+// TestCellsSet_AllowsValidMentionTypes confirms the guard lets through a
+// user @mention (mention_type 0) and a render-supported file type (22 = DOCX).
+func TestCellsSet_AllowsValidMentionTypes(t *testing.T) {
+	t.Parallel()
+	for _, mt := range []int{0, 22} {
+		cells := fmt.Sprintf(`[[{"rich_text":[{"type":"mention","text":"x","mention_type":%d,"mention_token":"t"}]}]]`, mt)
+		stdout, stderr, err := runShortcutCapturingErr(t, CellsSet, []string{
+			"--url", testURL, "--sheet-id", testSheetID,
+			"--range", "A1", "--cells", cells, "--dry-run",
+		})
+		if err != nil {
+			t.Errorf("mention_type %d: unexpected error: stdout=%s stderr=%s err=%v", mt, stdout, stderr, err)
+		}
+	}
+}
+
 // TestCellsSetImage_DryRun verifies the 2-step plan (upload + embed) is
 // rendered, including the parent_type=sheet_image upload metadata.
 func TestCellsSetImage_DryRun(t *testing.T) {

shortcuts/sheets/shortcuts.go

@@ -3,7 +3,11 @@
 
 package sheets
 
-import "github.com/larksuite/cli/shortcuts/common"
+import (
+	"github.com/larksuite/cli/shortcuts/common"
+	"github.com/spf13/cobra"
+	"github.com/spf13/pflag"
+)
 
 // Shortcuts returns all lark-sheets shortcuts. The list is grouped by
 // canonical skill to mirror the sheet-skill-spec layout
@@ -22,10 +26,46 @@ func Shortcuts() []common.Shortcut {
 		if _, ok := commandsWithSchema[all[i].Command]; ok {
 			all[i].PrintFlagSchema = printFlagSchemaFor(all[i].Command)
 		}
+		// Accept --token as a parse-time alias for --spreadsheet-token (the
+		// single highest-frequency reflex misspelling in eval traces) on every
+		// shortcut that registers --spreadsheet-token, so the typo costs zero
+		// round-trips instead of an unknown-flag failure. Wired through the
+		// existing PostMount hook and composed onto any prior PostMount, so the
+		// common framework needs no change at all.
+		if hasFlag(all[i].Flags, "spreadsheet-token") {
+			all[i].PostMount = withTokenAlias(all[i].PostMount)
+		}
 	}
 	return all
 }
 
+func hasFlag(flags []common.Flag, name string) bool {
+	for _, fl := range flags {
+		if fl.Name == name {
+			return true
+		}
+	}
+	return false
+}
+
+// withTokenAlias wraps an optional PostMount so that, after it runs, --token
+// resolves to --spreadsheet-token at parse time via pflag's normalize hook (no
+// duplicate flag in --help). It preserves any pre-existing PostMount — e.g.
+// +csv-put's --range / --start-cell flag-group setup — by running it first.
+func withTokenAlias(prev func(cmd *cobra.Command)) func(cmd *cobra.Command) {
+	return func(cmd *cobra.Command) {
+		if prev != nil {
+			prev(cmd)
+		}
+		cmd.Flags().SetNormalizeFunc(func(_ *pflag.FlagSet, name string) pflag.NormalizedName {
+			if name == "token" {
+				return pflag.NormalizedName("spreadsheet-token")
+			}
+			return pflag.NormalizedName(name)
+		})
+	}
+}
+
 func shortcutList() []common.Shortcut {
 	return []common.Shortcut{
 		// lark_sheet_workbook
@@ -38,8 +78,11 @@ func shortcutList() []common.Shortcut {
 		SheetHide,
 		SheetUnhide,
 		SheetSetTabColor,
+		SheetShowGridline,
+		SheetHideGridline,
 		WorkbookCreate,
 		WorkbookExport,
+		WorkbookImport,
 
 		// lark_sheet_sheet_structure
 		SheetInfo,
@@ -56,6 +99,7 @@ func shortcutList() []common.Shortcut {
 		CellsGet,
 		CsvGet,
 		DropdownGet,
+		TableGet,
 
 		// lark_sheet_search_replace
 		CellsSearch,
@@ -67,6 +111,7 @@ func shortcutList() []common.Shortcut {
 		CellsSetImage,
 		CsvPut,
 		DropdownSet,
+		TablePut,
 
 		// lark_sheet_range_operations
 		CellsClear,

shortcuts/sheets/shortcuts_alias_test.go

@@ -0,0 +1,53 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package sheets
+
+import (
+	"testing"
+
+	"github.com/spf13/cobra"
+)
+
+// TestWithTokenAlias verifies the PostMount-based --token → --spreadsheet-token
+// alias: it resolves at parse time, and it composes onto (rather than replaces)
+// any pre-existing PostMount — the property that lets it coexist with
+// +csv-put's --range/--start-cell flag-group setup.
+func TestWithTokenAlias(t *testing.T) {
+	t.Parallel()
+
+	// Alias resolves to the canonical flag.
+	cmd := &cobra.Command{Use: "x"}
+	cmd.Flags().String("spreadsheet-token", "", "")
+	withTokenAlias(nil)(cmd)
+	if err := cmd.Flags().Parse([]string{"--token", "shtABC"}); err != nil {
+		t.Fatalf("--token should resolve as an alias: %v", err)
+	}
+	if got := cmd.Flags().Lookup("spreadsheet-token").Value.String(); got != "shtABC" {
+		t.Errorf("--token should set --spreadsheet-token; got %q", got)
+	}
+
+	// Composes with an existing PostMount instead of dropping it.
+	prevRan := false
+	cmd2 := &cobra.Command{Use: "y"}
+	cmd2.Flags().String("spreadsheet-token", "", "")
+	withTokenAlias(func(_ *cobra.Command) { prevRan = true })(cmd2)
+	if !prevRan {
+		t.Error("pre-existing PostMount should still run")
+	}
+	if err := cmd2.Flags().Parse([]string{"--token", "shtZ"}); err != nil {
+		t.Fatalf("--token should still resolve when composed: %v", err)
+	}
+}
+
+// TestShortcuts_TokenAliasOnSpreadsheetTokenCommands asserts every shortcut that
+// takes --spreadsheet-token ends up with a PostMount (the composed token alias),
+// so the reflex typo is forgiven wherever the canonical flag exists.
+func TestShortcuts_TokenAliasOnSpreadsheetTokenCommands(t *testing.T) {
+	t.Parallel()
+	for _, s := range Shortcuts() {
+		if hasFlag(s.Flags, "spreadsheet-token") && s.PostMount == nil {
+			t.Errorf("%s takes --spreadsheet-token but has no PostMount (token alias missing)", s.Command)
+		}
+	}
+}

skills/lark-sheets/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: lark-sheets
-version: 2.0.0
-description: "飞书电子表格：创建和操作电子表格。支持创建表格、管理工作表与行列结构（增删/合并/调整尺寸/隐藏/冻结）、读写单元格（值/公式/样式/批注/单元格图片）、查找替换、多操作原子批量更新，以及图表、透视表、条件格式、筛选器、迷你图、浮动图片等对象的创建与维护。当用户需要创建电子表格、管理工作表、批量读写或编辑数据、统计汇总与可视化、表格美化、公式计算（含 Excel 公式迁移）等任务时使用。若用户是想按名称或关键词搜索云空间（云盘/云存储）里的表格文件，请改用 lark-drive 的 drive +search 先定位资源。当用户给出 doubao.com 的 /sheets/ URL/token 时，也应直接使用本 skill，不要因为域名不是飞书而回退到 WebFetch；路由依据是 URL 路径模式和 token，而不是域名。仅针对飞书在线电子表格，不适用于本地 Excel 文件。"
+version: 3.0.0
+description: "飞书电子表格：创建和操作电子表格。支持创建表格、管理工作表与行列结构（增删/合并/调整尺寸/隐藏/冻结）、读写单元格（值/公式/样式/批注/单元格图片）、查找替换、多操作原子批量更新，以及图表、透视表、条件格式、筛选器、迷你图、浮动图片等对象的创建与维护。当用户需要创建电子表格、管理工作表、批量读写或编辑数据、统计汇总与可视化、表格美化、公式计算（含 Excel 公式迁移）、金融/财务建模（DCF、三张表、预算、Sensitivity 等）等任务时使用。若用户是想按名称或关键词搜索云空间（云盘/云存储）里的表格文件，请改用 lark-drive 的 drive +search 先定位资源。当用户给出 doubao.com 的 /sheets/ URL/token 时，也应直接使用本 skill，不要因为域名不是飞书而回退到 WebFetch；路由依据是 URL 路径模式和 token，而不是域名。"
 metadata:
   requires:
     bins: ["lark-cli"]
@@ -38,20 +38,27 @@ metadata:
 
 | 你要做的事 | ✅ 正确写法 | ❌ 不存在（会被 cobra 拒） |
 | --- | --- | --- |
-| 读数据（纯值 / CSV） | `+csv-get`（范围用 `--range`） | — |
-| 读值 + 公式 / 样式 / 批注 | `+cells-get --include value,formula,style,comment,data_validation` | `--with-styles`、`--with-merges`、`--include-merged-cells` |
-| 写纯值（整块 CSV 平铺） | `+csv-put`（定位用 `--start-cell`，单个左上角锚点格；也接受 `--range` 别名，区间自动取左上角） | — |
+| 读数据（纯值 / CSV） | `+csv-get`（范围用 `--range`） | `+get-range`、`+range-get`、`+cells-read` |
+| 读值 + 公式 / 样式 / 批注 | `+cells-get --include value,formula,style,comment,data_validation` | `+get-cell`、`+cell-get`、`--with-styles`、`--with-merges`、`--include-merged-cells` |
+| 写纯文本值（整块 CSV 平铺，列里没有需保留的数值 / 日期语义） | `+csv-put`（定位用 `--start-cell`，单个左上角锚点格；也接受 `--range` 别名，区间自动取左上角） | — |
+| 写带类型的数据到**已有**表（列里有数字 / 金额 / 百分比 / 日期 / 计数，要可排序 / 求和 / 入图表 / 透视） | `+table-put`（列显式声明 `type` + `format`，类型保真；来源不限 DataFrame——Counter / dict / list 同理，详见 write-cells） | 在本地把数字拼成 `"$1,234"` / `"30.5%"` 字符串再 `+csv-put`（会落成文本、丢失计算能力） |
+| **新建**电子表格并写带类型的数据（类型保真需求同上，但目标表还不存在） | `+workbook-create --sheets`（协议与 `+table-put` 同构、一步建表 + typed 写入，无需先建空表再 `+table-put`；date / number 不丢，详见 workbook） | 用 `--values` 灌日期 / 数字（会落成文本、丢类型） |
 | 写值 / 公式 / 样式 | `+cells-set`（定位用 `--range`） | — |
+| 插图：图片**绑定到某条记录**、随行走（凭证 / 证件照 / 商品图 / 头像 / 二维码 / 每行配图） | `+cells-set-image`（单格 `--range`，嵌入单元格内） | — |
+| 插图：**自由摆放、不绑数据**的装饰 / 标识（logo / 水印 / 封面大图 / banner） | `+float-image-create`（浮动图片，自由定位 + 尺寸 + 层级） | — |
 | 查找单元格 | `+cells-search`（关键字用 `--find`） | `+cells-find`、`+find`、`--query` |
 | 查找并替换 | `+cells-replace` | — |
 | 看子表结构（合并 / 行高列宽 / 冻结 / 隐藏） | `+sheet-info` | `+sheet-get`、`+structure-get`、`+sheet-structure-get` |
-| 看工作簿 / 子表清单 | `+workbook-info` | — |
+| 看工作簿 / 子表清单 | `+workbook-info` | `+sheet-list`、`+workbook-get`、`+workbook-list` |
 | 导出 xlsx / 单表 csv | `+workbook-export` | — |
+| 导入本地 xlsx/xls/csv 文件为飞书电子表格 | `+workbook-import --file ./x.xlsx`（本地表格文件 → 飞书电子表格的正解；仅要导成多维表格 bitable 时才用 `drive +import --type bitable`） | `drive +import`（导电子表格时绕了 drive 通道、还要多给 `--type`，应直接用 `+workbook-import`）、把 .xlsx 在本地读成数据再 `+workbook-create` 重灌 |
 | 清除内容 / 格式 | `+cells-clear`（范围维度用 `--scope`，取值 content / formats / all） | `--type` |
 | 批量清除多区域 | `+cells-batch-clear`（`--scope`） | `--target` |
 | 调整列宽 / 行高 | `+cols-resize` / `+rows-resize`（行、列是两个独立命令） | `--dimension`（无此 flag） |
 | 分组汇总 / 透视 | `+pivot-create`（默认不传落点 flag → 自动新建子表，零覆盖） | 用 SUMIF / 本地脚本拼一张假透视表 |
 
+> ⚠️ **两种图片别选错**：图若**绑定某条记录、要随行排序 / 筛选 / 增删**（凭证 / 证件照 / 每行配图，话里带「对应 / 每行 / 这列」等绑定词）→ 单元格图片 `+cells-set-image`；只是自由摆放的装饰（logo / 水印 / 封面）→ 浮动图片 `+float-image-create`。别因「浮动图更好控制 / 更熟」默认选浮动图。
+> ⚠️ **纯文本还是数值语义**：要写的列里有数字 / 金额 / 百分比 / 日期 / 计数 → `+table-put`（写入已有表；声明 `type` + `format`，保留排序 / 求和 / 图表 / 透视能力；**目标表还不存在就用 `+workbook-create --sheets`**，同 typed 协议、一步建表 + 写入，别先建空表再 `+table-put`）；只有纯文本才用 `+csv-put`。两者写完显示可以完全相同，但 `+csv-put` 落的是文本、不能参与计算——别把数值在本地拼成带 `$` / `%` 的字符串再走 `+csv-put`。
 > ⚠️ **定位 flag**：`+cells-get` / `+cells-set` / `+csv-get` 用 `--range`；`+csv-put` 规范用 `--start-cell`（单个左上角锚点格），也接受 `--range` 别名（区间自动取左上角），二者择一即可。
 > ⚠️ **读取附加信息**一律走 `+cells-get --include …`，**没有** `--with-styles` 这类 flag；**看合并单元格**用 `+sheet-info` 的 `merged_cells`，不要在 `+cells-get` 里找 merge flag。
 
@@ -63,28 +70,28 @@ metadata:
 
 | Reference | 描述 |
 | --- | --- |
-| [飞书表格核心操作：分析、编辑与可视化](references/lark-sheets-core-operations.md) | 飞书表格核心操作工作流。当用户需要对已有的飞书表格进行查看、分析、编辑或可视化时使用。适用场景：数据查询与统计、公式计算、表格美化、创建图表/透视表、筛选排序、批量修改数据、调整表格结构等。即使用户没有明确说"飞书表格"，只要操作对象是已有的在线表格，都应触发此工作流。不适用于本地 Excel 文件操作。 |
-| [飞书表格样式与配色规范](references/lark-sheets-visual-standards.md) | 飞书表格样式与配色规范：表头/数据区/汇总行的颜色、字号、对齐、边框等取值标准，以及新增汇总行、追加行列继承原表风格、已有区域美化等典型场景的决策流程与样式要点。工具调用参数细节请参考对应的 lark-sheets-write-cells / lark-sheets-range-operations / lark-sheets-batch-update。条件格式（高亮、标红、数据条、色阶）请使用 lark-sheets-conditional-format。仅针对飞书表格，不适用于本地 Excel 文件。 |
-| [飞书表格公式生成规则](references/lark-sheets-formula-translation.md) | Excel 公式到飞书表格公式的迁移与生成规则。核心目标不是保留 Excel 原语法，而是按飞书表格可执行规则重写公式，并在结果上尽量对齐 Excel。当用户要求把 Excel 公式改写成飞书表格公式，或需要生成飞书公式（尤其涉及 ARRAYFORMULA、原生数组函数、INDEX/OFFSET、MAP/LAMBDA、日期差、多层范围结果与二次展开）时使用。仅针对飞书在线表格，不适用于本地 Excel 文件执行。 |
+| [飞书表格核心操作：分析、编辑与可视化](references/lark-sheets-core-operations.md) | 飞书表格核心操作工作流。当用户需要对已有的飞书表格进行查看、分析、编辑或可视化时使用。适用场景：数据查询与统计、公式计算、表格美化、创建图表/透视表、筛选排序、批量修改数据、调整表格结构等。即使用户没有明确说"飞书表格"，只要操作对象是已有的在线表格，都应触发此工作流。 |
+| [飞书表格样式与配色规范](references/lark-sheets-visual-standards.md) | 飞书表格样式与配色规范：表头/数据区/汇总行的颜色、字号、对齐、边框等取值标准，以及新增汇总行、追加行列继承原表风格、已有区域美化等典型场景的决策流程与样式要点。工具调用参数细节请参考对应的 lark-sheets-write-cells / lark-sheets-range-operations / lark-sheets-batch-update。条件格式（高亮、标红、数据条、色阶）请使用 lark-sheets-conditional-format。 |
+| [飞书表格公式生成规则](references/lark-sheets-formula-translation.md) | Excel 公式到飞书表格公式的迁移与生成规则。核心目标不是保留 Excel 原语法，而是按飞书表格可执行规则重写公式，并在结果上尽量对齐 Excel。当用户要求把 Excel 公式改写成飞书表格公式，或需要生成飞书公式（尤其涉及 ARRAYFORMULA、原生数组函数、INDEX/OFFSET、MAP/LAMBDA、日期差、多层范围结果与二次展开）时使用。 |
 
