Consider it done. The open-source AI agent that works out of the box · 想到,就能做到。开源、开箱即用的 AI Agent。
TypeScript agentai-agentai-assistantandroidclaude-codecodexelectroniosllmmacosreact-nativetypescriptwindows
feat(help): 让 Agent 读取产品帮助文档,并全面校订帮助内容
Dash706ba03ac5db
diff --git a/apps/desktop/help-knowledge/appearance.md b/apps/desktop/help-knowledge/appearance.mdindex 9e65b6b62b..96c06c9286 100644--- a/apps/desktop/help-knowledge/appearance.md+++ b/apps/desktop/help-knowledge/appearance.md@@ -1,19 +1,22 @@---id: appearancetitle: Appearance, theme, wallpaper and fonts-summary: Theme, theme family, wallpaper, fonts, sidebar styles, panel layout, and reply animation — all in Settings > Appearance.+summary: Theme, appearance mode, local and imported themes, wallpaper, UI / code / terminal fonts, sidebar styles, panel layout, and reply animation — all in Settings > Appearance.tab: appearance----Settings > Appearance controls how the app looks. It starts with the theme and appearance mode, followed by Wallpaper, Fonts, Sidebar & panels, and Reply Animation. Display language and notifications stay in Settings > General (see the language and notifications topic).+Settings > Appearance controls how the app looks. It starts with **Theme** and **Appearance Mode**, followed by **Wallpaper**, **Fonts**, **Sidebar & panels**, and **Reply Animation**. Display language and notifications stay in Settings > General (see the language and notifications topic).-- **Theme**: pick from the registered theme families; each family supplies its own colors for light and dark. The app uses a VSCode-style token system (see docs/design-rules/cindy-design-system.md) so themes only override what they need. Families with only one variant are tagged Light only / Dark only; while one is selected the appearance mode is fixed to that variant, and your saved mode returns when you pick a two-variant theme.-- **Appearance mode**: light, dark, or system (follows your OS).-- **Wallpaper**: choose a Cindy scene, your own image, or a video for the whole app. The wallpaper is independent of the theme.-- **Fonts**: choose fonts and sizes for interface text and code. Code font size changes code blocks, change comparisons, and editors; inline code, status numbers, badges, and path labels keep their size.-- **Pinned Style**: show pinned sessions as Text, Cards, or List. **List Style** controls unpinned sessions: Text uses one line; List adds a preview.-- **Local theme brand identity**: this is an optional power-user feature configured in the local theme JSON. The settings page intentionally keeps only create-copy, open-folder, and refresh actions. New copies include self-explanatory example paths directly in the JSON.-- **Export / open local theme files**: export the current theme's tokens to a file or open a local theme JSON for inspection / sharing. To replace the icon and logo used on the new-chat page:+**Theme:**++- **Theme**: pick a theme family; each family supplies its own light and dark colors. Local themes are marked **(Local)**. Families with only one variant are tagged **Light only** / **Dark only**; while one is selected the appearance mode is fixed to that variant, and your saved mode returns when you switch to a two-variant theme.+- **Appearance Mode**: **Light**, **Dark**, or **Auto** (follows your system).+- **Local themes** row:+ - **Create local copy** — writes an editable JSON copy of the current theme into your themes folder.+ - **Import VSCode / Obsidian theme** — pick a VSCode or Obsidian theme file and turn it into a local theme, then select it in the theme list. VSCode themes that inherit colors via `include` can't be imported on their own.+ - **Open themes folder** — opens the folder that holds local theme JSON files.+ - **Refresh** — reloads local themes after you edit them.+- **Local theme brand identity** (optional, for power users): in a local theme JSON you can replace the icon and logo shown on the new-chat page. New copies already contain example paths to replace:```json{@@ -24,11 +27,33 @@ Settings > Appearance controls how the app looks. It starts with the theme and a}```- The example filenames describe the final display areas; source images may be exported at 2x/3x while keeping the same aspect ratios. Paths must be absolute JSON strings; spaces do not need shell quotes. PNG, JPEG, and WebP are supported. Transparent padding is trimmed at render time without changing the source file. Only `brand.icon` and `brand.logo` are recognized; the previous single-image fields are not compatible with the redesigned two-asset layout.+ The example filenames describe the final display areas; source images may be exported at 2x/3x while keeping the same aspect ratios. Paths must be absolute JSON strings; spaces do not need shell quotes. Transparent padding is trimmed at render time without changing the source file. Only `brand.icon` and `brand.logo` are recognized. If a path doesn't exist, the default artwork is shown. Save the file, then click **Refresh**.++**Wallpaper:**++- Choose **None**, one of the built-in Cindy scenes / backgrounds, or **Custom wallpaper** for the whole app. The wallpaper is independent of the theme; **Reset** restores the default wallpaper, visibility, blur, and display settings.+- **Choose Wallpaper** / **Replace Wallpaper** / **Remove Wallpaper**: PNG, JPEG or WebP up to 20 MB, or an MP4 (H.264) video up to 100 MB. It's stored on this device.+- **Display**: **Static** (default) or **Dynamic** for animated scenes and custom videos. Motion pauses when the window is hidden and follows your system's reduced-motion setting; videos loop silently.+- **Wallpaper Visibility** (0% hides it, 100% shows it without an overlay) and **Wallpaper Blur** (0 turns it off; blurring animated wallpaper uses more GPU).++**Fonts:**+- **UI font** / **UI font size** — menus, settings, and chat text (12–24).+- **Code font** / **Code font size** — code blocks, change comparisons, and editors (10–24). Inline code, status numbers, badges, and path labels keep their size.+- **Terminal font** / **Terminal font size** — only the built-in terminal (10–24, default 13). A monospace font is recommended.+- Each font picker offers presets or a **Custom** font name; each size has a **Reset to default** button.++**Sidebar & panels:**++- **Pinned Style**: show pinned sessions as **Text**, **Cards**, or **List** (same as the view-style button on the sidebar's **Pinned** header).+- **List Style**: for unpinned sessions — **Text** uses one line; **List** adds a preview (same as **Display** in the sidebar's **Sidebar settings**; see the Sidebar topic).- **Panel layout**: panel positions and widths are saved automatically; **Restore default layout** puts them back.-- **Reply Animation**: new reply text fades in while it streams. Turn it off to show text immediately. It stays off when your system reduces motion.++**Reply Animation:**++- **Fade In Text**: new reply text fades in while it streams. Turn it off to show text immediately. It stays off when your system reduces motion.**Notes:**- Theme changes apply immediately; no restart needed.+- To zoom the whole window, use **Cmd/Ctrl + =** / **Cmd/Ctrl + -**, and **Cmd/Ctrl + 0** to reset (on macOS also under the **View** menu). The zoom level is remembered.diff --git a/apps/desktop/help-knowledge/attachments.md b/apps/desktop/help-knowledge/attachments.mdindex b9aed37e6e..8ed29dc54f 100644--- a/apps/desktop/help-knowledge/attachments.md+++ b/apps/desktop/help-knowledge/attachments.md@@ -1,30 +1,38 @@---id: attachmentstitle: Attaching files and images, and referencing code-summary: Paste / drag images, @-mention files, and add extra reference directories from the composer.+summary: Paste / drag images and files, use the composer's + / @ panel to attach files, reference files, tasks, browser tabs and more, and add extra directories the agent can work in.---You can give the agent more than plain text.-**Images:**+**Images and files:**-- **Paste** an image from the clipboard, or **drag** image files onto the composer.-- Accepted formats: **PNG, JPEG, WEBP, and GIF**.-- Thumbnails appear above the input; click one to open a full-screen lightbox preview (remove from there, or ESC to close).+- **Paste** an image from the clipboard (e.g. a screenshot), or **drag** files onto the composer / task window. You can also open the composer's **+** menu and choose **Add files, images, or videos…** to pick files from the OS dialog.+- Images (**PNG, JPEG, WEBP, GIF**) are sent to the model as images. Other files (PDF, Office documents, code, text, videos, etc.) are attached by path, and the agent opens or reads them as needed.+- Attachments appear above the input. Click an image to open a full-screen preview; click a text / code file to preview its contents. Hover a thumbnail and click **×** to remove it.+- Pasting a copied folder inserts it as a folder reference instead of an attachment.+- Pasting a very long block of text (about 24+ lines or 4,000+ characters) folds it into a **Pasted text** chip; click the chip to view or edit it.+- Local file attachments aren't available in tasks running on an SSH remote host — paste the content directly instead.-**`@`-mention to reference files / directories / agents:**+**The + menu and `@` mentions:**-- Type `@` to open the picker. It shows files, directories, and registered agents in one flat list.-- Scope: the picker scans your session's **working directory** (extra reference directories are given to the agent as context but aren't part of the `@` picker's file scan).-- Matching is fuzzy, with a boost for prefix matches on the filename.+- Clicking **+** (left of the composer toolbar) and typing `@` open the same panel. It lists actions — **Add files, images, or videos…**, **New goal**, **Plan mode**, **Collaboration**, and **Add directory…** — followed by your installed plugins and the task's extra directories. Type to filter; **↑ / ↓** and **Enter** select.+- Things you can reference with `@`:+ - **Files and folders** in the task's **working directory** (extra directories aren't scanned by the picker).+ - Agents defined in the project.+ - **Tasks** — type a keyword to find past tasks and reference them.+ - **Teammates** — "Message or delegate" to one of your teammates.+ - The current task's built-in **browser tabs** and open **app windows** on this computer (local tasks only).+ - **Plugin** commands and plugin resources.+- Matching favors names that start with what you typed, then falls back to in-order fuzzy matching on names and paths.-**Extra reference directories (read-only context for the agent):**+**Extra directories:**-- Open the composer's **+** menu (to the left of the permission selector — it also hosts New Goal, Plan Mode, and Plugins) and add a directory from the OS folder dialog.-- Per-session; up to **10** extra directories. This section shows for **Claude Code** sessions (Codex ignores extra directories).-- Directories that are subdirectories of the current working directory are silently skipped (already covered). Parent / ancestor directories are flagged with a warning.-- The agent gets read-only context for these directories — it can grep / read them, but writes still go to the working directory.+- Choose **Add directory…** in the **+** menu to pick a folder from the OS dialog. The agent can **read and write** in it, in addition to the working directory. Added directories are listed under **Directories** in the same panel; hover one to remove it.+- Per task; up to **10** extra directories (**Limit reached (10)** when full). Available for local Claude Code, Codex, and Pi tasks; not available for SSH remote or linked-device tasks.+- Choosing a subdirectory of the working directory is silently skipped (already covered). Choosing a parent / ancestor directory asks you to confirm first, because the agent could then modify everything inside it.**Notes:**-- Image attachments are sent with that user message only; they're not implicitly re-included on subsequent turns.-- The `@`-mention just inserts a reference string into your prompt — it's the agent's job to actually open / read the referenced item.+- Attachments are sent with that one message only; they're not implicitly re-included on later turns.+- An `@` reference just inserts a reference into your prompt — it's the agent's job to actually open / read the referenced item.diff --git a/apps/desktop/help-knowledge/collaboration.md b/apps/desktop/help-knowledge/collaboration.mdindex fdb0f3cc1e..5890c095c0 100644--- a/apps/desktop/help-knowledge/collaboration.md+++ b/apps/desktop/help-knowledge/collaboration.md@@ -1,33 +1,37 @@---id: collaborationtitle: Collaboration mode (lead + workers)-summary: Turn on Collab to bring one or more worker agents in beside your session and delegate tasks to them.+summary: Turn on Collaboration from the composer's + menu to bring one or more worker agents in beside your session — locally, in their own worktree, or on another computer — and delegate tasks to them.----Collaboration ("协同" / Collab) lets one session — the **lead** — bring in one or more **worker** agents to run alongside it (a "team"). The lead delegates tasks to workers through a built-in MCP, and you can watch and steer each worker.+Collaboration ("Collab") lets one session — the **lead** — bring in one or more **worker** agents to run alongside it (a "team"). The lead delegates tasks to workers through built-in collaboration tools, and you can watch and steer each worker.**Turning it on:**-- In any regular project or dialogue session, open the composer's **+** menu and choose **Collaboration mode**. You can also choose it before sending the first message of a new task. Claude Code, Codex, and Pi can lead locally or over SSH.-- Pick the first worker's agent (Claude Code, Codex, or Pi), and optionally its role / model / effort / an initial task. If you have not chosen a Worker permission before, the initial default is **Full access** — it is a selectable default, not a fixed mode. You can choose **Auto-review** instead. Once you create a Worker with either mode, Cindy remembers that choice for the next Worker instead of replacing it with the product default.-- Start. This creates the team plus that first worker.+- In a regular project or dialogue session, open the composer's **+** menu and choose **Collaboration**. You can also do this before sending the first message of a new session. Claude Code, Codex, and Pi sessions can lead, locally or over SSH. It isn't offered inside a worker or in a teammate chat.+- The **Start Collaboration** panel ("Bring in a Worker") sets up the first worker:+ - **Agent** (Claude Code, Codex, or Pi), **Model**, **Reasoning Effort**, an optional **Role Name** (just a label to tell workers apart), and an optional **Initial Task**.+ - **Permission Mode**: if you haven't chosen one before, it starts at **Full access** (skips routine approvals). You can choose **Auto-review** instead (safe work proceeds automatically; higher-risk actions may still be denied or need confirmation). Switching to Full access asks you to confirm first.+ - **Project** (local project sessions): work directly in the lead's project and branch, or **Create a worktree for this Worker** branching from a branch you choose. If the directory has uncommitted changes, the worker continues with them.+ - **Run on**: keep the worker **Local**, or run it on another computer on your account that's online and has remote control enabled. Its session, folder, and commands then stay on that computer (choose **Chat** or **Specify folder** for where it works there).+- The panel remembers your last Agent, model, effort, and permission choice for the next worker. Changing it doesn't affect the lead or workers that already exist.+- If the + menu says collaboration is disabled, turn **Collab Mode** back on under **Settings > Tools** (it can also be overridden per project).-**Adding more workers:**+**Adding and managing workers:**-- From the collaboration tab's worker list, use **Create new worker** (+) to add more.-- Default limits are a **soft cap of 5** and a **hard cap of 8** workers per lead — you'll see a warning past the soft cap, and creation is blocked at the hard cap. Both are adjustable (1–20) in the Collaboration section under **Settings > General**.+- Workers appear in the right sidebar's **Team** tab. The worker header lists all workers (choose **Tabs** or **Dropdown** under **Worker list layout**); use **Create new worker** there to add more.+- Default limits are a **Worker Soft Limit of 5** and a **Worker Hard Limit of 8** per lead — you'll see a warning past the soft limit, and creation is blocked at the hard limit. Both are adjustable (up to 20) in **Settings > General > Collaboration**, which also has **Idle Release (minutes)** to automatically release workers after they sit idle (0, the default, turns it off).+- Only one worker is shown at a time — switch between them in the worker header. You can message any worker directly, **edit** a worker's role name / label, or **archive** a worker individually (archived workers can't be restored; create a new one instead).**While collaborating:**-- Each worker is a full session with its own history, tools, working directory, and model. **Full access** skips routine approvals; **Auto-review** lets safe work proceed automatically while higher-risk actions may still be denied or require confirmation. The Worker creation panel remembers your last permission choice alongside its Agent and model preferences. The same remembered choice is used when an agent tool creates a Worker; an explicit `start_team` permission updates that shared choice. Raising it from Auto-review to Full access waits for your confirmation first; cancelling or letting the confirmation time out leaves the preference and team unchanged. Changing it does not affect the lead or workers that already exist. Workers inherit the lead's working directory by default (or share the worktree when the lead uses one).-- A worker's model defaults to your last new-session choice for that vendor, falling back to the lead's model.-- Only one worker is **focused** (shown) at a time — switch between them with the worker tabs / dropdown. You can intervene in any worker directly, and archive workers individually.+- Each worker is a full session with its own history, tools, and model. Workers use the lead's working directory by default (or the shared / new worktree chosen above).**Turning it off:**-- Closing collab **confirms with you first** (in-progress work is lost), then ends the workflow and archives **all** the team's workers. The conversation history stays searchable.+- Choose **Collaboration** again in the **+** menu, or close the **Team** tab. Cindy **asks you to confirm first** — running workers stop immediately and in-progress results are lost — then ends the team and archives **all** of its workers. The conversation history stays searchable.**Notes:**-- A lead can be a Claude Code, Codex, or Pi **project or dialogue** session, locally or over SSH. Remote workers are spawned on the **same remote host**, and the collaboration MCP reaches the remote side over an SSH remote-forward to the local bridge (with a persistent bearer token). A worker can't start its own sub-collab (no nesting).-- A lead can have only **one active collaboration workflow** at a time. If you have two clients open on the same account (e.g. dev + release) and start collab on the same session in both, the second fails with "开启协同失败" because a workflow is already active — close one client or end the existing collab.+- With an SSH lead, workers are created on the **same remote host**. A worker can't start its own collaboration (no nesting).+- A lead can have only **one active collaboration** at a time. If you have two Cindy clients open on the same account and start collaboration on the same session in both, the second fails with "Failed to start collaboration" because one is already active — close one client or stop the existing collaboration.diff --git a/apps/desktop/help-knowledge/commands.md b/apps/desktop/help-knowledge/commands.mdindex fba19278eb..952676ea7b 100644--- a/apps/desktop/help-knowledge/commands.md+++ b/apps/desktop/help-knowledge/commands.md@@ -1,27 +1,44 @@---id: commandstitle: Slash commands-summary: Type "/" in the composer to open a palette merging your skills, built-in app commands, and the agent's own commands.+summary: Type "/" in the composer to open a palette merging your skills, built-in app commands (/help, /clear, /side, /cmd, /issue, /review, /goal, /plan, /learn …), and the agent's own commands; type "$" for installed plugin commands.