快速开始Quick start
先完成工作区和模型两项基础设置，再开始第一个 Agent 任务。Set up a workspace and a model first, then start your first Agent task.
打开项目目录Open a project folder
使用顶部“文件 → 打开文件夹”，或点击左侧资源管理器的文件夹按钮。打开后，文件树、终端和 Agent 都以该目录作为当前工作区。Use File → Open Folder, or the folder button in Explorer. The file tree, terminal, and Agent will use that folder as the active workspace.
配置并测试模型Configure and test a model
打开左下角设置，进入“AI 设置”，填写供应商协议、地址、密钥和模型 ID。点击模型卡片上的测试按钮，确认连接成功后保存配置。Open Settings from the lower-left corner, choose AI Settings, then enter the provider protocol, base URL, key, and model ID. Test the model and save only after the connection succeeds.
发起第一个任务Start your first task
在右侧 Agent 面板选择 Ask、Plan 或 Agent 模式，确认模型、思考强度与文件权限，再输入目标。涉及修改时，建议明确要求“先检查、再修改、最后运行验证”。In the right Agent pane, choose Ask, Plan, or Agent, confirm the model, reasoning level, and file permission, then describe the goal. For code changes, ask it to inspect, edit, and verify in that order.
“先阅读项目根目录说明，概括项目结构和构建命令，不修改文件。”这能同时验证模型、工作区读取和中文输出是否正常。“Read the repository instructions, summarize the structure and build commands, and do not modify files.” This checks model access, workspace reading, and output quality without changing the project.
工作台与面板操作Workbench and panel controls
RoverCode 以“文件树—编辑器—Agent”为主轴，底部终端按需展开。RoverCode centers on Explorer, editor groups, and the Agent pane, with an on-demand terminal below.
左侧：活动栏与文件树Left: activity bar and Explorer
活动栏切换资源管理器、插件管理器和插件提供的侧视图。文件树包含工作区、全文搜索结果和最近文件。The activity bar switches Explorer, Plugin Manager, and plugin-provided views. Explorer includes the workspace tree, search results, and recent files.
中间：编辑器组Center: editor groups
支持多标签、左右或上下分屏、拖拽重排、Markdown 源码分栏，以及 PDF、Office、图片、网页和思维导图预览。Supports tabs, horizontal or vertical splits, drag reordering, Markdown source splits, and previews for PDF, Office, images, web pages, and mind maps.
右侧：Agent 面板Right: Agent pane
管理会话、上下文、模型、思考强度、权限、斜杠命令与执行状态。面板宽度可拖动分隔线调整。Manages sessions, context, model, reasoning, permissions, slash commands, and execution state. Drag the divider to resize it.
顶部三个面板按钮The three top panel buttons
显示或隐藏左侧面板；再次打开会恢复上次宽度和活动视图。Shows or hides the left pane and restores its last width and active view.
在编辑器、编辑器+终端、左右侧面板模式之间循环；进入侧面板模式时编辑区可暂时收起。Cycles through editor, editor plus terminal, and side-panels mode; side-panels mode can temporarily hide the editor.
显示或隐藏右侧 Agent；正在运行的会话不会因为面板隐藏而自动停止。Shows or hides the Agent pane. Hiding it does not automatically stop a running session.
左、右面板分隔线和终端上边缘都可拖动。窗口尺寸、面板开关、面板宽度、终端高度和编辑器缩放会保存到本机布局状态。Drag the left and right dividers and the terminal's top edge. Window size, pane visibility, pane widths, terminal height, and editor zoom are stored locally.
模型配置Model configuration
供应商负责连接方式，模型卡片负责具体能力和上下文参数。Providers define the connection; model cards define capabilities and token limits.
选择端点协议Choose an endpoint protocol
| 类型Type | 适用场景Use when | 注意事项Notes |
|---|---|---|
openai-chat | 服务兼容 OpenAI Chat Completions，或本地服务提供 OpenAI 风格接口。The service is compatible with OpenAI Chat Completions or exposes an OpenAI-style local API. | 兼容面最广，适合多数第三方中转和本地模型服务。Broad compatibility for gateways and local model servers. |
openai-responses | 服务原生支持 OpenAI Responses API，尤其适合对应的 GPT 模型。The service natively supports the OpenAI Responses API, especially for compatible GPT models. | 工具调用和流式行为由 Responses 协议处理，部分模型响应速度更好。Tool calls and streaming follow the Responses protocol; some models may respond faster. |
anthropic | 服务兼容 Anthropic Messages API。The service is compatible with Anthropic Messages. | 需要额外测试自定义请求头、代理和供应商兼容性。Test custom headers, proxy behavior, and provider compatibility. |
供应商字段Provider fields
- Name：列表中显示的供应商名称。: Display name in the provider list.