 ### 按对象的工具参考（含 shortcut）
 
 | Reference | 描述 |
 | --- | --- |
-| [Lark Sheet Workbook](references/lark-sheets-workbook.md) | 管理飞书表格的工作簿结构（子表列表及元数据）。当用户提到"看看这个表格有什么"、"表格结构"、"有哪些 sheet"、"新建一个 sheet"、"删除这个工作表"、"重命名"、"复制一份"、"移动到前面"时使用。仅针对飞书表格。 |
-| [Lark Sheet Sheet Structure](references/lark-sheets-sheet-structure.md) | 管理飞书表格的子表结构与布局。适用场景：查看行高、列宽、隐藏行列、合并单元格等布局信息，以及"插入一行"、"删除这列"、"隐藏行"、"冻结表头"、行列分组（大纲折叠/展开）等操作。行列大纲仅在用户明确提到"行分组"、"列分组"、"大纲"、"outline"时才触发，"按XXX分组"等数据分组场景请使用 lark-sheets-pivot-table。如需在表尾追加数据，应先通过此 skill 插入行，再通过 lark-sheets-write-cells 写入。仅针对飞书表格。 |
-| [Lark Sheet Read Data](references/lark-sheets-read-data.md) | 读取飞书表格中的单元格数据。当用户需要"看看数据"、"分析数据"、"统计/汇总"时使用；也适用于需要查看公式、样式、批注等详细信息的场景。仅针对飞书表格。 |
-| [Lark Sheet Search & Replace](references/lark-sheets-search-replace.md) | 在飞书表格中搜索和替换文本，支持限定范围、大小写匹配、精确匹配、正则表达式。当用户需要"查找"、"搜索"、"定位"某个值，或"替换"、"批量修改文本"、"把 A 改成 B"时使用。不要用于理解表格结构（应读取数据）、不要用于数据分析（应读取数据后计算）、不要把用户操作动作中的关键词（如"汇总金额""统计数量"）当作搜索词。仅针对飞书表格。 |
-| [Lark Sheet Write Cells](references/lark-sheets-write-cells.md) | 向飞书表格的指定区域批量写入值、公式、样式、批注或单元格图片。适用场景：填写数据、设置公式、修改格式、添加批注、嵌入单元格图片（如需操作浮动图片，请使用 lark-sheets-float-image）；若只需把一块 CSV 纯值批量铺到表格上（不带公式/样式），直接使用 `+csv-put` 更短更快。追加数据需先通过 lark-sheets-sheet-structure 插入行列。仅针对飞书表格。 |
-| [Lark Sheet Range Operations](references/lark-sheets-range-operations.md) | 对飞书表格中指定区域执行结构性操作（不涉及写入单元格数据值）。适用场景：清除内容或格式（"清空"、"删除内容"、"去掉格式"）、合并/取消合并单元格、调整行高列宽（"加宽列"、"自适应列宽"）、移动/复制/填充/排序数据（"移动数据"、"复制到"、"自动填充"、"按某列排序"）。写入单元格数据请使用 lark-sheets-write-cells。仅针对飞书表格。 |
-| [Lark Sheet Batch Update](references/lark-sheets-batch-update.md) | 将多个飞书表格写入操作合并为一次批量执行，按顺序依次完成。适合需要连续执行多个写入操作的场景（如先修改结构再写入数据）。仅针对飞书表格。 |
-| [Lark Sheet Chart](references/lark-sheets-chart.md) | 管理飞书表格中的图表（柱形图、折线图、饼图、条形图、面积图、散点图、组合图、雷达图等）。当用户需要创建图表、修改图表样式或数据源、查看已有图表配置、删除图表时使用。也适用于用户提到"数据可视化"、"画个图"、"趋势分析"、"对比图"、"占比分析"、"做个图表"等数据可视化相关场景。仅针对飞书表格。 |
-| [Lark Sheet Pivot Table](references/lark-sheets-pivot-table.md) | 管理飞书表格中的数据透视表。当用户需要创建透视表、修改透视表的行列字段/聚合方式/筛选条件、查看已有透视表配置、删除透视表时使用。也适用于用户提到"分组汇总"、"交叉分析"、"按XXX统计"、"按字段分组"、"再分下组"、"多维分析"、"数据透视"等场景。仅针对飞书表格。 |
-| [Lark Sheet Conditional Format](references/lark-sheets-conditional-format.md) | 管理飞书表格中的条件格式规则（重复值高亮、单元格值比较、数据条、色阶、排名、自定义公式等）。当用户需要创建条件格式、修改已有规则的范围或样式、查看当前条件格式配置、删除规则时使用。也适用于用户提到"高亮"、"标红"、"颜色标记"、"数据条"、"色阶"、"条件样式"等场景。仅针对飞书表格。 |
-| [Lark Sheet Filter](references/lark-sheets-filter.md) | 管理飞书表格中的筛选器（filter）。当用户需要筛选数据（按文本/数值/颜色/日期条件过滤行）、查看已有筛选配置、修改或删除筛选器时使用。也适用于"只看"、"筛选出"、"仅保留符合条件的"等场景。仅针对飞书表格。 |
-| [Lark Sheet Filter View](references/lark-sheets-filter-view.md) | 管理飞书表格中的筛选视图（filter view）。当用户需要"建一个 XX 视图"、"保存这个筛选状态"、"切换不同筛选"、维护一个 sheet 上多份独立筛选配置时使用。视图与筛选器（filter）相互独立，可在同一 sheet 共存；视图的隐藏行仅在用户进入该视图时本地生效，不影响其他协作者。仅针对飞书表格。 |
-| [Lark Sheet Sparkline](references/lark-sheets-sparkline.md) | 管理飞书表格中的迷你图（折线迷你图、柱形迷你图、胜负迷你图）。当用户需要在单元格内嵌入小型图表来展示数据趋势时使用。也适用于"趋势线"、"单元格内图表"、"迷你图"等场景。注意：不等同于被禁用的 SPARKLINE() 公式函数。仅针对飞书表格。 |
-| [Lark Sheet Float Image](references/lark-sheets-float-image.md) | 管理飞书表格中的浮动图片。当用户需要在表格中插入浮动图片、调整图片位置和大小、查看已有浮动图片、删除图片时使用。也适用于"插入图片"、"添加 logo"、"放一张图"等场景。注意：如果用户需要将图片嵌入到某个单元格内部（单元格图片），请阅读 lark-sheets-write-cells。仅针对飞书表格。 |
+| [Lark Sheet Workbook](references/lark-sheets-workbook.md) | 管理飞书表格的工作簿结构（子表列表及元数据）。当用户提到"看看这个表格有什么"、"表格结构"、"有哪些 sheet"、"新建一个 sheet"、"删除这个工作表"、"重命名"、"复制一份"、"移动到前面"时使用。 |
+| [Lark Sheet Sheet Structure](references/lark-sheets-sheet-structure.md) | 管理飞书表格的子表结构与布局。适用场景：查看行高、列宽、隐藏行列、合并单元格等布局信息，以及"插入一行"、"删除这列"、"隐藏行"、"冻结表头"、行列分组（大纲折叠/展开）等操作。行列大纲仅在用户明确提到"行分组"、"列分组"、"大纲"、"outline"时才触发，"按XXX分组"等数据分组场景请使用 lark-sheets-pivot-table。如需在表尾追加数据，应先通过此 skill 插入行，再通过 lark-sheets-write-cells 写入。 |
+| [Lark Sheet Read Data](references/lark-sheets-read-data.md) | 读取飞书表格中的单元格数据。当用户需要"看看数据"、"分析数据"、"统计/汇总"时使用；也适用于需要查看公式、样式、批注等详细信息的场景。 |
+| [Lark Sheet Search & Replace](references/lark-sheets-search-replace.md) | 在飞书表格中搜索和替换文本，支持限定范围、大小写匹配、精确匹配、正则表达式。当用户需要"查找"、"搜索"、"定位"某个值，或"替换"、"批量修改文本"、"把 A 改成 B"时使用。不要用于理解表格结构（应读取数据）、不要用于数据分析（应读取数据后计算）、不要把用户操作动作中的关键词（如"汇总金额""统计数量"）当作搜索词。 |
+| [Lark Sheet Write Cells](references/lark-sheets-write-cells.md) | 向飞书表格的指定区域批量写入值、公式、样式、批注或单元格图片。适用场景：填写数据、设置公式、修改格式、添加批注、嵌入单元格图片（如需操作浮动图片，请使用 lark-sheets-float-image）；若只需把一块 CSV 批量铺到表格上（值或公式，不带样式/批注），直接使用 `+csv-put` 更短更快。追加数据需先通过 lark-sheets-sheet-structure 插入行列。 |
+| [Lark Sheet Range Operations](references/lark-sheets-range-operations.md) | 对飞书表格中指定区域执行结构性操作（不涉及写入单元格数据值）。适用场景：清除内容或格式（"清空"、"删除内容"、"去掉格式"）、合并/取消合并单元格、调整行高列宽（"加宽列"、"自适应列宽"）、移动/复制/填充/排序数据（"移动数据"、"复制到"、"自动填充"、"按某列排序"）。写入单元格数据请使用 lark-sheets-write-cells。 |
+| [Lark Sheet Batch Update](references/lark-sheets-batch-update.md) | 将多个飞书表格写入操作合并为一次批量执行，按顺序依次完成。适合需要连续执行多个写入操作的场景（如先修改结构再写入数据）。 |
+| [Lark Sheet Chart](references/lark-sheets-chart.md) | 管理飞书表格中的图表（柱形图、折线图、饼图、条形图、面积图、散点图、组合图、雷达图等）。当用户需要创建图表、修改图表样式或数据源、查看已有图表配置、删除图表时使用。也适用于用户提到"数据可视化"、"画个图"、"趋势分析"、"对比图"、"占比分析"、"做个图表"等数据可视化相关场景。 |
+| [Lark Sheet Pivot Table](references/lark-sheets-pivot-table.md) | 管理飞书表格中的数据透视表。当用户需要创建透视表、修改透视表的行列字段/聚合方式/筛选条件、查看已有透视表配置、删除透视表时使用。也适用于用户提到"分组汇总"、"交叉分析"、"按XXX统计"、"按字段分组"、"再分下组"、"多维分析"、"数据透视"等场景。 |
+| [Lark Sheet Conditional Format](references/lark-sheets-conditional-format.md) | 管理飞书表格中的条件格式规则（重复值高亮、单元格值比较、数据条、色阶、排名、自定义公式等）。当用户需要创建条件格式、修改已有规则的范围或样式、查看当前条件格式配置、删除规则时使用。也适用于用户提到"高亮"、"标红"、"颜色标记"、"数据条"、"色阶"、"条件样式"等场景。 |
+| [Lark Sheet Filter](references/lark-sheets-filter.md) | 管理飞书表格中的筛选器（filter）。当用户需要筛选数据（按文本/数值/颜色/日期条件过滤行）、查看已有筛选配置、修改或删除筛选器时使用。也适用于"只看"、"筛选出"、"仅保留符合条件的"等场景。 |
+| [Lark Sheet Filter View](references/lark-sheets-filter-view.md) | 管理飞书表格中的筛选视图（filter view）。当用户需要"建一个 XX 视图"、"保存这个筛选状态"、"切换不同筛选"、维护一个 sheet 上多份独立筛选配置时使用。视图与筛选器（filter）相互独立，可在同一 sheet 共存；视图的隐藏行仅在用户进入该视图时本地生效，不影响其他协作者。 |
+| [Lark Sheet Sparkline](references/lark-sheets-sparkline.md) | 管理飞书表格中的迷你图（折线迷你图、柱形迷你图、胜负迷你图）。当用户需要在单元格内嵌入小型图表来展示数据趋势时使用。也适用于"趋势线"、"单元格内图表"、"迷你图"等场景。注意：不等同于被禁用的 SPARKLINE() 公式函数。 |
+| [Lark Sheet Float Image](references/lark-sheets-float-image.md) | 管理飞书表格中的浮动图片。当用户需要在表格中插入浮动图片、调整图片位置和大小、查看已有浮动图片、删除图片时使用。也适用于"插入图片"、"添加 logo"、"放一张图"等场景。注意：如果用户需要将图片嵌入到某个单元格内部（单元格图片），请阅读 lark-sheets-write-cells。 |
 
 ## 公共 flag 速查
 
@@ -100,18 +107,18 @@ metadata:
 **公共四件套** = `--url` / `--spreadsheet-token` / `--sheet-id` / `--sheet-name`，分成两组 XOR，**每组都必须给且只能给一个**（XOR = 二选一必填，不是"可选"）：
 
 1. **spreadsheet 定位（必填）**：`--url` 与 `--spreadsheet-token` 二选一，**必须给其中之一**。两个都不给 → 校验报错 `specify at least one of --url or --spreadsheet-token`；两个都给 → 互斥冲突。
-   - **`--url` 只解析 `/sheets/` 与 `/spreadsheets/` 两种链接**（从路径里抽出 token；也可以直接把裸 token 传给 `--spreadsheet-token`）。其它形态的链接不会被解析成表格 token。
-   - ⚠️ **`/wiki/` 知识库链接不能直接当表格定位用**：wiki 链接背后可能是电子表格，也可能是文档 / 多维表格等其它类型，`--url` **不会**自动把 wiki token 解析成 spreadsheet token，直接传会失败。必须先把它解析成真实文档 token —— `lark-cli wiki +node-get --node-token "<wiki 链接或 token>"`，确认返回的 `obj_type` 为 `sheet` 后，取其 `obj_token` 作为 `--spreadsheet-token` 传入（解析细节见 [`../lark-wiki/SKILL.md`](../lark-wiki/SKILL.md)）。
-   - **例外**：`+workbook-create` 是新建一个还不存在的表格，**不接受任何 spreadsheet / sheet 定位 flag**（只有 `--title` / `--folder-token` / `--headers` / `--values`）。
+   - **`--url` 解析 `/sheets/`、`/spreadsheets/` 与 `/wiki/` 三种链接**（从路径里抽出 token；也可以直接把裸 token 传给 `--spreadsheet-token`）。其它形态的链接不会被解析成表格 token。
+   - **`/wiki/` 知识库链接可直接传 `--url`**：会自动定位到链接背后的电子表格；若该链接背后不是电子表格（而是文档 / 多维表格等），则报错。
+   - **例外**：`+workbook-create`（新建表 + 可选写入数据）与 `+workbook-import`（把本地文件导入为新表）都产出一张**还不存在**的表格，**不接受任何 spreadsheet / sheet 定位 flag**——`+workbook-create` 只有 `--title` / `--folder-token` / `--values` / `--styles` / `--sheets`，`+workbook-import` 只有 `--file`（必填）/ `--folder-token` / `--name`。
 2. **sheet 定位（公共四件套 shortcut 必填）**：`--sheet-id` 与 `--sheet-name` 二选一，**必须给其中之一**。两个都不给 → 校验报错 `specify at least one of --sheet-id or --sheet-name`。
    - ⚠️ **不确定 sheet 名时禁止直接猜 `Sheet1`**：除非用户对话明确说出 sheet 名 / id，或上下文（之前的工具调用 / URL 锚点 `?sheet=xxx`）已经出现过具体值，否则**第一步先调 `+workbook-info --url "..."`**（或 `--spreadsheet-token`）拿 `sheets[].sheet_id` / `sheets[].title` 列表再选。中文环境下子表常叫"数据" / "Sheet"（无数字）/ "工作表 1" / 业务名，猜 `Sheet1` 大概率撞 `sheet not found`，比先查多耗一次失败调用 + 重试。
    - ⚠️ **`--range` 里的 `Sheet1!` 前缀不能替代 sheet 定位**：即使写了 `--range 'Sheet1!A1:B2'`，仍**必须**额外传 `--sheet-id` 或 `--sheet-name`，否则照样报上面的错。
-   - ⚠️ **A1 reference 含 `!`**（`--source` / `--range` / `--ranges`）**：shell session 起手先 `set +H`** 关 bash history expansion，否则 `"Sheet1!A1"` 会被拦成 `event not found`；含特殊字符（`-` / 空格 / 非 ASCII）的 sheet 名还要内部 single-quote 包，如 `--source "'Sales-2025'!A1:D100"`。
+   - ⚠️ **A1 reference 含 `!`**（`--source` / `--range` / `--ranges`）**：整段用单引号包裹**，如 `--range 'Sheet1!A1:B2'`——单引号能挡住 bash 的 history expansion（`!` 被拦成 `event not found`；双引号挡不住；别改用 `set +H`，原因见下方「复合 JSON / 大入参」）。sheet 名含特殊字符（`-` / 空格 / 非 ASCII）需在内部按 A1 标准再包一层单引号时，用 `'\''` 转义保持外层单引号，如 `--source ''\''Sales-2025'\''!A1:D100'`。
    - **例外**：徽章标为 `_公共：URL/token（无 sheet 定位）…_` 的 shortcut（如 `+workbook-info` / `+workbook-export` / `+batch-update` / `+dropdown-update|delete` / `+cells-batch-set-style` / `+cells-batch-clear` / `+sheet-create`）**不接受也不需要** sheet 定位，只给一组 spreadsheet 定位即可。`+pivot-create` 用 `--target-sheet-id` / `--target-sheet-name`（XOR，可都不传，落点细节见 `lark-sheets-pivot-table`）。
 
 | Flag | Type | 必填 | 说明 |
 | --- | --- | --- | --- |
-| `--url` | string | 二选一必填（与 `--spreadsheet-token`） | spreadsheet URL |
+| `--url` | string | 二选一必填（与 `--spreadsheet-token`） | spreadsheet 或 wiki URL |
 | `--spreadsheet-token` | string | 二选一必填（与 `--url`） | spreadsheet token |
 | `--sheet-id` | string | 二选一必填（与 `--sheet-name`；仅公共四件套 shortcut） | 工作表 reference_id |
 | `--sheet-name` | string | 二选一必填（与 `--sheet-id`；仅公共四件套 shortcut） | 工作表名称 |
@@ -153,4 +160,6 @@ flag 帮助里标注支持 **Stdin** 的入参，当 payload 较大、含换行
 lark-cli sheets +cells-set --url "..." --sheet-name "Sheet1" --range "A1:B2" --cells - < "$TMPFILE"
 ```
 
+**参数含特殊字符（`!` / 引号 / 空格 / 非 ASCII）时，用单引号包裹该参数即可，不要起手 `set +H` 之类的 shell 开关来防转义。** `set +H`（关 bash history expansion）在 `sh` / `dash` 下是非法选项（`set: Illegal option -H`）、会让整条命令直接失败；而单引号挡得住 `!` 的 history expansion（否则报 `event not found`），对 bash 与 `sh` / `dash` 一致安全。参数本身含单引号、或 payload 较大时，按上文走 stdin。
+
 **`@file` 接绝对路径会被拒，且被拒后不要照报错提示做。** `@file` 出于安全只接受 cwd 下的相对路径，传 cwd 之外的绝对路径会被拒。此时报错会建议"先 cd 到目标目录，或改用相对路径"——**两条都不要照做**：cd 过去、或把临时文件写进用户项目目录，都会污染工作目录。正解是改用 stdin（`--<flag> - < 文件`）。

skills/lark-sheets/references/lark-sheets-batch-update.md

@@ -22,7 +22,7 @@
 
 当同一工具需要对多个区域重复调用时，**必须**改用 `+batch-update` 合并为单次请求——`+batch-update` 是原子提交（要么全成功要么整批回滚）；逐个调用非原子，中途失败会留下半成品。
 
-**`+dropdown-update` 的选项模式（`--options` / `--source-range` 二选一）+ 配色规则**（`--colors` 长度可短不能长、必须配 `--highlight=true` 才生效、不传按内置 10 色色板循环补色）见 [`lark-sheets-write-cells`](./lark-sheets-write-cells.md) 的「Dropdown 选项 + 配色」节，本 skill 不重复。`+dropdown-delete` 不涉及这些 flag。
+**`+dropdown-update` 的选项模式（`--options` / `--source-range` 二选一）+ 配色规则**（`--colors` 长度可短不能长、必须配 `--highlight=true` 才生效、不传按内置 10 色色板循环补色）见 [`lark-sheets-write-cells`](./lark-sheets-write-cells.md) 的「Dropdown 选项 + 配色」节，本文不重复。`+dropdown-delete` 不涉及这些 flag。
 
 ## Shortcuts
 

skills/lark-sheets/references/lark-sheets-chart.md

@@ -36,7 +36,8 @@
   - **默认情况（inline 模式）**：`refs` 范围**应包含表头行**（首行/首列即系列名），且范围要精确覆盖目标数据，不要多选或少选。
   - **合并标题行要跳过**：如果表格在表头上方存在合并的标题行（如"员工统计表"横跨多列的大标题），`refs` 必须跳过标题行、从真正的列标题行开始。例如表头在第 3 行、数据在第 4-20 行，则 `refs` 应为 `A3:G20` 而非 `A1:G20`。包含合并标题行会导致列名识别错误、表头被当作数据参与聚合计算。
   - **数据与表头分离时必须用 detached 模式**：当 `refs` 只覆盖完整数据的一个子集（按筛选/分组只画其中一段），而真正的语义表头在该子集之外时，**必须**设置 `snapshot.data.headerMode='detached'`：refs 仅传纯数据范围，维度名/系列名通过 `snapshot.data.dim1.serie.nameRef` / `snapshot.data.dim2.series[].nameRef` 指向真正的表头单元格。详见下文"硬性规则：数据与表头分离场景必须使用 detached 模式"。
-- **axes[].label 不接受 `format` / `number_format` 字段**：想给坐标轴数值加千分位、百分号等格式化时，不要在 `axes[i].label` 里传 `format` 或 `number_format`（schema 未定义，会报 `unexpected property "format" is not defined in schema`）。数值格式化统一在源数据单元格的 `cell_styles.number_format` 里设置（写 `+cells-set` 时），图表会沿用单元格格式。
+- **axes[].label 不接受 `format` / `number_format` 字段**：想给坐标轴数值加千分位、百分号等格式化时，不要在 `axes[i].label` 里传 `format` 或 `number_format`（schema 未定义，会报 `unexpected property "format" is not defined in schema`）。数值格式化统一在源数据单元格的 `cell_styles.number_format` 里设置（写 `+cells-set` 时），图表会沿用单元格格式。**日期轴同理**：横轴显示成 `45297` 这类 Excel 序列号，是因为源日期列没设日期格式——给源列设 `number_format="yyyy-mm-dd"` 后横轴才会显示成日期（反例：折线图横轴日期显示为序列号）。大数值轴显示科学计数法同理，给源列设整数 / 千分位格式（反例：透视表数值轴显示科学计数法）。
+- **轴口径要对齐用户要的指标**：用户要"占比 / 比例"时，**纵轴应是百分比**——用饼图，或柱 / 条形图设 `stack.percentage: true` 让纵轴变 %，并把数据源指向占比列 / 让数据标签显示百分比；不要交付纵轴仍是原始计数的图（反例：要求看各类占比，却用普通堆积柱、纵轴是 0–350 的人数而非百分比）。
 - **创建后必须验证**：图表创建后必须调用 `+chart-list` 验证配置是否正确
 
 > **⚠️ 硬性规则：当用户通过列标题名称（而非列索引）指定横轴/纵轴系列时，必须先读取表格首行（表头）来确定列名与列索引的对应关系，再设置 `dim1`/`dim2` 的 `index`。**
@@ -66,7 +67,7 @@
 >
 > **反向约束**：场景 A 下不要写 `nameRef`——首行命名已经生效，多写反而冗余。`nameRef` 仅在场景 B 下使用（且必填）。
 
-## ⚠️ chart 数据源引用 pivot 时必须排除总计行（高频致命错误）
+## ⚠️ chart 数据源引用 pivot 时必须排除总计行
 
 当 chart 要基于刚创建的 pivot 产物画图时，**禁止凭猜写 `refs`**。pivot 默认启用 `show_row_grand_total` / `show_col_grand_total`，产物最后一行/一列通常是"总计"。如果 `refs` 把总计行一并框进去：
 - **柱形图**末尾会多一根天文数字柱子（=所有数据求和），把其他柱子压扁到看不见
@@ -84,15 +85,17 @@
 
 1. **查尺寸**：`+workbook-info` 拿该 sheet 的 `row_count` / `column_count`（下文记为 rowCount / columnCount；`+sheet-info` 只返回布局，不含行列总数）。
 2. **估跨度**：默认单元格 **105 px 宽 × 27 px 高**，`needCols = ceil(width/105)`，`needRows = ceil(height/27)`。
-3. **校验**：`position.row + needRows ≤ rowCount` 且 `col_idx + needCols ≤ columnCount`（col 按 A=0、B=1、…、Z=25、AA=26… 换算）。
+3. **校验**：`position.row + needRows ≤ rowCount` 且 `col_idx + needCols ≤ columnCount`（`position.row` 为 **0-based**：首行 = `row:0`，与 A1 区间 / `+dim-insert --position` 的 1-based 行号不同；col 按 A=0、B=1、…、Z=25、AA=26… 换算）。
 4. **不够就先扩表**，二选一，禁止硬塞越界位置：
    - **优先**放数据下方空区：`position = {row: data_end_row + 2, col: "A"}`；
    - 否则先调 `+dim-insert`（`lark-sheets-sheet-structure`）扩行/列，再 create。
 
+⚠️ **图表落点禁止压在已有数据矩形内**——必须落在数据区**右侧或下方的空白**，否则图表浮层会遮挡原始数据被判失败（反例：折线图落在数据区中间，遮挡了下方原始数据）。
+
 **示例**：21 列 sheet 放 600×400 图 → `needCols=6, needRows=15`
 - ❌ `{row: 0, col: "W"}` — col=22 越界
 - ✅ `{row: 42, col: "A"}` — 放数据下方
-- ✅ 先 `+dim-insert --dimension column --start 21 --end 27`（在 U 列后插 6 列；U=index 20，after 即从 21 起），再放图到 `{row: 0, col: "V"}`
+- ✅ 先 `+dim-insert --position V --count 6`（在 V 列前插 6 列，即 U 列之后），再放图到 `{row: 0, col: "V"}`
 
 ## Shortcuts
 
@@ -147,9 +150,9 @@ _公共四件套 · 系统：`--yes`、`--dry-run`_
 _创建/更新的图表属性_
 
 **顶层字段**：
-- `position` (object) — 必填 { row: number, col: string }
+- `position` (object?) — 必填 { row: number, col: string }
 - `offset` (object?) — 可选 { row_offset?: number, col_offset?: number }
-- `size` (object) — 必填 { width: number, height: number }
+- `size` (object?) — 必填 { width: number, height: number }
 - `snapshot` (object?) — 图表快照配置 { title?: object, subTitle?: object, style?: object, legend?: oneOf, plotArea: object, …共 6 项 }
 
 ## Examples
@@ -164,24 +167,28 @@ _创建/更新的图表属性_
 
 > **`snapshot.data` 必填 `dim1.serie.index` 或 `dim2.series[].index` 之一**（1-based，对应 `refs.value` 范围内的列序）。schema 允许传空 `{}` 但 server 运行时强制：缺则被拒为 `snapshot.data.dim1.serie.index and dim2.series[].index are both missing; at least one must be set`，即便侥幸通过也只会渲染空图。
 
+> ⚠️ **含 `'Sheet'!` 前缀的 `--properties` 必须走 stdin 或 `@file`，不要用 inline 单引号**。`refs` / `nameRef` 里的 sheet 前缀带单引号（`'Sheet1'!A1`），若塞进 inline 的 `--properties '{...}'`，bash 会把内层那对单引号吃掉（sheet 名带空格还会被拆成多个词），JSON 直接被破坏。下面示例统一用 `--properties - <<'JSON' … JSON`（heredoc 定界符加引号 = 不做 shell 替换），或 `--properties @file.json`（`@` 只接 cwd 下相对路径）。
+
 最小可用列图（inline 模式：refs 含表头行）：
 
 ```bash
 lark-cli sheets +chart-create --url "https://example.feishu.cn/sheets/shtXXX" \
-  --sheet-name "Sheet1" --properties '{
-    "position":{"row":42,"col":"A"},
-    "size":{"width":600,"height":400},
-    "snapshot":{
-      "data":{
-        "refs":[{"value":"'Sheet1'!A1:B10"}],
-        "dim1":{"serie":{"index":1}},
-        "dim2":{"series":[{"index":2}]}
-      },
-      "plotArea":{"plot":{"type":"column"}}
-    }
-  }'
+  --sheet-name "Sheet1" --properties - <<'JSON'
+{
+  "position":{"row":42,"col":"A"},
+  "size":{"width":600,"height":400},
+  "snapshot":{
+    "data":{
+      "refs":[{"value":"'Sheet1'!A1:B10"}],
+      "dim1":{"serie":{"index":1}},
+      "dim2":{"series":[{"index":2}]}
+    },
+    "plotArea":{"plot":{"type":"column"}}
+  }
+}
+JSON
 
-# 走文件（推荐配置较多时）
+# 或落到 cwd 下相对路径文件再用 @file
 lark-cli sheets +chart-create --url "..." --sheet-name "Sheet1" --properties @chart-config.json
 ```
 
@@ -190,7 +197,8 @@ lark-cli sheets +chart-create --url "..." --sheet-name "Sheet1" --properties @ch
 饼图比 column / bar 更复杂：`sectors` 是 object，里面再包一个**单数** `sector` 数组——CLI 不替你 normalize，写错路径会被 server schema 直接拒。
 