----Type `/` at the start of the composer to open the command palette.+Type `/` at the start of a line (or after a space) in the composer to open the command palette.**What's in the palette (priority order on collisions):**-1. **Your installed skills and Cindy's built-in Skills** (highest priority) — anything in your global or project skills directory, including `/cindy-skill-creator` and `/learn` when enabled (see the Skills topic).-2. **Built-in app commands** — `/help` (show available commands), `/clear` (clear the current session context — resets the conversation in place without creating a new session), `/cmd` (run a shell command in the working directory), `/issue` (file an issue), `/goal` (start a goal-driven run), `/workflows`, and `/jump-session`.-3. **The agent's own commands** — e.g. `/compact`, `/agents`, `/memory` (the exact list depends on the agent; Claude Code and Codex each contribute their own).+1. **Your installed skills and Cindy's built-in Skills** (highest priority) — anything in your global or project skills directory, including the built-in `/cindy-skill-creator` (see the Skills topic).+2. **Built-in app commands:**+ - `/help` — show a help card listing every available command with usage examples.+ - `/clear` — clear the current session context in place (wipes its messages and state without creating a new session; the session stays in the sidebar).+ - `/side` — open a side chat with this session's context; any text after it becomes the draft question.+ - `/cmd <command>` — run a shell command in the working directory and show the output in the chat (for a session on another of your computers, it runs on that computer).+ - `/issue [description]` — send feedback or a bug report to the Cindy team (see below).+ - `/review [focus or path]` — review the current session's work in a fresh, read-only reviewer session (local sessions only; see the Reviewing changes topic).+ - `/goal <condition>` — set a goal the agent keeps working toward across turns; `/goal` alone opens the **New goal** dialog, and `/goal clear` clears it.+ - `/plan` — toggle **Plan mode** on or off (when available for the session; see the Plan mode topic).+ - `/learn [description]` — distill a reusable Skill from this session or a described workflow, reviewed as a diff before saving (when enabled; see the Skills topic).+ - `/workflows` — show the progress of workflows running in this session.+ - `/jump-session` — jump straight to a session by its ID.+3. **The agent's own commands** — e.g. `/compact` and `/context` for Claude Code. The exact list depends on the engine (Claude Code, Codex, or Pi).++**Plugin commands (`$`):**++- Type `$` (a full-width `$` or `¥` also works) to open a separate palette listing commands from your installed, enabled plugins.-`/issue` does not require a GitHub plugin or GitHub account. The confirmation card uses Cindy's official bot by default; if a working account is already configured under **Plugins > Cindy GitHub**, that account appears as an optional submission identity.+**Using the palette:**-Before submitting, Cindy helps clarify missing details so the issue is actionable. For bugs, it can collect reproduction steps and user-approved, redacted diagnostic summaries; the confirmation card always shows the final public content. After creation, Cindy returns the issue link and can help interested users reproduce or fix the problem from source because Cindy is open source.+- Start typing to filter. Matching is substring-based (case-insensitive) on the command name — exact matches rank first, then prefix matches, then other matches; Cindy's official commands rank ahead on ties. It is not fuzzy.+- Up to 25 matches are shown at a time.+- **↑ / ↓** move focus, **Enter** or **Tab** inserts the focused command into the composer (add any arguments, then send), **Esc** closes the palette.-**Using the palette:**+**`/issue` feedback:**-- Start typing to filter. Matching is **contains-based** (case-insensitive substring matching on the command name) — not fuzzy.-- Up to 25 matches shown at a time.-- **↑ / ↓** move focus, **Enter** runs the focused command, **Esc** closes the palette.+- `/issue` doesn't require a GitHub plugin or GitHub account. Before submitting, the agent helps clarify missing details so the issue is actionable; for bugs it can collect reproduction steps and user-approved, redacted diagnostic summaries.+- The confirmation card always shows the final public **Title** and **Body**, lets you choose the **Submission Identity** — **Cindy official bot (default)**, or **My GitHub account** if a working account is connected on the **GitHub** plugin's detail page under **Plugins** — and set a **Public Attribution** (or **Anonymous**). You must confirm the content contains nothing private before **Submit Feedback**.+- After creation, Cindy returns the issue link and can help you reproduce or fix the problem from source, because Cindy is open source. To see issues you've submitted, open **Issues** from the app menu (top-left **Menu**, or **Help > Issues** in the macOS menu bar).**Notes:**-- If a name collides across sources (same command in a skill and in a built-in), the higher-priority source wins; the lower-priority one is silently skipped (logged as a warning).-- Most built-in app commands run inside the desktop app (a few, like `/workflows` and `/jump-session`, are handled directly in the UI); agent commands are sent as prompt prefixes to the agent.+- If a name collides across sources (same command in a skill and in a built-in), the higher-priority source wins; the lower-priority one is silently skipped.+- Built-in app commands run inside the desktop app; agent commands and skills are sent to the agent as the start of your prompt.diff --git a/apps/desktop/help-knowledge/connections.md b/apps/desktop/help-knowledge/connections.mdindex f64b4b52ce..75e0842fdc 100644--- a/apps/desktop/help-knowledge/connections.md+++ b/apps/desktop/help-knowledge/connections.md@@ -1,23 +1,26 @@---id: connectionstitle: Connecting external accounts-summary: External accounts (Google, Jira / Confluence, GitHub, GitLab) are connected on their plugin detail pages under Plugins; Slack and Telegram are bound under Settings > IM bots; Codex sign-in lives in Settings > Model Providers.+summary: External accounts (the Gmail / Google Drive / Google Calendar / Google Sheets plugins, GitHub, GitLab, and Jira / Confluence via XD Atlassian for organizations that provide it) are connected on each plugin's detail page under Plugins; Slack, Telegram, X and personal chat bots are set up under Settings > IM bots; ChatGPT / Claude / Grok subscriptions and Cindy AI live in Settings > Providers.----External account connections have moved out of the old "Third-Party Platforms" section — each integration now lives on its own plugin detail page under **Plugins**, and the plugin's tools use the account you connect there.+There is no separate "Third-Party Platforms" settings page any more — each integration lives on its own plugin detail page under **Plugins**, and the plugin's tools use the account you connect there. Chat apps are set up under **Settings > IM bots**, and model accounts under **Settings > Providers**.**Where to connect what:**-- **Slack** — connect under **Settings > IM bots** (the "Cindy" tab). Each workspace has **Slack Communications on This Device** (本机 Slack 通讯) when the server supports it. Multiple authorized devices can search, read, and post as the Slack user, while only one device receives Bot chats and tasks. Turning off local communications does not revoke the account or affect other devices. Enabling communications on a previously displaced device may open Slack authorization, but never takes over the Bot. Old servers retain the original binding-based behavior.-- **Telegram** — under **Settings > IM bots** (the "Cindy" tab), turn on Telegram and choose **Link Telegram**. Open the one-time link, start the official Cindy bot, and complete confirmation if prompted. The same card lets you cancel, retry, open the bot, add it to a group, or unlink; no personal bot token is required.-- **Google** (Gmail / Calendar / Drive / Sheets) — handled by the Filo Google plugin (Plugins > Filo Google).-- **GitHub** — handled by the Cindy GitHub plugin (Plugins > Cindy GitHub; existing PAT connections are migrated automatically). This connection is optional for `/issue`: feedback uses Cindy's official bot by default, and a working GitHub connection only adds the option to submit as yourself.-- **GitLab** — handled by the Cindy GitLab plugin (Plugins > Cindy GitLab; existing PAT connections are migrated automatically).-- **Codex** — sign in under Settings > Model Providers (separate from the Codex CLI binary the app ships with).-- **FeiShu** — bot only (App ID / Secret under Settings > IM bots; see the FeiShu bot topic). FeiShu is no longer a sign-in method for Cindy itself.-- **Cindy AI** — Settings > Model Providers. Cindy Web Search uses this account by default and needs no key; optional Brave / Tavily / DeepSeek search keys are configured on that plugin's detail page.+- **Slack** — **Settings > IM bots**, **Official** section (requires a Cindy account). Turn on the Slack card to authorize; use **Add Slack workspace** to bind more workspaces. Each workspace has **Slack Communications on This Device** when the server supports it: multiple authorized devices can search, read, and post as the Slack user, while only one device receives Bot chats and tasks. Turning off local communications does not revoke the account or affect other devices. Enabling communications on a previously displaced device may open Slack authorization, but never takes over the Bot.+- **Telegram (official bot)** — **Settings > IM bots**, **Official** section: turn on Telegram and choose **Link Telegram**. Open the one-time link, start the official Cindy bot, and confirm if prompted. The same card lets you cancel, retry, open the bot, add it to a group, or unlink; no personal bot token is required.+- **X (Twitter)** — **Settings > IM bots**, **Official** section: authorize your X account, then mention the Cindy bot in a post to send a task to this device. Replies are posted publicly, so use it only for public questions.+- **Your own chat bots** — **Settings > IM bots**, **Personal** section: **Personal WeChat**, **WeCom Intelligent Bot**, **Feishu / Lark Bot** (App ID / App Secret; see the Feishu bot topic), **DingTalk Bot**, **Discord Bot**, and **Telegram Bot** (your own bot token). You manage these credentials yourself.+- **Google** — the separate **Gmail**, **Google Drive**, **Google Calendar**, and **Google Sheets** plugins (**Plugins > Gmail**, and so on). Each one is authorized separately on its own detail page. Some organization accounts get a single combined **Filo Google** plugin instead.+- **Jira / Confluence** — the **XD Atlassian** enterprise plugin (**Plugins > XD Atlassian**), only for organizations that provide it. Earlier Jira / Confluence connections carry over automatically.+- **GitHub** — the **GitHub** plugin (**Plugins > GitHub**; earlier token connections carry over automatically). This connection is optional for `/issue`: feedback uses Cindy's official bot by default, and a working GitHub connection only adds the option to submit as yourself.+- **GitLab** — the **GitLab** plugin (**Plugins > GitLab**; earlier token connections carry over automatically).+- **ChatGPT (Codex), Claude, and Grok subscriptions** — **Settings > Providers**: **OpenAI** (use your local Codex account or sign in to a separate ChatGPT account), **Anthropic** (**Sign in with Claude Code**), and **xAI** (SuperGrok). This is separate from the Codex / Claude Code programs the app ships with.+- **Cindy AI** — **Settings > Providers**, provisioned automatically after you sign in to Cindy. The **Web Search** plugin uses this account by default and needs no key; optional Brave / Tavily / DeepSeek search keys are configured on that plugin's detail page.+- **Feishu / Lark** is bot-only (see above); it is not a sign-in method for Cindy itself.**Managing connections:**-- Each plugin detail page shows the connected accounts (with status) and lets you reconnect, disconnect, or switch the default account.+- Each plugin detail page shows the connected account(s) and their status, and lets you reconnect or disconnect. If a plugin later needs new permissions, it asks you to reconnect.- Disconnect removes the local token only; you may also want to revoke access on the provider's side.- A connection error usually means the token expired or was revoked on the provider's side; reconnect to refresh.diff --git a/apps/desktop/help-knowledge/feishu-bot.md b/apps/desktop/help-knowledge/feishu-bot.mdindex 3c534558dc..ee1a1671c2 100644--- a/apps/desktop/help-knowledge/feishu-bot.md+++ b/apps/desktop/help-knowledge/feishu-bot.md@@ -1,47 +1,63 @@---id: feishu-bot-title: Setting up the FeiShu (Lark) bot-summary: Configure the FeiShu bot (App ID / App Secret) in Settings > IM bots to get DM notifications and bot interaction.+title: Setting up the Feishu / Lark bot+summary: Connect your own Feishu or Lark bot (App ID / App Secret) under Settings > IM bots > Personal to chat with Cindy, get session DMs, and use it in group chats.tab: im-bot----Cindy can send DMs and take commands through a FeiShu (Lark) bot. (This is unrelated to signing into Cindy — you sign in with your Cindy account; FeiShu is no longer a login method.)+Cindy can take requests and send you DMs through a Feishu or Lark bot that you create yourself. (This is unrelated to signing into Cindy — Feishu is not a sign-in method.)**Setting up the bot:**-- Open Settings > IM bots, then the FeiShu Bot section.+- Open **Settings > IM bots**. In the **Personal** section, expand **Feishu / Lark Bot**.+- Under **Service**, choose **Feishu** or **Lark**.+- Don't have a bot yet? Click **Create on Feishu Open Platform ↗** (or **Lark Open Platform**) to create one, then come back with its credentials.- Paste the bot's **App ID** and **App Secret**, then click **Bind**.-- The status badge updates: **needs-config** → **connecting** → **connected** (or **conflict** / **error**).+- The status badge shows **Needs configuration** → **Connecting…** → **Connected**. Other states: **Reconnecting…**, **Used by another device**, **Connection failed**.**Binding the bot to you:**-- After the badge shows **connected**, the page instructs you to DM the bot a code (or click an action) to complete the binding. Until you bind, the bot won't know which FeiShu user to DM.+- After it connects, send the bot any private message once. The first person to DM the bot becomes its **Owner**; until then the card shows "Waiting for first private-message bind".+- DMs from anyone other than the owner are silently ignored.+- The bound card shows the service, App ID, and owner, with **Reconnect Bot** and **Unbind**. Unbinding disconnects the bot and removes the saved App ID / App Secret / owner from this computer.-**Notifications:**+**New conversation settings:**-- Toggle **message notifications** on the same page to get bot DMs when sessions finish, or when the bot has updates for you.-- **System Notifications** is a separate toggle in **Settings > General > Notifications**. It covers sessions finishing, encountering an error, or needing a reply. See the language and notifications topic for background-window and Agent Island behavior.+- **New Conversation Configuration** in the same card picks the Agent and model for new Feishu conversations (or **Automatic**, which uses an available recommended model). Existing chats switch on their next message unless their model was changed manually.+- **Permission mode for new group tasks** sets the permission mode for every session created in a group chat (@-ing the bot, `/new`, `/ctr`). It defaults to **Auto-review**. Group messages can come from other members, so choose **Full access** only for groups you trust. Use `/permission` in the group to change an existing one.++**Notifications (shown after the bot is bound):**++- **Online/offline announcement → Message notification** — the bot posts a message when this computer connects or disconnects.+- **Session notifications → Send a Feishu / Lark DM when a session finishes or needs a reply** — DMs you only while the Cindy window is unfocused. You can turn it on only after the owner is bound.+- **System Notifications** (desktop pop-ups) is a separate toggle in **Settings > General > Notifications** — see the language and notifications topic.++**Chat commands:**++- `/new` starts a fresh context, `/model` switches the model, `/permission` changes the permission mode, `/ctr` takes over a desktop session (`/exctr` ends the takeover), and `/help` lists the commands.+- Send `!stop` to stop the current run and clear queued messages. The session is kept, so you can continue it.**Group chats:**-- In a group, @ the bot to ask a question. The bot opens a new thread (话题) for the reply — the main stream stays clean and each thread is its own session, like Slack threads.-- The bot pages back through recent chat history (newest first, until the messages stop being relevant to your question) and answers with that history as context — including images and files shared in the chat. For a fresh thread opened from the main stream, the context still comes from the main stream history.+- In a group, @ the bot to ask a question. The bot opens a new thread (话题) for the reply, so the main chat stays clean and each thread is its own session.+- When @-ed in the main chat, the bot reads back through recent group messages (newest first, up to about 250) and stops once they're no longer relevant to your question. It answers with that history as context, including images and files shared in the chat.- @-ing the bot inside an existing thread uses that thread's own history as context.-- Thread sessions are titled `[飞书·{group name}·{topic summary}] {id suffix}` in the sidebar.+- These sessions appear in the sidebar's Chat group: thread sessions are titled like `[飞书·{group name}·{topic summary}] {id suffix}`, group sessions `[飞书·群] {group name}`, and DMs `[飞书·DM] {summary}`. Lark bots use "Lark" instead of "飞书".**Permissions for group features:**-Group chat context needs an extra app permission that is not part of the default setup:+Group chat context needs extra app permissions that are not part of the default setup:-1. Open the **FeiShu (Lark) Open Platform** → Developer Console → your app → **Permissions & Scopes** (权限管理).-2. Search for and enable **both** of the following (searching by the scope id works regardless of the console language):- - **"Read all messages in associated group chat" (获取群组中所有消息, `im:message.group_msg`)** — pulls the chat history via API.- - **"Read user and bot messages in group chats" (获取群组中用户和机器人消息, `im:message.group_msg.include_bot:read`)** — needed for group message events, so the history covers messages from other bots too.- These are sensitive permissions: on FeiShu they are only available to self-built apps (自建应用), not store apps.+1. Open the **Feishu (or Lark) Open Platform** → Developer Console → your app → **Permissions & Scopes** (权限管理).+2. Search for and enable **both** of these (searching by the scope ID works in any console language):+ - **"Read all messages in associated group chat" (获取群组中所有消息, `im:message.group_msg`)** — lets the bot read chat history.+ - **"Read user and bot messages in group chats" (获取群组中用户和机器人消息, `im:message.group_msg.include_bot:read`)** — includes messages from other bots in that history.+ These are sensitive permissions: on Feishu they are only available to self-built apps (自建应用), not store apps.3. Create and publish a new app **version** (self-built apps need admin approval).-Without it the bot still answers when @-ed, but with no chat context, and it DMs the owner a one-line heads-up pointing here. Optionally, enabling **"Obtain basic group information" (获取群基本信息, `im:chat:readonly`)** lets session titles show the real group name instead of an id suffix.+Without these permissions the bot still answers when @-ed, but without chat context, and it DMs the owner a short notice with a link to the app's permission page. Optionally, enabling **"Obtain basic group information" (获取群基本信息, `im:chat:readonly`)** lets session titles show the real group name instead of an ID suffix.**Notes:**-- The "conflict" badge state means another Cindy instance is already bound to the same bot for this user — typical when running dev + release simultaneously. Disconnect on one side to clear it.