- Type：三种协议之一。: One of the three protocols.
- Base URL：接口根地址；本地 OpenAI 兼容服务常见为 : API root. A common local OpenAI-compatible URL is
http://127.0.0.1:11434/v1。. - Key：API 密钥，仅保存在用户本机配置。: API key stored only in the local user config.
- Proxy：可选 HTTP 代理地址。: Optional HTTP proxy URL.
- Headers：JSON 格式的自定义请求头。: Custom request headers in JSON format.
模型字段Model fields
- API model：请求实际使用的模型 ID，必须与供应商完全一致。: Exact model ID sent to the provider.
- Display name：界面中的友好名称。: Friendly name shown in the UI.
- Reasoning：默认思考强度及可选档位。: Default reasoning level and allowed levels.
- Context / Max context：普通与 Max 模式上下文上限。: Context limits for normal and Max modes.
- Max output：单次最大输出 token。: Maximum output tokens per response.
- Capabilities：Agent、图像、Cmd+K、快速应用、思考、启用、Subagent 优先。: Agent, images, Cmd+K, fast apply, thinking, enabled, and preferred subagent.
添加模型的三种方式Three ways to add models
填写 Base URL 和 Key 后点击“拉取模型”，RoverCode 请求供应商的模型列表，再让你勾选导入。After entering Base URL and Key, fetch the provider's model list and select what to import.
按名称搜索内置目录，自动带入上下文、输出、思考和图像等公开参数；导入后仍应核对供应商实际限制。Search the built-in catalog to prefill public context, output, reasoning, and image metadata, then verify the provider's actual limits.
适合私有部署、新模型或目录未收录的 ID。至少填写 API model，并手动确认能力开关与 token 上限。Best for private deployments, new models, or uncatalogued IDs. Enter the API model and verify every capability and token limit.
先测试供应商地址与密钥，再测试每个模型。若普通对话成功但 Agent 失败，重点检查模型的 Agent / tool calling 能力、协议类型和自定义请求头。保存后的源文件位于 Test the provider URL and key first, then each model. If chat works but Agent fails, check Agent/tool-calling capability, protocol type, and custom headers. The saved source file is %USERPROFILE%\.rovercode\config\providers.json，不要将其提交到仓库。 and must not be committed to a repository.
Agent 工作流Agent workflow
让模型回答、规划或真正执行任务，并把文件、选区和图片作为明确上下文。Ask, plan, or execute real work with explicit file, selection, and image context.
问答与分析Questions and analysis
适合解释代码、比较方案、阅读资料。默认不应主动修改文件。For explanations, comparisons, and reading. It should not modify files unless requested.
计划与拆解Planning and breakdown
先明确假设、步骤、风险和验证标准，适合范围较大或需要决策的任务。Clarifies assumptions, steps, risks, and verification before implementation.
执行与验证Execution and verification
读取、搜索、编辑、运行允许的命令，并根据真实工具结果继续下一轮。Reads, searches, edits, runs approved commands, and continues from real tool results.
输入框工具栏从左到右Composer toolbar, left to right
添加上下文Add context
- 从文件树把文件或目录拖到 Agent 面板。Drag files or folders from Explorer into the Agent pane.
- 文件树右键选择“添加到 Agent 聊天”。Use Add to Agent Chat from the Explorer context menu.
- 在编辑器选中文本后按 Select text in the editor and press Alt + L。.
- 粘贴图片或从 Markdown 图片右键加入上下文。Paste images or add Markdown images from the context menu.
- 自动上下文会附带当前文件与选区状态，可在发送前移除不需要的引用。Auto context can include the active file and selection; remove irrelevant references before sending.
会话管理Session management
- 顶部按钮用于新建、搜索和导出会话。Top buttons create, search, and export sessions.
- 会话选择器显示运行中与未读完成状态。The session picker shows running and unread-complete states.
- 右键会话可复制 ID、重命名或删除。Right-click a session to copy its ID, rename, or delete it.
- 可重新生成、编辑用户消息、重试失败消息和恢复中断任务。Regenerate, edit user messages, retry failures, and resume interrupted runs.
- 检查点可恢复任务前的文件状态；修改文件可直接打开差异视图检查。Checkpoints restore earlier file state, and modified files can open directly in diff view.