 ```bash
-lark-cli sheets +chart-create --url "..." --sheet-name "Sheet1" --properties '{
+lark-cli sheets +chart-create --url "..." --sheet-name "Sheet1" --properties - <<'JSON'
+{
   "position":{"row":24,"col":"F"},
   "size":{"width":600,"height":450},
   "snapshot":{
@@ -208,7 +216,8 @@ lark-cli sheets +chart-create --url "..." --sheet-name "Sheet1" --properties '{
       "dim2":{"series":[{"index":2,"aggregateType":"sum"}]}
     }
   }
-}'
+}
+JSON
 ```
 
 **数据与表头分离（必须用 `detached` + `nameRef`）**：
@@ -216,7 +225,8 @@ lark-cli sheets +chart-create --url "..." --sheet-name "Sheet1" --properties '{
 场景：周度销量明细表，真实表头在第 1 行（A1=周次、C1=订单量、D1=退款量），数据按 B 列"店铺"分段；用户只要"3 号店"那一段（第 11–17 行）。
 
 ```bash
-lark-cli sheets +chart-create --url "..." --sheet-name "Sheet2" --properties '{
+lark-cli sheets +chart-create --url "..." --sheet-name "Sheet2" --properties - <<'JSON'
+{
   "position":{"row":7,"col":"F"},
   "size":{"width":600,"height":360},
   "snapshot":{
@@ -233,7 +243,8 @@ lark-cli sheets +chart-create --url "..." --sheet-name "Sheet2" --properties '{
       ]}
     }
   }
-}'
+}
+JSON
 ```
 
 约束：

skills/lark-sheets/references/lark-sheets-conditional-format.md

@@ -34,7 +34,7 @@
 - **日期/空值比较必须防空**：用户说"过期的标红"时，除了 `TODAY()`，公式必须排除空单元格，否则空白格也会被误判为"早于今天"而全表标红。正确公式：`=AND(E1<>"", E1<=TODAY())`；错误公式：`=E1<=TODAY()`（空值会被当作 0 判为过期）
 - **公式条件注意引用方式**：自定义公式条件中的单元格引用需要根据实际场景选择相对/绝对引用（如 `=E1<=TODAY()` 而非 `=$E$1<=TODAY()`，后者只比较一个格）
 
-⚠️ **用户明确要求"辅助列+条件格式"两步走时，禁止用 `expression` 绕过（高频致命错误）**：当用户说以下任意一种表达时，必须按两步走（先建辅助列 → 再基于辅助列做条件格式），**禁止**直接用一个 `rule_type: "expression"` 公式一步完成：
+⚠️ **用户明确要求"辅助列+条件格式"两步走时，禁止用 `expression` 绕过**：当用户说以下任意一种表达时，必须按两步走（先建辅助列 → 再基于辅助列做条件格式），**禁止**直接用一个 `rule_type: "expression"` 公式一步完成：
 
 - "**增加辅助列**，再/然后标记……"
 - "**先计算/判断** XX **是否** YY，**再**标记……"

skills/lark-sheets/references/lark-sheets-core-operations.md

@@ -2,7 +2,7 @@
 
 ## 概览
 
-面向"已有飞书表格"的核心工作流，核心原则：**先了解，再分析或写入，最后验证**。本文是方法论总纲；具体工具的参数细节、边界陷阱在对应子 skill，本文用指针引到那里，不重复展开。
+面向"已有飞书表格"的核心工作流，核心原则：**先了解，再分析或写入，最后验证**。本文是方法论总纲；具体工具的参数细节、边界陷阱在对应 reference，本文用指针引到那里，不重复展开。
 
 **三份「通用方法与规范」如何分工**（都不含 shortcut，按主题单一归属）：
 
@@ -12,21 +12,21 @@
 
 > **下面的铁律对所有任务一律生效**，即使你是被索引直接路由进 visual 或 formula 而没经过本文——编辑类任务务必先回到这里过一遍铁律。
 
-## 铁律（所有编辑类任务必须满足，子 skill 不得放宽）
+## 铁律（所有编辑类任务必须满足，各 reference 不得放宽）
 
-1. **最小改动**：除用户明示要改的单元格 / 列外，原表其它单元格、行列结构、Sheet 名、合并区、格式必须 1:1 保持。中间结果优先放原数据**右侧**；会与原数据混淆或要承载透视表 / 图表时才**新建空白 Sheet**。**禁止**擅自删 / 改名 / 隐藏 / 移动**已存在**的 Sheet（新建允许，节制使用）。
-2. **真实写回 + 回读校验**：交付必须是对在线表格的真实写入，并 `+csv-get` / `+cells-get` / `+<对象>-list` 回读校验。**严禁**只在文本里描述"已完成"、用普通公式 / 文本假装结构化对象、或只给占位而无真实写入。
-3. **读全再写，禁止只探前 N 行**：批量填充 / 补齐 / 修正类任务必须先确认**真实数据末行**再写，否则会漏写表尾（高频致命错误）。完整的"按表格形态分流读取 + `current_region` / `has_more` 兜底 + 真实末行确认"流程见 `lark-sheets-read-data` 的「确定数据范围的正确流程」。
-4. **公式优先于硬编码**：能用飞书公式表达的计算（总计 / 占比 / 增长率 / 提取 / 查找等）一律写公式而非静态值，源数据变化才能自动重算。用户口头的"分列 / 排序 / 求和 / 提取"也要落地为公式或原生工具（SORT / `TEXTBEFORE` / `MID` / 透视表 等）。Excel 公式迁移、数组语义、不支持函数清单一律以 `lark-sheets-formula-translation` 为唯一权威。
+1. **最小改动**：除用户明示要改的单元格 / 列外，原表其它单元格、行列结构、Sheet 名、合并区、格式必须 1:1 保持。中间结果优先放原数据**右侧**；会与原数据混淆或要承载透视表 / 图表时才**新建空白 Sheet**。**禁止**擅自删 / 改名 / 隐藏 / 移动**已存在**的 Sheet（新建允许，节制使用）。**改写 / 转换类任务要精确圈定适用行列**：只对任务真正要求的对象做变换，**不该转的行 / 列保持原值 1:1**（典型反例：要求"统一翻译"时把本就是中文、应原样保留的评论也重新翻译；要求"改写某列格式"时连原始测量值也一并改动 → 应保留的原文被篡改）。
+2. **真实写回 + 回读校验**：交付必须是对在线表格的真实写入，并 `+csv-get` / `+cells-get` / `+<对象>-list` 回读校验。**严禁**只在文本里描述"已完成"、用普通公式 / 文本假装结构化对象、或只给占位而无真实写入。**收尾前必须确认产物文件真实存在 / 可导出**——别在没真正生成产物时只凭文本"已完成"就结束（反例：文本称已完成，实际没生成产物文件，等于没交付）。
+3. **读全再写，禁止只探前 N 行**：批量填充 / 补齐 / 修正类任务必须先确认**真实数据末行**再写，否则会漏写表尾。完整的"按表格形态分流读取 + `current_region` / `has_more` 兜底 + 真实末行确认"流程见 `lark-sheets-read-data` 的「确定数据范围的正确流程」。
+4. **公式优先于硬编码**：能用飞书公式表达的计算（总计 / 占比 / 增长率 / 提取 / 查找等）一律写公式而非静态值，源数据变化才能自动重算。用户口头的"分列 / 排序 / 求和 / 提取"也要落地为公式或原生工具（SORT / `TEXTBEFORE` / `MID` / 透视表 等）。Excel 公式迁移、数组语义、不支持函数清单一律以 `lark-sheets-formula-translation` 为唯一权威。**即使用户没说"联动 / 自动更新"，凡是可由表内其它单元格推导的派生值（年龄=当年-出生年、占比=本类数/总数、达标=阈值判断、排名、各类分组汇总）默认就必须用公式**——用户默认期望派生列能随源数据重算，**离线 Python / 脚本算完写静态值，即便当前数值正确，改了源数据也不会自动更新，等于没满足"派生"的本意**（反例：年龄、月度汇总、占比、分组求和等派生列写死值，源数据一改结果就过时）。
 5. **续写 / 扩展必须继承样式**：续写、补齐、复制区块、新增行列时，**禁止**只读值只写值。必须连带 `cell_styles` + `border_styles` + 合并 + 行高一起继承。完整继承清单与做法见 `lark-sheets-write-cells` 的「新增列 / 新增行的样式继承」（`border_styles` 四边最易漏）。
 6. **多步写入优先 `+batch-update`**：多个连续写入、或同一工具对多个区域重复调用（多次 merge / resize / cells-set），必须合并为单次原子 `+batch-update`。语义与不可嵌套的限制见 `lark-sheets-batch-update`。
 7. **分组汇总必须用透视表**："按 X 统计 Y / 分组汇总 / 各部门数量金额"必须用 `+pivot-{create|update|delete}`（推荐省略 sheet_id 自动新建子表），**禁止**用 SUMIF / COUNTIF 或本地脚本覆盖原表替代。
-8. **任务拆成可验证 checklist**：落地前把指令拆成所有"独立可验证子要点"，每点一个 `assert`，全部通过才交付：多维度操作（按部门一/二/三级排序）每维一个 assert；多目标（删 N 行）每目标一个；多格式兼容（多种日期格式）每种至少一个样本；范围类（A1:H11 加边框）起 / 末行 / 末列三边界都核。只完成第一个要点（只排一级、只删 1 行）属违规。
+8. **任务拆成可验证 checklist**：落地前把指令拆成所有"独立可验证子要点"，每点一个 `assert`，全部通过才交付：多维度操作（按部门一/二/三级排序）每维一个 assert；多目标（删 N 行）每目标一个；多格式兼容（多种日期格式）每种至少一个样本；范围类（A1:H11 加边框）起 / 末行 / 末列三边界都核。只完成第一个要点（只排一级、只删 1 行）属违规。**题面 / 表头里写明的格式规范也是子要点**：表头注明"需标注某字段"就必须给对应单元格加规定前缀并逐条 assert 前缀存在（反例：漏加规定前缀，该要点即不达标）；"相同编号连续行合并"必须遍历所有相同编号组全部合并（反例：只合并了其中一部分组）。
 9. **全量处理要前置断言条数**：翻译 / 打标 / 批量公式落地等逐条任务，落地前把"预期处理条数"硬编码进代码，处理完 `assert actual == expected`。**严禁**输出"已完成前 N 条，剩余将继续"的半成品。
 
 ## 推荐工作流程
 
-1. **规划 skill 清单**：开工前一次性列出本任务要读的子 skill（避免读一个调一个），本轮已读过的不重复读。本 skill + `lark-sheets-workbook` 几乎每次都要。
+1. **规划 reference 清单**：开工前一次性列出本任务要读的 reference（避免读一个调一个），本轮已读过的不重复读。本文 + `lark-sheets-workbook` 几乎每次都要。
 2. **了解结构**：先 `+workbook-info` 拿子表列表 / 行列数 / 冻结位置（不可猜测，猜错会越界覆盖）；涉及合并 / 隐藏 / 分组 / 行高列宽再用 `lark-sheets-sheet-structure` 的 `+sheet-info`。
 3. **读取数据（按任务类型选路径，细则见 `lark-sheets-read-data`）**：
 
@@ -37,7 +37,7 @@
    | 需要公式 / 样式 / 批注 | C：`+cells-get` |
    | 续写 / 扩展 / 完善已有内容 | D：`+csv-get` 看结构 + `+cells-get` 读源区样式 + `+sheet-info --include row_heights,merges`（见铁律 5） |
 
-   **【高频致命错误】** 对"完善 / 补齐 / 填空"类任务用路径 B 探 10 行就写入，实测会漏写表尾多行。写入前必须按 `lark-sheets-read-data`「确定数据范围的正确流程」确认真实数据末行。按关键字定位区域用 `lark-sheets-search-replace` 的 `+cells-search`。
+   **注意**：对"完善 / 补齐 / 填空"类任务用路径 B 探 10 行就写入，实测会漏写表尾多行。写入前必须按 `lark-sheets-read-data`「确定数据范围的正确流程」确认真实数据末行。按关键字定位区域用 `lark-sheets-search-replace` 的 `+cells-search`。
 
 4. **理解数据语义（写入前必做）**：读表头 + 3-5 行样本确认各列含义与格式（文本 / 数字 / 日期 / 混合）；写公式前先分析样本值格式模式再选提取策略；建透视表前先列清"行字段=分组维度、值字段=聚合指标"。需求模糊时（如"加入加减乘除"未说逻辑）基于表头与已有公式推断，不确定就问用户，禁止臆造业务逻辑。
 
@@ -67,6 +67,7 @@
 - **喂给 CLI 的 CSV / JSON 用 UTF-8、不带 BOM**：BOM 会污染首格的值或触发 `invalid character` 解析错；脚本读写文件时显式指定 `encoding='utf-8'`。
 - **临时文件交给运行时的标准库**：用 `tempfile.gettempdir()` / `os.tmpdir()` 等取临时目录，不要硬编码固定路径；放在用户项目目录之外。
 - **命令失败先读错误再调整**：同一条命令失败后不要原样重发；先看 stderr 的报错（参数错误、缺依赖、解释器不可用等）定位原因，再决定换写法、补依赖或退回原生工具。
+- **写回的必须是纯单元格值，禁止把"值+样式标注"串当值写回**：本地脚本或某些 xlsx 解析库会把单元格渲染成 `甲方支行(V-Align: bottom)` 这种"值(样式)"字符串，CSV 字段还可能带包裹双引号。回写前必须**剥离括号样式标注、去掉残留引号**，只写原始值——否则样式描述会变成单元格的字面文本污染原数据（反例：排序后单元格值里被写进 `(V-Align: bottom)` 这类样式后缀文本，末尾还多一个双引号）。**排序本身优先用 `+range-sort` 原生工具**，不要"读出来本地排完再整列写回"，从根上避免这类回写污染。
 
 ## 公式策略
 

skills/lark-sheets/references/lark-sheets-filter-view.md

@@ -109,6 +109,17 @@ lark-cli sheets +filter-view-create --url "..." --sheet-id "$SID" \
   --properties '{"rules":[{"column_index":"C","conditions":[{"type":"number","compare_type":"greaterThan","values":[100]}]}]}'
 ```
 
+**`conditions[].type` × `compare_type` 取值**（`type` 决定可用的 `compare_type`；两者均必填）：
+
+| `type` | 可用 `compare_type` | `values` |
+|---|---|---|
+| `text` | `contains` / `doesNotContain` / `beginsWith` / `doesNotBeginWith` / `endsWith` / `doesNotEndWith` / `equals` / `notEquals` | 字符串数组 |
+| `number` | `equal` / `notEqual` / `greaterThan` / `greaterThanOrEqual` / `lessThan` / `lessThanOrEqual` / `between` / `notBetween` | 数值（或数值字符串）数组；`between` / `notBetween` 传两个边界 |
+| `multiValue` | `equal` / `notEqual` | 字符串数组（精确匹配其中任一值） |
+| `color` | `backgroundColor` / `foregroundColor` | 不传 `values`（按单元格颜色筛选） |
+
+> ⚠️ `text` 用 `equals` / `notEquals`（**带 s**），`number` / `multiValue` 用 `equal` / `notEqual`（**不带 s**）——别混。完整 schema 跑 `+filter-view-create --print-schema --flag-name properties`。
+
 > `--range` **必须覆盖表头行**（如 `A1:F1000`），不能只包含数据行；`--view-name` 重名时服务端自动改名。
 