-- Settings > IM bots also hosts the Slack and Telegram bindings (the "Cindy" tab) and a bring-your-own Discord bot, alongside this FeiShu bot.+- **Used by another device** means this App ID is already connected on another device or Cindy instance (for example, running two Cindy builds at once). Unbind it on one side to clear the conflict.+- A bot can live in only one place. You can also give a bot to a teammate from that teammate's **Messaging platforms** page. Choosing the bot already set up in Cindy moves it out of Settings > IM bots, and its chats continue with the teammate.+- Settings > IM bots also has an **Official** section (Slack, Telegram, and X through Cindy's official bots, no credentials needed). The **Personal** section also has Personal WeChat, WeCom, DingTalk, Discord, and Telegram bots. See the integrations topic.diff --git a/apps/desktop/help-knowledge/getting-started.md b/apps/desktop/help-knowledge/getting-started.mdindex 02f1a337b5..95c075038e 100644--- a/apps/desktop/help-knowledge/getting-started.md+++ b/apps/desktop/help-knowledge/getting-started.md@@ -1,29 +1,34 @@---id: getting-startedtitle: Getting started with Cindy-summary: Create a session, choose an agent and working directory, send your first prompt, and iterate.+summary: Start a new session with New (⌘N / Ctrl+N), choose a project and a model (the engine — Claude Code, Codex, or Pi — follows the model), send your first prompt, and iterate.----Cindy is a desktop app that runs AI coding agents (Claude Code and Codex) for you.+Cindy is a desktop app that runs AI agents (Claude Code, Codex, and Pi) for you.++**Before your first message:**++- You need at least one model. If none is available, Cindy prompts you to sign in to Cindy (Cindy AI, the official service) or connect a provider with your own API key. You can do this later in **Settings > Providers**.**Your first session:**-- Click **+ New** in the top of the left sidebar.-- Choose the **agent** (Claude Code or Codex).-- For code work, choose a **working directory** — the agent runs its file ops there. (You can skip this for plain-Q&A sessions.)-- Optionally pick a **model** in the composer toolbar.-- Type your goal and send.+- Click **New** (the ⊕ icon) at the top of the left sidebar, or press ⌘N / Ctrl+N.+- Use the folder picker (**Chat or select project**) to choose a project folder for code work, where the agent reads and edits files. Choose **Chat** for plain Q&A that doesn't need a project.+- Pick a **model** in the composer's model selector. There's no separate agent picker: the engine (Claude Code, Codex, or Pi) follows the model you choose and is shown next to it (see the Choosing a model topic). A new session starts with the model you last used.+- Optionally set a **permission mode** in the same toolbar. If you have remote devices connected, you can also choose which device runs the session.+- Type your goal and send. By default **Enter** sends and **Shift+Enter** adds a new line. You can switch to ⌘↵ / Ctrl+Enter under **Settings > General > Message input**.**Iterating:**- A session is a persistent chat thread: ask, review the result, then ask the next thing.-- While the agent is replying, the **Send** button becomes **Stop** — click it to interrupt. You can keep typing in the meantime; follow-ups queue up and send in order.+- While the agent is working, the composer shows a separate **Stop** button; click it to stop the current run.+- You can keep typing while the agent is busy: the send button becomes **Queue message**, and queued messages go out in order after the current turn. With the default Enter-to-send setting, press ⌘↵ (Ctrl+Enter on Windows/Linux) instead to steer the running turn without stopping it (see the Sessions and the chat thread topic).- Type `/` in the composer to open the slash-command palette (see the Commands topic).**Organizing:**-- All sessions are listed in the left sidebar (see the Sidebar topic). Switch between them by clicking; each session remembers its agent, model, and working directory.-- Keep unrelated tasks in **separate** sessions — sharing one session for everything dilutes context.+- All sessions are listed in the left sidebar (see the Sidebar topic). Click one to switch to it. Each session remembers its engine, model, and working directory.+- Keep unrelated work in **separate** sessions. Putting everything in one session dilutes the context.**Tone tip:**-- Describe the goal and context first, then what result you want. That works better than a single short command. The agents respond well to "here's what I'm trying to do, here are the constraints, here's what 'done' looks like".+- Describe the goal and context first, then the result you want. That works better than a single short command. The agents respond well to "here's what I'm trying to do, here are the constraints, here's what 'done' looks like".diff --git a/apps/desktop/help-knowledge/import-sessions.md b/apps/desktop/help-knowledge/import-sessions.mdindex d1fb144a1d..322acbf128 100644--- a/apps/desktop/help-knowledge/import-sessions.md+++ b/apps/desktop/help-knowledge/import-sessions.md@@ -1,23 +1,32 @@---id: import-sessionstitle: Importing Codex and Claude conversations-summary: Scan and import existing Codex or Claude Code chat history from Settings > Session Import.+summary: Import existing local Codex or Claude Code history, or a shared .cshare session file, from Settings > Session Import.tab: import----If you already have Codex or Claude Code conversations on disk (from the CLIs), Import brings them into Cindy so they show up in the sidebar alongside your other sessions.+If you already have Codex or Claude Code conversations on this computer (from the CLIs or the Codex app), Session Import brings them into Cindy so they show up in the sidebar with your other sessions.-**Importing:**+**Importing local history:**-- Open Settings > Session Import.-- Click **Scan** to discover sessions in the standard Codex / Claude Code locations on your machine.+- Open **Settings > Session Import** (the page is titled **Local Session Import**).+- Cindy scans the standard Codex and Claude Code history locations when you open the page. Click **Scan** to rescan. Scanning is read-only: it doesn't change your database or the sidebar.- Filter the results:- - by **source** — Codex or Claude Code;- - by **placement** — by project (sessions inside a specific working directory) or by chat (free-form sessions not tied to a project).-- Tick the sessions or projects you want, then click **Import**.-- They appear in the left sidebar.+ - **Source** — All / Codex / Claude;+ - **Placement** — All / **Existing projects** (sessions that will go under a project group) / **Chat** (Codex sessions without a project).+- Tick individual sessions, a whole project group, or **Select all in current filter**, then click **Import selected**.+- A toast reports how many sessions were inserted or updated. They appear in the left sidebar.++**Importing a shared session (.cshare):**++- Click **Import from .cshare…** on the same page, or drag a `.cshare` file onto the Cindy window.+- If the file is encrypted, enter the password from the sender. Then review the preview, choose the matching working directory on this computer, and optionally create it in a worktree.+- If the same session already exists here, you can **Overwrite & import** or cancel.+- To create a `.cshare` file yourself, open a session's menu (right-click it in the sidebar, or use its ⋮ menu) and choose **Export session…**. You can protect the file with a password. Exporting a collaboration lead includes its workers. Sessions that run on a remote machine or another device can't be exported.**Notes:**-- Importing is **one-way and non-destructive** — the originals on disk are left in place; nothing is synced back. Re-importing the same session is a no-op.-- Imported threads are read-only history at first; you can continue them just like any other Cindy session.-- This is for bringing in past conversations — it's unrelated to archiving / deleting sessions (see the Sidebar topic for those).+- Importing local history is **one-way and non-destructive**: the original files stay where they are, and nothing is synced back.+- Sessions that are already in Cindy, plus internal sessions, are skipped and don't appear in the list (the summary counts them as **Skipped**).+- You can continue imported sessions just like any other Cindy session.+- Only import shared files from people you trust. The imported history can influence the agent, and a continued session can still read and write local files and run commands.+- This is for bringing in past conversations. It's unrelated to archiving or deleting sessions (see the Sidebar topic for those).diff --git a/apps/desktop/help-knowledge/integrations.md b/apps/desktop/help-knowledge/integrations.mdindex 68c8d33407..cdea7db659 100644--- a/apps/desktop/help-knowledge/integrations.md+++ b/apps/desktop/help-knowledge/integrations.md@@ -1,23 +1,50 @@---id: integrationstitle: What Cindy integrates with-summary: Supported external services — FeiShu, Slack, Telegram, Discord, Google, Jira / Confluence, GitHub, GitLab — and what's not supported.+summary: Supported external services — chat apps (Slack, Telegram, X, Feishu / Lark, WeChat, WeCom, DingTalk, Discord), Google (Gmail / Drive / Calendar / Sheets plugins), Jira / Confluence (XD Atlassian, some organizations), GitHub, GitLab, custom MCP servers — and what's not supported.----At a glance, Cindy integrates with:+At a glance, Cindy integrates with the services below.-- **FeiShu (Lark)** — DM-style bot notifications and interaction (Settings > IM bots). See the FeiShu bot topic. (FeiShu is no longer a sign-in method for Cindy itself.)-- **Slack** — Settings > IM bots (the "Cindy" tab) manages workspace connections. With a compatible server, **Slack Communications on This Device** enables agent tools (search / read history / post as the authorized user) independently on multiple devices; Bot chats and tasks still go to a single receiving device. Enabling communications does not change that receiver.-- **Telegram** — link your Telegram account to the official Cindy bot under Settings > IM bots (the "Cindy" tab). No bot token is required. Telegram tasks use the same local projects and Cindy sessions as Slack, while keeping channel-specific project and model preferences separate.-- **Discord** — a Discord bot you bring yourself (your own Discord App + bot token) under Settings > IM bots.-- **Google** — Gmail / Calendar / Drive / Sheets, via the Filo Google plugin (Plugins > Filo Google).-- **GitHub / GitLab** — via the Cindy GitHub and Cindy GitLab plugins (Plugins).-- **The coding agents** — Claude Code and Codex can reach pay-per-token models through **Cindy AI** in Settings > Model Providers.+**Chat apps (Settings > IM bots):**++The IM bots page has two sections:++- **Official** — Cindy's own bots. You don't create a bot or enter credentials; just turn on the channel's switch and follow the linking steps. This section isn't available in local mode.+ - **Slack** — bind one or more Slack workspaces and send tasks to Cindy from Slack. Each workspace can also turn on **Slack Communications on This Device** where supported (see the Connections topic).+ - **Telegram** — link your Telegram account to the official Cindy bot with **Link Telegram**. No bot token is required.+ - **X** — link your X (Twitter) account and send tasks by mentioning the Cindy bot on X (shown when the server offers it).+ - Official channels share this computer's project list and sessions, while each channel keeps its own account and project / model preferences.+- **Personal** — bots you create and manage yourself, each with its own credentials and settings:+ - **Personal WeChat** — scan to connect and use Cindy from a private chat.+ - **WeCom Intelligent Bot** — Bot ID + Secret. Also hosts **WeCom group notifications** (a group-bot webhook for automation completion alerts).+ - **Feishu / Lark Bot** — App ID + App Secret. See the Feishu bot topic.+ - **DingTalk Bot** — Client ID + Client Secret (Stream mode), for DMs and group chats.+ - **Discord Bot** — your own bot token + your Discord user ID, for Discord DMs.+ - **Telegram Bot** — your own BotFather token + your Telegram user ID, for DMs and group chats.+- Each channel has a **New Conversation Configuration** that picks the Agent and model for new conversations from that channel.+- Teammates can also have their own bots on their **Messaging platforms** page. A bot can be connected in only one place.++**Accounts and tools (Plugins):**++- **Google** — four separate built-in plugins: **Gmail**, **Google Drive**, **Google Calendar**, and **Google Sheets**, each authorized separately. Some organization accounts get a single combined **Filo Google** plugin instead.+- **GitHub / GitLab** — the built-in **GitHub** and **GitLab** plugins.+- **Jira / Confluence** — the **XD Atlassian** enterprise plugin, only for organizations that provide it.+- Other built-in plugins include **Art**, **Web Search**, **Mermaid**, and **Notion** (see the Plugins topic).+- Connect accounts on each plugin's detail page. That connection is what gives the plugin's tools access to your account (see the Connections topic).++**Models:**++- The agents can use pay-per-token models through **Cindy AI** (the official service) or through your own providers and API keys, all under **Settings > Providers** (see the Providers topic).++**Custom MCP servers:**++- Add any remote MCP server (http / sse) under **Settings > Tools > External tools > Add MCP server**. Use this for services without a built-in integration.**Not supported:**-- There is no built-in Microsoft Teams integration. For chat notifications, use the FeiShu, Slack, Telegram, or Discord bot.-- There's no native Linear integration as a first-party feature — those would have to come via an MCP a user installs themselves.+- There is no built-in Microsoft Teams integration. For chat, use one of the IM bots above.+- There's no first-party Linear integration. Add a Linear MCP server yourself under **Settings > Tools > External tools** if you need it.**Notes:**-- Most of these connections are accessed by the agents through their tools (plugins / MCPs), not directly by the desktop app. Connecting on the plugin detail page is what gives its tools access to your account.+- The agents mostly reach these services through their tools (plugins / MCPs), not through the desktop app directly.diff --git a/apps/desktop/help-knowledge/keyboard-shortcuts.md b/apps/desktop/help-knowledge/keyboard-shortcuts.mdnew file mode 100644index 0000000000..750db9d4b8--- /dev/null+++ b/apps/desktop/help-knowledge/keyboard-shortcuts.md@@ -0,0 +1,40 @@+---+id: keyboard-shortcuts+title: Keyboard shortcuts and accessories+summary: Settings > Keyboard Shortcuts lists Cindy's default shortcuts and lets you rebind, clear, or reset them; Accessories sets up Work Louder keyboards and Xbox, PlayStation, Switch, or generic game controllers.+---+Open **Settings > Keyboard Shortcuts** to see and change Cindy's app shortcuts. Below, **Mod** means ⌘ on macOS and Ctrl on Windows / Linux.++**Default shortcuts:**++- **New session** — Mod+N+- **Toggle sidebar** — Mod+B+- **Cycle permission mode** — Shift+Tab (while the message box is focused)+- **Previous / Next right sidebar tab** — macOS: ⌘⇧[ / ⌘⇧] or Ctrl+Shift+Tab / Ctrl+Tab; Windows / Linux: Ctrl+PageUp / Ctrl+PageDown or Ctrl+Shift+Tab / Ctrl+Tab+- **Open terminal** — Ctrl+` (opens a terminal in the right sidebar, or focuses the existing one)+- **Find in page** — Mod+F+- **Search in project** — Mod+Shift+F (while browsing documents)+- **Zoom in / Zoom out / Reset zoom** — Windows / Linux: Ctrl+= / Ctrl+- / Ctrl+0. On macOS, zoom uses the standard View menu shortcuts and isn't listed here.+- Built-in browser: **Focus address bar** Mod+L, **Browser back** / **Browser forward** Alt+← / Alt+→ (Option on macOS), **Reload page** Mod+R or F5++These also work but aren't shown on the page: Mod+1 … Mod+9 jump to the first nine visible sessions in the sidebar; Mod+S saves the document you're editing; Mod+W closes the active right sidebar tab when focus is there (on macOS it otherwise closes the window); ⌘, opens Settings on macOS.++**Changing a shortcut:**++- Click the pencil on a row, press the new key combination, or click **Cancel**. The trash button removes the binding (the row shows **Not set**).+- Cindy rejects combinations that can't be used, are **Reserved by the operating system**, conflict with another Cindy shortcut (it names which one), or conflict with the voice input shortcut.+- A changed row gets **Reset to default**; **Reset all to defaults** appears at the top once anything is customized.+- Other shortcuts live elsewhere: the voice input shortcut is in **Settings > Voice Input** (see the voice input topic), and the send key (**Use ⌘↵ to send**) is under **Settings > General > Message input**.++**Accessories (keyboards and controllers):**++- Open **Accessories** at the top of the page. A device also gets its own entry on the main page once it's connected or turned on.+- **Work Louder Codex Micro** and **Work Louder Creator Micro 2** — open tasks from the keys and show their status with key lighting.+ - Connect with a data-capable USB cable and turn on **Enable**. Cindy uses the official Work Louder software that comes with a supported ChatGPT or Codex app, so keep one installed; close ChatGPT if it's holding the device. On macOS, allow Cindy under **Input Monitoring**.+ - **Lighting**: **Brightness** and **Auto-dim**.+ - **Agent keys**: **Task order** (**Sidebar order**, **Recent tasks**, **Priority**, or **Custom**), **Focus Cindy with a single tap**, and each key's role — a **Task key** switches tasks and shows status (only the first 6 task keys can light independently), an **Action key** runs a shortcut.+ - Also **Microphone** (separate mic keys), **Joystick** directions, **Encoder** mode (switch tasks, navigate the composer, reasoning effort, scroll, or custom), the keycap **Layout** editor, and **Reset layout**.+ - **Codex compatibility** (macOS only): **Prevent Codex from occupying the device** while Cindy runs; quit and reopen Codex to apply.+- **Xbox controller**, **PlayStation controller** (DualSense / DualShock), **Switch controller** (Switch Pro / Joy-Con), and **Generic controller** — scroll, talk, send, and switch tasks with a gamepad over USB or Bluetooth.+ - Turn on **Enable**; Cindy only listens while it's in the foreground.+ - Click a control to assign an action (a command, a skill, or **Voice input** — hold to talk, release to stop). A stick can **Scroll conversation** or use **Custom directions**. **Reset mapping** restores the defaults.diff --git a/apps/desktop/help-knowledge/language-and-notifications.md b/apps/desktop/help-knowledge/language-and-notifications.mdindex 9a16fbe163..11bdb9c67f 100644--- a/apps/desktop/help-knowledge/language-and-notifications.md+++ b/apps/desktop/help-knowledge/language-and-notifications.md@@ -1,21 +1,27 @@---id: language-and-notificationstitle: Display language and notifications-summary: Display language and desktop / FeiShu notifications — in Settings > General.+summary: Display language and desktop notifications in Settings > General, plus where IM-bot notifications, Agent Island, and other General settings (app behavior, send shortcut) live.tab: general----Settings > General controls the display language and how the app notifies you. Theme, wallpaper, and fonts live in Settings > Appearance (see the appearance topic).+Settings > General controls the display language, desktop notifications, and a few app-wide behaviors. Theme, wallpaper, and fonts live in Settings > Appearance (see the appearance topic).**Language:**-- Display language: System / English / 简体中文 / 繁體中文 / 日本語 / 한국어. Used for the interface and AI error reports. Other replies follow your message or language request. On mobile, the phone's display language is used for AI error reports for messages sent from that phone.+- **Display language**: Auto (follow system) / English / 简体中文 / Traditional Chinese (繁體中文) / 日本語 / 한국어. It's used for the interface and AI error reports. Other replies follow your message or language request.+- Changes apply immediately. With **Auto (follow system)**, Cindy reads the system language at startup, so restart Cindy after changing your computer's language.+- On mobile, the phone app's display language is used for AI error reports for messages sent from that phone.