内置斜杠命令Built-in slash commands
/web-search为本次请求显式启用联网搜索和公开信息核验。Explicitly enables web search and public-source verification for the request./init为当前仓库生成贡献者指南 AGENTS.md。Generates repository contributor guidance in AGENTS.md./subagent要求使用 Subagent 并行处理可独立拆分的任务。Requests parallel Subagents for independent subtasks./compact主动压缩会话与工作区上下文后继续。Compacts conversation and workspace context before continuing./create-skill创建或更新工作区 AI 技能；项目技能和用户技能也会出现在同一菜单中。Creates or updates a workspace AI skill; workspace and user skills appear in the same menu.默认使用“请求审批”。“完全访问”允许 Agent 读写工作区外文件并运行允许的外部命令，只应在目标目录和影响范围明确时临时启用。Use Request approval by default. Full access allows reading and writing outside the workspace and running approved external commands; enable it only when the target and impact are clear.
文件树与编辑器Explorer and editor
从创建文件到拖拽分屏、全文搜索和大文件浏览，日常编辑不需要离开工作台。Create files, split by drag, search the workspace, and inspect large files without leaving the workbench.
文件树操作Explorer operations
- 顶部按钮：打开文件夹、刷新、新建文件、新建文件夹、新建思维导图、重命名、删除。Top actions: open folder, refresh, new file, new folder, new mind map, rename, and delete.
- 右键菜单：打开、分屏打开、剪切、粘贴、重命名、删除、在资源管理器中显示、加入 Agent 上下文。Context menu: open, open in split, cut, paste, rename, delete, reveal in Explorer, and add to Agent context.
- 拖动条目到文件夹可移动；拖到编辑器边缘可按方向分屏打开；拖到 Agent 可附加上下文。Drag entries onto folders to move, onto editor edges to open in a split, or into Agent to attach context.
- 搜索结果和最近文件区域可折叠、调整高度，并支持右键操作。Search Results and Recent sections can collapse, resize, and expose context actions.
标签与分屏Tabs and splits
- 拖动标签可重新排序、移动到其他编辑器组，或放到边缘创建新分屏。Drag tabs to reorder, move between groups, or drop on an edge to create a split.
- 按住 Hold Alt 点击标签，会向下分屏打开同一标签。 while clicking a tab to split it downward.
- 标签栏可用滚轮横向浏览；右键标签可关闭当前、关闭其他或选择分屏方向。Use the wheel to scroll the tab strip. Right-click for close, close others, and split actions.
- 工具栏提供向右分屏、向下分屏、保存、Markdown 源码分栏和跳转定义。The toolbar provides split right, split down, save, Markdown source split, and go to definition.
按 Press Ctrl + Shift + F 聚焦工作区搜索。输入内容后按 Enter，点击结果跳转到准确行列。 to focus workspace search. Press Enter, then click a result to jump to its exact line and column.
支持 UTF-8、UTF-8 BOM 与 GBK。状态栏显示当前文件信息；保存前确认编码，避免中文文件被错误转换。Supports UTF-8, UTF-8 BOM, and GBK. Check file metadata in the status bar before saving Chinese text files.
超大文本以分块只读方式加载，工具栏提供上一块和下一块。大文件模式下保存、替换和部分语言服务会被禁用。Very large text files load in read-only chunks with Previous and Next controls. Save, replace, and some language services are disabled.
跳转定义、搜索结果、文件打开和光标位置会形成编辑器导航历史。使用 Definitions, search results, file opens, and cursor locations build editor history. Use Alt + ← / →，或鼠标侧键，在历史位置之间前进后退。 or the mouse side buttons to move backward and forward.
Markdown 富编辑Markdown rich editing
直接在所见即所得界面修改 Markdown，同时保留源码分栏和格式往返。Edit Markdown directly in a rich view while keeping a synchronized source split and round-trip formatting.
富编辑视图Rich editor
选择文字后会出现浮动工具栏，可设置标题、链接、粗体、斜体、下划线、删除线、文字颜色、背景色、字号和文字效果。还支持列表、任务列表、引用、代码块、表格和图片。Selecting text opens a floating toolbar for headings, links, bold, italic, underline, strike, colors, font size, and effects. Lists, tasks, quotes, code blocks, tables, and images are supported.
源码分栏Source split
点击编辑器工具栏的代码图标，会在旁边打开同一文件的 Markdown 源码。两侧共享文档内容和滚动比例；再次点击即可关闭源码分栏。Click the code icon in the editor toolbar to open the same file as Markdown source beside the rich view. Both panes share content and scroll position; click again to close.