 ### `+filter-view-update`

skills/lark-sheets/references/lark-sheets-filter.md

@@ -102,6 +102,17 @@ lark-cli sheets +filter-create --url "..." --sheet-id "$SID" \
   --properties '{"rules":[{"column_index":"B","conditions":[{"type":"multiValue","compare_type":"equal","values":["北京","上海"]}]}]}'
 ```
 
+**`conditions[].type` × `compare_type` 取值**（`type` 决定可用的 `compare_type`；两者均必填）：
+
+| `type` | 可用 `compare_type` | `values` |
+|---|---|---|
+| `text` | `contains` / `doesNotContain` / `beginsWith` / `doesNotBeginWith` / `endsWith` / `doesNotEndWith` / `equals` / `notEquals` | 字符串数组 |
+| `number` | `equal` / `notEqual` / `greaterThan` / `greaterThanOrEqual` / `lessThan` / `lessThanOrEqual` / `between` / `notBetween` | 数值（或数值字符串）数组；`between` / `notBetween` 传两个边界 |
+| `multiValue` | `equal` / `notEqual` | 字符串数组（精确匹配其中任一值） |
+| `color` | `backgroundColor` / `foregroundColor` | 不传 `values`（按单元格颜色筛选） |
+
+> ⚠️ `text` 用 `equals` / `notEquals`（**带 s**），`number` / `multiValue` 用 `equal` / `notEqual`（**不带 s**）——别混。完整 schema 跑 `+filter-create --print-schema --flag-name properties`。
+
 ### `+filter-update`
 
 > ⚠️ update 是覆盖式：`--properties` 中传新 `rules` 会替换旧组。如只想加一条，要带上已有的全部条件再追加。必填 `--range`。

skills/lark-sheets/references/lark-sheets-float-image.md

@@ -1,12 +1,13 @@
 # Lark Sheet Float Image
 
-> **单元格图片 vs 浮动图片**：飞书表格有两种图片类型，请根据需求选择正确的工具：
-> - **单元格图片**：图片嵌入在单元格内部，随单元格移动，属于单元格内容的一部分。→ 使用 `+cells-set`，在 `rich_text` 中设置 `type: "embed-image"`（见 lark-sheets-write-cells）。
-> - **浮动图片**（本 Skill）：图片悬浮在单元格上方，可自由指定位置、大小和层级，不属于任何单元格的内容。→ 使用本 Skill 的 `+float-image-{create|update|delete}`。
+> **选浮动图还是单元格图？只看一条**：这张图是不是**属于某条记录、要随那行一起排序 / 筛选 / 增删**？
+> - **是 → 单元格图片**（不在本 reference）：嵌进单元格、随行走。用 `+cells-set-image`（或 `+cells-set` 的 `rich_text` + `type: "embed-image"`，见 lark-sheets-write-cells）。典型：凭证 / 证件照 / 商品图 / 头像 / 二维码 / 每行配图；话里带「对应 / 每行 / 每条 / 这列」等绑定词即属此类。
+> - **否 → 浮动图片**（本 reference）：自由摆放、不绑数据的装饰 / 标识（logo / 水印 / 封面大图 / banner）。
+> - ⚠️ 别凭"浮动图位置尺寸更好控制 / 更熟"就选它——那是按操作便利选，不是按场景选；用浮动图承载"对应某记录"的图会在增删行 / 排序后错位。
 
 ## 真对象硬约束
 
-当用户要求"插入图片 / 添加 logo / 放一张图"时，**必须**通过 `+float-image-{create|update|delete}`（浮动图片）或 `+cells-set` 的 `embed-image`（单元格图片）创建真实的图片对象。**禁止**只在文本回复中给出图片链接 / 描述图片内容代替插入。判断标准：交付后 `+float-image-list` 或单元格 `rich_text` 必须能读到该图片对象。
+当用户要求"插入图片 / 添加 logo / 放一张图"时，**必须**通过 `+float-image-{create|update|delete}`（浮动图片）或 `+cells-set-image` / `+cells-set` 的 `embed-image`（单元格图片）创建真实的图片对象。**禁止**只在文本回复中给出图片链接 / 描述图片内容代替插入。判断标准：交付后 `+float-image-list` 或单元格 `rich_text` 必须能读到该图片对象。
 
 ## 使用场景
 
@@ -20,7 +21,7 @@
 典型工作流：先读取现有浮动图片了解配置 → 执行创建/更新/删除 → **必须再次读取验证结果**。
 
 **常见配置错误（必须注意）**：
-- **单元格图片 vs 浮动图片选择错误**：如果用户希望图片嵌入单元格内部（随单元格移动），应使用 `+cells-set` 的 `rich_text` + `embed-image`，而非本 Skill
+- **单元格图片 vs 浮动图片选择错误（最易选错）**：图与某条记录一一对应、要随行排序 / 筛选 / 增删时，应走 `+cells-set-image`（见顶部判别），用浮动图会错位。
 - **图片位置参数要精确**：锚点单元格的行列索引和偏移量决定了图片位置，设置不当会导致图片遮挡数据
 - **创建后必须验证**：调用 `+float-image-list` 确认图片位置和大小正确
 
@@ -30,7 +31,7 @@
 - `--image-token`：复用**已存在**的图片 file_token。常见来源：① `+float-image-list` 返回的 `image_token`（适合"换皮不换位置"复用同一张图）；② `+cells-set-image` 成功返回里的 `file_token`（它也是 `sheet_image` 上传句柄）。适合"同一张图复用到多处"，省去重复上传。
 - `--image-uri`：图片 reference_id（image URI），由系统自动转 file_token。
 
-> ⚠️ **`--image` 仅 `+float-image-create` 支持**。`+float-image-update` 换图仍只接受 `--image-token` / `--image-uri`，而且**图片源是 update 唯一可省的部分**——三者全不传则保留原图。但 `--image-name` / `--position-{row,col}` / `--size-{width,height}` 在 update 时和 create 一样**必填**（`manage_float_image` 工具强制要求这套核心字段，且 `+float-image-list` 不回传 `image_name` 供 CLI 回填）。要在 update 里换一张本地新图，先用 `+cells-set-image` 上传到任意临时单元格、从返回取 `file_token`，再把它传给 update 的 `--image-token`。
+> ⚠️ **`--image` 仅 `+float-image-create` 支持**。`+float-image-update` 换图仍只接受 `--image-token` / `--image-uri`，而且**图片源是 update 唯一可省的部分**——三者全不传则保留原图。但 `--image-name` / `--position-{row,col}` / `--size-{width,height}` 在 update 时和 create 一样**必填**（`+float-image-update` 强制要求这套核心字段，且 `+float-image-list` 不回传 `image_name` 供 CLI 回填）。要在 update 里换一张本地新图，先用 `+cells-set-image` 上传到任意临时单元格、从返回取 `file_token`，再把它传给 update 的 `--image-token`。
 
 ## Shortcuts
 
@@ -129,7 +130,7 @@ lark-cli sheets +float-image-create --url "..." --sheet-id "$SID" \
 
 ### `+float-image-update`
 
-> **update ≈ create，只有图片源可省**：`manage_float_image` 工具的 update 要求和 create 相同的核心字段——`--image-name`、`--position-{row,col}`、`--size-{width,height}` **全部必填**；唯一区别是**图片源（`--image-token` / `--image-uri`）可以全省**，省略即保留原图。这**不是**"只发改动字段"的 patch：缺任一核心字段会被工具拒绝（`+float-image-list` 不回传 `image_name`，CLI 无法替你回填）。
+> **update ≈ create，只有图片源可省**：`+float-image-update` 的 update 要求和 create 相同的核心字段——`--image-name`、`--position-{row,col}`、`--size-{width,height}` **全部必填**；唯一区别是**图片源（`--image-token` / `--image-uri`）可以全省**，省略即保留原图。这**不是**"只发改动字段"的 patch：缺任一核心字段会被拒绝（`+float-image-list` 不回传 `image_name`，CLI 无法替你回填）。
 >
 > 推荐流程：先 `+float-image-list --float-image-id <id>` 回读当前 position / size，再带上 `--image-name` 和完整的 position / size 调一次 `+float-image-update`。
 

skills/lark-sheets/references/lark-sheets-formula-translation.md

@@ -209,7 +209,7 @@ Excel：`{=A1:A10*B1:B10}`（Ctrl+Shift+Enter 输入）
 
 飞书日期序列：`0 = 1899-12-30`，`1 = 1899-12-31`，没有 Excel 的 1900 年闰年兼容问题。
 
-**高频错误写法（不要用）：**
+**错误写法（不要用）：**
 
 - `=DAY(B2-A2)` ✗ — 差值会被当成日期序列号再拆字段
 - `=MONTH(B2-A2)` ✗

skills/lark-sheets/references/lark-sheets-range-operations.md

@@ -40,6 +40,10 @@
 
 **典型反例**：默认列宽 11 但内容含 12+ 字符的中文 / 含单位的数值（如 `109.10μmol/L`）/ 长数字未设 `number_format` 显示为科学计数法 —— 用户在结果表里看不到完整原值。
 
+**打印场景控制总宽（用户说"适合打印 / A4 / 打印范围"时必做）**：扩单列宽防截断的同时，**所有列宽之和要落在纸张可打印宽度内**——A4 横向约 ≤ 102 个半角字符（约 1000px），纵向约 ≤ 70 个字符。超宽时不要无限加宽，改用 `cell_styles.word_wrap="auto-wrap"` + 调高行高，或缩窄非关键列，让整表在一页内（反例：总列宽远超 A4 可打印宽度，且长文本行高不够被截断）。
+
+**只加宽承载新内容的列，不改动原有列的列宽**：列宽自适应**只针对新增 / 真正放不下新内容的列**；原表已有列的列宽**禁止重新计算、禁止缩小**——即便你估算的"理想宽度"与原值不同，只要原内容没被截断就不要动它。无差别地把所有列重设一遍宽度（哪怕只 ±1）都属于破坏原文件视觉格式（反例：填完数据后顺手把原有列的列宽从 16 改成 17，与原附件不一致，破坏了原视觉格式）。
+
 **⚠️ 合并单元格安全操作规则**（`+cells-{merge|unmerge}` 必读）：
 
 1. **先读后写**：操作前必须用 `+sheet-info --include merges` 或 `+cells-get` 识别已有合并区域（特征：多个连续单元格中只有左上角有值，其余为空）。
@@ -192,7 +196,7 @@ _排序条件列表（仅 sort 操作）_
 
 ## Examples
 
-> ⚠️ 本 skill 派生的 shortcut 跨 3 个分组：`+rows-resize` / `+cols-resize` → 工作表，`+cells-*` → 单元格，`+range-*` → 区域。skill 视角统一在这里讲解。
+> ⚠️ 本 reference 派生的 shortcut 跨 3 个分组：`+rows-resize` / `+cols-resize` → 工作表，`+cells-*` → 单元格，`+range-*` → 区域。这里统一从区域操作视角讲解。
 
 公共四件套：所有 shortcut 顶部排列 `--url` / `--spreadsheet-token` / `--sheet-id` / `--sheet-name`（XOR）。
 

skills/lark-sheets/references/lark-sheets-read-data.md

@@ -13,61 +13,66 @@
 
 预探后必须在公式 / 筛选条件里用 `IFERROR` / `IFS` / 提取数值的辅助列处理所有变体；不能为了通过 head(10) 的样本就直接落地。一旦设计的逻辑只覆盖 sample 中出现的格式，就属于违规。
 
+⚠️ **大数字（15 位以上的身份证 / 参考号 / 流水号）做去重 / 比较时禁止用 `+csv-get` 的显示值**：`+csv-get` 返回的是**格式化显示值**，15 位以上数字会被显示成 `1.04E+14` 这类科学计数法——多个本不相同的号在显示层全变成同一个 `1.04E+14`，拿去判重会**整列误判为重复**。比较 / 去重 / 匹配大数字时必须改用 `+cells-get`（取原始精确值）或把该列读为文本，禁止用 csv-get 的科学计数显示值（反例：大批长参考号被显示成科学计数后，互不相同的号全变成同一个值，被当成整列重复并错误高亮）。
+
 ## 使用场景
 
-读取。从飞书表格中读取单元格数据。本 reference 覆盖 3 个 shortcut，按读取目的选择：
+读取。从飞书表格中读取单元格数据。本 reference 覆盖 4 个 shortcut，按读取目的选择：
 
 | 读取目的 | 用这个 shortcut | 数据去向 | 说明 |
 |---------|----------------|---------|------|
-| 快速查看纯值数据、批量处理 | `+csv-get` | 对话上下文 | 返回 CSV 文本（加 `--rows-json` 改为结构化 rows `{row_number, values:{列字母→值}}`）；大表请按 `--range` 行窗口分批读（截断时看 `has_more`） |
+| 快速查看纯值数据、批量处理 | `+csv-get` | 对话上下文 | 返回 CSV 文本（每行带 `[row=N]` 前缀）；大表请按 `--range` 行窗口分批读（截断时看 `has_more`） |
+| 按列类型结构化读出（喂 DataFrame / round-trip 回 `+table-put`） | `+table-get` | 对话上下文 | 返回 typed 协议（`columns:[列名]` + `data` + `dtypes`/`formats` + `range`），输出形状对齐 pandas split；可一行 `pd.DataFrame(sheet["data"], columns=sheet["columns"]).astype(sheet["dtypes"])` 还原 DataFrame，或直接 round-trip 回 `+table-put`。不带 `--range` 时读**完整 used range**（跨过表中部空行 / 空列），每个子表回传实际读取范围 `range` 供完整性校验 |
 | 查看公式、样式、批注、数据验证 | `+cells-get` | 对话上下文 | 返回单元格完整信息，token 开销较大 |
 | 查看某区域的下拉框（数据验证）选项 | `+dropdown-get` | 对话上下文 | 返回该 A1 范围已配置的下拉列表选项 |
 
 **选择原则**：
 - 只看值或做数据处理 → `+csv-get`；大表分批读取，避免一次拉全表撑爆上下文
-- 要结构化、按 `row_number` / 列字母定位的输出 → `+csv-get --rows-json`（默认 CSV 串更省 token，超大表批量仍用默认）
+- 要按列类型结构化读出（喂 DataFrame / round-trip 回 `+table-put`）→ `+table-get`
 - 需要公式/样式/批注 → `+cells-get`
 - 只想知道某区域下拉框有哪些选项 → `+dropdown-get`
 
-⚠️ 超大数据请走"`+csv-get` 按 `--range` 行窗口（如 `A1:Z500` / `A501:Z1000` …）分批读到本地文件 + 本地脚本处理 + `+csv-put` 分批回写"。
+⚠️ **大数据优先落盘、别灌进上下文**：`+csv-get` / `+cells-get` 都受调用方 Bash / 终端的单命令 stdout 输出上限约束（常见默认约 30000 字符，超过会被截断或转存为文件）。纯值分析优先 `+csv-get --format csv` 按 `--range` 行窗口（`A1:Z500` / `A501:Z1000` …）分批重定向到文件 + 本地脚本处理 + `+csv-put` 分批回写；若确实要让结果直接进上下文又不想触发转存，给任一命令把 `--max-chars`（默认 500000）调小到略低于该上限（如 `25000`），CLI 改为优雅截断 + `has_more` 分页。
 
 **`+csv-get` 返回值核心设计**：
 - `annotated_csv` — **CSV 数据唯一入口**。每一逻辑行前加 `[row=N] ` 前缀（N = 真实表格行号）。任何需要行号的下游操作（合并、写入、清空、格式化、插入/删除、条件格式、筛选、图表/透视表范围、搜索替换等），**行号一律直接从 `[row=N]` 读取**。若需要纯 CSV（如喂给本地脚本做解析），去前缀即可：`line.replace(/^\[row=\d+\] /, '')`。
 - `col_indices` — **定位列字母唯一入口**。在表头中找到目标字段是第 j 个（0-based），用 `col_indices[j]` 取列字母。**禁止手数逗号**——列数超过 10 时极易 off-by-one（例如把 W 误判为 X）。
 - `row_indices` — 程序化引用的备用数组。LLM 推理请用 `annotated_csv` 的前缀，不要查这个数组里的 index（把行号当数值用容易心算出错）。
-- `current_region` — 从请求范围扩展到被空行空列包围的连续数据区域（等价于 Excel Ctrl+Shift+*），适合先读少量行探表头、同时获知整表实际范围。
+- `current_region` — 从请求范围扩展到被空行空列包围的连续数据区域（等价于 Excel Ctrl+Shift+*），适合先读少量行探表头。⚠️ 它**遇表中部整行空行 / 整列空列就截断**，可能小于真实数据范围（漏掉空行之后的行）；**不能**直接当整表末行用，判断整表是否读全要拿 `+workbook-info` 的物理 `row_count` / `column_count` 当上界交叉核对（见下方「按 row_count 盲读空行」与「确定数据范围的正确流程」）。
 
 注意：
 
 - `+csv-get` 和 `+cells-get` 支持分页/截断，注意检查 `has_more` / `truncated` 标志；使用 `+cells-get` 时，在读取 `cells` 之前还必须先看 `warning_message`，并用每个 range 的 `actual_range` / `row_indices` / `col_indices` 判断真实位置
-- 隐藏行列默认包含在返回结果中（`--skip-hidden=false`），如需只看可见数据设为 `true`
+- 隐藏行列默认包含在返回结果中（`--skip-hidden=false`），如需只看可见数据设为 `true`。读取原语本身不标注哪些行列被隐藏：若要识别隐藏区间（以决定是否过滤、或如何解读混入的隐藏数据），用 `+sheet-info --include hidden_rows,hidden_cols` 取隐藏行列集合，再结合 `+csv-get` / `+cells-get` 返回的 `row_indices` / `col_indices` 判断每行 / 每列是否隐藏
 
 **常见配置错误（必须注意）**：
-- **全量读取导致上下文溢出（高频致命错误）**：不要对大表（数百行以上）直接用 `+csv-get` 或 `+cells-get` 读取全部数据到上下文。大表场景必须分批读取：用 `--range` 切行窗口逐块读（`+csv-get` / `+cells-get` 单次返回量由 `--max-chars` 自动兜底，截断时返回 `has_more`）；过大时考虑导出到本地文件后用脚本处理再分批回写
+- **全量读取导致上下文溢出**：不要对大表（数百行以上）直接用 `+csv-get` 或 `+cells-get` 读取全部数据到上下文。大表场景必须分批读取：用 `--range` 切行窗口逐块读（`+csv-get` / `+cells-get` 单次返回量由 `--max-chars` 自动兜底，截断时返回 `has_more`）；过大时考虑导出到本地文件后用脚本处理再分批回写
 - **了解结构 ≠ 读取全量数据**：探表不用读全表，但必须同时探两个方向的表头：
   - **横向（列头）**：先读前几行，且**列范围必须覆盖所有列**——用 `+workbook-info` 拿总列数，`range` 末列填到最后一列（例如总列数是 N，则 `range: "A1:[列N]10"`）。列范围截短会遗漏右侧字段、后续写入列定位错误。
-  - **纵向（行标）**：若左侧 1-2 列是行标签（日期/类别/编号枚举每行含义，典型交叉表/透视布局），**必须再读 `A:A` 或 `A:B` 把行标列读到底**，拿全部行标。只读前几行会看不全表尾的行，导致批量写入漏改——这是"只改前 N 行、其余未更新"的主要成因。扁平列表（每行独立记录、列是字段）可跳过这一步，但仍要靠 `current_region` 兜底。
+  - **纵向（行标）**：若左侧 1-2 列是行标签（日期/类别/编号枚举每行含义，典型交叉表/透视布局），**必须再读 `A:A` 或 `A:B` 把行标列读到底**，拿全部行标。只读前几行会看不全表尾的行，导致批量写入漏改——这是"只改前 N 行、其余未更新"的主要成因。扁平列表（每行独立记录、列是字段）可跳过这一步，但仍要按下方「确定数据范围的正确流程」用 `+workbook-info` 的物理 `row_count` 交叉核对末行（`current_region` 遇空行会截断，不能单独兜底）。
   - 数据量大或会进入上下文上限时，分批读 + 本地处理 + 分批回写，不要一口气拉全表到上下文。
 - **`+cells-get` 滥用**：当只需要数据值时，使用 `+csv-get`（token 开销约为 `+cells-get` 的 1/5）。只有确实需要公式、样式或批注时才用 `+cells-get`
 - **忽略分页标志**：读取返回 `has_more=true` 时，说明还有更多数据。如果任务需要完整数据，必须继续分页读取，不能只处理第一页就开始写入
-- **直接按 `+cells-get` 返回二维数组下标推导真实位置（高频错误）**：`ranges[n].cells[i][j]` 里的 `i/j` 只是返回数组下标，不等于真实表格行列。定位真实行号必须用 `ranges[n].row_indices[i]`，定位真实列字母必须用 `ranges[n].col_indices[j]`；若 `--skip-hidden=true`、请求范围越界被裁剪，或最后一行是部分返回，错误地自己数下标会立刻错位
-- **CSV 行号计数错误（高频致命错误）**：`+csv-get` 返回的 CSV 遵循 RFC 4180 标准，被双引号 `"..."` 包裹的字段中的换行符属于**字段内容的一部分**（即单元格内换行），不代表新的一行。计算行号时必须按**逻辑记录**计数，而非按物理换行符 `\n` 计数
-- **手动数列确定列号（高频致命错误）**：禁止通过在 CSV 表头中手动数逗号/字段来确定目标列的列字母。当列数超过 10 时，手动计数极易产生 off-by-one 偏移（例如把 W 列误判为 X 列）。**必须使用 `col_indices`**：先在 CSV 表头中找到目标字段名是第 j 个字段（0-based），再用 `col_indices[j]` 获取该列的实际列字母
-- **用数据列的值推导行号（高频致命错误，常被巧合掩盖）**：CSV 中常见"序号 / ID / 编号 / No."等形似行号的列，其值与实际表格行号**没有任何绑定关系**——序号可能跳号（1,2,3,5,6...）、可能从非 1 开始、可能有重复或被中途重置。此规则适用于**所有需要行号的下游操作**：合并单元格、区间写入/清空/格式化、插入/删除行、条件格式范围、筛选器范围、图表数据源、透视表范围、搜索替换范围等等——**凡是要把行号填进任何工具参数的场景，行号一律从 `annotated_csv` 中目标行开头的 `[row=N]` 前缀直接读取**，禁止用"序号=行号"、"表头占 1 行所以数据从第 2 行开始"、"第 N 个序号就在第 N+1 行"等心算，也禁止先心算再"事后核对"。**危险特征**：前几十行中序号恰好等于表格行号（典型成因：表头 +1 与一次跳号 -1 的偏移互相抵消形成巧合），模型一旦把这个巧合当作规律，会在后续所有行沿用；而中间再出现跳号时，从该行起整块区域全部错位，且错位不自查很难发现。**正确工作流**：①在 `annotated_csv` 里定位目标逻辑行（按字段内容匹配）；②直接读取该行开头的 `[row=N]` 前缀得到真实表格行号；③把这个行号填进下游工具参数。区间操作时，起始行用 start 行的 `[row=N]`、结束行用 end 行的 `[row=N]`。**自检**：动手前，在 `annotated_csv` 靠后位置再抽 1~2 行，核对 `[row=N]` 是否与首列"序号"一致——不一致（典型：`[row=57] 58,...`）即说明有跳号/隐藏行，更要严格从 `[row=N]` 取值，不要被序号列迷惑