**Notifications:**-- **System Notifications** — notify when a session finishes, encounters an error, or needs a reply. Regular sessions notify only while the window is in the background. When Agent Island is enabled, it handles these alerts instead.-- **FeiShu DM on session finished** — requires the FeiShu bot to be configured (see the FeiShu bot topic).+- **System Notifications** — notify when a session finishes, encounters an error, or needs a reply. Regular sessions notify only while the window is in the background.+- **Agent Island** (macOS 14 or later, in its own **Settings > Agent Island** tab) shows running, needs-reply, and unread completed tasks at the top of the display. When it's on, it handles these alerts instead of system notifications.+- **Feishu / Lark DMs** are no longer set here. Turn on **Send a Feishu / Lark DM when a session finishes or needs a reply** in the Feishu / Lark Bot card under **Settings > IM bots** (see the Feishu bot topic). It needs a bound bot and only sends while the window is unfocused.+- **WeCom group notifications** (automation completion alerts via a WeCom group-bot webhook) are in the WeCom Intelligent Bot card under Settings > IM bots.-**Notes:**+**Other settings in General:**-- Language changes apply immediately, but a few static strings may need a window reload to update.+- **App Behavior** — **Keep computer awake** (so background agents and scheduled tasks keep running), **Start at Login** (off by default), **Show remaining usage in the menu bar** (macOS), **When closing the main window** (Windows: Minimize to tray / Quit Cindy; Linux: Minimize window / Quit Cindy), and **Prevent Accidental Clicks When Returning** (macOS / Windows).+- **Message input** — **Use ⌘↵ to send** (Ctrl+Enter on Windows / Linux). When it's off (the default), Enter sends and Shift+Enter inserts a new line.+- General also holds your account info, Teammates, Collaboration limits, Agent Resource Usage, Git Safety, Pi extensions, Experimental, and Logout.diff --git a/apps/desktop/help-knowledge/memory.md b/apps/desktop/help-knowledge/memory.mdindex 99b67fe059..9383131e51 100644--- a/apps/desktop/help-knowledge/memory.md+++ b/apps/desktop/help-knowledge/memory.md@@ -1,40 +1,45 @@---id: memorytitle: Memory across conversations-summary: The "Cindy" master switch turns on shared cross-agent memory (default on); when off, each agent uses its own native memory — all in Settings > Personalization > Memory.+summary: Settings > Personalization > Memory — the "Cindy" switch (on by default) turns on shared memory per working directory; when it's off, Claude Code and Codex use their own native memory. Teammates such as Cindy keep their own memory, viewable in the teammate's settings.tab: personalization---Memory lets the agents carry context across your conversations — facts about you, your projects, and how you like to work. There are two kinds, and one master switch chooses between them.**Where it lives:**-- Settings > Personalization > Memory.+- Settings > Personalization > **Memory**.**The "Cindy" master switch (default ON):**-- Turning **Cindy** on enables a **shared, cross-agent memory** scoped to each working directory — Claude Code and Codex sessions in the same directory draw on the same memory.-- While it's on, the per-agent native toggles below are **disabled / grayed out**, because the shared memory supersedes them.-- It's **on by default** for new installs.+- Turning **Cindy** on enables a **shared memory** scoped to each working directory — Claude Code, Codex, and Pi tasks in the same working directory draw on the same memory.+- While it's on, the **Claude Code** and **Codex** native-memory rows below are **disabled / grayed out**, because the shared memory replaces them.+- The **Pi** row is the opposite: it's only available while Cindy is on. It controls whether Pi saves its long-session compression summaries to Cindy memory for future Pi tasks.**Per-agent native memory (used when the master switch is OFF):**-- When Cindy is off, each agent falls back to **its own native memory** — so turning the master off does not mean "no memory," it means Claude Code and Codex each remember things their own way.-- Separate toggles let you enable / disable native memory for **Claude Code** and **Codex** independently (these are only interactive while the master switch is off).+- When Cindy is off, Claude Code and Codex each fall back to **their own native memory** — so turning the master off doesn't mean "no memory"; it means each agent remembers things its own way.+- The **Claude Code** and **Codex** rows each have their own switch (only usable while Cindy is off).**Resetting:**-- **Per-agent reset** clears a single agent's native memory.-- **Reset all Cindy memory** clears the shared cross-agent memory for every working directory.-- A section-level **reset to defaults** restores the switches themselves.-- Resets ask for confirmation and aren't undoable.+- The reset button on the **Cindy** row deletes all Cindy memories for every project.+- The reset button on an agent row deletes all memories that agent has accumulated across every project.+- When you've changed the switches, a **Custom** badge and a **Restore default** button appear next to the Memory title; it restores the switches (not the memories).+- Resets ask for confirmation and can't be undone.++**Teammates (such as Cindy) have their own memory:**++- A teammate keeps its own memory, separate from the task memory above. To see or edit it, open the teammate's chat, click its name / avatar at the top to open its settings, and choose **Memory**. There you can search memories, open one to read or edit it, delete it, or turn **Use memory** off.**What memory is, what it isn't:**-- It's the agents' **long-term memory across sessions** — notes about you and your projects.-- It's **not** the in-session chat history (that's always persisted to the session itself).-- It's **not** your Personalization custom instructions (those are standing instructions you write; memory is what the agents record about you).+- It's the agents' **long-term memory across tasks** — notes about you and your projects.+- It's **not** the chat history of a task (that is always saved with the task).+- It's **not** your custom instructions in Personalization (those are standing instructions you write; memory is what the agents record).+- To make the agent remember something specific, just tell it in the chat (for example "remember that we use pnpm in this repo").**Notes:**-- Changes take effect on the **next** chat / agent turn — toggling mid-conversation doesn't retroactively scrub current context.-- Memory is stored locally and does not migrate if you change machines or wipe the app's user data.+- After changing a memory switch you usually need to **start a new task** for it to take effect; running tasks keep the setting they started with (running Codex tasks switch after their current turn finishes).+- Memory and these switches belong to your signed-in account on this computer. Memory is stored locally and doesn't move to another computer.diff --git a/apps/desktop/help-knowledge/mobile-app.md b/apps/desktop/help-knowledge/mobile-app.mdnew file mode 100644index 0000000000..f0adf9d6cc--- /dev/null+++ b/apps/desktop/help-knowledge/mobile-app.md@@ -0,0 +1,48 @@+---+id: mobile-app+title: Using Cindy on your phone (mobile app, notifications, remote desktop)+summary: Control your computer's Cindy tasks from the Cindy mobile app — no pairing, just sign in with the same account and allow device control; get push notifications when a task finishes, fails, or needs your reply; and view or operate your computer with Remote desktop.+tab: remote-control+---+The Cindy mobile app is a remote control for Cindy on your computer. The agents still run on the computer; the phone shows the same tasks and lets you work with them.++**Connecting your phone (no pairing code):**++1. Install the mobile app. On the computer, click the phone icon (**Download Cindy Mobile**) in the account area of the left sidebar to get a download QR code.+2. Sign in on the phone with the **same Cindy account** as the computer.+3. On the computer, turn on **Allow same-account devices to control this machine** (**Settings > Remote Connection**, or the switch in the download dialog).++Your computer then shows up on the phone automatically. If the phone says the computer is offline, open Cindy on the computer and stay signed in; the phone reconnects on its own. Revoked phones can't reconnect until you restore their access in **Settings > Remote Connection** (see the Remote topic).++**What you can do from the phone:**++- Browse tasks by computer, open them, and start a **new task** (choose the computer, then a project or **Chat (no project)**, the agent, and optionally a worktree).+- Send messages (messages sent while the agent is working are queued), **Stop**, attach photos, files, or screenshots, use voice input, and switch model, reasoning effort, and permission mode. **Plan Mode** and **Goal Mode** are available too.+- Approve permission prompts, answer the agent's questions, and review plans.+- Rewind (**Back to here**), fork (**Start new session from here**), ask in a side chat, and manage Collaboration workers.+- Browse a task's files (read-only), use plugins, and chat with Teammates.+- Share a task with other accounts, or join one (**Settings > Share Task > Manage Sharing** on the phone; see the Sharing tasks topic).+- On iPhone, share content from other apps into Cindy; it's placed in the composer, not sent automatically.+- The phone can't change the computer's global settings, sign-in, or API keys, and it can't be controlled itself.++**Push notifications:**++- On the phone, open **Settings** and turn on **Message and task notifications** (off by default; the system asks for permission). You're notified about chat messages and tasks that **finish, fail, or need your reply** (a question, permission prompt, or plan review).+- **Show chat previews** controls whether notifications include message text.+- No push is sent while you're watching that task live on another device, or while the Cindy window on the computer is focused. While the phone app is open, task notifications don't pop up as banners.+- An automation's runs notify the phone only if that automation has desktop or Feishu notifications turned on.+- There's no separate setting on the computer: the phone's switch is the only control.+- On Android phones without Google push, turn on **Keep chat connected in background** to keep receiving chat messages (it shows an ongoing notification you can stop).++**Remote desktop (see and operate the computer's screen):**++- On the computer, turn on **Allow remote desktop** in **Settings > Remote Connection** (device control must also be on). On macOS, grant **Screen recording** (to view) and **Accessibility** (to control) when prompted; the phone can open the permission guide on the computer.+- On the phone, tap **Remote desktop** in a task's header, on the computer's task list, or in its device details. Control is requested when you connect; switch on **View only** to just watch.+- Phone tools include **Keyboard**, **Trackpad** / **Touch** input, **All windows** / **Desktop**, clipboard copy and paste between phone and computer, **Sound**, display quality and resolution, and a Security section (**Lock Computer on Exit**, **Computer privacy screen**, **Clipboard sync**, and on iPhone with a Mac, **Automatically unlock computer**).+- From another computer on your account, open **Remote desktop** from the sidebar's device scope menu, the button next to a task that runs on that computer, or **Settings > Remote Connection**. It opens in its own window with **Controls** for display, sound, clipboard, and security.+- Only one device can use a computer's remote desktop at a time; **Take over connection** disconnects the other device. The controlled computer shows who is viewing or controlling it, with **Disconnect**.++**Notes:**++- The phone has no agent of its own: if the computer is asleep, offline, or signed out, its tasks can't run. Turn on **Keep computer awake** in **Settings > General > App Behavior** to keep it reachable.+- The Mainland China edition is a separate app with its own accounts (it signs in with a phone number). Use the same edition and the same account on the phone and the computer.diff --git a/apps/desktop/help-knowledge/models-and-compat.md b/apps/desktop/help-knowledge/models-and-compat.mdindex 6e41812697..20ee51d263 100644--- a/apps/desktop/help-knowledge/models-and-compat.md+++ b/apps/desktop/help-knowledge/models-and-compat.md@@ -1,23 +1,33 @@---id: models-and-compattitle: Choosing a model-summary: Pick a session's model from the composer dropdown; non-Anthropic models work automatically (no compat toggle to flip).+summary: Pick a task's model from the model selector in the composer; Cindy picks the engine (Claude Code / Codex / Pi) for you, and Customize lets you change engine, reasoning effort, and Fast. Compatibility-mode engines are off by default and are turned on per model in Settings > Providers.tab: providers----You pick the model a session uses from the **model dropdown in the composer toolbar**. In the dropdown, models are organized under the source providers you've connected. The brand groups they belong to are **Anthropic**, **OpenAI**, **Budget GPT**, **xAI**, **Google**, and **Domestic** (Kimi / GLM / Qwen / DeepSeek), plus non-chat model kinds (image / audio / video / …).+You pick the model a task uses from the **model selector in the composer toolbar** (the pill showing the current model). It works the same for a new task and inside an existing one.-The default depends on the agent: Claude Code defaults to **Opus 4.8**, Codex defaults to a **GPT** model.+**The model panel:**-**Switching mid-session:**+- A search box at the top; a rail on the left with **Recent**, **Favorites**, **All**, and one icon per connected provider (source); the model list on the right; and **Configure models** at the bottom to connect more sources.+- You choose a **model**; Cindy automatically picks the recommended **engine** (Claude Code, Codex, or Pi) for it and shows it in the pill.+- Hover a model row and click **Customize** (or right-click the row) to change its engine, **Reasoning Effort**, and **Speed** (Fast, where supported). **Reset to recommended** undoes your changes.+- Click ☆ on a row to **Save to favorites** — a favorite keeps that exact setup (source, model, engine, effort, Fast). **Recent** lists the last few setups you used.-- Open the dropdown and pick a different model. It takes effect on the **next turn**.+**Default model for a new task:**++- It depends on what you've connected: with a ChatGPT (OpenAI) subscription it's **GPT-6.1 Sol** on Codex; with a Claude (Anthropic) subscription it's **Opus 5.5** on Claude Code; otherwise it's Cindy AI's **Auto** on Pi. After that, a new task starts with the model you last picked.++**Switching mid-task:**++- Open the selector and pick a different model. It applies from your **next message**; while a switch is pending, the pill's tooltip shows "Current" and "Next message".+- If the new model runs on a **different engine**, Cindy asks you to confirm (**Switch Agent?**). The new engine continues from a handoff summary of the conversation, so some reasoning or details may not carry over perfectly.**Using non-Anthropic models:**-- You don't need to flip anything. Claude Code's SDK emits Anthropic-specific request fields (`thinking`, `cache_control`, etc.) that non-Anthropic backends would reject, so the app **always** routes requests through a local proxy that scrubs those fields per-model before forwarding. Claude / Anthropic models pass through untouched.-- (This used to be a manual "Compat Mode" toggle in Settings. That toggle has been retired — the handling is now automatic and always on, so there's nothing to configure.)+- There's no global compatibility toggle to flip. Each model lists the engines that support it natively, and the recommended one is chosen automatically.+- Some engines can only run a model in **Compatibility mode** (requests are converted). These are **off by default**, because they can be unstable. To turn one on, open **Settings > Providers**, open the model's **Advanced settings**, and enable that engine under **Engine support**. It is then shown with a "Compatibility mode" label in the model's Customize options.**Notes:**-- Which models you can pick depends on the providers connected in **Settings > Model Providers** (see the Providers topic).-- "Budget GPT" is a lower-cost routing tier for GPT models; "Domestic" groups the China-hosted models.+- Which models you can pick depends on the sources connected in **Settings > Providers** (see the Providers topic). Use **Configure models** at the bottom of the panel to get there.+- Models from your other computers or shared with you can also appear in the panel, labeled with where they run.diff --git a/apps/desktop/help-knowledge/personalization.md b/apps/desktop/help-knowledge/personalization.mdindex b6666539bf..a52ce6580b 100644--- a/apps/desktop/help-knowledge/personalization.md+++ b/apps/desktop/help-knowledge/personalization.md@@ -1,25 +1,34 @@---id: personalization-title: Custom instructions (personalization)-summary: Standing instructions the agents follow across every session — in Settings > Personalization.+title: Custom instructions and the Personalization page+summary: Settings > Personalization — custom instructions the agents follow in every new task, plus Memory, Subagents, Auxiliary models, Vision Bridge, Smart Contacts, context compaction, default terminal shell, where links open, and Tips toggles.tab: personalization----Personalization is where you give the agents standing instructions that apply to **every** session — tone, language, conventions, things you want the agent to always keep in mind.+Personalization is where you give the agents standing instructions that apply to **every** new task — tone, language, conventions, things you want the agent to always keep in mind. The same page also holds several agent-behavior settings.-**Setting it up:**+**Custom instructions:**-- Open Settings > Personalization.-- Write your instructions in the text area, then save.+- Open **Settings > Personalization**. The first section (**Personalization · Custom Prompts**) has a text box for your instructions.+- Write your instructions, then click **Save**. The limit is 8,000 characters; a counter shows how much you've used.+- They're appended to the end of the agent's system prompt in Cindy's own tasks.- Useful examples: "respond in Chinese", "default to TypeScript", "prefer pnpm over npm", "always check tests before reporting done".-**Other things on this page:**+**Other sections on this page:**-- **Memory** — long-term memory across sessions (see the Memory topic).-- **Subagent models** — choose which models the agents' subagents use.-- **Tips** — a few functional toggles (e.g. silent encrypted retry, chat embedding).+- **Memory** — long-term memory across tasks (see the Memory topic).+- **Subagents** — the default model Claude Code uses for subagents, and **Codex smart routing** (lets Codex pick from more connected models per subtask; off uses Codex's native assignment).+- **Auxiliary models** — the small models used for background chores such as automatic task naming, composer input suggestions, the Help assistant, pinned-task summaries, and IM titles. Keep **Automatic recommendation**, or choose **Custom** to set a **Preferred** model and up to two fallbacks.+- **Vision Bridge** — lets text-only models (e.g. DeepSeek) read images: images are first described by a multimodal model you choose (a **Primary** plus an optional **Backup**). Turn on **Enable Vision Bridge**, then pick which models route through it.+- **Smart Contacts** — a people / organization directory the agents can look up and add to as they work (uncertain entries wait in **Pending** for your review). You can browse, search, and edit entries here, and optionally turn on **Sync across my devices**.+- **Context compaction** — the context-usage percentage at which Claude Code and Pi automatically compact a long conversation (Codex sets its own threshold).+- **Default terminal shell** — the shell used for new terminal tabs in the right sidebar (existing tabs aren't affected).+- **Open Links** — whether links in messages open in Cindy's **Built-In Browser** or your **System Default Browser**, set separately for web pages and local pages. Right-click a link to pick another option just for that link.