常用 Markdown 快捷键Common Markdown shortcuts
切换 H1–H6 标题。处于有序列表时会保留原编号，并把该项转换为独立标题。Toggle H1–H6. Inside an ordered list, the item number is preserved and converted into a standalone heading.
粗体。Bold.
斜体。Italic.
下划线。Underline.
有序列表。Ordered list.
无序列表。Bullet list.
段内换行。Line break inside a block.
撤销 / 重做；标题转换可一次撤销。Undo / redo; heading conversion is a single undo step.
图片与剪贴板Images and clipboard
可粘贴剪贴板图片、插入本地图片、复制或删除图片，并保持工作区相对路径。图片可直接加入 Agent 上下文。Paste clipboard images, insert local images, copy or delete them, and keep workspace-relative paths. Images can be added directly to Agent context.
搜索与 PDF 导出Search and PDF export
富编辑顶部工具栏提供当前文档搜索、区分大小写、上一个/下一个匹配和 PDF 导出。导出前先保存文档和相关图片资源。The rich-editor toolbar provides document search, case sensitivity, previous/next match, and PDF export. Save the document and image assets first.
文档、图片与思维导图Documents, images, and mind maps
常用资料可以直接留在项目目录，并在编辑器标签中预览或继续编辑。Keep working material inside the project and open it directly in editor tabs.
PDF 预览PDF preview
页码导航、缩放、适合宽度/页面和文本选择；适合查看规范、报告与导出结果。Page navigation, zoom, fit modes, and text selection for specifications, reports, and exports.
Word 预览Word preview
保留常用文本、段落、表格和分页布局；内部书签链接可在文档内跳转。Preserves common text, paragraph, table, and page layout; internal bookmark links navigate within the document.
Excel 预览Excel preview
切换工作表、查看格式化单元格和大表格；工作表标签支持键盘方向键、Home 与 End。Switch sheets and inspect formatted cells and large tables; sheet tabs support arrows, Home, and End.
PowerPoint 预览PowerPoint preview
缩略图、上一页/下一页、滚轮翻页、缩放和重置视图。Thumbnails, previous/next slide, wheel paging, zoom, and view reset.
图片预览Image preview
缩放、拖动、适合窗口、查看原始尺寸和复制图片，适合检查界面截图与设计素材。Zoom, pan, fit, inspect original dimensions, and copy images for UI screenshots and design assets.
思维导图Mind maps
从文件树新建，编辑主题、分支、链接、样式和图片，并支持 Markdown 大纲转换与图片导出。Create from Explorer, edit topics, branches, links, styles, and images, and convert Markdown outlines or export images.
Office/PDF/PPTX 主要用于查看，不等同于完整桌面办公套件。需要深度修改时使用原生应用；需要让 Agent 分析时，可把文件路径或选中的文字作为上下文。Office, PDF, and PPTX views are primarily for inspection, not replacements for full desktop office suites. Use native apps for deep editing, and pass paths or selected text to Agent for analysis.
快捷键与鼠标操作Keyboard and mouse controls
以下为 Windows 下已实现的主要操作；编辑器还能切换默认、Notepad++ 和 VS Code 三套按键预设。These are the main implemented Windows controls. The editor also offers Default, Notepad++, and VS Code keymap presets.
| 范围Area | 快捷键 / 操作Shortcut / action | 功能Function |
|---|---|---|
| 全局Global | Ctrl + S | 保存当前可编辑文件。Save the active editable file. |
| 全局Global | Ctrl + Z / Y | 撤销 / 重做。Undo / redo. |
| 工作区Workspace | Ctrl + Shift + F | 打开并聚焦全文搜索。Open and focus workspace search. |
| 编辑器Editor | Ctrl + \ | 把当前标签向右分屏。Split the active tab to the right. |
| 编辑器Editor | Tab / Shift + Tab | 插入两个空格或缩进 / 减少缩进。Insert two spaces or indent / outdent. |
| 导航Navigation | Alt + ← / → | 编辑位置后退 / 前进。Go back / forward through editor locations. |
| 导航Navigation | 鼠标侧键 后退 / 前进Mouse Back / Forward buttons | 与 Alt+方向键相同，跨文件和行列恢复历史位置。Same as Alt+arrows, restoring history across files and locations. |
| LSP | F12 | 跳转到定义；按住 Ctrl 点击符号也可跳转。Go to definition; Ctrl-click a symbol works too. |
| LSP | Shift + F12 | 查找引用。Find references. |
| Agent | Alt + L | 把当前编辑器选区加入 Agent 聊天。Add the current editor selection to Agent chat. |
| 文件树Explorer | F2 | 重命名选中条目。Rename the selected entry. |
| 文件树Explorer | Ctrl + X / V | 剪切选中条目 / 粘贴到选中目录。Cut the selected entry / paste into the selected directory. |
| 缩放Zoom | Ctrl + 鼠标滚轮mouse wheel | 调整中间编辑区域缩放。Change the center editor zoom. |
| Markdown | Ctrl + Alt + 1…6 | 切换 H1–H6 标题。Toggle H1–H6 headings. |
按键预设差异Keymap preset differences
CodeMirror 默认编辑、历史和搜索按键；包含保存、选择、撤销、重做、查找等常规操作。CodeMirror's default editing, history, and search bindings, including common save, selection, undo, redo, and find behavior.