-- **按 `row_count` 盲读空行（高频低效）**：`+workbook-info` 的 `row_count` 是 sheet 的**网格物理行数**（常是 200 / 1000 等默认值），不是数据末行；按它把 `--range` 拉到 `S200`（实际数据可能只到 `S32`）会读回大片空行，浪费上下文又干扰判断。真实数据末行以 `+csv-get` 返回的 `current_region` 为准（它就是数据边界），再按下方「确定数据范围的正确流程」确认末行。
-- **current_region 当作纯数据范围（高频致命错误）**：`current_region` 返回的是从请求范围向四周扩展到被空行空列包围的**连续非空区域**，等价于 Excel 的 Ctrl+Shift+\*。它包含该区域内**所有非空行**——不仅包含数据行，还可能包含标题行、汇总行（如"总计"）、签名行（如"编制人/审批人"）、脚注等非数据内容。**严禁直接将 `current_region` 的末尾行作为数据范围的结束行**。正确做法见下方「确定数据范围的正确流程」
+- **直接按 `+cells-get` 返回二维数组下标推导真实位置**：`ranges[n].cells[i][j]` 里的 `i/j` 只是返回数组下标，不等于真实表格行列。定位真实行号必须用 `ranges[n].row_indices[i]`，定位真实列字母必须用 `ranges[n].col_indices[j]`；若 `--skip-hidden=true`、请求范围越界被裁剪，或最后一行是部分返回，错误地自己数下标会立刻错位
+- **CSV 行号计数错误**：`+csv-get` 返回的 CSV 遵循 RFC 4180 标准，被双引号 `"..."` 包裹的字段中的换行符属于**字段内容的一部分**（即单元格内换行），不代表新的一行。计算行号时必须按**逻辑记录**计数，而非按物理换行符 `\n` 计数
+- **手动数列确定列号**：禁止通过在 CSV 表头中手动数逗号/字段来确定目标列的列字母。当列数超过 10 时，手动计数极易产生 off-by-one 偏移（例如把 W 列误判为 X 列）。**必须使用 `col_indices`**：先在 CSV 表头中找到目标字段名是第 j 个字段（0-based），再用 `col_indices[j]` 获取该列的实际列字母
+- **用数据列的值推导行号（常被巧合掩盖）**：CSV 中常见"序号 / ID / 编号 / No."等形似行号的列，其值与实际表格行号**没有任何绑定关系**——序号可能跳号（1,2,3,5,6...）、可能从非 1 开始、可能有重复或被中途重置。此规则适用于**所有需要行号的下游操作**：合并单元格、区间写入/清空/格式化、插入/删除行、条件格式范围、筛选器范围、图表数据源、透视表范围、搜索替换范围等等——**凡是要把行号填进任何工具参数的场景，行号一律从 `annotated_csv` 中目标行开头的 `[row=N]` 前缀直接读取**，禁止用"序号=行号"、"表头占 1 行所以数据从第 2 行开始"、"第 N 个序号就在第 N+1 行"等心算，也禁止先心算再"事后核对"。**危险特征**：前几十行中序号恰好等于表格行号（典型成因：表头 +1 与一次跳号 -1 的偏移互相抵消形成巧合），模型一旦把这个巧合当作规律，会在后续所有行沿用；而中间再出现跳号时，从该行起整块区域全部错位，且错位不自查很难发现。**正确工作流**：①在 `annotated_csv` 里定位目标逻辑行（按字段内容匹配）；②直接读取该行开头的 `[row=N]` 前缀得到真实表格行号；③把这个行号填进下游工具参数。区间操作时，起始行用 start 行的 `[row=N]`、结束行用 end 行的 `[row=N]`。**自检**：动手前，在 `annotated_csv` 靠后位置再抽 1~2 行，核对 `[row=N]` 是否与首列"序号"一致——不一致（典型：`[row=57] 58,...`）即说明有跳号/隐藏行，更要严格从 `[row=N]` 取值，不要被序号列迷惑
+- **`row_count` 与 `current_region` 都不能单独定末行**：`+workbook-info` 的 `row_count` 是 sheet 的**网格物理行数**（常是 200 / 1000 等默认值），通常**大于**真实数据末行——直接按它把 `--range` 拉到 `S200` 会读回大片空行，浪费上下文。反过来，`+csv-get` 返回的 `current_region` 是从锚点扩展、被空行空列围住的连续块，**遇表中部整行空行就截断**，可能**小于**真实数据范围（漏掉空行之后的行，典型反例：1–80 行有数据、81 行空、82 行起还有数据，`current_region` 只到 80，82 行起整段被漏读）。正确做法：把 `row_count` 当**上界**、`current_region` 当**起点参考**，在二者之间按下方「确定数据范围的正确流程」确认真实末行（含跨过中间空行的核对），不要只信其一。
+- **current_region 当作纯数据范围**：`current_region` 返回的是从请求范围向四周扩展到被空行空列包围的**连续非空区域**，等价于 Excel 的 Ctrl+Shift+\*。它包含该区域内**所有非空行**——不仅包含数据行，还可能包含标题行、汇总行（如"总计"）、签名行（如"编制人/审批人"）、脚注等非数据内容。**严禁直接将 `current_region` 的末尾行作为数据范围的结束行**。正确做法见下方「确定数据范围的正确流程」
 
 ### 确定数据范围的正确流程（排序、筛选、批量写入等操作前必做）
 
-当后续操作需要精确的数据范围（如排序、筛选、删除、批量写入）时，仅靠 `current_region` 探测到的范围是不够的——必须同时确认数据的**起始行**和**结束行**。具体步骤：
+当后续操作需要精确的数据范围（如排序、筛选、删除、批量写入）时，仅靠 `current_region` 探测到的范围是不够的——它**两头都可能不准**：表中部有整行空行时会被截断（末行偏小、漏数据），表尾有汇总 / 签名行时又会偏大。必须同时确认数据的**起始行**和**结束行**。具体步骤：
 
 1. **确认起始行**：读取前 5~10 行，识别表头行位置，数据起始行 = 表头行 + 1
-2. **确认结束行**（关键步骤，不可跳过）：读取 `current_region` 末尾附近的若干行（建议读取末尾 5~10 行），逐行检查内容，排除非数据行：
-   - **汇总行**：内容为"合计"、"总计"、"小计"、"总计:"等
-   - **签名/审批行**：内容为"编制人"、"审核人"、"部门负责人"等
-   - **空行或分隔行**：整行为空或仅有边框
-   - **备注/脚注行**：注释性文字、说明文字等
-3. **最终数据范围** = 起始行 ~ 最后一条有效数据行（排除非数据行）
+2. **确认结束行**（关键步骤，不可跳过）：
+   - **先防截断（漏数据）**：拿 `+workbook-info` 的物理 `row_count` 当上界，与 `current_region` 末行对比。若 `current_region` 末行 **远小于** `row_count`（差出很多空间），不要直接采信——在 `current_region` 末行之后再探一段（如往下读到 `row_count`，或分段扫到首个连续空白区），确认空行之后确实没有数据；典型反例：`row_count=327`、`current_region` 只到第 80 行，第 81 行空、82 行起还有数据，只读到 80 就漏了一大段。
+   - **再排尾部非数据行**：读取确认到的末行附近若干行（建议末尾 5~10 行），逐行排除：
+     - **汇总行**：内容为"合计"、"总计"、"小计"、"总计:"等
+     - **签名/审批行**：内容为"编制人"、"审核人"、"部门负责人"等
+     - **空行或分隔行**：整行为空或仅有边框
+     - **备注/脚注行**：注释性文字、说明文字等
+3. **最终数据范围** = 起始行 ~ 最后一条有效数据行（跨过中间空行、排除尾部非数据行）
 
 **示例**：`current_region` 返回 `A1:N51`，读取 Row 48~51 发现：
 
@@ -83,6 +88,7 @@
 | `+cells-get` | read | 单元格 |
 | `+dropdown-get` | read | 对象 |
 | `+csv-get` | read | 单元格 |
+| `+table-get` | read | 单元格 |
 
 ## Flags
 
@@ -94,7 +100,7 @@ _公共四件套 · 系统：`--dry-run`_
 | --- | --- | --- | --- |
 | `--range` | string | required | A1 范围，如 `A1:F10`（不带 sheet 前缀；用 `--sheet-id` / `--sheet-name` 指定 sheet） |
 | `--include` | string_slice | optional | 要返回的信息类别，逗号分隔多个（可选值：`value` / `formula` / `style` / `comment` / `data_validation`） |
-| `--max-chars` | int | optional | 防爆，默认 200000（隐藏 flag：不在 `--help` 列出，但可正常传入） |
+| `--max-chars` | int | optional | 单次返回字符上限，默认 500000（兜底防爆）。大数据通常宜重定向落盘做分析；仅当要让结果直接进上下文、又不触发文件转存时才调小（如 25000），以 has_more 分页 |
 | `--skip-hidden` | bool | optional | 跳过隐藏行列，默认 `false` |
 
 ### `+dropdown-get`
@@ -112,10 +118,21 @@ _公共四件套 · 系统：`--dry-run`_
 | Flag | Type | 必填 | 说明 |
 | --- | --- | --- | --- |
 | `--range` | string | required | A1 范围，如 `A1:F30`（不带 sheet 前缀；用 `--sheet-id` / `--sheet-name` 指定 sheet） |
-| `--max-chars` | int | optional | 防爆，默认 200000（隐藏 flag：不在 `--help` 列出，但可正常传入） |
+| `--max-chars` | int | optional | 单次返回字符上限，默认 500000（兜底防爆）。大数据通常宜重定向落盘做分析；仅当要让结果直接进上下文、又不触发文件转存时才调小（如 25000），以 has_more 分页 |
 | `--include-row-prefix` | bool | optional | 是否在每行前加 `[row=N]` 前缀，默认 `true` |
 | `--skip-hidden` | bool | optional | 跳过隐藏行列，默认 `false` |
-| `--rows-json` | bool | optional | 返回结构化 rows（`{row_number, values:{列字母→值}}`）而非 CSV 文本，默认 `false` |
+
+### `+table-get`
+
+_公共：URL/token（无 sheet 定位） · 系统：`--dry-run`_
+
+| Flag | Type | 必填 | 说明 |
+| --- | --- | --- | --- |
+| `--sheet-id` | string | optional | 只读该子表（按 id）；省略则读所有子表 |
+| `--sheet-name` | string | optional | 只读该子表（按名）；省略则读所有子表 |
+| `--range` | string | optional | 读取的 A1 范围；省略则读每个子表的完整 used range（会跨过表中部的整行空行 / 整列空列，不会被截断） |
+| `--no-header` | bool | optional | 把第一行当数据而非表头（列名取 col1/col2 …） |
+| `--dataframe-out` | string | optional | 以一份 Arrow IPC 文件（Feather v2）格式输出 typed 表格，替代默认的 JSON 输出。用 `@<path>` 传文件或 `-` 写二进制 stdout（同其他 binary I/O flag 的约定）。是 `+table-put` / `+workbook-create` 入口 `--dataframe` 的镜像 —— pandas 端 `pd.read_feather("x.arrow")` 或 `pd.read_feather(io.BytesIO(stdout))` 一行读回。仅支持单 sheet：必须给 `--sheet-id` 或 `--sheet-name`；读整本 workbook 仍走默认 JSON。列类型沿用 typed 读回（string/number/date/bool）；`number_format` 以 Arrow Field metadata 保留，Arrow 文件可直接喂回 `+table-put --dataframe`。 |
 
 ## Examples
 
@@ -137,20 +154,11 @@ lark-cli sheets +csv-get --spreadsheet-token shtXXX --sheet-name "销售明细"
 
 - `annotated_csv` — 含 `[row=N]` 前缀的 CSV 主入口
 - `col_indices` / `row_indices` — 列字母 / 行号映射数组
-- `current_region` — 自动扩展到非空连续区域的 A1 范围。它是**真实数据边界**，**优先于 `+workbook-info` 的 `row_count`**（`row_count` 是网格物理行数，常是 200 / 1000 等默认值、远大于实际数据；按它盲读会拉回大片空行）
-- `has_more` — 是否截断；截断后续读用 `--range` 接着读
+- `current_region` — 从锚点扩展到被空行空列包围的连续区域的 A1 范围。⚠️ **它不是整表真实边界**：遇表中部整行空行 / 整列空列会截断、可能小于真实数据范围；表尾的汇总 / 签名 / 脚注又可能让它大于纯数据范围。判断整表是否读全须拿 `+workbook-info` 的物理 `row_count` 当上界交叉核对（见上方「`row_count` 与 `current_region` 都不能单独定末行」）
+- `row_count` / `col_count` — **本次返回的行 / 列数**（= `actual_range` 的尺寸，随 `--range` 变），**不是整表物理总行列数**；整表物理尺寸取 `+workbook-info`
+- `has_more` — 当前 `--range` 是否因 `--max-chars` 被截断（截断后续读接着用 `--range`）；它**只反映本次 range 内是否读完**，`has_more=false` **不代表整表已读全**（range 之外的数据不在判断内）
 
-**加 `--rows-json`：返回结构化 rows（而非 CSV 字符串）**
-
-```bash
-lark-cli sheets +csv-get --url "https://example.feishu.cn/sheets/shtXXX" --sheet-name "Sheet1" --range "A1:G20" --rows-json
-```
-
-`--rows-json` 下的输出契约（替换 `annotated_csv` / `col_indices` / `row_indices`）：
-
-- `rows` — 数组，每元素 `{row_number, values}`。`row_number` 是真实表格行号（整数，下游需要行号的操作直接取它）；`values` 按**列字母** key（如 `values["D"]`，绝对列字母）。**所有逻辑行都在 `rows` 里**。引号内换行已解析进单元格值，无需自己按 RFC-4180 拆行。
-- `data_not_fully_read` — **仅当没读全时出现**：`{read_through_row, data_extends_through_row, unread_rows, reread_range}`。出现即表示真实数据超出本次读取范围；批量写入前必须按 `reread_range` 重读全区，否则漏行。
-- 其余字段（`current_region` / `actual_range` / `has_more`）同上。
+> 要按列类型结构化读出（喂 DataFrame、或 round-trip 回 `+table-put`）用 `+table-get`（见下）；`+csv-get` 给的是带 `[row=N]` 前缀的纯值快照，下游需要行号/列坐标时直接从前缀与 `col_indices` 取。
 
 ### `+cells-get`
 
@@ -164,6 +172,91 @@ lark-cli sheets +cells-get --url "https://example.feishu.cn/sheets/shtXXX" --she
 
 > ⚠️ 调用方在 `cells[i][j]` 中**不能**用下标推真实行列：必须读 `ranges[n].row_indices[i]` / `ranges[n].col_indices[j]`。
 
+### `+table-get`（飞书 → DataFrame，类型保真读出）
+
+`+table-put`（写入侧，见 write-cells reference）的镜像：把表格读回与 `--sheets` 完全同构的 typed 协议（`sheets[]` + `columns:[列名]` + `data:[[行]]` + `dtypes:{列名:pandas_dtype}` + `formats?:{列名:number_format}` + `range`），可直接喂回 `+table-put` 或一行还原 DataFrame。
+
+**默认（不带 `--range`）读取整张子表的完整 used range**：会跨过表中部的整行空行 / 整列空列，覆盖到真实数据边界。每个子表都回传实际读取的 `range`（如 `A1:F10`）——`+table-get` 不返回分页 / 截断标志，这个 `range` 是判断是否读全的唯一信号：拿它和源 xlsx 行列数、关键末行 / 末日期交叉核对，确认读取完整。仍要精确控制范围时显式传 `--range`。
+
+列类型从每列 `number_format` 推断（日期格式→`date`/`datetime64[ns]`、数值→`number`/`float64`、bool→`bool`），`date` 列的序列号转回 ISO `yyyy-mm-dd`——日期、数字往返不丢类型。**列类型只在该列所有非空值一致时才定（`number` / `date` / `bool`）；一列混了类型（如数字列混入「暂无」、日期列混入裸数字）会降为 `string`（dtypes 输出 `object`），让 `dtypes` 与 `data` 里每个值自洽——能 round-trip 回 `+table-put`、不让 pandas `astype` 崩。降级是无损的（脏值原样保留为文本）；若要把零星脏值转成数值列，交给调用方在 pandas 侧做（`to_numeric(errors='coerce')`），那里原始值仍在、可追溯。** 默认读所有子表、第一行当表头（`--no-header` 把首行当数据、列名取 `col1` / `col2` …）。
+
+```bash
+# 默认读所有子表 → sheets[]（与 +table-put 的 --sheets 同构，可喂回或转 DataFrame）
+lark-cli sheets +table-get --url "<表URL>"
+# 可选：--sheet-name / --sheet-id 限定只读某一个子表（不给则读全部）
+lark-cli sheets +table-get --url "<表URL>" --sheet-name "销售"
+```
+
+#### 输出 → DataFrame（2 行 helper）
+
+输出形状对齐 pandas split：`columns` 是列名数组、`data` 是二维数据、`dtypes` 是 `{列名: pandas_dtype_str}` 映射。直接喂给 `pd.DataFrame(...).astype(...)` 就能一次性还原所有列类型（不必逐列 `to_datetime` / `to_numeric`），写入侧 `df_to_sheet` 的镜像 helper：
+
+```python
+import pandas as pd
+def sheet_to_df(sheet):
+    return pd.DataFrame(sheet["data"], columns=sheet["columns"]).astype(sheet["dtypes"])
+
+# 单 sheet
+df = sheet_to_df(out["data"]["sheets"][0])
+
+# 多 sheet——按名字取
+sheets = {s["name"]: sheet_to_df(s) for s in out["data"]["sheets"]}
+df_sales = sheets["销售"]
+```
+
+> 显示格式（千分位、百分比、自定义日期）在 `sheet["formats"]`，pandas 不消费；改完数据 round-trip 回去时透传给 `+table-put` 即可，飞书侧显示不变。
+
+#### `--dataframe-out`（Arrow IPC / Feather v2 二进制读出）
+
+`--dataframe-out` 是 `+table-put` 入口 `--dataframe` 的镜像：把 typed 读回直接编码成 Arrow IPC 文件，pandas 端一行 `pd.read_feather()` 读回——省掉 JSON 解析 + `astype(dtypes)`，列类型 / `number_format` 走 Arrow schema + Field metadata 保真。**仅支持单 sheet**（Arrow 文件一 schema 容器），必须给 `--sheet-id` 或 `--sheet-name`；读整本 workbook 仍走默认 JSON。
+
+```bash
+# 文件
+lark-cli sheets +table-get --url "<表URL>" --sheet-name "销售" --dataframe-out @./out.arrow
+# binary stdout（不落盘）
+lark-cli sheets +table-get --url "<表URL>" --sheet-name "销售" --dataframe-out -
+```
+
+```python
+import io, pandas as pd, subprocess
+
+# 1) 文件
+subprocess.run(["lark-cli","sheets","+table-get","--url",URL,
+                "--sheet-name","销售","--dataframe-out","@./out.arrow"], check=True)
+df = pd.read_feather("./out.arrow")
+
+# 2) stdin/stdout 管道（不落盘）—— 跟 --dataframe 写入侧对称的一行
+res = subprocess.run(["lark-cli","sheets","+table-get","--url",URL,
+                      "--sheet-name","销售","--dataframe-out","-"],
+                     capture_output=True, check=True)
+df = pd.read_feather(io.BytesIO(res.stdout))
+```
+
+> `number_format` 进 Arrow Field metadata（key=`number_format`），Arrow 文件可以直接喂回 `+table-put --dataframe` round-trip 写回，types / formats 一路保真。
+
+#### round-trip：读 → 改 → 写回（写读对偶）
+
+`sheet_to_df` 和 write-cells reference 里的 `df_to_sheet` 是一对镜像 helper，round-trip 三段读 / 改 / 写各一行：
+
+```python
+import json, subprocess
+# 1. 读
+out = json.loads(subprocess.check_output(
+    ["lark-cli","sheets","+table-get","--url",URL,"--sheet-name","销售"]))
+sheet = out["data"]["sheets"][0]
+df = sheet_to_df(sheet)
+
+# 2. 改（pandas 操作）
+df["营收"] = df["营收"] * 1.1
+
+# 3. 写回（formats 是飞书侧显示格式，pandas 不消费，透传保留显示）
+payload = {"sheets": [df_to_sheet(df, sheet["name"], formats=sheet.get("formats"))]}
+subprocess.run(["lark-cli","sheets","+table-put","--url",URL,"--sheets","-"],
+               input=json.dumps(payload).encode(), check=True)
+```
+
+`sheet_to_df(sheet)` 消费 `(columns, data, dtypes)`，`df_to_sheet(df, name, formats=...)` 重新生成同样三个字段——读 / 写完全对偶，只有 `formats` 需要手工透传一次。
+
 ### Validate / DryRun / Execute 约束
 
 - `Validate` 阶段只做 XOR 检查、Enum 合法性、防爆参数上限校验；**禁止**联网（如不能用 `--sheet-name` 提前去查 `sheet-id`）。

skills/lark-sheets/references/lark-sheets-sparkline.md

@@ -17,7 +17,7 @@
 
 **常见配置错误（必须注意）**：
 - **数据源范围要精确**：迷你图的数据源范围必须与实际数据行列精确对应，范围偏移会导致图形展示错误
-- **不要与 SPARKLINE() 公式混淆**：飞书表格的 `SPARKLINE()` 公式函数已被禁用，迷你图只能通过本 Skill 的对象方式创建