+- **Tips** — a few functional toggles: **Input Suggestions** (suggests your next message after a reply; press Tab to insert it), **Silently drop protocol-encrypted content and retry** (helps when switching providers mid-task; on by default), **Enable chat semantic indexing** (not available in local mode), and **Show question navigation rail** (tick marks beside the chat to jump between your questions).**Notes:**-- Changes apply to **new turns**; they don't rewrite previous messages in current sessions.-- Per-session overrides (a stricter rule for one specific session) are best done in that session's first message, not here.-- Don't put secrets here — instructions are sent with every prompt.+- Custom instructions apply to **new tasks**; tasks that are already running keep the instructions they started with and don't rewrite earlier messages.+- They apply only to Cindy's built-in tasks, not to Claude Code, Codex, or Pi you run yourself in a terminal or another IDE.+- Custom instructions are saved on this computer for your signed-in account; they don't sync to other devices.+- For a stricter rule in one specific task, put it in that task's first message instead.+- Don't put secrets here — the instructions are sent to the model with every task.diff --git a/apps/desktop/help-knowledge/plan-mode.md b/apps/desktop/help-knowledge/plan-mode.mdindex a5011d732e..37f14a8577 100644--- a/apps/desktop/help-knowledge/plan-mode.md+++ b/apps/desktop/help-knowledge/plan-mode.md@@ -1,24 +1,31 @@---id: plan-mode-title: Reviewing and approving plans-summary: When the agent proposes a plan, approve it, edit it, or send feedback before it proceeds.+title: Plan mode and reviewing plans+summary: Turn on Plan mode from the composer's "+" menu (or type /plan) so the agent proposes a plan first; then approve it, edit it, send feedback, or dismiss it before any work starts.----For larger tasks the agent may first propose a **plan**, shown as a card in the chat. You decide whether to approve, edit, or send feedback before any code runs.+In **Plan mode** the agent first proposes a **plan** and waits for your approval before it starts doing the work. It's available for Claude Code, Codex, and Pi tasks.-**Card controls:**+**Turning Plan mode on:**-- **Approve** — the agent receives the (possibly edited) plan and starts executing.-- **Feedback** — opens a text row; what you write is sent back to the agent as a denial reason. The agent can then propose a revised plan or take a different path.-- **Edit** — open the plan markdown in the card and edit it directly. Edits are auto-saved (debounced) to a plan file on disk; on Approve the edited version is what the agent uses.-- **Cancel** (the toolbar **✕** or **Esc**) — dismiss the review and end the plan loop without approving or sending feedback.-- **Expand / Half / Minimize** — change the card's display size to read or skim it.+- Click the **+** button in the composer (or type `@`) and choose **Plan mode**, or type `/plan` and pick it.+- While it's on, a **Plan mode** chip appears above the composer. Click the ✕ on the chip (**Exit plan mode**) to turn it off.+- After you approve a plan, Plan mode turns off automatically and the agent starts executing. To plan again, turn it back on.-**How it appears:**+**Reviewing the plan:**-- The agent emits a `plan_review` event with the plan markdown; the app renders the card in the chat thread and pauses the agent until you respond.+When the plan is ready, a **Review the agent's plan** card replaces the composer at the bottom of the task, with the plan and a **Content** outline beside it.++- **Approve the plan and start coding** — click it or press **Enter**. The agent receives the plan (including your edits) and starts working.+- **Tell the agent what to do instead** — click this row, type your feedback, and press Enter or click send. The agent revises the plan and shows it again.+- **Edit plan** (toolbar) — edit the plan's Markdown directly. Edits are saved automatically, and Approve uses the edited version. **Exit edit mode** goes back to the formatted view.+- **Dismiss this plan** (the toolbar **✕**, or **Esc**) — closes the review without approving or sending feedback; the agent doesn't start the work.+- **Expand / Collapse to half / Minimize** — change the card's size so you can read the chat above it; a minimized card can be restored from its bar.++**In the chat history:**++- Each plan stays in the task as a **Plan review** entry marked **Approved**, **Revision requested** (with your feedback), **Cancelled**, or **Expired**. Use **Show full plan** to read an approved plan again.+- If the agent's turn is stopped or the task closes before you decide, the review is marked **Expired** and can no longer be answered — send a new message to continue.**Notes:**-- For Claude Code, this card is exactly how the SDK's `ExitPlanMode` tool surfaces: the app intercepts that tool call into a `plan_review` interaction and, on Approve, auto-exits plan mode. Cindy's renderer ↔ main `resolveInteraction` layer is what renders the card and returns your decision.-- An approved plan card stays in your chat history with an "approved" state; you can scroll back to it.-- If you close the window while a plan card is open, the agent is still paused waiting for your decision — reopen the session to resolve it.+- You can also review and approve plans from the Cindy mobile app when you control this computer from your phone.diff --git a/apps/desktop/help-knowledge/plugins.md b/apps/desktop/help-knowledge/plugins.mdindex 7157b9a3e4..17fac20fc1 100644--- a/apps/desktop/help-knowledge/plugins.md+++ b/apps/desktop/help-knowledge/plugins.md@@ -1,63 +1,78 @@---id: pluginstitle: Plugins (extending what the agents can do)-summary: What plugins are, how to browse / install / enable / disable them, per-working-directory control, the permission model, and invoking one with a $command.+summary: Where the Plugins page is, the built-in plugins, installing (marketplace, .cindy file, custom marketplaces), enabling / disabling globally or per project, updates, uninstalling, account / key setup, $commands, panels, and the permission model.tab: ghosts----Plugins (internally "意识" / Ghost) extend what the agents can do — they add AI-callable tools, panels that live as pinned tabs in the right sidebar, full-page app views, and "ask Cindy to make it" media generation. Each plugin is a self-contained `.cindy` capability pack that runs in its own isolated sandbox.+Plugins extend what the agents can do — they add tools the agent can call, panels that live as tabs in the right sidebar, full-page app views, and integrations with outside services. Each plugin is a self-contained `.cindy` package that runs in its own isolated sandbox.**The Plugins page:**-- Open it from the **Plugins** entry in the left sidebar.-- Browse by origin: **all / built-in / team-shared / external**. The team-shared (enterprise) group only shows if your account belongs to an organization.-- Each plugin has a detail page — that's where you enable it, connect an account / enter an API key, and read what it can do.+- Open it from **Plugins** in the left sidebar, or from **Settings > Plugins**. **Plugins** and **Skills** are the two tabs at the top of the same page (see the Skills topic for skills).+- **Installed** lists your plugins; **Recommended** lists plugins from the marketplace, filtered by source: **All / Public / Team shared / Custom**. **Team shared** only appears if your account belongs to an organization.+- Use **Search plugins** to find one. Each card has a quick action — **Use** opens its panel, **Chat** starts a new task with the plugin, **Manage** (or clicking the card) opens its detail page.+- The detail page is where you enable / disable it, connect an account or enter an API key, see what it contains and its version, view **Your calls** (usage), and uninstall it.-**Built-in plugins that ship with the app:**+**Built-in plugins:**-- **Cindy Art** (`$cindy-art`) — generate and edit images and short videos.-- **Cindy Web Search** (`$cindy-web-search`) — public web search. **No key needed**: by default it searches through Cindy AI and uses your signed-in Cindy account's balance (unavailable in local mode). Results come with a title, link, and short summary. To search through your own service instead, add a Brave, Tavily, or DeepSeek key on its detail page. A failed search is reported as-is; it never silently switches to another source.-- **Cindy Mermaid** (`$cindy-mermaid`) — build / repair Mermaid diagrams.-- **Cindy GitHub** (`$cindy-github`) / **Cindy GitLab** (`$cindy-gitlab`) — read / write issues, PRs / MRs, reviews, CI, releases.-- **Filo Google** (`$filo-google`) — Gmail / Calendar / Drive / Sheets.+- **Art** (`$cindy-art`) — create and edit images or generate short videos.+- **Web Search** (`$cindy-web-search`) — public web search. **No key needed**: by default it searches through Cindy AI using your signed-in account's balance (not available in local mode). To use your own service instead, add a Brave, Tavily, or DeepSeek key on its detail page. A failed search is reported as-is; it never silently switches to another source.+- **Mermaid** (`$cindy-mermaid`) — fix and normalize Mermaid diagrams written by the agent.+- **GitHub** (`$cindy-github`) / **GitLab** (`$cindy-gitlab`) — repositories, issues, pull / merge requests, reviews, CI. GitLab also supports self-hosted instances.+- **Gmail**, **Google Drive**, **Google Calendar**, **Google Sheets** — each is a separate plugin and connects its own Google account authorization. Some organization accounts get a single combined **Filo Google** plugin instead.+- **XD Atlassian** (Jira / Confluence) — an enterprise plugin, only for organizations that provide it.+- **Notion**, plus mail plugins such as **iCloud Mail**, **QQ Mail**, and **163 Mail**.+- The exact list can vary by account and organization.-**Enabling and disabling:**+**Installing:**-- **Enable / disable (globally)** from the plugin's detail page. Disabling puts its tools and panel to sleep but keeps the plugin installed — its right-sidebar tab is kept and comes back when you re-enable it. A plugin's panel shows up as a pinned tab in the right sidebar that stays across tasks; close it from the tab, and reopen it later by right-clicking the plugin's entry in the left sidebar and choosing "Open in the sidebar".-- **Disable for one working directory only** — a plugin can be turned off just for a specific project. In that project's sessions it behaves as if it isn't installed (its tools and `$command` are unavailable there); everywhere else it stays on.+- Click **Install** on a Recommended plugin, or open **Add plugin** and choose **Install from file…** to pick a `.cindy` file. You can also drag a `.cindy` file anywhere onto the Cindy window, or double-click it in your file manager.+- **Add plugin > Add Plugin Marketplace…** adds a custom marketplace from a GitHub repository, Git URL, or local folder; its plugins then show under **Custom**.+- Before anything is installed, a **confirmation dialog** lists everything the plugin will be able to do (network addresses, credentials, background tasks, skills, bundled code, panels…). Nothing is installed until you confirm; after that the plugin is enabled right away.+- When you ask an agent to install a plugin, the same confirmation appears as a card in that task — whatever the task's permission mode, including Full Access. This also works from your phone when you control this computer remotely. If you deny it, nothing is installed.+- Plugins installed for you by default (for example, ones your organization rolls out) are installed and updated without a confirmation.-**Installing:**+**Updates:**++- Updates from the marketplace are applied automatically when the plugin is idle — **as long as the new version doesn't need more permissions**. If it does, the old version keeps working and the card shows "New permissions need confirmation"; click Update to review and confirm the added permissions.+- When updates are waiting, the page shows "N plugin updates available" with **Update all** (or **Ignore this round**). Failed updates retry later; you can retry right away from there.++**Enabling, disabling, and uninstalling:**++- **Enable / disable** with the switch on the plugin's detail page. Disabling puts its tools and panel to sleep but keeps the plugin installed (its card shows **Disabled**).+- **Off for one project only** — use the scope picker at the top of the Plugins page to choose a project (**Project Scope**), then turn the plugin off there. In that project's tasks it behaves as if it isn't installed (its tools and `$command` are unavailable); everywhere else it stays on. Its card shows **Off in this project**. A plugin disabled globally must be enabled in the **Global** view first.+- **Uninstall** from the detail page (**Uninstall plugin**). Its panel and capabilities are removed; its layout spot is restored if you install it again.+- Enabling or disabling takes effect on the agent's **next** query, not in the current turn.-- Install from the marketplace, or add a `.cindy` file manually (drag it in / pick it from disk).-- Choosing Install, dragging a package, or picking one from disk first shows a **confirmation dialog** listing everything the plugin declares it can do (network domains, credentials, background tasks, skills, bundled code, panels…). Nothing is installed until you confirm; after that the plugin is enabled right away.-- When you ask an agent to install a plugin (`ghost_market_install` / `ghost_forge_install`), the same confirmation appears as a card in that task — whatever the task's permission mode, including Full Access. It also works from the phone when you control this computer remotely. If you deny it, nothing is installed.-- Plugins pushed by your organization (server default installs) are installed and updated silently without a confirmation.-- The server selects the release appropriate for this Cindy version before marketplace delivery. Desktop does not apply a second `minCindyVersion` gate; custom marketplaces and local `.cindy` packages follow the same install policy.-- Updates from stable marketplace sources are applied automatically when the plugin is idle — **as long as the new version doesn't need more permissions**. If it does, the automatic update pauses, the old version keeps working, and the plugin card shows "New permissions need confirmation"; clicking Update shows the added permissions for you to confirm. Failed attempts retry later; you can still click Update to retry immediately.+**Panels and app views:**-**Creating or updating your own plugin:**+- A plugin's panel appears as a tab in the right sidebar and stays across tasks. If you close it, reopen it with **Open in the sidebar** on the plugin's detail page, or from the right sidebar's **+** menu.+- Plugins with a full-page view appear in the left sidebar (under **More** by default; use **Customize** to move one to the top). Right-click one to **Open as page**, **Open in the sidebar**, or **Open in a separate window**.-- Choose **Create with Cindy** from the Plugins page's add menu. Cindy opens a new-task draft with the built-in Plugin Creator's actual command, selecting the built-in version even if a user Skill has the same name. Describe what you want and send it using your chosen Agent and model.-- The built-in `cindy-plugin-creator` Skill supports creating, editing, debugging, packaging, and publishing plugins. Its bundled references contain the authoring documentation for the current Cindy version. If it is unavailable, the page reports that and keeps your existing draft.-- Creating and packaging a plugin produces a `.cindy` file. Installation and publication are separate actions that you can request when ready.+**Configuring a plugin's account / API key:**++- Do it on the plugin's **detail page**. Plugins that use a key or token (optional search keys for Web Search, GitHub / GitLab / Notion tokens, mail app passwords) have a field there; the value is stored encrypted on this computer and can't be read back.+- Plugins that use account authorization (for example the Google plugins) show a button to connect your account instead.+- If you try to use a plugin that isn't set up yet, Cindy asks you to open its settings first.**Using a plugin with `$command`:**-- Type its command at the **start** of a message — e.g. `$cindy-art a poster for…`. Only enabled plugins that declare that command match.+- Type its command at the **start** of a message — e.g. `$cindy-art a poster for…`. Only enabled plugins that declare that command match; a matched command turns into a pill with the plugin's icon.+- You can also pick a plugin from the composer's **+** menu, which puts its command at the start for you.- You don't have to use `$command`: the agents can also pick an enabled plugin's tools on their own when a task calls for it.-**The permission / sandbox model (why plugins are safe):**+**Creating your own plugin:**-- A plugin's sandbox has **no direct filesystem, network, or system access**. The host mediates every operation and enforces its actual execution context.-- Autonomous/background plugin behavior must stay within the capabilities and scopes declared directly in its manifest. Whether an in-flight plugin tool runs is decided by the existing Agent authorization flow; network and working-directory operations bind to that call, while bundled code and CLIs continue to use the existing Node worker. Plugin authors do not need a client change or a manifest Slot to pre-register a specific command, host, or path.-- **Autonomous network** access is limited to a domain allowlist the plugin declares; the host proxies every request. Credentials never enter the sandbox — the host injects them only into declared requests whose target matches that declaration.-- Because of this, a plugin **can't see your files** on its own. A file reaches a plugin only when you or the agent explicitly hand one over (an image attachment, or a directory the host passes through after an ownership check).+- Choose **Add plugin > Create with Cindy**. Cindy opens a new-task draft that uses the built-in plugin creator Skill (`cindy-plugin-creator`); describe what you want and send it with your chosen agent and model.+- That Skill can create, edit, debug, package, and publish plugins. Packaging produces a `.cindy` file; installing and publishing are separate steps you ask for when ready.-**Configuring a plugin's account / API key:**+**The permission / sandbox model (why plugins are safe):**-- Do it on the plugin's **detail page**. Plugins that use a key (optional search-service keys for Web Search, GitHub/GitLab tokens) have a field there; the value goes straight into the host's encrypted store and can't be read back.-- Plugins that use OAuth (for example Filo Google) show a **Connect account** button instead.+- A plugin's sandbox has **no direct file, network, or system access**. Cindy carries out every operation on its behalf.+- Background behavior is limited to what the plugin declared and you approved at install. Whether a plugin tool call in a task runs follows that task's normal permission mode.+- Network access on its own is limited to the addresses the plugin declares. Credentials never enter the sandbox — Cindy adds them only to requests going to the declared address.+- So a plugin **can't see your files** on its own. A file reaches a plugin only when you or the agent explicitly hand one over (for example an image attachment, or a folder you approve).**Notes:**-- Enabling or disabling a plugin takes effect on the **next** agent query, not retroactively in the current turn.-- Skills (`/` slash commands) and Plugins (`$` commands) are two tabs of the same management surface — see the Skills topic for skills.+- You can manage the plugins installed on your other computers from the same page by switching the computer at the top (the target computer must be online and on a recent Cindy version).diff --git a/apps/desktop/help-knowledge/providers.md b/apps/desktop/help-knowledge/providers.mdindex 3b868f7677..fe587e710b 100644--- a/apps/desktop/help-knowledge/providers.md+++ b/apps/desktop/help-knowledge/providers.md@@ -1,27 +1,42 @@---id: providers-title: Model providers and Cindy AI-summary: Cindy AI is auto-provisioned after sign-in for pay-per-token model access; manage it (and custom providers) in Settings > Model Providers.