Ctrl+L 选中行；select line; Ctrl+D 删除行；delete line; Ctrl+] / [ 缩进 / 反缩进。indent / outdent.
Alt+↑/↓ 移动行；move line; Shift+Alt+↑/↓ 复制行；copy line; Ctrl+Shift+K 删除行。delete line.
导航历史记录文件路径、行号和列号。因此即使在同一个文件中多次跳转定义，侧键也能回到先前的精确编辑位置。History stores file path, line, and column. Even after several jumps within one file, the side buttons return to the exact earlier location.
终端、LSP 与插件Terminal, LSP, and plugins
把构建、代码导航和扩展能力放在同一工作区里。Keep builds, code navigation, and extensions inside the same workspace.
集成终端Integrated terminal
使用“终端 → 新建终端”或顶部中间面板按钮打开。终端位于编辑器底部，默认以当前工作区为目录，可拖动上边缘调整高度并单独关闭。Open from Terminal → New Terminal or the middle panel button. It appears below the editor, starts in the workspace, resizes from its top edge, and closes independently.
语言服务Language services
LSP 提供定义、声明、实现与引用定位；文档面包屑和符号路径帮助理解当前位置。Tree-sitter 与 CodeMirror 回退共同提供语法高亮。LSP provides definition, declaration, implementation, and reference navigation. Breadcrumbs and symbol paths show location, while Tree-sitter and CodeMirror fallback provide highlighting.
插件管理器Plugin Manager
可从目录或插件包安装、刷新、启用、禁用和卸载插件。插件可以贡献语言包、活动栏视图、配置、任务和 Agent 工作区工具。Install from a directory or package, refresh, enable, disable, and uninstall plugins. Plugins may contribute language packs, activity views, configuration, tasks, and Agent workspace tools.
先在终端手动确认项目的安装、构建和测试命令，再交给 Agent 自动执行。涉及删除、部署、系统配置或工作区外路径时，应保持审批并检查命令目标。Confirm install, build, and test commands manually before delegating them to Agent. Keep approval enabled for deletion, deployment, system changes, or paths outside the workspace.
联网搜索与远程服务Web search and remote services
让不自带联网能力的模型核验公开信息，并从 Android 继续查看和推进任务。Verify public information with models that lack native browsing, and continue tasks from Android.
联网搜索 BrokerWeb Search Broker
在“设置 → 服务设置”启用并测试。回退策略包括：从不自动使用、供应商搜索不可用时使用、所有显式联网请求都使用。发送前可选择 /web-search 明确要求查询最新公开信息。Enable and test it under Settings → Service Settings. Fallback modes are: never, when provider search is unavailable, or always for explicit web-search requests. Use /web-search to require current public information.
Remote AgentRemote Agent
服务设置中可选择默认或自定义服务器，登录或注册账号并设置设备名。连接后，移动端可查看在线电脑、已注册工作区和 Agent 会话，并继续发送指令。Choose the default or a custom server, sign in or register, and name the device. Once connected, mobile can view online computers, registered workspaces, and Agent sessions, then send more instructions.
要求 Agent 优先引用官方文档、发布说明和可信来源，并说明访问时间。涉及下载、部署、价格、版本或安全信息时，至少进行一次交叉核验。Ask Agent to prioritize official documentation, release notes, and trusted sources, and to state when they were accessed. Cross-check downloads, deployments, prices, versions, and security information.
安全、数据与排错Safety, data, and troubleshooting
把密钥留在本机，把权限保持在最小范围，把问题定位建立在测试与日志上。Keep keys local, permissions narrow, and troubleshooting grounded in tests and logs.