+- **不要与 SPARKLINE() 公式混淆**：飞书表格的 `SPARKLINE()` 公式函数已被禁用，迷你图只能通过 `+sparkline-{create|update|delete}` 的对象方式创建
 - **创建后必须验证**：调用 `+sparkline-list` 确认迷你图配置正确
 
 ## Shortcuts

skills/lark-sheets/references/lark-sheets-visual-standards.md

@@ -24,6 +24,8 @@
 
 **差异化标注场景**：用户要求"重复行 / 异常值 / 重要项视觉区分"时，标注列 / 行必须设置与普通数据**显著不同**的 `cell_styles`（背景色 + 加粗 + 字体色至少改一项），不能与普通数据格式完全一致。
 
+**显式要求边框 / 表头 / 对齐时同样按上面标准落地**（不必等用户说"美化"）：① 用户说"给某矩形区域加边框"必须**整个矩形含表头行、数据行、汇总行全部加内外框**，落地后核起 / 末行、末列三边界（反例：要求加边框的区域实际无任何边框）；② **新建表头前先确认哪一行才是表头**——别把已有的第一行数据误当表头刷成蓝底白字，真正该加的表头列也要建出来（反例：把第一行数据误设成了表头样式）；③ 新增 / 编辑区域的字号必须与原表一致，禁止 13 号与 14 号、10 号与 11 号混杂（反例：新列字号与原表不一致）。
+
 ## 通用样式规范
 
 > 以下取值标准都在「最高优先级原则」的**继承原表风格 / 扩展而非覆盖**前提下生效：凡涉及"沿用原表"的条目，遵循该原则即可，本节不再逐条复述。
@@ -201,4 +203,3 @@ Step 3 — 微调收尾：`+batch-update` + `+rows-resize / +cols-resize` / `+ce
 - 合并区域样式只写左上角，不要对合并内的其他单元格重复写入样式。
 
 > 合并单元格完整的安全操作规则（含数据保护、样式占位等 5 条）见 `lark-sheets-range-operations` 的 `+cells-{merge|unmerge}` 章节。
-

skills/lark-sheets/references/lark-sheets-workbook.md

@@ -10,7 +10,7 @@
 
 ## 使用场景
 
-读写。管理工作簿结构。本 reference 覆盖 11 个 shortcut：
+读写。管理工作簿结构。本 reference 覆盖 14 个 shortcut：
 
 | 操作需求 | 使用工具 | 说明 |
 |---------|---------|------|
@@ -41,8 +41,11 @@
 | `+sheet-hide` | write | 工作簿 |
 | `+sheet-unhide` | write | 工作簿 |
 | `+sheet-set-tab-color` | write | 工作簿 |
+| `+sheet-hide-gridline` | write | 工作簿 |
+| `+sheet-show-gridline` | write | 工作簿 |
 | `+workbook-create` | write | 工作簿 |
 | `+workbook-export` | read | 工作簿 |
+| `+workbook-import` | write | 工作簿 |
 
 ## Flags
 
@@ -59,7 +62,7 @@ _公共：URL/token（无 sheet 定位） · 系统：`--dry-run`_
 | Flag | Type | 必填 | 说明 |
 | --- | --- | --- | --- |
 | `--title` | string | required | 新工作表名称 |
-| `--index` | int | optional | 插入位置；省略时附加到末尾 |
+| `--index` | int | optional | 插入位置（0-based）；省略时附加到末尾 |
 | `--row-count` | int | optional | 初始行数（默认 200，上限 50000） |
 | `--col-count` | int | optional | 初始列数（默认 20，上限 200） |
 
@@ -115,6 +118,18 @@ _公共四件套 · 系统：`--dry-run`_
 | --- | --- | --- | --- |
 | `--color` | string | required | Hex 色值如 `#FF0000`，传空 `""` 清除 |
 
+### `+sheet-hide-gridline`
+
+_公共四件套 · 系统：`--dry-run`_
+
+_仅含公共 / 系统 flag。_
+
+### `+sheet-show-gridline`
+
+_公共四件套 · 系统：`--dry-run`_
+
+_仅含公共 / 系统 flag。_
+
 ### `+workbook-create`
 
 _系统：`--dry-run`_
@@ -123,8 +138,10 @@ _系统：`--dry-run`_
 | --- | --- | --- | --- |
 | `--title` | string | required | 新 spreadsheet 标题 |
 | `--folder-token` | string | optional | 目标文件夹 token；省略时放在云空间根目录 |
-| `--headers` | string + File + Stdin（简单 JSON） | optional | 表头行 JSON 数组：`["列A","列B"]` |
-| `--values` | string + File + Stdin（简单 JSON） | optional | 初始数据 JSON 二维数组：`[["alice",95]]` |
+| `--values` | string + File + Stdin（简单 JSON） | optional | untyped 初始数据，一个 JSON 二维数组（表头并入第一行）：`[["列A","列B"],["alice",95]]`；值原样写入、类型由飞书自动识别，走与 --sheets 相同的分批 `+cells-set`；配 --styles 控制格式/颜色/合并/行列尺寸 |
+| `--sheets` | string + File + Stdin（复合 JSON） | optional | 建表后写入的 typed 表格协议 JSON（同 +table-put）：顶层 sheets 数组，每项 `{name, start_cell?, mode?, header?, allow_overwrite?, columns:["colA","colB",...], data:[[...]], dtypes?:{colA:pandasDtype, ...}, formats?:{colA:numberFormat, ...}}`。Agents 通常用 `{**json.loads(df.to_json(orient="split")), "dtypes": df.dtypes.astype(str).to_dict()}` 一行构造。与 --values、--dataframe 互斥；新表默认子表复用为第一个子表，日期/数字类型保真。 |
+| `--styles` | string + File + Stdin（复合 JSON） | optional | 建表时同时写入的视觉处理操作 JSON：顶层 `{styles:[...]}`，每项对应一个目标子表、含 `name`，并至少给 `cell_styles` / `row_sizes` / `col_sizes` / `cell_merges` 之一。`cell_styles` 用 A1 单元格 range + 扁平样式字段（字段同 +cells-set-style，含 number_format / 颜色 / 对齐 / border_styles）；row/col sizes 用行/列范围 + type/size；merges 用单元格 range + 可选 merge_type。与 --sheets 搭配时 styles 数组长度/顺序/name 必须与 --sheets.sheets 对应；与 --values 搭配时只给一个 styles 项（其 name 忽略）。完整 cell_styles 字段结构跑 `+workbook-create --print-schema --flag-name styles`。 |
+| `--dataframe` | string | optional | 单 sheet 类型保真表格的二进制入口，从一个 Arrow IPC 文件（Feather v2，pandas `df.to_feather()` 直接写出）读入，与 --values / --sheets 互斥。用 `@<path>` 传文件或 `-` 读二进制 stdin（同其他输入 flag 的约定）。Arrow 字节按原样读 —— 不做 TrimSpace / BOM strip，IPC magic 字节完整保留（区别于文本类输入 flag）。列类型从 Arrow schema 推导；每列的 `number_format` 可写在 Arrow Field metadata 里。建表后写入默认子表（`Sheet1` —— 直接复用，不残留空 Sheet1）。要多子表或换落点，请改用 `--sheets`。 |
 
 ### `+workbook-export`
 
@@ -136,6 +153,43 @@ _公共：URL/token（无 sheet 定位） · 系统：`--dry-run`_
 | `--sheet-id` | string | optional | 仅 csv 模式必填：指定要导出哪张 sheet 为 CSV。这是 `+workbook-export` 专有 flag，与公共四件套的 sheet 定位无关（本 shortcut 不接受公共 sheet 定位） |
 | `--output-path` | string | optional | 本地保存路径；省略时只触发导出不下载 |
 
+### `+workbook-import`
+
+| Flag | Type | 必填 | 说明 |
+| --- | --- | --- | --- |
+| `--file` | string | required | 本地文件路径（.xlsx / .xls / .csv） |
+| `--folder-token` | string | optional | 目标文件夹 token；省略则导入到云空间根目录 |
+| `--name` | string | optional | 导入后表格名称；省略则用本地文件名（去掉扩展名） |
+
+## Schemas
+
+> 复合 JSON flag 字段速查（只列顶层 + 一层嵌套）。深层结构看下方 `## Examples`，或用 `--print-schema` 读完整 JSON Schema（用法见 SKILL.md「公共 flag 速查」与「Agent 使用提示」）。
+
+### `+workbook-create` `--sheets`
+
+_一个或多个子表的 typed 数据，每个数组元素写入一张子表；支持多 DataFrame → 多子表一次写入_
+
+**数组项**（类型 object）：
+- `name` (string) — 目标子表名
+- `start_cell` (string?) — 写入起点单元格（A1 记法，如 "B2"），默认 "A1"
+- `mode` (enum?) — overwrite（默认）：从 start_cell 起写「表头 + 数据」块；append：把数据追加到子表已有数据下方（默认不重复表头） [overwrite / append]
+- `header` (boolean?) — 是否写一行列名表头
+- `allow_overwrite` (boolean?) — 为 false 时，若写入会落在非空单元格则拒写以保护原数据（返回 partial_success）
+- `columns` (array<string>) — 列名字符串数组，顺序与 `data` 中每行取值一一对应
+- `data` (array<array<string|number|boolean|null>>) — 数据行；每行是一个数组，长度必须等于 `columns` 数
+- `dtypes` (object?) — 可选
+- `formats` (object?) — 可选
+
+### `+workbook-create` `--styles`
+
+
+**数组项**（类型 object）：
+- `cell_merges` (array<object>?) — 单元格合并操作数组；range 使用 A1 单元格范围，merge_type 默认 all each: { merge_type?: enum, range: string }
+- `cell_styles` (array<object>?) — 单元格样式操作数组；每项用 A1 单元格 range 指定范围，字段名与 +cells-set-style 对齐 each: { background_color?: string, border_styles?: object, font_color?: string, font_line?: enum, font_size?: number, …共 12 项 }
+- `col_sizes` (array<object>?) — 列宽操作数组；range 使用列范围如 A:C，type 为 pixel/standard，pixel 需要 size each: { range: string, size?: number, type: enum }
+- `name` (string) — 子表名
+- `row_sizes` (array<object>?) — 行高操作数组；range 使用行范围如 1:3，type 为 pixel/standard/auto，pixel 需要 size each: { range: string, size?: number, type: enum }
+
 ## Examples
 
 公共四件套：所有 shortcut 顶部排列 `--url` / `--spreadsheet-token` / `--sheet-id` / `--sheet-name`（XOR）。`+workbook-info` 只用前两者；`+sheet-*` 系列对单个工作表操作，需 `--sheet-id` 或 `--sheet-name`。
@@ -144,6 +198,119 @@ _公共：URL/token（无 sheet 定位） · 系统：`--dry-run`_
 
 输出契约：返回 `sheets[]`，每个含 `sheet_id` / `title`（工作表显示名；旧 payload 用 `sheet_name`，读取时优先取 `title`、缺失再回退 `sheet_name`）/ `row_count` / `column_count` / `index` / `is_hidden`，以及计数字段 `merged_cells_count` / `chart_count` / `pivot_table_count` / `float_image_count`（无 `frozen_*` 字段，冻结信息请用 `+sheet-info` 读取）。是操作飞书表格的第一步——任何后续 sheet 级动作都需要先拿这里的 sheet_id。
 
+### `+workbook-create`
+
+新建电子表格，可选预填数据。三种数据入口（untyped `--values` / typed `--sheets` JSON / typed `--dataframe` Arrow 二进制）**三方互斥**，按需选一——三者都走同一条分批写入：
+
+```bash
+# 1) untyped：--values（一个二维数组，表头并入第一行；值原样写、类型由飞书自动识别，
+#    日期会落成文本，配 --styles 控制格式）
+lark-cli sheets +workbook-create --title "销售" \
+  --values '[["门店","销售额"],["北京",259874]]'
+
+# 2) typed JSON：--sheets（一步建表 + 类型保真）。date 列落成真日期（可排序/透视）、
+#    number 不丢精度、string 列保前导零（如订单号 00123）；多子表一次建。
+lark-cli sheets +workbook-create --title "交易" --sheets '{
+  "sheets":[
+    {"name":"明细",
+     "columns":["日期","金额","单号"],
+     "dtypes":{"日期":"datetime64[ns]","金额":"float64","单号":"object"},
+     "formats":{"金额":"#,##0.00"},
+     "data":[["2024-01-15",1234.5,"00123"]]}
+  ]}'
+
+# 3) typed binary：--dataframe（pandas df.to_feather 直接出，Arrow IPC / Feather v2）。
+#    单子表（落点固定为新表的默认子表，原地复用、不残留空 Sheet1），列类型从 Arrow
+#    schema 自动恢复，无需手填 dtypes/formats；要多子表回到 --sheets。
+lark-cli sheets +workbook-create --title "交易" --dataframe @./in.arrow
+# 或走 stdin（不落盘）：
+python prepare.py | lark-cli sheets +workbook-create --title "交易" --dataframe -
+```
+
+`--sheets` 协议与 `+table-put` 完全同构（字段含义见 lark-sheets-write-cells 的 `+table-put`，大 payload 走 stdin / `@file`）；`--dataframe` 是同一份 typed 数据的二进制 wire（Arrow IPC，详见同 reference 的 `+table-put` 段落的 `--dataframe` 小节），按 producer 已有的 API 选——pandas 走 `--dataframe`，多子表 / 手拼 JSON 走 `--sheets`。关键差异：**新建工作簿的默认子表会被复用为第一个子表**（重命名后承载数据），不会残留空 `Sheet1`；其余子表按需新建。它把 `+table-put` 单独做不到的"建表 + typed 写入"合到一条命令，是「pandas 算完直接落地一张带真日期的新表」的首选。回读校验用 `+table-get`（与 `--sheets` 同构、可 round-trip；pandas 用户也可走 `--dataframe-out` 直拿 Arrow 文件）。
+
+`--styles` 可在建表写入时同时写视觉处理。它和 `--sheets` 一样只有一种外层写法：顶层对象里放 `styles` 数组；数组每项对应一个子表，含 `name`，并按能力拆成四类可选数组：
+
+- `cell_styles`：像 `+cells-set-style`，用 A1 单元格 `range` 加扁平样式字段（`font_weight` / `background_color` / `horizontal_alignment` / `vertical_alignment` / `number_format` 等）和可选 `border_styles`；这些样式会随内容在同一次写入里一并应用。完整字段跑 `+workbook-create --print-schema --flag-name styles`。
+- `cell_merges`：用 A1 单元格 `range` 设置合并，`merge_type` 默认为 `all`，可选 `rows` / `columns`。
+- `row_sizes`：用行范围（如 `1:3`）设置行高，`type` 为 `pixel` / `standard` / `auto`；`pixel` 需要 `size`。
+- `col_sizes`：用列范围（如 `A:C`）设置列宽，`type` 为 `pixel` / `standard`；`pixel` 需要 `size`。
+
+同一单元格命中多个 `cell_styles` 项时，后面的操作继续合并覆盖已传字段。`cell_merges` / `row_sizes` / `col_sizes` 在内容写入后顺序执行。
+
+```bash
+# 3) untyped：仍用 {"styles":[...]}，只有一个子表样式项（name 忽略）；range 覆盖 --values 初始区域
+lark-cli sheets +workbook-create --title "销售" \
+  --values '[["门店","销售额"],["北京",259874],["上海",198320]]' \
+  --styles '{
+    "styles":[
+      {"name":"Sheet1","cell_styles":[
+        {"range":"A1:B1","font_weight":"bold","background_color":"#f5f5f5","horizontal_alignment":"center","vertical_alignment":"middle"},
+        {"range":"B2:B3","number_format":"#,##0"}
+      ]}
+    ]
+  }'
+
+# 4) typed 单子表：--styles.styles[0].name 必须对应 --sheets.sheets[0].name
+lark-cli sheets +workbook-create --title "交易" --sheets '{
+  "sheets":[
+    {"name":"明细",
+     "columns":["日期","金额"],
+     "dtypes":{"日期":"datetime64[ns]","金额":"float64"},
+     "formats":{"金额":"#,##0.00"},
+     "data":[["2024-01-15",1234.5]]}
+  ]}' --styles '{
+    "styles":[
+      {"name":"明细",
+       "cell_styles":[
+        {"range":"A1:B1","font_weight":"bold","background_color":"#f5f5f5",
+          "border_styles":{"bottom":{"style":"solid","weight":"thin","color":"#000000"}}},
+        {"range":"A2:A2","number_format":"yyyy-mm-dd"},
+        {"range":"B2:B2","number_format":"#,##0.00","font_color":"#0f7b0f"}
+       ],
+       "cell_merges":[{"range":"A1:B1"}],
+       "col_sizes":[{"range":"A:B","type":"pixel","size":120}],
+       "row_sizes":[{"range":"1:1","type":"pixel","size":28}]}
+    ]
+  }'
+
+# 5) typed 多子表：styles 数组和 sheets 数组长度、顺序、name 都必须一致
+lark-cli sheets +workbook-create --title "经营看板" --sheets '{
+  "sheets":[
+    {"name":"收入","columns":["月份","收入"],"dtypes":{"收入":"int64"},"formats":{"收入":"#,##0"},"data":[["2026-05",1200000]]},
+    {"name":"成本","columns":["月份","成本"],"dtypes":{"成本":"int64"},"formats":{"成本":"#,##0"},"data":[["2026-05",730000]]}
+  ]}' --styles '{
+    "styles":[
+      {"name":"收入","cell_styles":[
+        {"range":"A1:B1","font_weight":"bold","background_color":"#f0f7ff"},
+        {"range":"B2:B2","font_color":"#0f7b0f"}
+      ]},
+      {"name":"成本","cell_styles":[
+        {"range":"A1:B1","font_weight":"bold","background_color":"#fff7ed"},
+        {"range":"B2:B2","font_color":"#b42318"}
+      ]}
+    ]
+  }'
+```
+
+> ⚠️ **`+workbook-create` 是把内存里的数据写成新表；要把已有的本地 Excel/CSV 文件原样导入成新表，用 `+workbook-import`**（见下），不要先在本地读出文件再 `+workbook-create` 重灌。
+
+### `+workbook-import`
+
+把已有的本地 `.xlsx` / `.xls` / `.csv` 文件导入为一个**新的**飞书电子表格（异步任务 + 内置轮询），与 `+workbook-export`（导出）对称，固定导入为电子表格类型。
+
+```bash
+# 导入到云空间根目录；表格名默认取本地文件名（去掉扩展名）
+lark-cli sheets +workbook-import --file ./data.xlsx
+
+# 指定目标文件夹与导入后表格名
+lark-cli sheets +workbook-import --file ./report.csv --folder-token <FOLDER_TOKEN> --name "月度报表"
+```
+
+- **不接受任何 spreadsheet / sheet 定位 flag**（它是新建，不操作已有表）：只有 `--file`（必填）/ `--folder-token` / `--name`。
+- 本地表格文件 → 飞书电子表格一律用本命令，**不要**用 `drive +import` 导电子表格——它是 sheets 之外的通用导入、还需额外指定 `--type`，绕路且更易错。只有要把本地表格导入成**多维表格**（bitable）时，才改用 `lark-cli drive +import --type bitable`。
+- 返回 `token` / `url`（导入完成的新表格）/ `ticket` / `ready` / `job_status`；未在内置轮询窗口内完成时返回 `timed_out=true` 与续查命令 `next_command`。
+
 ### `+sheet-create`
 
 示例：
@@ -153,6 +320,8 @@ lark-cli sheets +sheet-create --url "https://example.feishu.cn/sheets/shtXXX" \
   --title "汇总" --index 0
 ```
 