+title: Providers and Cindy AI+summary: Model sources live in Settings > Providers — Cindy AI is provisioned automatically after sign-in for pay-per-token model access; add, reorder, disable, and share other providers (subscription accounts, API keys, local models, custom endpoints) there.tab: providers----Claude Code, Codex, and lightweight utility models can use **Cindy AI** for pay-per-token model access. Voice input is included for signed-in Cindy users and uses the separate voice service; it does not require or receive a Cindy AI key.+A provider is a source of models (with its own billing). Every model you can pick in the composer comes from a provider connected in **Settings > Providers**. Voice input is included for signed-in Cindy users and uses the separate voice service; it does not require or receive a Cindy AI key.++**The Providers page:**++- The left column lists your connected providers; click one to see its details and models on the right. **Drag to reorder** the list.+- **Add provider** is at the bottom of the list. **Enter Share Link…** (top right) is where you paste a provider share link someone sent you.**Cindy AI:**-- Open **Settings > Model Providers** (the tab is labeled "Providers"). Cindy AI is pinned at the top.-- You don't paste a key — it's **provisioned automatically after you sign in**. The page fetches its credentials for you ("Fetching credentials…" while it does).-- Controls: **Refresh credentials**, **Rotate key**, and **Disconnect** (which clears the Cindy AI key saved locally). Because it's tied to your account, it's available on each device you sign into.+- Cindy AI is the official service: all models, for Claude Code, Codex, and Pi, billed per token from your balance — no keys to manage.+- If you're not signed in, a **Cindy AI** row marked **Recommended** appears at the top of the list with a **Sign in to Cindy** button.+- You don't paste a key — it's **provisioned automatically after you sign in** ("Fetching credentials…" while that happens, then **Connected**). Because it's tied to your account, it works on every computer you sign into. Organization accounts may show "Not enabled for your organization".+- The detail pane shows your remaining balance or plan allowance, with **View usage** and **Buy a plan** / **Upgrade plan** (these open **Settings > Usage and billing**).+- Credential actions are in the **···** (More Actions) menu: **Refresh credentials**, **Rotate key**, **Disconnect** (clears the locally saved Cindy AI key), plus the masked key (click to copy).+- **Rotate key** revokes the old key immediately: running sessions, remote sessions, and mobile voice that use it need to be restarted.++**Adding other providers (Add provider):**++- The wizard has three steps: **Choose provider → Connect → Choose models**. Options are grouped as **Recommended**, **Detected on this computer**, **Subscription account** (sign in through your browser, e.g. Claude Code, ChatGPT, SuperGrok), **Local models** (run models on this computer — Cindy can install llama.cpp or Ollama and download models — or connect a local server that's already running), **More providers** (API key), and **Custom**.+- CLIs already signed in on this computer appear under **Detected on this device** in the list, with **Use local account**.+- **Custom endpoint** connects any OpenAI- or Anthropic-compatible service: display name, which runtime(s) use it (Claude Code / Codex / Pi), base URL, authentication (API key, subscription account sign-in, or none), and models. **Fetch model list** pulls available models; **Test connection** sends short requests (may cost a little); **Quick fill from a template** pre-fills known services.-**Status badges:**+**Managing a provider:**-- **Connected** — this device has usable Cindy AI credentials.-- If it's still fetching or disconnected, use Refresh credentials / sign in.+- In the detail pane, turn individual models on or off (**Disable This Model** / **Enable All**); only enabled models appear in the model picker.+- The **···** menu has **Rename**, **Edit connection settings** (custom providers), **Disable This Provider** (keeps credentials; enable it again later), and **Delete** (message history is kept).-**Custom providers:**+**Sharing a provider (Remote & Sharing):**-- You can also add your own provider (a base URL + key) — that's the flow with a paste-a-key dialog. Cindy AI itself never uses that dialog.+- A connected provider shows **Allow remote use**: when on, your other computers and people you share with can use its models; the agent runs on this computer using its sign-in and quota. This needs remote control allowed on this computer (**Settings > Remote Connection**).+- **Remote & Sharing** opens sharing management: **Create Share Link** (one-time link; you approve the request, optionally checking a pairing code), see **People You Share With** with usage, and **Pause Sharing** / **Remove** people.+- Providers from your other computers appear in the list with a remote badge; providers others shared with you appear under **Providers Shared with Me**.**Notes:**-- Direct service keys such as Brave Search, Tavily, and DeepSeek search are **not** managed here — they live on the matching plugin detail pages under Plugins (Cindy Web Search). They are optional; web search works through Cindy AI without them.-- Rotating or replacing credentials takes effect for new turns; in-flight requests finish on the previous key.+- Direct service keys such as Brave Search, Tavily, and DeepSeek search are **not** managed here — they live on the **Web Search** plugin's detail page under **Plugins**. They are optional; web search works through Cindy AI without them.diff --git a/apps/desktop/help-knowledge/remote.md b/apps/desktop/help-knowledge/remote.mdindex 210e61368d..0a0fdcba00 100644--- a/apps/desktop/help-knowledge/remote.md+++ b/apps/desktop/help-knowledge/remote.md@@ -1,31 +1,44 @@---id: remotetitle: Remote workspaces and device control-summary: Run agents on a remote machine over SSH, and let another signed-in device (e.g. your phone) control this desktop — both in Settings > Remote & device control.+summary: Run agents on a remote machine over SSH, let your other signed-in devices (e.g. your phone) control this computer or view its desktop, and manage local app connections — all in Settings > Remote Connection.tab: remote-control----Settings > **Remote & device control** has two parts: **SSH remote workspaces** (run an agent on another machine) and **My Devices** (let other devices you're signed into control this one).+**Settings > Remote Connection** has three parts: **My devices** (control your other signed-in devices, or let them control this one), **SSH remote hosts** (run an agent on another machine), and **Local App Connections**.-**SSH remote workspaces:**+**SSH remote hosts:**-- **Add a host** with the **+** button: give it an alias, hostname (IP or domain), user, port (default 22), and an auth method (SSH agent or a key file). Hosts from your `~/.ssh/config` also show up automatically; you can re-read that file at any time.-- Once a host connects (status dot turns green), expand it to:- - **Install the agents** (Claude Code / Codex) on the remote machine, and run a quick test prompt against them.- - **Start a remote session** — enter a working directory on the remote (defaults to `~`); the app creates the directory if needed and opens a normal session whose agent runs **on the remote machine** with that remote working directory. It appears in your sidebar like any other session.-- The agent-facing SSH tools (via the `cindy-ssh` plugin) reuse these **same** configured hosts — if a tool reports the host isn't found, add it here first.+- **Add** opens "Add a remote machine": alias, optional display name, hostname (IP or domain), user, port (default 22), and authentication — **SSH agent (recommended)**, optionally pinned to one key, or an **Identity file** (encrypted keys aren't supported; use ssh-agent for those). Optionally turn on **Route agent traffic via proxy** (a Cindy-managed tunnel back to your local proxy, or your own proxy URL reachable from the remote).+- Hosts from your `~/.ssh/config` show up automatically (their connection details are read-only in Cindy); the refresh icon (**Reload ~/.ssh/config**) re-reads the file. **Manage keys** lets you browse, generate, or unlock SSH keys; **Setup key** helps install a key on a host.+- Each host has **Connect** / **Disconnect**, **Auto-connect** (connect when Cindy starts), **Edit**, and **Remove**. Status reads Disconnected / Connecting… / Authenticating… / **Ready**.+- Once a host is Ready, **Show agent setup** lets you **Install** / **Reinstall** / **Uninstall** Claude Code, Codex, and Pi on it, **Sync auth** (copy your local Codex sign-in to the remote), and run a **Quick test** prompt.+- To work on the remote: on the new-session page, open the project picker and choose **Add remote project**, pick the host under **Remote hosts (SSH)**, then choose (or create) a folder. Claude Code, Codex, and Pi can all run over SSH; if the agent isn't installed on the host yet, Cindy installs it automatically when you send the first message. Agent setup also has **Start a remote Codex session** (working directory defaults to `~`; you're asked before a missing directory is created). The session appears in your sidebar like any other, but its agent runs **on the remote machine**.+- Models that only work locally (e.g. Ollama) or subscription-direct models may not be available in SSH sessions; Cindy AI works.+- The agent's built-in **SSH commands** tool (Settings > Tools) reuses these **same** hosts — if it reports a host isn't found, add it here first.-**My Devices (controlling this desktop from another device):**+**My devices (shown when you're signed in):**-- **No manual pairing** — any device signed into the **same Cindy account** is discovered automatically and appears in the list (this is how the phone app connects to your desktop).-- The **top card is this machine**: rename it, see its relay-connection status, and toggle the master switch **"allow other devices to control this machine"**.-- Each **other device** has two independent switches: **you control it** (outbound) and **allow it to control this machine** (inbound). You can rename devices, and delete ones that are offline.+- **No manual pairing** — every device signed into the **same Cindy account** appears automatically (this is how the phone app connects to your computer). If none appear, sign in with the same account on the other device.+- The **top card is this device**: rename it, see its connection status, and the master switch **"Allow same-account devices to control this machine"** (no per-connection approval; a visible indicator shows while it's being controlled, and you can disconnect at any time).+- Below it, **Allow remote desktop** lets your signed-in devices view and operate all apps on this computer (device control must also be on). On macOS this needs Screen Recording and Accessibility permission; on Windows, **Lock screen control** additionally needs administrator authorization.+- Each **other device** has two independent switches: **I control it** and **Let it control this device**, plus rename and delete. Desktops also have a **Remote desktop** button to open their screen. Phones can control this computer but can't be controlled.+- While another device is connected, a "<device> is connected to Cindy" notice appears on this computer with a **Revoke** button (with several devices, **View** opens this page). A revoked device can't reconnect until you restore its access here.++**Using another computer:**++- On the new-session page, the **Device** picker chooses which computer the session runs on. The sidebar's device scope menu shows sessions from your other computers.**What a controlling device can and can't do:**-- A controlling device (like the phone app, which is control-only — it has no local agent or database of its own) can drive sessions: create / close / fork them, send and steer input, approve interaction prompts, switch model / effort / permission mode, rewind, manage collaboration workers and scheduled tasks, and read session / message lists.-- It can **not** touch auth or API keys, change global settings, write the database directly, trigger updates, or run local shell / file dialogs on the controlled machine. A new remote session's working directory must be currently accessible on the controlled desktop; disconnected network drives are rejected.+- A controlling device (like the phone app, which is control-only — it has no local agent of its own) can drive sessions: create / close / fork them, send and steer input, approve prompts, switch model / effort / permission mode, rewind, manage collaboration workers and Automations, and read session / message lists.+- It can **not** touch sign-in or API keys, change the controlled computer's global settings, trigger updates, or open local shell / file dialogs there. A new remote session's working directory must be currently accessible on the controlled computer; disconnected network drives are rejected.++**Local App Connections:**++- Lists other local apps you've allowed to use Cindy's models and agent capabilities (Cindy keeps the provider credentials). Revoke a single connection, or all of them, here.**Notes:**- Remote project sessions **can** be the lead in a collaboration workflow — workers then run on the same remote host (see the Collaboration topic).+- To keep a computer reachable while you're away, turn on **Keep computer awake** in **Settings > General > App Behavior**.- Device control works only between devices on the **same account**; the relay just routes between them — your session data isn't stored on it.diff --git a/apps/desktop/help-knowledge/reviewing-changes.md b/apps/desktop/help-knowledge/reviewing-changes.mdindex e54744a98f..0f105834c7 100644--- a/apps/desktop/help-knowledge/reviewing-changes.md+++ b/apps/desktop/help-knowledge/reviewing-changes.md@@ -1,26 +1,40 @@---id: reviewing-changestitle: Reviewing the agent's file changes-summary: Open the Review tab in the right sidebar to see what changed as git-style diffs, and stage / revert hunks.+summary: See what changed from a message's "Changed N files" card (Review / Undo / Reapply) or in the right sidebar's Review tab (git diffs by source, stage / revert, commit and push); use /review for an independent read-only review.----When you want to see what the agent changed, use the **Review** tab in the right sidebar — it shows the working directory's changes as git diffs.+There are two places to see what the agent changed: the **changes card** under each reply, and the **Review** tab in the right sidebar.-**Opening it:**+**Per-message changes card:**-- In the right sidebar, open the **+** menu and pick **Review** to add the Review tab.-- The diffs come from a real **git diff** of your working directory against a base ref — not from replaying the agent's individual Edit / Write calls. You can change the base branch it compares against.+- When a reply edits files, a **"Changed N files"** card appears under it, listing the files.+- **Review** opens those exact changes ("Changes from this message") in the Review tab.+- **Undo** reverts that message's file changes; **Reapply** puts them back. These are blocked while the agent is still replying, and if the files have changed since then, Cindy asks you to review the current changes first. Undo needs Git installed.+- If some writes (e.g. from shell commands) couldn't be captured exactly, the card says so and only the captured part can be reviewed or undone. Exact per-message changes aren't available for remote sessions yet.++**Opening the Review tab:**++- Show the right sidebar with the panel icon at the top-right of the window, click **+** (Add tab), and pick **Review**.+- The source dropdown picks what to compare: **Unstaged** (default), **Staged**, **Commit** (pick a commit on this branch), **Branch** (everything since the merge-base with a **Compare base** branch you choose), or **Last turn** (only files the agent touched in its latest turn).+- Diffs are read from real git state, so the working-tree views include edits you made yourself, not only the agent's.**Inside the tab:**-- Changed files are shown as diffs, **expanded by default**; collapse the ones you don't care about.-- Toggle between **unified** and **split** views. There's also inline word-diff, word-wrap, hide-whitespace, a file tree, and a rich Markdown preview for `.md` files.-- Colors follow git convention: **red removed / green added**.+- Toggle **Unified** / **Split** views. Diffs are expanded by default; **Expand all diffs** / **Collapse all diffs** and per-file collapse are available.+- **Show files** opens a file tree (with a filter); **Jump to file** filters and jumps.+- The view-options menu has word wrap, word diffs, hide whitespace, rich text preview for Markdown files (on by default), and **Copy git apply command**. Changed images show a Before / After preview.+- Colors follow git convention: **red removed / green added**. Each file can be opened or revealed in its folder.**Acting on changes:**-- Each hunk has action pills to **stage / unstage / revert** it — so you can review the agent's edits and selectively keep or roll back individual hunks right from the tab.+- **Stage** / **Unstage** / **Revert** whole files or single hunks, or use **Stage all** / **Unstage all** / **Revert all**. Revert asks for confirmation (untracked files are deleted).+- **Commit or push** (top of the tab): write a commit message, then **Commit** or **Commit and push** (optionally **Include unstaged changes**), or **Push** local commits. Cindy asks before a force push.+- Writes are paused while the agent is running, and during detached HEAD, merge conflicts, or an in-progress git operation. Sessions on SSH hosts or another computer are view-only.++**Independent review (`/review`):**++- Type `/review` (optionally with a focus or path) to start a fresh, read-only reviewer session with no memory of the work; it checks code changes, files, documents, or images against the task. It currently works for local sessions only.**Notes:**-- Because the tab reflects real git state, it shows everything in your working tree — including edits you made yourself — not only what the agent touched this turn.- This is separate from **Plan mode** (approving a plan before work starts) and from the **collaboration** worker panes.diff --git a/apps/desktop/help-knowledge/scheduler.md b/apps/desktop/help-knowledge/scheduler.mdindex 201a3379c8..65c0f785cb 100644--- a/apps/desktop/help-knowledge/scheduler.md+++ b/apps/desktop/help-knowledge/scheduler.md@@ -1,30 +1,35 @@---id: scheduler-title: Scheduling automated tasks-summary: Run a prompt automatically on a schedule (recurring or one-off) against a chosen working directory.+title: Scheduling automations+summary: Automations run a prompt (or a script) for you on a schedule, once, or on demand — open Automations in the left sidebar to create, pause, run, and review them.----The Scheduler runs prompts for you on a schedule — recurring (cron-style) or one-off (a specific time).+**Automations** run a prompt for you on a schedule — recurring, once, or only when you trigger it. Open **Automations** from the top of the left sidebar (if you've customized the sidebar, it may be under **More**).-**Creating a task:**+**Creating an automation:**-- Open the Scheduler.-- Give the task a **prompt** (what the agent should do), a **working directory** (where it runs), and a **schedule** — recurring (interval / cron) or one-time (specific datetime).-- Pick which **agent** runs it (Claude Code or Codex) and the model / mode you want.-- Save. The task fires according to its schedule.+- Click **New Automation**. Start from a template (**Start from a template** — e.g. nightly test self-healing, PR gatekeeper, domain radar) or choose **Blank**.+- Give it a title and a **prompt**, and choose where it runs: a **project** folder (optionally an isolated **worktree**, under **Advanced**) or a plain chat with no project.+- Pick the **model** (and reasoning effort); the engine — Claude Code, Codex, or Pi — follows the model.+- **Schedule**: **Automatic** or **Manually** (runs only when you click Run now). For Automatic, choose a preset — every N minutes, hourly, daily, weekdays, weekly, or monthly — in **Cron** mode (fixed clock times) or **Relative interval** mode (waits that long after the previous run finishes). Tick **Once** to run a single time at the next scheduled slot.