本机配置与密钥Local configuration and keys
供应商配置位于 %USERPROFILE%\.rovercode\config\providers.json。不要截图、导出、提交或粘贴真实 API Key；共享配置示例时用占位符。Provider settings live at %USERPROFILE%\.rovercode\config\providers.json. Never screenshot, export, commit, or paste real API keys; use placeholders in examples.
运行诊断日志Runtime diagnostic log
AI 请求失败、流式中断、助手无文本或工作区工具异常时，检查 %USERPROFILE%\.rovercode\logs\debug.log。日志用于分类连接、协议、解析和工具错误。For request failures, interrupted streams, missing assistant text, or workspace-tool errors, inspect %USERPROFILE%\.rovercode\logs\debug.log to classify connection, protocol, parsing, and tool failures.
模型测试失败Model test fails
依次检查 Type、Base URL、Key、API model、代理和 Headers。确认地址没有重复的路径段，并验证供应商是否真的兼容所选协议。Check Type, Base URL, Key, API model, proxy, and Headers. Ensure paths are not duplicated and the provider truly supports the selected protocol.
普通对话成功，Agent 失败Chat works, Agent fails
检查模型是否启用 Agent、工具调用是否受支持、上下文与最大输出是否合理，以及供应商流式工具事件是否兼容。Check that Agent is enabled, tool calling is supported, token limits are realistic, and streamed tool events are compatible.
没有助手文本或任务停在工具后No assistant text or run stops after a tool
保留会话 ID，查看 debug.log 中的 provider、stream、parser 与 trajectory 记录，再尝试恢复中断任务或重新生成。Keep the session ID, inspect provider, stream, parser, and trajectory entries in debug.log, then resume the interrupted run or regenerate.
文件显示乱码Text encoding looks wrong
停止保存，确认原文件是 UTF-8、UTF-8 BOM 还是 GBK。不要在未确认前转换编码；必要时从版本控制或备份恢复。Do not save. Confirm whether the source is UTF-8, UTF-8 BOM, or GBK. Avoid conversion until verified, and restore from version control or backup if needed.
推荐工作流Recommended workflows
把功能组合成可重复的操作顺序，比单独记住每个按钮更有效。Repeatable sequences are more useful than memorizing isolated buttons.
完成一次代码修改Complete a code change
- 打开工作区并确认构建命令。Open the workspace and confirm build commands.
- 把相关文件或选区加入 Agent 上下文。Add relevant files or selections to Agent context.
- 使用 Plan 明确范围和验证标准，再切到 Agent 执行。Use Plan to define scope and verification, then switch to Agent.
- 打开修改文件差异，检查无关变更。Inspect diffs and reject unrelated changes.
- 在终端运行测试或构建，最后保存。Run tests or builds in the terminal, then save.
编写 Markdown 文档Write a Markdown document
- 新建或打开
.md，在富编辑视图完成正文。Create or open a.mdfile and draft in rich view. - 用 Ctrl+Alt+1…6 快速调整标题层级。Use Ctrl+Alt+1…6 for heading levels.
- 需要检查格式时打开源码分栏。Open the source split to inspect formatting.
- 粘贴图片并确认相对路径。Paste images and confirm relative paths.
- 保存后导出 PDF，检查分页和图片。Save, export PDF, and inspect pagination and images.
核验最新公开信息Verify current public information
- 在服务设置中测试联网搜索 Broker。Test the Web Search Broker in Service Settings.
- 选择
/web-search，明确要核验的时间范围和来源类型。Select/web-searchand state the time range and source types. - 要求优先官方来源并进行交叉核验。Require official sources and cross-checking.
- 让 Agent 在结论中保留链接和访问时间。Keep links and access dates in the conclusion.
离开电脑后继续任务Continue after leaving the PC
- 在服务设置登录 Remote Agent 并确认设备在线。Sign in to Remote Agent and confirm the device is online.
- 启动长任务前给出清晰完成标准和验证命令。Give a clear done condition and verification command before a long task.
- 在 Android 查看会话状态和完成通知。Review session status and completion notices on Android.
- 需要人工决策时再发送补充指令，不盲目扩大权限。Send follow-up decisions when needed without broadening permissions blindly.
先从一个真实工作区开始Start with a real workspace
打开项目、测试模型、给出明确目标，并让每一步都落在可检查的文件差异和验证结果上。Open a project, test the model, state a clear goal, and keep every step grounded in reviewable diffs and verification results.