+> 💡 `+sheet-create` 只建一张**空子表**。要在已有工作簿里建子表并一步写入 typed 数据和/或样式，用 `+table-put`（payload 里命名的子表缺则自动新建）配合它的 `--sheets` / `--styles`，省掉先建表再 `+cells-set` / `+cells-set-style` 的二次往返。
+
 ### `+sheet-delete`
 
 > ⚠️ 工作表删除不可逆；先 `--dry-run` 看输出 sheet_id + title 确认是要删的那张。
@@ -190,8 +359,16 @@ lark-cli sheets +sheet-unhide --url "..." --sheet-id "$SID"
 lark-cli sheets +sheet-set-tab-color --url "..." --sheet-id "$SID" --color "#FF0000"
 ```
 
+### `+sheet-show-gridline` / `+sheet-hide-gridline`
+
+```bash
+# 切换子表网格线显隐；二态语义在命令名里，无需额外参数（同 +sheet-hide/+sheet-unhide）
+lark-cli sheets +sheet-show-gridline --url "..." --sheet-id "$SID"
+lark-cli sheets +sheet-hide-gridline --url "..." --sheet-id "$SID"
+```
+
 ### Validate / DryRun / Execute 约束
 
-- `Validate`：XOR 公共四件套；`+sheet-create` 校验 `--title` 非空、`--row-count` ≤ 50000、`--col-count` ≤ 200；`+sheet-delete` 必须 `--yes` 或 `--dry-run`。
+- `Validate`：XOR 公共四件套；`+sheet-create` 校验 `--title` 非空、`--row-count` ≤ 50000、`--col-count` ≤ 200；`+sheet-delete` 必须 `--yes` 或 `--dry-run`；`+workbook-create` 的 `--sheets` 与 `--values` **互斥**，给了 `--sheets` 则按 typed 协议校验 payload（其余约束同 `+table-put`）。
 - `DryRun`：`+sheet-*` 写操作输出"将要 PATCH 的 sheet metadata"；`--sheet-name` 在 dry-run 输出里生成为 `<resolve:Sheet1>` 占位符，不实际解析为 sheet-id。
 - `Execute`：写操作不自动回读；如需确认目标 sheet 的新状态，自行调用 `+workbook-info`。

skills/lark-sheets/references/lark-sheets-write-cells.md

@@ -5,6 +5,7 @@
 1. **明确写入边界**：写入前必须能回答"目标 range 的起止行列号是多少？是否落在用户授权范围内？"。除用户明示要修改的区域外，禁止扩张到原数据列以外或新建 Sheet。
 2. **完整性断言**：批量写入前先把"预期写入条数"硬编码到代码里（如要填 106 条翻译 → `expected = 106`），写完后回读断言 `actual == expected`。少于预期就继续写，禁止交付半成品。
 3. **回读抽样校验**：写完关键值 / 公式后，用 `+csv-get` 或 `+cells-get` 重新读取写入区域，至少抽样 3-5 个代表性单元格（首 / 中 / 末），核对值与预期一致（与本地脚本计算的预期值对照）。公式特定的"先验证模板再 --copy-to-range / 修完再读回"细则见下方相关章节。
+4. **护原表 · 派生产物落点（写排名 / 标记 / 汇总 / 改写列时易丢数据）**：派生结果一律写到**真实末列 +1 的全新空列**或新建子表，**禁止复用任何已有原数据列**——哪怕该列看起来"空"，也要先 `+csv-get` 回读确认整列无原始数据再写。三条铁律：① 不把新公式 / 新值写进原数据列（典型反例：把新算的排名公式写进了原本存放另一份原始数据的列，整列原始数据被覆盖丢失）；② 不改写、不合并原表头字段名（典型反例：把几个独立表头字段合并成一列，原字段名丢失）；③ 慎用 `--allow-overwrite`：它一旦让写入区盖到相邻原始列 / 行就是不可逆数据丢失，加它之前必须用 `+sheet-info` / `+csv-get` 核清目标 range 不含任何原始数据。
 
 ## 新增列 / 新增行的样式继承（防止视觉风格不一致）
 
@@ -13,11 +14,11 @@
 **完整继承清单**（写新列 / 新行时 cells 数组必须同时携带）：
 
 1. `cell_styles.font_size` / `cell_styles.font_weight` / `cell_styles.font_color` / `cell_styles.font_style`（字号 / 粗细 / 颜色 / 斜体等）
-2. `cell_styles.horizontal_alignment` / `cell_styles.vertical_alignment`（H-Align / V-Align）—— 漏继承会导致新列对齐与原列不一致（高频）
+2. `cell_styles.horizontal_alignment` / `cell_styles.vertical_alignment`（H-Align / V-Align）—— 漏继承会导致新列对齐与原列不一致（常见）
 3. `cell_styles.number_format`（小数位 / 千分位 / 百分比 / 日期格式）—— 漏继承会导致同列数值格式混乱
 4. `cell_styles.background_color`（背景色）
 5. `border_styles`（四边框）
-6. **`merged_cells`（合并范围）**——续写场景必查（高频致命错误）：用 `+sheet-info --include merges` 读原数据区域的合并信息。**原行有跨列合并**（如标题行 `A1:G1` 合并）时，新行**必须**用 `+cells-{merge|unmerge}` 工具复制相同合并模式到新行（如续写第 3 个周报块的标题行 `A23:G23` 必须合并）。仅传 cells 数组的 5 类样式不够——合并范围要单独靠 `+cells-{merge|unmerge}` 工具落地（典型反例：续写多周记录表时，新增周次的标题行未合并，视觉上与原前几周风格不一致）
+6. **`merged_cells`（合并范围）**——续写场景必查：用 `+sheet-info --include merges` 读原数据区域的合并信息。**原行有跨列合并**（如标题行 `A1:G1` 合并）时，新行**必须**用 `+cells-{merge|unmerge}` 工具复制相同合并模式到新行（如续写第 3 个周报块的标题行 `A23:G23` 必须合并）。仅传 cells 数组的 5 类样式不够——合并范围要单独靠 `+cells-{merge|unmerge}` 工具落地（典型反例：续写多周记录表时，新增周次的标题行未合并，视觉上与原前几周风格不一致）
 
 **采样模板的正确做法**：
 - 表头新列 → 读相邻表头单元格（如新加 D1 → 读 A1/B1/C1 任一）
@@ -44,13 +45,34 @@
 
 ## 使用场景
 
-写入。为一块单元格区域设置值、公式、批注/备注和/或格式。也支持通过 `rich_text` 中 `type: "embed-image"` 在单元格内嵌入图片（单元格图片）。关键：数组维度必须严格匹配——`cells` 二维数组必须与 `range` 的行列维度完全一致，range 是闭区间，否则会触发 `InvalidCellRangeError`。计算示例：区域 `A1:D3` = 3 行 × 4 列 = `[[r1c1,r1c2,r1c3,r1c4],[r2c1,r2c2,r2c3,r2c4],[r3c1,r3c2,r3c3,r3c4]]`；区域 `A41:N48` = 8 行 × 14 列 = 8 个数组且每个数组 14 个单元；单个单元格 `A1` = `[[cell]]`；单列区域 `B5:B7` = `[[cell1],[cell2],[cell3]]`。空单元请使用 `{}`。**如果填写的区域存在大量重复内容，务必优先使用 `--copy-to-range` 字段复制，可大幅减少 `cells` 长度。**
+写入。向飞书表格的单元格区域写入值、公式、样式、批注、图片或下拉，也可批量写入 CSV / DataFrame。本 reference 覆盖 6 个 shortcut，按数据来源 + 内容形态选：
 
-> **单元格图片 vs 浮动图片**：
-> - **单元格图片**（本工具）：图片嵌入在单元格内部，属于单元格内容，随单元格移动。通过 `rich_text` 中 `type: "embed-image"` 写入。
-> - **浮动图片**：图片悬浮在单元格上方，可自由定位和调整大小，不属于单元格内容。→ 使用 lark-sheets-float-image。
+| 场景 | 用这个 shortcut | 原因 |
+|------|----------------|------|
+| 模型手里已经有 CSV 文本（小规模手动构造、从 `+csv-get` 取到后简单加工） | `+csv-put` | 直接传 CSV 文本 + `--start-cell`，不用自己拼二维 cells 数组；必要时自动扩容行列 |
+| 列里有数值语义的数据（数字 / 金额 / 百分比 / 日期 / 计数）→ 飞书，要类型保真（来源不限：DataFrame、Counter、dict、list 都算） | `+table-put` | 列显式声明 `type`：date 落真日期、**金额 / 百分比 / 计数等数值列保精度且带 `number_format`（可排序 / 求和 / 入图表）**、string 保前导零，多 sheet 一次写。**只要列有数值语义就走这里**，不要在本地把数字拼成带 `$` / `%` 的字符串再走 `+csv-put` |
+| 写入含样式、批注、图片、数据校验等任意富写入 | `+cells-set` | 唯一支持完整富字段的 shortcut（公式 `+csv-put` 也能写） |
+| 只改已有 cell 的样式，不动 value/formula | `+cells-set-style` | 拍平 10 个样式字段为独立 flag；不触发不必要的值写入 |
+| 单 cell 嵌入图片 | `+cells-set-image` | 比 `+cells-set` 参数更简短 |
+| 大量纯值 + 需要表头样式/边框 | 先用 `+csv-put` 写值，再用 `+cells-set-style` 补样式 | 分工配合，入参最短 |
 
-高频模式（**必须遵守，禁止逐行写入替代**）：
+**优先级**：常规批量写入（纯值或公式）优先 `+csv-put`（最短入参，直接传 CSV 文本）；含样式/批注/图片才用 `+cells-set`。⚠️ 这里"纯值"特指**已是文本、无需保留数值语义**的内容；只要列里是金额 / 百分比 / 日期 / 计数等有数值语义的数据，应优先 `+table-put`（声明 `number` / `date` 类型 + `number_format`），而不是 `+csv-put`。
+
+⚠️ `+csv-put` 可写值或公式：以 `=` 开头的单元格会被当作公式计算（读回时 `formula` 字段保留、`value` 为计算结果）。**公式内部含逗号 / 引号 / 换行时必须按 RFC 4180 转义**——含逗号的字段整格用双引号包裹、字段内部的引号再翻倍：如 `=COUNTIF(D5:D22,"及格")` 必须写成 `"=COUNTIF(D5:D22,""及格"")"`（外层双引号包裹整格，内部 `"及格"` 的引号翻倍成 `""及格""`）。漏转义会被 CSV 解析器按逗号拆列、整块写入区域错位（如本该 `G4:H6` 错成 `G4:K4`），详见下方 `+csv-put` 示例。**因此含逗号 / 引号 / 换行的公式优先改用 `+cells-set`（JSON 二维数组）/ `+table-put` 结构化写入——直接传 `formula` 字段、零 CSV 转义负担，从根上避免拆列错位。** 此外 `+csv-put` **不会**携带样式/批注/图片，也无法把 `=` 开头的内容当字面量文本写入；需要样式/批注/图片用 `+cells-set`（或"写值 + 补样式"两步法）。
+
+⚠️ **别把本该是数值的列格式化成字符串用 `+csv-put` 写入**：金额 / 百分比 / 市值 / 计数等列，若在本地拼成带 `$` / `%` / 千分位的字符串（如 `"$1,234.50"` / `"+30.5%"`）再 `+csv-put` 灌进去，单元格会变成**文本**——丢失排序 / 求和 / 图表 / 透视能力，且与 `number` 列混排时无法参与计算。正解是 `+table-put` 声明该列 `type:"number"`（百分比存小数，如 `0.305`）+ `format`（如 `"$#,##0.00"` / `"0.0%"` / `"#,##0"`），**显示效果完全相同、数值无损**。判断信号：**当你准备把一个数字 format 成字符串再写时，几乎总该用 `+table-put` 而非 `+csv-put`**。
+
+⚠️ 大数据回写走"`+csv-get` 按 `--range` 行窗口分批读到本地 + 本地脚本处理 + `+csv-put` 分批回写"。
+
+## `+cells-set` 写入要点（常用模式 / 公式 / 样式）
+
+> 以下是用 `+cells-set`（及 `+cells-set-style`）做富写入时的常用模式与铁律；选哪个 shortcut 见上方「使用场景」。
+
+`+cells-set` 为一块区域设置值 / 公式 / 批注 / 样式，也支持 `rich_text` 的 `type: "embed-image"` 嵌入单元格图片。**关键：`cells` 二维数组的行列维度必须与 `range`（闭区间）严格一致，否则触发 `InvalidCellRangeError`**——维度计算示例见文末 `## Schemas` 的 `--cells`。
+
+> **单元格图片 vs 浮动图片（最易选错）**：图若**属于某条记录、要随那行排序 / 筛选 / 增删**（凭证 / 证件照 / 每行配图，话里带「对应 / 每行 / 这列」等绑定词）→ **单元格图片**（本工具）：用 `+cells-set-image`（最短）或 `+cells-set` 的 `rich_text` + `type: "embed-image"`。只是自由摆放的装饰（logo / 水印 / 封面）→ 浮动图片，见 lark-sheets-float-image。别因「浮动图更好控制 / 更熟」默认选浮动图——它承载"对应某记录"的图会随增删行 / 排序错位。
+
+常用模式（**必须遵守，禁止逐行写入替代**）：
 
 - 整列公式：先在 `H2` 写一个公式，再用 `--copy-to-range "H2:H100"` 或 `--copy-to-range "H:H"` 向下填充。**禁止对每一行单独调用 `+cells-set` 写入相同结构的公式**
 - 整列格式：先在 `J1` 写一个带样式的模板单元格，再用 `--copy-to-range "J:J"`
@@ -93,24 +115,25 @@ Step 2: `+cells-set` — range="A2", cells 含 value + cell_styles + border_styl
 2. **看到 `#` 开头的错误值**立即修公式：`#NAME?` 多半是函数名拼错或用了飞书不支持的函数（如 `GOOGLETRANSLATE` / CUBE 系列；注意 `UNIQUE` / `FILTER` / `SPLIT` 飞书是支持的）；`#VALUE!` 多半是类型不匹配或括号错位；`#REF!` 是引用错误；`~CIRCULAR~REF~` 是循环引用（公式引用了自身或会闭环）
 3. **`--copy-to-range` 扩展前先验证模板**：模板单元格公式自己都算错，`--copy-to-range` 复制到 100 行就是 100 个错误
 4. **去重 / 筛选函数**：飞书**支持** `UNIQUE` / `FILTER` / `SPLIT`（原生数组函数，详见 `lark-sheets-formula-translation`），可直接用；`DISTINCT` 不是飞书函数，去重用 `UNIQUE`。大数据量去重 / 分组也可用透视表（`+pivot-{create|update|delete}`，值字段聚合方式选 count）
-5. **循环引用预检（高频致命错误）**：写聚合公式（SUM / AVERAGE / COUNT 等）前必须明确**引用范围不包含目标单元格自身或其传递依赖**。典型反例：在 C3 写 `=SUMIF(B:B,LEFT(B3,9)&"*",C:C)`，B 列匹配 B3 前 9 位时 C3 自己也命中，导致 C3 自引用 → `~CIRCULAR~REF~`。修法：用辅助列 / 显式排除自身（`SUMIFS(C:C, B:B, ..., A:A, "<>"&A3)`）/ 缩小范围避开自己
+5. **循环引用预检**：写聚合公式（SUM / AVERAGE / COUNT 等）前必须明确**引用范围不包含目标单元格自身或其传递依赖**。典型反例：在 C3 写 `=SUMIF(B:B,LEFT(B3,9)&"*",C:C)`，B 列匹配 B3 前 9 位时 C3 自己也命中，导致 C3 自引用 → `~CIRCULAR~REF~`。修法：用辅助列 / 显式排除自身（`SUMIFS(C:C, B:B, ..., A:A, "<>"&A3)`）/ 缩小范围避开自己
 6. **REGEX 模式覆盖率验证**：公式里的 `REGEXEXTRACT` / `REGEXMATCH` / `REGEXREPLACE` 等正则模式落地前必须用本地脚本在源列上跑一遍命中率统计（`df[col].str.contains(pattern).mean()`）；命中率 < 100% 时必须扩展 pattern 或加多分支（IFS / 多个 IFERROR 串联）兜底，**禁止**只覆盖样本前 N 行就交付（典型反例：用 `REGEXEXTRACT(D5,"长(\d+)")` 只匹配带"长"前缀的尺寸文本，对"宽×高"、"×"、"*"等其它分隔符直接漏匹配）
 7. **公式范围与用户指令字面对齐**：用户说"对 F 至 L 列求和"就必须写 `SUM(F2:L2)` 或 `F2+G2+H2+I2+J2+K2+L2`，**不能漏列、多列、错列**。写完用 `+cells-get` 拿回 `formula` 字符串，与用户原话逐字对照（参与求和的列名一致 / 起止列号一致 / 运算符一致），不一致就是违规
+8. **量纲 / 单位换算 / 数量乘项预检（公式不报错但结果整体偏倍数）**：从文本提取数字做计算前，先核对**单位是否统一、是否漏乘数量、口径是否一致**——这类错误公式能跑通、无 `#` 报错，回读也看不出（值"像对的"）。必须用本地脚本对 3–5 个代表行**离线手算一遍预期值**，与公式结果逐格比对量级：① 单位不一致先统一再算（典型反例：尺寸 `320CM*337CM` 直接取数相乘除以 1e6 得 0.11，正确是 CM→MM 换算后得 10.78，**差 100 倍**）；② 按"单件×数量"的量必须乘数量列（典型反例：侧面板面积漏乘 F 列数量，F=2 的行只算了一半）；③ 标准值口径对齐（典型反例：营养成分 mg/kg 与 g/100g 口径混用，整列放大 100 倍）。**口径 / 单位 / 数量任一项错，整列计算结果就是错的；这类错误公式不报错、回读也不易看出，必须靠离线手算对照。**
 
-⚠️ **收到 `formula_errors` 反馈后不要只打补丁（高频致命错误）**：`+cells-set` 返回值里若出现 `formula_errors: [{cell, formula, error_type, detail}]`，说明某些 cell 公式编译失败（`error_type=compile_failed` 通常是函数语法错如 `SPLIT(x)[1]` 的下标取值飞书不支持（SPLIT 本身支持，取第 N 项用 `INDEX(SPLIT(...),N)`）；`non_formula` 是 `=` 开头但解析不通过）。此时**禁止只聚焦修报错点的局部语法**（如仅把 `[1]` 换成 `INDEX(..,1)`），必须：
+⚠️ **收到 `formula_errors` 反馈后不要只打补丁**：`+cells-set` 返回值里若出现 `formula_errors: [{cell, formula, error_type, detail}]`，说明某些 cell 公式编译失败（`error_type=compile_failed` 通常是函数语法错如 `SPLIT(x)[1]` 的下标取值飞书不支持（SPLIT 本身支持，取第 N 项用 `INDEX(SPLIT(...),N)`）；`non_formula` 是 `=` 开头但解析不通过）。此时**禁止只聚焦修报错点的局部语法**（如仅把 `[1]` 换成 `INDEX(..,1)`），必须：
 
 1. **重新审视整条公式的完整性**：被 formula_errors 标出的那一行，公式除了下标语法错，还可能有其他先天缺陷（字符清洗不全、IFERROR 兜底漏条件、引用列写错），修完语法错后立即整体复核
 2. **同步对称修复所有相似列**：如果同一任务涉及多列相似处理（如"算 H 列面积"用 D 列尺寸、"算 I 列面积"用 E 列尺寸），**修完一列必须把同样的清洗/兜底逻辑同步到所有相似列**，禁止出现 H 列用 `SUBSTITUTE(长)+SUBSTITUTE(高)+SUBSTITUTE(×)` 而 I 列只用 `SUBSTITUTE(×)` 这种不对称处理——会导致一列编译通过有值、另一列编译通过但 IFERROR 全返回空，用户看到的是"数据为空"而非"公式错"
 3. **修完再读回验证**：不只看 `formula_errors` 为空（这只证明编译通过，不证明运行时有值），必须 `+csv-get` 读目标列前 3-5 行，确认**非空源数据对应的目标列有非空计算结果**
 4. **核心心智**：`formula_errors` 是"帮你暴露编译错"的工具，不是"修掉它就收工"的通行证。编译通过 + 运行时 IFERROR 兜底空 = 用户视角的"没算出来"
 
-⚠️ **新增行的边框/样式禁止用 `{}` 跳过（高频致命错误）**：`cells` 数组里 `{}` 的语义是"**此单元格不做任何修改、保留原状态**"。这在写入**已有行**时是安全的（原有边框/样式保持不变），但在写入**新行**（比如表尾追加汇总行、扩展行）时是灾难：新行底子里本来就没边框，`{}` 不修改 = 保留无边框状态，导致该 cell 视觉断裂。
+⚠️ **新增行的边框/样式禁止用 `{}` 跳过**：`cells` 数组里 `{}` 的语义是"**此单元格不做任何修改、保留原状态**"。这在写入**已有行**时是安全的（原有边框/样式保持不变），但在写入**新行**（比如表尾追加汇总行、扩展行）时是灾难：新行底子里本来就没边框，`{}` 不修改 = 保留无边框状态，导致该 cell 视觉断裂。
 
 ⚠️ **"汇总行"识别 → 读 `lark-sheets-visual-standards` 拿完整样式规范**：下述双重条件**同时满足**才是汇总行，禁止仅凭"有 AVERAGE"就判定：
 - **语义信号**（二选一）：用户 prompt 含"合计/汇总/总计/统计/各科平均分/最下面加一行算…/底部总计"等意图词；或上下文明确是"表尾追加一行做聚合"
 - **结构信号**：新行全行都在做聚合（含 `=SUM/AVERAGE/COUNT/MAX/MIN/SUBTOTAL(...)`，支持 IFERROR 包裹），**不是**单个 cell 算个参考值或每行都算的派生列
 
-满足上述时，**不要在本 skill 里猜样式**，直接去读 `lark-sheets-visual-standards` 的「场景一 → 1A. 添加汇总行 / 表头行」章节，按那里的样式要点配齐 `font.bold / horizontal_alignment / background_color / border_styles`。
+满足上述时，**不要在本文里猜样式**，直接去读 `lark-sheets-visual-standards` 的「场景一 → 1A. 添加汇总行 / 表头行」章节，按那里的样式要点配齐 `font.bold / horizontal_alignment / background_color / border_styles`。
 
 反例（**不是**汇总行，禁止自动加粗）：
 - 用户说"在 H5 帮我算个 AVERAGE 参考"→ 单 cell 计算
@@ -208,24 +231,6 @@ lark-cli sheets +dropdown-set \
 
 `+dropdown-update`（多 range 批量更新）的所有 flag 语义与 `+dropdown-set` 完全一致；只是目标 `--ranges` 由单值变成 JSON 数组（每项带 sheet 前缀），同一份选项 + 配色应用到所有 range。
 
-## 工具选择
-
-本 skill 提供以下 CLI shortcut，按数据来源 + 内容形态选：
-
-| 场景 | 用这个 shortcut | 原因 |
-|------|----------------|------|
-| 模型手里已经有 CSV 文本（小规模手动构造、从 `+csv-get` 取到后简单加工） | `+csv-put` | 直接传 CSV 文本 + `--start-cell`，不用自己拼二维 cells 数组；必要时自动扩容行列 |
-| 写入含公式、样式、批注、图片、数据校验等任意富写入 | `+cells-set` | 唯一支持完整字段的 shortcut |
-| 只改已有 cell 的样式，不动 value/formula | `+cells-set-style` | 拍平 10 个样式字段为独立 flag；不触发不必要的值写入 |
-| 单 cell 嵌入图片 | `+cells-set-image` | 比 `+cells-set` 参数更简短 |
-| 大量纯值 + 需要表头样式/边框 | 先用 `+csv-put` 写值，再用 `+cells-set-style` 补样式 | 分工配合，入参最短 |
-
-**优先级**：常规纯值写入优先 `+csv-put`（最短入参，直接传 CSV 文本）；含公式/样式/批注/图片才用 `+cells-set`。
-
-⚠️ `+csv-put` 只写纯值，**不会**携带公式/样式/批注/图片；公式字符串以 `=` 开头会被当作字面量文本落地。如果数据里需要公式或样式，**必须**用 `+cells-set`（或"写值 + 补样式"两步法）。
-
-⚠️ 大数据回写走"`+csv-get` 按 `--range` 行窗口分批读到本地 + 本地脚本处理 + `+csv-put` 分批回写"。
-
 ## Shortcuts
 
 | Shortcut | Risk | 分组 |
@@ -235,6 +240,7 @@ lark-cli sheets +dropdown-set \
 | `+cells-set-image` | write | 单元格 |
 | `+dropdown-set` | write | 对象 |
 | `+csv-put` | write | 单元格 |
+| `+table-put` | write | 单元格 |
 
 ## Flags
 
@@ -299,10 +305,20 @@ _公共四件套 · 系统：`--dry-run`_
 | Flag | Type | 必填 | 说明 |
 | --- | --- | --- | --- |
 | `--start-cell` | string | required | 目标区域起点 A1（如 `A1`、`B5`，不带 sheet 前缀；用 `--sheet-id` / `--sheet-name` 指定 sheet）；必须是单个单元格，不接受范围写法；终点按 CSV 实际行列数自动推断 |
-| `--csv` | string + File + Stdin（非 JSON 文本） | required | RFC 4180 CSV 文本；只写纯值，不带公式/样式/批注 |
+| `--csv` | string + File + Stdin（非 JSON 文本） | required | RFC 4180 CSV 文本；可写值或公式（以 = 开头的单元格按公式计算）；不带样式 / 批注 / 图片，需要这些用 +cells-set。 |
 | `--allow-overwrite` | bool | optional | 允许覆盖（默认 true）；设为 false 时若目标非空报错 |
 | `--range` | string | optional | --start-cell 的别名（与 +csv-get / +cells-set 一致，用 --range 定位）；传区间（如 A1:H17）时自动取其左上角单元格（隐藏 flag：不在 `--help` 列出，但可正常传入） |
 
+### `+table-put`
+
+_公共：URL/token（无 sheet 定位） · 系统：`--dry-run`_
+
+| Flag | Type | 必填 | 说明 |
+| --- | --- | --- | --- |
+| `--sheets` | string + File + Stdin（复合 JSON） | xor | Typed 表格协议（pandas-DataFrame-shaped）JSON，与 `--dataframe` 互斥：顶层 sheets 数组，每项 `{name, start_cell?, mode?, header?, allow_overwrite?, columns:["colA","colB",...], data:[[...]], dtypes?:{colA:pandasDtype, ...}, formats?:{colA:numberFormat, ...}}`。Agents 通常用 `{**json.loads(df.to_json(orient="split")), "dtypes": df.dtypes.astype(str).to_dict()}` 一行构造。`dtypes` 值是 pandas dtype 字符串（`int64`、`float64`、`Int64`、`bool`、`boolean`、`datetime64[ns]`、`object`、...），CLI 端映射成内部 string/number/date/bool —— 省略 `dtypes` 时该列按文本写入（适合原始 CSV-shaped 数据）。`formats[col]` 是 Excel number_format 字符串（如 `#,##0.00`、`0.0%`、`yyyy-mm`）；缺省时 date 列用 `yyyy-mm-dd`，string 列用文本格式 `@`。 |
+| `--dataframe` | string | xor | 单 sheet 类型保真表格的二进制入口，从一个 Arrow IPC 文件（即 Feather v2，pandas `df.to_feather()` 直接写出）读入，与 `--sheets` 互斥。用 `@<path>` 传文件或 `-` 读二进制 stdin（同其他输入 flag 的约定）。Arrow 字节按原样读 —— 不做 TrimSpace / BOM strip，IPC magic 字节完整保留（区别于文本类输入 flag）。列类型从 Arrow schema 推导（int*/uint*/float* → number，date32/date64/timestamp → date，utf8/large_utf8 → string，bool → bool）；每列的 `number_format` 可写在 Arrow Field metadata 里（`pa.field("price", pa.float64(), metadata={b"number_format": b"$#,##0.00"})`）。子表走默认落点：名为 `Sheet1`（缺则新建），从 A1 起覆盖写并带表头。要换子表名 / 起始位置 / 写入方式，或要写多子表，请改用 `--sheets`。 |
+| `--styles` | string + File + Stdin（复合 JSON） | optional | 类型保真写入后再应用的视觉处理操作 JSON：顶层 `{styles:[...]}`，每项对应一个被写入的子表、含 `name`，并至少给 `cell_styles` / `row_sizes` / `col_sizes` / `cell_merges` 之一。`cell_styles` 用 A1 单元格 range + 扁平样式字段（字段同 +cells-set-style，含 number_format / 颜色 / 对齐 / border_styles）；row/col sizes 用行/列范围 + type/size；merges 用单元格 range + 可选 merge_type。styles 数组的长度/顺序/name 必须与被写入的子表对应：配 --sheets 时与 --sheets.sheets 对应；配 --dataframe（单子表，名为 Sheet1）时只给一个 name 为 `Sheet1` 的 styles 项。完整 cell_styles 字段结构跑 `+table-put --print-schema --flag-name styles`。 |
+
 ## Schemas
 
 > 复合 JSON flag 字段速查（只列顶层 + 一层嵌套）。深层结构看下方 `## Examples`，或用 `--print-schema` 读完整 JSON Schema（用法见 SKILL.md「公共 flag 速查」与「Agent 使用提示」）。
@@ -338,20 +354,45 @@ _列表选项_
 **数组项**（类型 string）：
 - 标量：string
 
+### `+table-put` `--sheets`
+
+_一个或多个子表的 typed 数据，每个数组元素写入一张子表；支持多 DataFrame → 多子表一次写入_
+
+**数组项**（类型 object）：
+- `name` (string) — 目标子表名
+- `start_cell` (string?) — 写入起点单元格（A1 记法，如 "B2"），默认 "A1"
+- `mode` (enum?) — overwrite（默认）：从 start_cell 起写「表头 + 数据」块；append：把数据追加到子表已有数据下方（默认不重复表头） [overwrite / append]
+- `header` (boolean?) — 是否写一行列名表头
+- `allow_overwrite` (boolean?) — 为 false 时，若写入会落在非空单元格则拒写以保护原数据（返回 partial_success）
+- `columns` (array<string>) — 列名字符串数组，顺序与 `data` 中每行取值一一对应
+- `data` (array<array<string|number|boolean|null>>) — 数据行；每行是一个数组，长度必须等于 `columns` 数
+- `dtypes` (object?) — 可选
+- `formats` (object?) — 可选
+
+### `+table-put` `--styles`
+
+
+**数组项**（类型 object）：
+- `cell_merges` (array<object>?) — 单元格合并操作数组；range 使用 A1 单元格范围，merge_type 默认 all each: { merge_type?: enum, range: string }
+- `cell_styles` (array<object>?) — 单元格样式操作数组；每项用 A1 单元格 range 指定范围，字段名与 +cells-set-style 对齐 each: { background_color?: string, border_styles?: object, font_color?: string, font_line?: enum, font_size?: number, …共 12 项 }
+- `col_sizes` (array<object>?) — 列宽操作数组；range 使用列范围如 A:C，type 为 pixel/standard，pixel 需要 size each: { range: string, size?: number, type: enum }
+- `name` (string) — 子表名
+- `row_sizes` (array<object>?) — 行高操作数组；range 使用行范围如 1:3，type 为 pixel/standard/auto，pixel 需要 size each: { range: string, size?: number, type: enum }
+
 ## Examples
 
 公共四件套：所有 shortcut 顶部排列 `--url` / `--spreadsheet-token` / `--sheet-id` / `--sheet-name`（XOR）。
 
 ### `+cells-set` 的拆分与转介绍
 
-"工具选择"段已讲清纯值（`+csv-put`）vs 富写入（`+cells-set`）。下表补 CLI 侧的 `+cells-set` **兄弟拆分**，以及不属于本 skill 的**跨 skill 转介绍**——避免 agent 用 `+cells-set` 硬扛所有写入场景。
+"工具选择"段已讲清纯值（`+csv-put`）vs 富写入（`+cells-set`）。下表补 CLI 侧的 `+cells-set` **兄弟拆分**，以及不属于本 reference 的**跨 reference 转介绍**——避免 agent 用 `+cells-set` 硬扛所有写入场景。
 
 | 写入场景 | 用这个 | 不要用 |
 |---------|--------|--------|
 | 只改**已有 cell 的样式**，不动 value/formula | `+cells-set-style` | `+cells-set`（会触发不必要的值写入） |
 | 把**单张图片嵌入**到某个 cell | `+cells-set-image` | `+cells-set`（参数更繁琐） |
-| **插行/列 + 写入** 这种多步组合，且要原子 | `+batch-update`（跨 skill） | 多次独立 `+cells-set`（非原子；插入会扰动后续 range） |
-| 在**多个不连续 range** 上应用同一组样式 | `+cells-batch-set-style`（跨 skill） | 多次 `+cells-set-style`（非原子） |
+| **插行/列 + 写入** 这种多步组合，且要原子 | `+batch-update`（见 lark-sheets-batch-update） | 多次独立 `+cells-set`（非原子；插入会扰动后续 range） |
+| 在**多个不连续 range** 上应用同一组样式 | `+cells-batch-set-style`（见 lark-sheets-batch-update） | 多次 `+cells-set-style`（非原子） |
 
 ### `+cells-set`
 
@@ -413,15 +454,35 @@ lark-cli sheets +csv-put --spreadsheet-token shtXXX --sheet-id "$SID" \
   --start-cell "A1" --csv @data.csv
 ```
 
-> `+csv-put` 比 `+cells-set` 短得多——只想批量灌纯值时优先用它。需要公式/样式才换 `+cells-set`。
+> `+csv-put` 比 `+cells-set` 短得多——批量灌值或公式时优先用它。需要样式/批注/图片才换 `+cells-set`。
 >
-> ⚠️ `=` 开头的字符串会被当字面量写入（**不会变公式**）：
+> ✅ `=` 开头的单元格会被当作公式计算（不是字面量文本）：
 >
 > ```bash
 > lark-cli sheets +csv-put --url "..." --sheet-name "Sheet1" \
 >   --start-cell "A1" \
 >   --csv $'name,score\nalice,=SUM(B2:B10)'
-> # ↑ A2 实际写入字符串 "=SUM(B2:B10)"，**不是公式**。需要写公式请用 +cells-set。
+> # ↑ B2 写入公式 =SUM(B2:B10)，读回 formula 保留、value 为计算结果。
+> # 反过来：无法用 +csv-put 写「= 开头的字面量文本」（会被当公式）；样式/批注/图片仍用 +cells-set。
+> ```
+>
+> ⚠️ **公式内部含逗号 / 引号必须 RFC 4180 转义**：CSV 用逗号分隔字段，公式里的逗号（如 `COUNTIF(D5:D22,"及格")` 的参数分隔逗号）会被解析器当成字段分隔符，把一格拆成多格、整块二维结构压扁错位。规则：**含逗号的字段整格用双引号包裹，字段内部的引号再翻倍**：
+>
+> ```bash
+> # 从 G4 写一个 2 列 3 行的统计块；=COUNTIF 含逗号 + 内部引号，必须转义
+> lark-cli sheets +csv-put --url "..." --sheet-name "Sheet1" \
+>   --start-cell "G4" \
+>   --csv $'统计项,结果\n成绩总和,=SUM(C5:C22)\n及格人数,"=COUNTIF(D5:D22,""及格"")"'
+> # ↑ "=COUNTIF(D5:D22,""及格"")"：外层双引号包裹整格，内部 "及格" 的引号翻倍成 ""及格""。
+> # 裸写 =COUNTIF(D5:D22,"及格") 会被 CSV 按逗号拆成两格、写入区域从 G4:H6 错位成 G4:K4。
+> ```
+>
+> 💡 **含逗号 / 引号 / 换行的公式优先用 `+cells-set`（JSON 二维数组）或 `+table-put` 写入**——结构化写入直接传 `formula` 字段，没有 CSV 转义负担，从根上杜绝拆列错位。准备给 `+csv-put` 的公式加逗号时，先考虑换结构化写入：
+>
+> ```bash
+> # 同样的统计块，结构化写入无需任何转义
+> lark-cli sheets +cells-set --url "..." --sheet-name "Sheet1" --range "G4:H6" \
+>   --cells '[[{"value":"统计项"},{"value":"结果"}],[{"value":"成绩总和"},{"formula":"=SUM(C5:C22)"}],[{"value":"及格人数"},{"formula":"=COUNTIF(D5:D22,\"及格\")"}]]'
 > ```
 