+- **Run session**: **New each run**, **Continuous session** (every run continues the same session), or **Bind existing session**.+- Optional: **Stay quiet when checks find nothing** (only report changes or problems), send the result to a Feishu bot or WeCom group, and a **Pre-run check** — a command that decides whether a round runs or is skipped (**Generate with AI** can write it for you).+- **Execution** can also be **Script only**: run a command in a project directory without an AI turn.+- You can also ask an agent in any session to set up an automation for you.-**Managing existing tasks:**+**Managing automations:**-- Edit (prompt, schedule, working directory).-- **Pause / resume** without losing the task definition.-- Delete.+- The list groups automations by project and filters by status (All / Active / Paused).+- **Run now**, **Edit**, **Rename**, **Pause** / **Resume**, and **Delete Automation**. Deleting asks what to do with the sessions it created: keep, archive, or delete them. Pausing or deleting stops any run in progress.+- **Promote to project automation** saves it into the project's repository so collaborators share it; project automations are edited through their config file.**Run history:**-- Each task tracks its past runs — when it fired, whether it succeeded, the resulting session (you can open it to see what the agent did).+- Select an automation to see **Recent runs**: when each ran, its status (Success, Skipped, Aborted, Interrupted, …), duration, cost, and pre-run check output. **Open session** shows what the agent did; **Restart** re-runs, and a run record can be deleted.**Notes:**-- The task only fires while the desktop app is running on your machine — it's a local scheduler, not a cloud cron. Closing the app means tasks won't fire until you reopen it.-- Tasks run as normal sessions under the hood; you can find their resulting sessions in the sidebar (often grouped by the scheduler).-- If you have two instances open on the same account (e.g. dev + release), they coordinate so a due task fires only once — you won't get duplicate runs.-- A failing task shows the failure in run history but doesn't auto-pause itself — you may want to disable a chronically failing task manually.+- Automations run locally: they only fire while Cindy is running on this computer (closing the window to the tray is fine). Runs missed while the app was closed are not made up — the next one fires at the next scheduled time. To stop the computer from sleeping, turn on **Keep computer awake** in **Settings > General > App Behavior**.+- Automation runs are normal sessions; in the sidebar they're grouped under the automation's name (marked "Created by automation"), with Run now / Edit / Pause there too.+- A limited number of runs execute at the same time; others show "Waiting for a slot" and start as soon as one frees up.+- If several Cindy instances share the same account data on this computer, they coordinate so a due run fires only once.+- A failing automation shows the failure in its run history but doesn't pause itself — disable a chronically failing one manually.diff --git a/apps/desktop/help-knowledge/sessions-and-chat.md b/apps/desktop/help-knowledge/sessions-and-chat.mdindex 994b18c925..f29a3dbd22 100644--- a/apps/desktop/help-knowledge/sessions-and-chat.md+++ b/apps/desktop/help-knowledge/sessions-and-chat.md@@ -1,34 +1,43 @@---id: sessions-and-chattitle: Sessions and the chat thread-summary: Create sessions, send / stop / queue messages, clear context, and switch between sessions in the sidebar.+summary: Create sessions, pick model / engine / project, send / stop / queue / steer messages, compact or clear context, rewind, edit, fork, and switch between sessions.----Each **session** is a persistent chat thread driven by **one agent** (Claude Code or Codex). All your chats live as sessions in the left sidebar.+Each **session** is a persistent chat thread run by an agent engine (Claude Code, Codex, or Pi). All your sessions live in the left sidebar.**Starting a session:**-- Click **+ New** at the top of the sidebar.-- Pick the agent (Claude Code or Codex) and — for code work — a working directory.-- Pick a model in the composer toolbar if you want something other than the default.-- Type your first message and send.+- Click **New** at the top of the left sidebar (or press ⌘N / Ctrl+N).+- Choose a project folder for code work, or stay in **Chat** for plain Q&A. If you have other computers or SSH hosts, the **Device** picker and **Add remote project** choose where it runs (see the Remote topic).+- Pick a model in the composer's model selector. There's no separate agent picker: the engine (Claude Code, Codex, or Pi) follows the model and is shown next to it; to run a model on a different engine, use that model's **Customize** options (see the Choosing a model topic). You can also set the permission mode in the composer toolbar.+- Type your first message and send. **Enter** sends and **Shift+Enter** adds a new line; to send with ⌘↵ / Ctrl+Enter instead, use **Settings > General > Message input**.**During a turn:**-- While the agent is replying, **Send** becomes **Stop** — click it to interrupt the current turn cleanly.-- You can keep typing while the agent is busy — follow-ups are **queued** in order and sent one after another when the previous turn finishes.+- While the agent is replying, the composer shows a separate **Stop** button (labeled **Stop generating**) — click it to interrupt the current turn.+- You can keep typing: the send button becomes **Queue message**, and follow-ups are **queued** and sent one after another when each turn finishes. With the default Enter-to-send setting, press ⌘↵ / Ctrl+Enter instead to **steer** — deliver the message into the running turn without interrupting it.+- Queued messages appear above the composer, where you can **Steer**, **Edit**, **Delete**, or drag to reorder them. If you stop the turn, the queue pauses until you click **Continue sending**.-**Chat controls:**+**Context:**-- **`/clear`** — clears the current session's conversation context in place (resets messages and state without creating or switching to a new session). The session itself stays in the sidebar; only its content is wiped.-- **`/compact`** — ask the agent to summarize and compress earlier turns into a shorter context (Claude Code).-- **Edit** — you can edit your **last** user message. Doing so **rewinds** the conversation: it drops that message and everything after it, then resends your edited message. In sessions with file-rewind support (local git working directories with savepoints), file changes are also rolled back. In sessions without savepoints (remote sessions, non-git directories, or Codex sessions that haven't created savepoints), only the conversation is rewound — filesystem changes remain.+- The context ring in the composer shows how full the context is; click it to compact (Claude Code and Pi; Codex manages its own context, so the ring is read-only there). **`/compact`** also asks the agent to summarize earlier turns.+- **`/clear`** — clears the current session's conversation context in place (resets messages and state without creating or switching to a new session). The session stays in the sidebar; only its content is wiped.++**Message actions (hover a message):**++- **Copy message**; the **More message actions** menu also has **Copy current message link**.+- **Edit** — available on your **last** message. Editing **rewinds** the conversation to that message and resends your edited text.+- **Back to here** — rewind to an earlier message of yours: later messages are removed and, when a Git file recovery point exists, project files are rolled back too (the dialog shows how many files). Without one (non-git folders, recovery points turned off in **Settings > General > Git Safety**), only the conversation is rewound. Rewind isn't supported yet for remote Claude Code sessions.+- **Start a new session** — fork a separate session from that point; the original stays unchanged.+- **Delete this message** — removes it from the history; the agent's context is rebuilt from what remains on the next send.+- Replies that changed files show a **Changed N files** card with Review / Undo / Reapply (see the Reviewing changes topic).**Switching between sessions:**-- Click any session in the sidebar to switch to it. State is persisted — each session remembers its agent, model, working directory, attachments, etc.-- Keep **unrelated tasks in separate sessions** — sharing one session for everything dilutes the agent's context and confuses it.+- Click any session in the sidebar, or press ⌘1–⌘9 / Ctrl+1–Ctrl+9 to jump to the first nine visible sessions. Each session remembers its engine, model, working directory, attachments, and draft.+- Picking a model that runs on a different engine mid-session switches the engine after you confirm **Switch Agent?** (see the Choosing a model topic).+- Keep **unrelated tasks in separate sessions** — sharing one session for everything dilutes the agent's context.**Notes:**-- Sessions are **per local user** — they live in the local SQLite DB at `<userData>/cindy-<userId>.db` (installs migrated from the legacy app get their sessions copied from the old `xdt-maker-<userId>.db` on first login). They don't sync to other machines automatically.-- If you have two desktop instances open (e.g. dev + release) on the same user account, both read/write the same DB and will see each other's sessions.+- Sessions are stored locally on this computer, separately for each signed-in account. They aren't copied to your other computers, but with **My devices** (Settings > Remote Connection) you can view and drive sessions on your other signed-in computers from the sidebar.diff --git a/apps/desktop/help-knowledge/shared-tasks.md b/apps/desktop/help-knowledge/shared-tasks.mdnew file mode 100644index 0000000000..8af6aea250--- /dev/null+++ b/apps/desktop/help-knowledge/shared-tasks.md@@ -0,0 +1,46 @@+---+id: shared-tasks+title: Sharing a task with other people+summary: Share a task with other Cindy accounts through an invitation link (Share Task in the task menu); guests see the full history and message the same Agent. Manage sharing in Settings > Share Task; to send a copy instead, export a .cshare file.+---+**Share Task** lets people on **other Cindy accounts** join one of your tasks. They see its full history, tool results, and attachments, and can message the same Agent. The task keeps running on your computer (the "host"), using your models and connections. To use your own phone or other computers on the **same account**, you don't need sharing; see the Remote topic.++**Starting to share (host):**++- Open the task's menu (the **⋯** next to the task title, or right-click it in the sidebar) and choose **Share Task**, then **Start Sharing**. You must be signed in, and the task must be active (not archived). Tasks running on your other computers can be shared too.+- **Copy invitation** copies a message with the invitation link. Send it to your teammate. Anyone holding the link can join after signing in, with the same Agent access as you, and without needing your approval.+- A task can have up to **2 guests**. There's also a limit on how many tasks you can host or join at once.+- Sharing ends automatically if the host computer stays offline. To resume, start sharing again and send a new invitation link.++**Managing a shared task (host):**++- In the task menu, **Manage Sharing** has **Copy invitation**, **Manage Members…** (see members and **Remove** a guest), and **Stop Sharing**.+- Removing a guest or stopping sharing cuts off access right away and makes old invitation links invalid. The task history and the Agent's work are kept.+- Stop sharing before you archive or delete a shared task.++**Joining (guest):**++- Open the invitation link: it opens Cindy and joins you right away, then you **Enter Task**. If Cindy doesn't open, choose **Join Shared Task** and paste the link (an older invitation code also works).+- Where to find **Join Shared Task**: the **+** next to the **Share Task** group in the left sidebar (the group appears once you host or join something), or **Settings > Share Task**.+- As a guest you can send messages, add attachments, use voice input, stop the Agent, and answer the Agent's permission prompts, questions, and plan reviews. The task's Agent, model, and reasoning settings can only be changed by the host. Side chats and forking aren't available to guests.+- The task's menu only shows **Leave** for guests. To rejoin later, you need a new invitation link from the host.+- An invitation only works in a Cindy app that uses the same service — for example, an invitation from the Mainland China edition can't be used in the other edition.++**Settings > Share Task:**++- **Shared by Me**: every task you share (including from your other computers) — click a title to open it, **Manage** to see members and the invitation, or **Stop All Sharing**.+- **Joined by Me**: tasks you joined — click a title to open it, or **Leave**.+- **Join Shared Task** button: paste an invitation link or code.++**In the sidebar:**++- Shared tasks you host or joined appear together in the **Share Task** group. Tasks you host show a gold crown.++**Sending a copy instead of live sharing:**++- To give someone a snapshot of a task, use **Export session…** in the task menu to create a `.cshare` file (optionally password-protected). The other person imports it on their own computer and continues independently. See the Importing sessions topic.++**Notes:**++- Only invite people you trust: guests use the Agent with the task's existing permissions, tools, and plugins, and can read and change files in the task's working folder.+- Ending sharing doesn't take back anything guests already saw or downloaded.diff --git a/apps/desktop/help-knowledge/sidebar.md b/apps/desktop/help-knowledge/sidebar.mdindex 4db4ca7ecf..69ec71bd08 100644--- a/apps/desktop/help-knowledge/sidebar.md+++ b/apps/desktop/help-knowledge/sidebar.md@@ -1,75 +1,84 @@---id: sidebartitle: Finding and organizing sessions-summary: Browse, group, filter, sort, search, and pin sessions or projects in the left sidebar.+summary: Browse, group, sort, filter, search, tag, and pin sessions or projects in the left sidebar, and find and restore archived sessions.----The left sidebar lists your sessions. It's the main way to navigate and organize. The **Sidebar display settings** menu holds grouping, sorting, filters, and display options — open it from the sliders icon on the main list header, from **Sidebar display settings** in the device-scope menu, or by right-clicking any blank area of the sidebar.+The left sidebar lists your sessions. It's the main way to navigate and organize. The top of the sidebar has navigation rows (**New** — ⌘N / Ctrl+N starts a new session — then **Automations**, **Plugins**, **Search**; use **More → Customize** to show, hide, or reorder them). **Teammates** have their own section above **Pinned** (see the Teammates topic). Below them is the main session list.-**Device scope (the list header itself):**+**Finding and restoring an archived session:**-- The main list header doubles as the device scope switcher. Its text names what you're looking at — **All sessions**, **This device's sessions**, a remote device's name, or **N machines** — and clicking it opens the scope menu: all / this device / each connected remote device (single-click switches; checkboxes multi-select), plus **Remote connection settings** and **Sidebar display settings**.-- The scope applies to the whole sidebar, including the Pinned section.-- With no remote devices, the header still shows **All sessions ▾**. Clicking it opens a short menu with only **Remote connection settings** and **Sidebar display settings** — no device list. The header stays visible even when the list is empty, only pinned items remain, remote devices/tasks are loading or failed, or the selected remote device is still connecting. There is no separate device row in the top navigation.+- Archived sessions are hidden from the default list, but they are not deleted. Two ways to find one:+ 1. **Search** (top of the sidebar): search covers archived sessions by default, because its **Status** filter starts at **All**. Click a result to open it.+ 2. **Sidebar settings → Session status → Archived** (or **All**): the main list then shows archived sessions. Open **Sidebar settings** from the sliders button at the right end of the list header (it appears when you hover over the header), from **Sidebar settings** in the list-header dropdown, or by right-clicking a blank area of the sidebar.+- While **Archived** is selected, the sliders button stays visible and shows a small dot so you remember the list isn't the normal view. **All** shows no dot. Switch back to **Active** to return to the normal list.+- To restore: hover over an archived row and click the **Unarchive** (undo arrow) button, or right-click the row (or use its **⋮** menu, or the **⋯** menu in the open session's header) and choose **Unarchive**. Archived rows show an archive icon in place of the usual status icon.+- Sending a new message in an archived session restores it to **Active** automatically. An automation linked to an archived session also restores it the next time it runs.+- To restore many at once: with **Session status** set to **Archived**, a project's **⋮** menu shows **Restore all**, and the Chat group's menu shows **Restore All**.-**Grouping (independent checkboxes, freely combinable):**+**Archiving and deleting:**-- **Group by project** (default on) — sessions with a project fold into project rows; project rows and stray chats compete for position on the same timeline (a chat active two minutes ago sorts above a project last touched yesterday). Turn it off for a flat list.-- **Group by device** (default on) — only appears when a remote device is connected and the scope isn't narrowed to a single machine; splits the list into device sections (this device first), each with a collapsible header showing the device name and online state. When the last remote device disconnects — or the scope is a single machine — the option disappears and the effect is lifted; the saved preference is untouched and comes back automatically.-- **Group chats together** (default on) — collects projectless sessions into one collapsible **Chat** group instead of interleaving them. With device grouping on, each device section gets its own Chat group with independent collapse state.-- Grouping **by date** has been removed.+- Hover over a session row and click the archive button; it turns into a red **Archived** button — click that to confirm (it reverts after about 4 seconds otherwise). You can also choose **Archived** from the right-click menu. A running session can't be archived until you stop it; a session being remote-controlled must be taken back first.+- Cindy asks for confirmation when the session's worktree has uncommitted changes (they're saved as a git snapshot first) or when a linked automation will keep running and restore the session.+- Archiving removes the session's pin.+- **Delete** asks for confirmation and can't be undone from the UI.+- A project's **⋮** menu has **Archive all** (pinned and running sessions are skipped); the Chat group's menu has the same action.+- **Multi-select:** Cmd-click (macOS) / Ctrl-click (Windows) toggles rows; Shift-click selects a range. A bar above the list then offers **Archive selected sessions**, **Delete selected sessions**, and **Clear selection**.+- Sessions are **active**, **archived**, or **deleted**. **Settings > Storage > Database cleanup** can permanently clear the history of old archived (and optionally unarchived) sessions; scan first, and Cindy asks before cleaning.-**Sorting:**+**Device scope (the list header title):**-- **By time** (default) — one activity timeline for the whole list, most recent first. There is no oldest-first option.-- **Priority** — four tiers, same as Codex: sessions waiting on you (pending reply / approval / error) first, then finished-unread sessions, then running sessions, then the rest by time. A project row inherits the highest priority among its sessions. Unread state is in-memory: after a restart most sessions fall back to time order.-- **Manual** — drag project rows to set a persistent custom order. Manual order covers **project rows only**; stray chats and the Chat group always follow by recency. Switch to Manual first to drag; other sort modes do not start a drag.-- Sorting **alphabetically** and **oldest first** have been removed.