 > **定位 + 写入边界（关键，避免误覆盖）**：
@@ -430,8 +491,116 @@ lark-cli sheets +csv-put --spreadsheet-token shtXXX --sheet-id "$SID" \
 > - dry-run 与成功响应都回显 `writes_range`（实际落区，如 `B2:D4`）：**写前先 `--dry-run` 看一眼落区**，确认不会盖到相邻数据。
 > - 要保护非空 cell：`--allow-overwrite=false`（落区内出现非空 cell 即报错）。
 
+### `+table-put`（DataFrame → 飞书，类型保真写入）
+
+把结构化数据（DataFrame、list of dict、Counter）类型保真写入**已有**表（写入语义同 `+cells-set`）。协议形状**对齐 pandas `to_json(orient="split")`**：`columns:[列名]` + `data:[[行...]]`，可选 `dtypes:{列名:pandas_dtype}` 决定每列类型（number 保精度、date 落真日期），可选 `formats:{列名:number_format}` 覆盖显示格式（千分位 / 百分比 / 自定义日期）。dtypes 缺失时整张表按 string 写入（带 `@` 文本格式，邮编 / 订单号等含前导零的 id 保真）。
+
+只写入**已有**表（`--url` / `--spreadsheet-token` 二选一必填），不新建工作簿——**要新建表格直接用 `+workbook-create --sheets`**（同协议、一步建表 + 类型保真写入，详见 workbook reference）。读回用镜像命令 `+table-get`（见 read-data reference），输出与 `--sheets` 同构、可 round-trip。
+
+```bash
+# sheet 按 name 匹配、缺则新建；多 DataFrame 经 stdin 一次写多 sheet
+python export.py | lark-cli sheets +table-put --url "<表URL>" --sheets -
+# 某 sheet 带 "mode":"append" 追加到已有数据末尾、默认不重复表头
+lark-cli sheets +table-put --spreadsheet-token "<token>" --sheets @payload.json
+```
+
+每个 sheet 还可带 `"allow_overwrite": false`（遇非空拒写、保护原数据）、`"header": false`（只写数据不写表头）。完整字段跑 `+table-put --print-schema --flag-name sheets`。
+
+#### DataFrame → 协议（5 行 helper）
+
+pandas 的 `df.to_json(orient="split", date_format="iso")` 一步完成所有清洗（NaN→null、Timestamp→ISO 字符串、numpy 标量→原生数字），helper 只要把 dtypes 拼上去——5 行覆盖单 / 多 sheet：
+
+```python
+import json
+def df_to_sheet(df, name, formats=None):
+    return {"name": name,
+            **json.loads(df.to_json(orient="split", date_format="iso")),
+            "dtypes": df.dtypes.astype(str).to_dict(),
+            **({"formats": formats} if formats else {})}
+
+# 单 sheet（显式 format 覆盖默认显示）
+payload = {"sheets": [df_to_sheet(df, "销售", {"营收": "#,##0.00", "毛利率": "0.0%"})]}
+
+# 多 sheet——helper 让每个 sheet 一行，不再重复 boilerplate
+payload = {"sheets": [df_to_sheet(df1, "销售"),
+                      df_to_sheet(df2, "成本"),
+                      df_to_sheet(df3, "利润")]}
+```
+
+> **CSV-shaped 全文本数据**（不需要类型保真、含前导零的 id 也要保留）省掉 dtypes 即可，inline 一行写完，不必走 helper（注意保留 `date_format="iso"`，否则 datetime 列会被序列化成 epoch 毫秒数字，CLI 拒绝）：
+> ```python
+> payload = {"sheets": [{"name": "原始",
+>                        **json.loads(df.to_json(orient="split", date_format="iso"))}]}
+> ```
+> **别把 `to_json + json.loads` 换成 `df.to_dict(orient="split")`**：会留 `numpy.int64` 让 `json.dumps` 后续报 "not serializable"——这一步是清洗的关键。
+
+不用 pandas 也行——typed 协议就是纯 JSON。手写场景：
+
+```python
+# Counter / dict / 手拼数据：直接写 columns + data，按需加 dtypes/formats
+payload = {"sheets": [{
+    "name": "渠道",
+    "columns": ["channel", "count", "rate"],
+    "data": [["app", 1240, 0.62], ["web", 760, 0.38]],
+    "dtypes": {"count": "int64", "rate": "float64"},
+    "formats": {"rate": "0.0%"},
+}]}
+```
+
+> **dtype 速查**：`int64`/`float64`（数值）、`Int64`（含空值的整数，nullable）、`bool`/`boolean`、`datetime64[ns]`（date，默认 `yyyy-mm-dd`）、`object`（string）。pandas dtype 字符串原样塞进 dtypes 即可，CLI 端按前缀匹配（`int*`/`uint*`/`Int*`/`float*` → number 等）。未识别 dtype 兜底为 string。
+
+#### `--dataframe`（Arrow IPC / Feather v2 二进制入口）
+
+`--dataframe` 与 `--sheets` 互斥、功能等价，但走二进制 wire——pandas `df.to_feather()` 写出的 Arrow IPC 文件直接喂 CLI，类型从 Arrow schema 自动恢复，**不用再手填 dtypes/formats**，也自动绕过 NaT / NaN / `datetime64[ns, tz]` 的 JSON 序列化坑。子表落点固定为 `Sheet1`、A1 起覆盖写、带表头；要换子表名 / 起始位置 / 多子表，回到 `--sheets` JSON 协议。
+
+```bash
+# 文件（cwd 相对路径；受 SafePath 沙箱约束，不接受绝对路径）
+lark-cli sheets +table-put --url "<表URL>" --dataframe @./in.arrow
+# stdin 二进制（不落盘）
+python prepare.py | lark-cli sheets +table-put --url "<表URL>" --dataframe -
+```
+
+```python
+import io, subprocess, pandas as pd
+df = pd.DataFrame({"date": pd.to_datetime(["2024-01-15"]), "amount": [1234.5], "id": ["00123"]})
+
+# 1) 文件
+df.to_feather("./in.arrow")  # 写到当前目录
+subprocess.run(["lark-cli","sheets","+table-put","--url",URL,"--dataframe","@./in.arrow"], check=True)
+
+# 2) stdin（不落盘）—— pandas 写 BytesIO，subprocess 把 buf 灌进去
+buf = io.BytesIO(); df.to_feather(buf)
+subprocess.run(["lark-cli","sheets","+table-put","--url",URL,"--dataframe","-"],
+               input=buf.getvalue(), check=True)
+```
+
+> 每列的 `number_format` 写在 Arrow Field metadata 里，CLI 端自动透传到飞书显示格式（千分位 / 百分比 / 自定义日期等）：
+> ```python
+> import pyarrow as pa, pyarrow.feather as feather
+> table = pa.Table.from_pandas(df)
+> schema = table.schema.set(
+>     table.schema.get_field_index("amount"),
+>     pa.field("amount", pa.float64(), metadata={b"number_format": b"#,##0.00"}))
+> feather.write_feather(table.cast(schema), "./in.arrow")
+> ```
+
+#### `--styles`（写入时同时套样式）
+
+`--styles` 在 typed 写入后顺带应用视觉处理，省掉一次 `+cells-set-style` 往返。协议与 `+workbook-create --styles` **完全同构**（详见 workbook reference）：顶层 `{styles:[...]}`，数组每项对应一个被写入的子表、含 `name`，并按能力拆成四类可选数组——`cell_styles`（A1 单元格 range + 扁平样式字段，含 `number_format` / 颜色 / 对齐 / `border_styles`，随内容在同一次写入里一并应用）、`cell_merges`、`row_sizes`、`col_sizes`。styles 数组的长度 / 顺序 / name 必须与被写入的子表对应：配 `--sheets` 时与 `--sheets.sheets` 对齐；配 `--dataframe`（单子表，名为 `Sheet1`）时只给一个 name 为 `Sheet1` 的 styles 项。
+
+```bash
+lark-cli sheets +table-put --url "<表URL>" \
+  --sheets '{"sheets":[{"name":"明细","columns":["日期","金额"],"dtypes":{"日期":"datetime64[ns]","金额":"float64"},"formats":{"金额":"#,##0.00"},"data":[["2024-01-15",1234.5]]}]}' \
+  --styles '{"styles":[{"name":"明细",
+    "cell_styles":[{"range":"A1:B1","font_weight":"bold","background_color":"#f5f5f5","horizontal_alignment":"center"}],
+    "cell_merges":[{"range":"A1:B1"}],
+    "col_sizes":[{"range":"A:B","type":"pixel","size":120}]}]}'
+```
+
+完整字段跑 `+table-put --print-schema --flag-name styles`。
+
 ### Validate / DryRun / Execute 约束
 
-- `Validate`：XOR 公共四件套；`+cells-set` 的 `--cells` 必须能解析为 JSON 二维矩阵且行列数与 `--range` 完全一致；`+cells-set-style` 的样式 flag 至少一个非空（或带 `--border-styles`）；`+cells-set-image` 的 `--range` 必须是单 cell（起止 cell 相同）；`+csv-put` 的 `--csv` 必须能按 RFC 4180 解析；防爆参数上限校验。
+- `Validate`：XOR 公共四件套；`+cells-set` 的 `--cells` 必须能解析为 JSON 二维矩阵且行列数与 `--range` 完全一致；`+cells-set-style` 的样式 flag 至少一个非空（或带 `--border-styles`）；`+cells-set-image` 的 `--range` 必须是单 cell（起止 cell 相同）；`+csv-put` 的 `--csv` 必须能按 RFC 4180 解析；`+table-put` 给了 `--styles` 则按子表名 / 顺序 / 数量与 `--sheets`（或 `--dataframe` 的单子表 `Sheet1`）对齐校验；防爆参数上限校验。
 - `DryRun`：输出目标 range + 推断尺寸 + 是否覆盖非空 cell 警告，零网络副作用。
 - `Execute`：写后不自动回读；如需确认，自行调用 `+cells-get --range <写入区域> --include value,formula` 抽样核对。

tests/cli_e2e/sheets/sheets_crud_workflow_test.go

@@ -143,14 +143,14 @@ func TestSheets_CRUDE2EWorkflow(t *testing.T) {
 		assert.True(t, len(matchedCells.Array()) > 0, "should find at least one cell containing 'Alice'")
 	})
 
-	t.Run("export spreadsheet with +export as bot", func(t *testing.T) {
+	t.Run("export spreadsheet with +workbook-export as bot", func(t *testing.T) {
 		require.NotEmpty(t, spreadsheetToken, "spreadsheet token is required")
 		outputDir := t.TempDir()
 		outputPath := filepath.Join(outputDir, "export.xlsx")
 
 		result, err := clie2e.RunCmd(ctx, clie2e.Request{
 			Args: []string{
-				"sheets", "+export",
+				"sheets", "+workbook-export",
 				"--spreadsheet-token", spreadsheetToken,
 				"--file-extension", "xlsx",
 				"--output-path", "./export.xlsx",

tests/cli_e2e/sheets/sheets_table_get_dryrun_test.go

@@ -0,0 +1,76 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package sheets
+
+import (
+	"context"
+	"testing"
+	"time"
+
+	clie2e "github.com/larksuite/cli/tests/cli_e2e"
+	"github.com/stretchr/testify/require"
+	"github.com/tidwall/gjson"
+)
+
+// TestSheets_TableGetDefaultDryRun pins the request structure +table-get emits
+// when no --range is given: it must first read get_workbook_structure (to learn
+// each sheet's grid dimensions, which anchor the used-range probe over the full
+// grid) and then read cells via get_cell_ranges. This guards the pro016 / pro025
+// fix — the default read must span internal blank rows/columns, not stop at the
+// A1 current region.
+func TestSheets_TableGetDefaultDryRun(t *testing.T) {
+	setSheetsDryRunEnv(t)
+
+	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
+	t.Cleanup(cancel)
+
+	result, err := clie2e.RunCmd(ctx, clie2e.Request{
+		Args: []string{
+			"sheets", "+table-get",
+			"--spreadsheet-token", "shtDryRun",
+			"--dry-run",
+		},
+		DefaultAs: "user",
+	})
+	require.NoError(t, err)
+	result.AssertExitCode(t, 0)
+
+	out := result.Stdout
+
+	// api.0 — the structure read that supplies the grid dimensions.
+	require.Equal(t, "get_workbook_structure", gjson.Get(out, "api.0.body.tool_name").String(), "stdout:\n%s", out)
+	require.Equal(t, "/open-apis/sheet_ai/v2/spreadsheets/shtDryRun/tools/invoke_read",
+		gjson.Get(out, "api.0.url").String(), "stdout:\n%s", out)
+
+	// api.1 — the cells read.
+	require.Equal(t, "get_cell_ranges", gjson.Get(out, "api.1.body.tool_name").String(), "stdout:\n%s", out)
+}
+
+// TestSheets_TableGetSingleSheetDryRun confirms the single-sheet selector path
+// also reads get_workbook_structure now (previously it did not): the grid
+// dimensions are needed even when only one sheet is read, so the used-range
+// probe can anchor over the full grid.
+func TestSheets_TableGetSingleSheetDryRun(t *testing.T) {
+	setSheetsDryRunEnv(t)
+
+	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
+	t.Cleanup(cancel)
+
+	result, err := clie2e.RunCmd(ctx, clie2e.Request{
+		Args: []string{
+			"sheets", "+table-get",
+			"--spreadsheet-token", "shtDryRun",
+			"--sheet-name", "Sheet1",
+			"--dry-run",
+		},
+		DefaultAs: "user",
+	})
+	require.NoError(t, err)
+	result.AssertExitCode(t, 0)
+
+	out := result.Stdout
+	require.Equal(t, "get_workbook_structure", gjson.Get(out, "api.0.body.tool_name").String(),
+		"single-sheet path must still read the structure for grid dimensions; stdout:\n%s", out)
+	require.Equal(t, "get_cell_ranges", gjson.Get(out, "api.1.body.tool_name").String(), "stdout:\n%s", out)
+}

tests/cli_e2e/sheets/sheets_table_get_workflow_test.go

@@ -0,0 +1,118 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package sheets
+
+import (
+	"context"
+	"encoding/json"
+	"testing"
+	"time"
+
+	clie2e "github.com/larksuite/cli/tests/cli_e2e"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"github.com/tidwall/gjson"
+)
+
+// TestSheets_TableGetUsedRangeWorkflow is the live regression for the pro016 /
+// pro025 incident: data with an internal blank row (and a blank separator
+// column) must read back in full when +table-get is run without --range. Before
+// the fix, the default read used the A1 current region, which stopped at the
+// first blank row/column and silently truncated everything past it.
+//
+// Layout written (header + 9 data rows, blank row at sheet row 6, blank column
+// D between the A:C block and the E:F block):
+//
+//	row 1:  name  age  city  <blank>  x1  x2
+//	rows 2-5: data
+//	row 6:  <entirely blank>
+//	rows 7-10: data
+//
+// The true used range is A1:F10. The default +table-get must return all 9 data
+// rows and 6 columns and report a range covering row 10 / column F.
+func TestSheets_TableGetUsedRangeWorkflow(t *testing.T) {
+	parentT := t
+	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Minute)
+	t.Cleanup(cancel)
+
+	suffix := clie2e.GenerateSuffix()
+	spreadsheetToken := createSpreadsheet(t, parentT, ctx, "lark-cli-e2e-tableget-"+suffix, "bot")
+
+	infoRes, err := clie2e.RunCmd(ctx, clie2e.Request{
+		Args:      []string{"sheets", "+info", "--spreadsheet-token", spreadsheetToken},
+		DefaultAs: "bot",
+	})
+	require.NoError(t, err)
+	infoRes.AssertExitCode(t, 0)
+	sheetID := gjson.Get(infoRes.Stdout, "data.sheets.sheets.0.sheet_id").String()
+	require.NotEmpty(t, sheetID, "sheet_id should not be empty, stdout: %s", infoRes.Stdout)
+
+	// Write the full A1:F10 block in one shot. Column D and row 6 are left blank
+	// (empty strings) so the A1 current region would truncate at them.
+	blank := ""
+	values := [][]any{
+		{"name", "age", "city", blank, "x1", "x2"},
+		{"Alice", 30, "NY", blank, "p", "q"},
+		{"Bob", 25, "LA", blank, "r", "s"},
+		{"Carol", 40, "SF", blank, "t", "u"},
+		{"Dave", 22, "TX", blank, "v", "w"},
+		{blank, blank, blank, blank, blank, blank}, // row 6: entirely blank
+		{"Eve", 33, "BOS", blank, "a", "b"},
+		{"Frank", 28, "SEA", blank, "c", "d"},
+		{"Grace", 45, "DEN", blank, "e", "f"},
+		{"Hank", 50, "PHX", blank, "g", "h"},
+	}
+	valuesJSON, _ := json.Marshal(values)
+
+	writeRes, err := clie2e.RunCmd(ctx, clie2e.Request{
+		Args: []string{
+			"sheets", "+write",
+			"--spreadsheet-token", spreadsheetToken,
+			"--sheet-id", sheetID,
+			"--range", "A1:F10",
+			"--values", string(valuesJSON),
+		},
+		DefaultAs: "bot",
+	})
+	require.NoError(t, err)
+	writeRes.AssertExitCode(t, 0)
+	writeRes.AssertStdoutStatus(t, true)
+
+	t.Run("default table-get spans the internal blank row/column", func(t *testing.T) {
+		result, err := clie2e.RunCmd(ctx, clie2e.Request{
+			Args: []string{
+				"sheets", "+table-get",
+				"--spreadsheet-token", spreadsheetToken,
+				"--sheet-id", sheetID,
+			},
+			DefaultAs: "bot",
+		})
+		require.NoError(t, err)
+		result.AssertExitCode(t, 0)
+		result.AssertStdoutStatus(t, true)
+
+		sheet := gjson.Get(result.Stdout, "data.sheets.0")
+		rows := sheet.Get("data").Array()
+		require.Equal(t, 9, len(rows),
+			"default table-get must return all 9 data rows (not truncate at the blank row 6); stdout:\n%s", result.Stdout)
+
+		// The last data row must be present — the regression dropped everything
+		// after the blank row.
+		lastRow := rows[len(rows)-1].Array()
+		assert.Equal(t, "Hank", lastRow[0].String(), "last data row should be Hank; stdout:\n%s", result.Stdout)
+
+		// Columns must span past the blank separator column D to reach x1 / x2.
+		cols := sheet.Get("columns").Array()
+		require.Equal(t, 6, len(cols), "must read all 6 columns across the blank column D; stdout:\n%s", result.Stdout)
+		assert.Equal(t, "x1", cols[4].String())
+		assert.Equal(t, "x2", cols[5].String())
+
+		// The reported range must cover the true used range (row 10, column F),
+		// so a caller can detect truncation by inspecting it.
+		rng := sheet.Get("range").String()
+		require.NotEmpty(t, rng, "table-get output must report the range actually read; stdout:\n%s", result.Stdout)
+		assert.Contains(t, rng, "10", "reported range should reach row 10; got %q", rng)
+		assert.Contains(t, rng, "F", "reported range should reach column F; got %q", rng)
+	})
+}

tests/cli_e2e/sheets/sheets_table_put_dryrun_test.go

@@ -0,0 +1,85 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package sheets
+
+import (
+	"context"
+	"strings"
+	"testing"
+	"time"
+
+	clie2e "github.com/larksuite/cli/tests/cli_e2e"
+	"github.com/stretchr/testify/require"
+	"github.com/tidwall/gjson"
+)
+
+// TestSheets_TablePutStylesDryRun pins the request structure +table-put emits
+// when --styles is supplied: the set_cell_range write carries cell_styles merged
+// into the matrix, and the structural styles (merge / resize) render as their
+// own invoke_write tool calls afterward — the same shape +workbook-create uses.
+func TestSheets_TablePutStylesDryRun(t *testing.T) {
+	setSheetsDryRunEnv(t)
+
+	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
+	t.Cleanup(cancel)
+
+	result, err := clie2e.RunCmd(ctx, clie2e.Request{
+		Args: []string{
+			"sheets", "+table-put",
+			"--spreadsheet-token", "shtDryRun",
+			"--sheets", `{"sheets":[{"name":"数据","columns":["a","b"],"data":[["x","y"]]}]}`,
+			"--styles", `{"styles":[{"name":"数据","cell_styles":[{"range":"A1:B1","font_weight":"bold"}],"cell_merges":[{"range":"A1:B1"}],"col_sizes":[{"range":"A:A","type":"pixel","size":120}]}]}`,
+			"--dry-run",
+		},
+		DefaultAs: "user",
+	})
+	require.NoError(t, err)
+	result.AssertExitCode(t, 0)
+
+	out := result.Stdout
+
+	// api.0 — the typed write, with cell_styles merged into the cells matrix.
+	require.Equal(t, "POST", gjson.Get(out, "api.0.method").String(), "stdout:\n%s", out)
+	require.Equal(t, "/open-apis/sheet_ai/v2/spreadsheets/shtDryRun/tools/invoke_write",
+		gjson.Get(out, "api.0.url").String(), "stdout:\n%s", out)
+	require.Equal(t, "set_cell_range", gjson.Get(out, "api.0.body.tool_name").String(), "stdout:\n%s", out)
+	firstInput := gjson.Get(out, "api.0.body.input").String()
+	require.Contains(t, firstInput, `"font_weight":"bold"`, "cell_styles should merge into the matrix; input:\n%s", firstInput)
+	require.Contains(t, firstInput, `"range":"A1:B2"`, "write range should cover header + data; input:\n%s", firstInput)
+
+	// api.1 — the merge op.
+	require.Equal(t, "merge_cells", gjson.Get(out, "api.1.body.tool_name").String(), "stdout:\n%s", out)
+	require.Contains(t, gjson.Get(out, "api.1.body.input").String(), `"range":"A1:B1"`, "stdout:\n%s", out)
+
+	// api.2 — the column resize.
+	require.Equal(t, "resize_range", gjson.Get(out, "api.2.body.tool_name").String(), "stdout:\n%s", out)
+	require.Contains(t, gjson.Get(out, "api.2.body.input").String(), `"type":"pixel"`, "stdout:\n%s", out)
+}
+
+// TestSheets_TablePutStylesNameMismatchRejected confirms a --styles item whose
+// name does not match the --sheets payload sheet is rejected up front (no write
+// lands), so a typo surfaces as a validation error rather than a silent skip.
+func TestSheets_TablePutStylesNameMismatchRejected(t *testing.T) {
+	setSheetsDryRunEnv(t)
+
+	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
+	t.Cleanup(cancel)
+
+	result, err := clie2e.RunCmd(ctx, clie2e.Request{
+		Args: []string{
+			"sheets", "+table-put",
+			"--spreadsheet-token", "shtDryRun",
+			"--sheets", `{"sheets":[{"name":"数据","columns":["a"],"data":[["x"]]}]}`,
+			"--styles", `{"styles":[{"name":"其他","cell_styles":[{"range":"A1","font_weight":"bold"}]}]}`,
+			"--dry-run",
+		},
+		DefaultAs: "user",
+	})
+	require.NoError(t, err)
+	result.AssertExitCode(t, 2)
+	combined := result.Stdout + "\n" + result.Stderr
+	if !strings.Contains(combined, "must match") {
+		t.Fatalf("expected name-mismatch error, got:\nstdout:\n%s\nstderr:\n%s", result.Stdout, result.Stderr)
+	}
+}