+- The main list header title is a dropdown that names what you're looking at: **All sessions**, **Local**, a remote device's name, or **N devices**. Click it to switch scope: **All**, **Local**, or a connected remote device (click a row to switch to that device alone; click the checkbox at the right of a row, or Cmd/Ctrl-click, to select several). The menu also has **Remote desktop** (when another desktop computer can be controlled), **Remote connection settings**, and **Sidebar settings**.+- The scope applies to the whole sidebar, including **Pinned**.+- With no remote devices, the title still reads **All sessions ▾** and the menu only has **Remote connection settings** and **Sidebar settings**. The header stays visible even when the list is empty or remote sessions are still loading.-**Filters (one row that opens a submenu):**+**Sidebar settings menu** (each row shows its current value and opens a submenu):-- **Status** — active / archived / all.-- **Project** — multi-select working directories plus **Chat** (projectless sessions), or "all". Selecting specific projects hides chats unless **Chat** is also checked; check only **Chat** to see projectless sessions alone.-- **Agent** — Claude Code / Codex / all.-- **Recent activity** — last 1 day / 3 days / 7 days / 30 days / all.-- A **Reset filters** action at the bottom clears all four at once. Filters do **not** apply to the Pinned section, except Status (archived pins stay hidden while Status is Active). The list header does not show a standing “filtered” mark; open Sidebar display settings to see which filters are on.--**Display:**--- The main list has **Text** (compact single row) and **List** (two-line row with a preview) styles. Fresh installs default to List; installs upgraded from an older version keep Text so the sidebar looks the way it did before the upgrade. This is separate from the Pinned section's style.-- **Session info** — choose what shows at the right edge of each row: **Time** (default), **PR status** (a status icon plus a monospace `#number`; only the icon is colored — green open / gray draft / red closed / accent merged; the open green follows the row surface so it stays readable on both the rest background and the inverse selected pill; a small orange dot on the icon means the PR still has unresolved review threads; unknown state shows a gray icon; sessions without a PR show nothing), **Tokens** (compact total like `1.4M`), and **Cost** (`$`/`¥` by currency). Multi-select; items render in the order you selected them.+- **Group by** — independent checkboxes:+ - **Project** (default on): sessions with a project fold into project rows; project rows and stray chats share one timeline. Turn it off for a flat list.+ - **Chat** (default on): collects projectless sessions into one collapsible **Chat** group.+ - **Device** (default on): appears when you have a remote device. Splits the list into device sections (this computer first), each with a collapsible header showing the device name and online state.+- **Session sort** — **Priority** (waiting on you first, then unread, then running, then by time; a project row ranks by its most urgent session), **Recent Activity** (default, most recent first), or **Creation Time** (newest first).+- **Project sorting** (only while grouping by project) — **Follow sessions** (default) or **Manual**: drag project rows to reorder them; sessions inside still follow **Session sort**.+- **Session status** — **Active** (default) / **Archived** / **All**. See above.+- **Filters** — **Projects** (multi-select, plus **Chat** for projectless sessions; selecting specific projects hides chats unless **Chat** is also checked), **Harness** (All / Claude Code / Codex / Pi), and **Last activity** (1d / 3d / 7d / 30d / All). **Reset filters** clears these three; it doesn't change **Session status**. The **Filters** row shows how many are active.+- **Display** — **Text** (compact single row) or **List** (two-line row with a preview). Fresh installs default to List; upgraded installs keep Text. This is the same setting as **List Style** in **Settings > Appearance**.+- **Session info** — what appears at the right edge of each row (multi-select, shown in the order you picked): **Time** (default), **PR status** (a status icon plus `#number`; a small orange dot means unresolved review threads), **worktree** (the worktree's short name; click it to open the folder), **Tokens**, **Cost**.+- Filters and **Session status** don't apply to the **Pinned** section; it only follows the device scope.**Pinning:**-- Right-click a session and pick **Pin** to keep it at the top. There's no cap on the number of pinned sessions; a newly pinned session goes to the **front** of the pin order (most recent pin first).-- Right-click a pinned session and pick **Unpin** to remove the pin.-- Open a project's overflow menu and pick **Pin project** to move the whole project into **Pinned**. Project pins are independent from session pins, remain available after a restart or project rename, and can be expanded or collapsed inside **Pinned**.-- Pinned projects and sessions share one draggable order. Newly pinned items move to the front, while the other pinned items keep their relative order. Use **Unpin project** to return a project to its position under the current project sorting mode.-- The **Pinned** section keeps its own display style: **Text**, **List**, and **Card** — Card mode is exclusive to Pinned.+- Right-click a session and choose **Pin** / **Unpin**. There's no limit; a new pin goes to the front.+- In a project's **⋮** menu, choose **Pin project** / **Unpin project** to pin the whole project. Pinned projects and sessions share one draggable order.+- The **Pinned** header has its own view-style button: **Text**, **Cards**, or **List** (Cards is only available in Pinned).**Collapse / expand all groups:**-- The fold button next to the display-settings button collapses or expands every group at once. With both device and project grouping on, it cycles through levels: collapse project groups → collapse device groups → expand all. The tooltip always names the next step.+- The fold button in the list header (shown on hover) collapses or expands groups. With both device and project grouping on, it steps through: collapse project groups → collapse device groups → expand all. The tooltip names the next step.**Searching:**-- The search box matches against the **session title only** — not message content and not the working directory path. If you can't remember a session's title, browse via project / agent filters instead.+- Click **Search** at the top of the sidebar and type. It matches session **titles and message content** (keyword and semantic matches), shows matching snippets, and jumps to the matching message when you click a result.+- The sliders button in the search box opens **Search criteria**: **Sort by** (Relevance / Recently active / Oldest active), **Status** (Active / Archived / **All**, default All), **Harness** (All / Claude Code / Codex / Pi), **Last activity**, and **Projects**. A dot on the button means a filter is narrowing results. Filters other than sort reset when you close search.+- To search inside one project, choose **Search sessions in this project** from the project's **⋮** menu; the Chat group's menu has **Search This Chat Group**.-**Right-click actions:**+**Right-click / ⋮ menu on a session:**-- On a regular session: **Pin / Unpin**, **Rename**, **Move to project** (submenu), **Copy session link** (a `cindy://session/<id>` deep link), **Open in new window**, **Archive**, **Delete**.-- On an archived session: **Rename**, **Unarchive**, **Copy session link**, **Delete**.+- Regular session: **Pin / Unpin**, **Rename**, **Tags**, **Copy session link** (a `cindy://session/<id>` link), **Share Task**, **Move task** (to another project, to Chat, or copy to another computer), **Export session…**, **Open in new window**, **Archived**, **Delete**. Some items are hidden for remote sessions.+- Archived session: **Rename**, **Tags**, **Copy session link**, **Export session…**, **Unarchive**, **Delete**.+- **Tags:** pick up to seven favorite colored tags directly in the menu, or choose **More tags** to create, edit, or reorder tags. Tags show as colored dots after the session title.+- Double-click a session title to rename it.-**Removing a project from the sidebar:**+**Project menu (⋮ or right-click on a project row):**-- Open a local project's overflow menu and choose **Remove Project from Sidebar**. Cindy asks for confirmation, removes the project from the sidebar, and keeps its existing sessions available as projectless chats in the main list. Individually pinned sessions remain in **Pinned**. Those sessions are not archived or stopped, and files on your computer are not deleted.-- To restore the project, choose **Add Project** and select the same directory again. Cindy restores the existing sessions under their original project grouping instead of creating an empty session.+- **Rename project**, **Pin project**, **Search sessions in this project**, **Show Files**, **Open in file manager**, **Copy Project Link**, **Sync Codex sessions for this project**, **Remove Project from Sidebar**, **Archive all** / **Restore all**. Remote projects show a shorter menu. The **+** button on a project row starts a new session in that directory.-**Session statuses:**+**Removing a project from the sidebar:**-- Sessions are **active**, **archived**, or **deleted** — there are no other states. Archiving hides a session from the default view without removing it; delete is unrecoverable from the UI.+- Choose **Remove Project from Sidebar** on a local project. Cindy asks for confirmation, then moves its sessions to the **Chat** section. They aren't archived or stopped, pinned sessions stay in **Pinned**, and no files are deleted.+- To bring the project back, start a new session and pick the same folder in the project picker (**Choose another project folder** if it's not listed). Cindy restores the existing sessions under that project.**Notes:**-- Click the sidebar collapse arrow to shrink it to an icon-only strip — that's purely visual, you don't lose any features.-- **Pin state, order, hidden projects, project filters, and identity-based collapse/selection state are isolated per Cindy account.** Pin and hidden-project changes sync across open windows for the same account. Display-only preferences (list style, grouping toggles, sorting, session info) remain local to each window, so they don't sync between, say, a dev window and the installed app.+- Click the sidebar collapse arrow to shrink it to an icon-only strip; nothing is lost.+- Pins, pin order, hidden projects, project filters, and collapse state are kept separately for each Cindy account. Display preferences (list style, grouping, sorting, session info) are local to each installation.diff --git a/apps/desktop/help-knowledge/skills.md b/apps/desktop/help-knowledge/skills.mdindex e3a1ad889c..73b6a72bee 100644--- a/apps/desktop/help-knowledge/skills.md+++ b/apps/desktop/help-knowledge/skills.md@@ -1,9 +1,11 @@---id: skillstitle: Skills (reusable agent capabilities)-summary: Create, search, browse, install, disable, uninstall, publish, and update agent Skills; Skills appear as "/" slash commands and are shared across Claude Code, Codex, and Pi.+summary: Where the Skills page is; create, import, search, install, edit, disable, uninstall, publish, update, and manage visibility of agent Skills; Skills appear as "/" slash commands and are shared across Claude Code, Codex, and Pi.----Skills are reusable agent capabilities you package as a folder and load into your sessions. They're managed on the same page as Plugins — **Skills** and **Plugins** are two tabs of one management surface (the Skills tab is the in-app browser for finding, installing, publishing, and updating skills).+Skills are reusable agent capabilities you package as a folder and load into your sessions. They're managed on the same page as Plugins: click **Plugins** in the sidebar (or **More → Plugins** if you've hidden it), then the **Skills** tab (the same page is also under **Settings > Plugins > Skills**). The Skills tab lists **Local Skills** on this computer and the **SkillHub** catalog (filters **Public**, **Organization**, **Managed**) for finding, installing, publishing, and updating skills.++- When another computer is linked, use the **Manage computer** picker at the top of the page to view, install, update, or uninstall skills on that computer instead.**Skills work across local agents:**@@ -13,76 +15,74 @@ Skills are reusable agent capabilities you package as a folder and load into you- User-created or imported global skills (available to every session): the shared root `~/.agents/skills/<name>/`, which is cross-linked to `~/.claude/skills/` and `~/.codex/skills/` so every engine sees the same skills. On Windows that's under `C:\Users\<you>\`.- Project-scoped (only inside one working directory): `<working-dir>/.agents/skills/<name>/` or `<working-dir>/.claude/skills/<name>/`.-- Cindy’s preinstalled skills and plugin-provided skills are kept in Cindy-managed directories and loaded into local Claude Code, Codex, and Pi tasks. Cindy does not add them to the user’s shared skill directories, so external CLIs do not discover them automatically. User-created and imported skills keep the locations above.+- Cindy’s preinstalled skills and plugin-provided skills are kept in Cindy-managed directories and loaded into local Claude Code, Codex, and Pi sessions. Cindy does not add them to the user’s shared skill directories, so external CLIs do not discover them automatically. User-created and imported skills keep the locations above.- Each skill is its own folder with a required `SKILL.md` at the root (the prompt / spec the agent reads). Sibling files and subfolders in that folder are also visible to the agent.-- Global and project discovery roots may also contain one namespace/container level (for example an author folder): Cindy checks each discovery root's direct children, then their direct children, and stops there. A folder with `SKILL.md` is a Skill; its child folders are skill resources, not separate Skills.+- Skills can also be grouped one folder deep (for example inside an author folder); deeper nesting isn't scanned. A folder with `SKILL.md` is a Skill; its subfolders are that Skill's resources, not separate Skills.**Importing a local skill:**- On the Skills page, use **Import skill** (top-right) to pick a `.zip` package or a standalone `SKILL.md` file.-- A zip must contain a `SKILL.md` (at the package root, or inside a single top-level folder). The YAML frontmatter must include non-empty `name` and `description` fields; `name` must match `^[a-z0-9-]{1,200}$`.+- A zip must contain a `SKILL.md` (at the package root, or inside a single top-level folder). The YAML frontmatter must include non-empty `name` and `description` fields; `name` may contain only lowercase letters, digits, and hyphens.- You then choose where to install: global, a known project, or another directory. Cindy extracts metadata from the file automatically and lists the skill with that name and description.- Imported skills can be uninstalled from the detail page, and you can publish them to SkillHub later if you want.**Using an installed skill:**-- Hover over a local skill in the `/` suggestions and choose **View skill details** in its information panel to open that exact skill. This does not insert or send the command. Pi package skills remain usable from `/`, but do not show this local detail shortcut; manage them through their package.--- Type `/` in the composer to open the slash-command palette; your installed skills show up there alongside built-in and agent commands. Pick one to run it.+- Type `/` in the composer to open the slash-command palette; your installed skills show up there alongside built-in and agent commands (see the Slash commands topic). Pick one to run it.+- Hover over a local skill in the `/` suggestions and choose **View skill details** in its information panel to open that exact skill. This does not insert or send the command. In a new-session draft this works for global skills; project skill details become available once the session has been created. Pi package skills remain usable from `/`, but don't show this shortcut; manage them through their package.**Creating or learning a Skill from the composer:**-- To create a Cindy **plugin** (`.cindy` with `ghost.json`), choose **Create with Cindy** from the Plugins page's add menu. This selects the separate built-in `cindy-plugin-creator` Skill, including when a user Skill has the same name. Its references cover plugin APIs and packaging; see the Plugins topic.-+- To create a Cindy **plugin** (a `.cindy` package) rather than a Skill, choose **Add plugin > Create with Cindy** on the Plugins page; it uses the separate built-in plugin creator (see the Plugins topic).- Cindy includes `/cindy-skill-creator`. Invoke it with a description of the Skill you want, such as `/cindy-skill-creator Create a Skill that checks release notes`. Invoking it without a description asks what you want to create.- The current Agent creates or updates the Skill directly with its normal file tools. It does not use the `/learn` staging flow.- Ask for a Skill in the composer, for example “Find a release-notes Skill” or `/cindy-skill-creator Search our organization's code-review Skills`. The Agent can search SkillHub's public marketplace or your organization's catalog. Without a specified scope, organization identities search both; personal identities search the public marketplace. Results identify their source, including when two Skills share a name. Search alone does not install anything.- You can also ask `/cindy-skill-creator` to upload your own Skill to SkillHub or publish a new version directly in the conversation. Give the local Skill folder (or the Skill just created) and, for a first upload, who should be able to access it. Ordinary local creation/editing does not automatically publish it.-- Uploads use Cindy's signed-in account, which determines ownership; choosing sharing targets does not change that ownership. The Agent can list your published Skills, upload the folder, and check scan/review status. Upload success does not mean public approval. Updating requires confirmed original authorship and management access, and preserves visibility and ownership; a missing authorship flag is not treated as permission. This works in a local desktop task, including when controlled from a phone; SSH files must first be made available on the Cindy host.-- Each Agent upload or version update follows the task's permission mode: Ask requests approval for that publication, Auto reviews it against your request, and Full Access bypasses operation approval. A previous helper-tool approval does not authorize later publications. Search, listing, and status queries remain automatic.+- Uploads use Cindy's signed-in account, which determines ownership; choosing sharing targets does not change that ownership. The Agent can list your published Skills, upload the folder, and check scan/review status. Upload success does not mean public approval. Updating requires that you're the original author and can manage the Skill, and preserves visibility and ownership. This works in a local desktop session, including when controlled from your phone; Skills on an SSH host must first be copied to this computer.+- Each upload or version update follows the session's permission mode: Ask requests approval for that publication, Auto reviews it against your request, and Full Access doesn't ask. Each publication is approved on its own. Search, listing, and status checks don't need approval.- Unless you request another location, new Skills go to `~/.agents/skills/<name>/`. Ask for a project-specific Skill to create it under `<working-dir>/.agents/skills/<name>/`.-- Cindy also includes `/learn`. Bare `/learn` distills the current task; `/learn <description>` learns from a described workflow; `/learn hub:<scope>:<slug> [instructions]` learns from a SkillHub Skill. Cindy gathers evidence, runs the distillation in a separate task, and shows a diff for review before saving the proposed Skill.-- The built-in `cindy-skill-creator` and `learn` entries appear in the local Skill list with an Official badge. You can disable or enable either entry there; because they ship with Cindy, they cannot be uninstalled. The setting applies to new or restarted Agent sessions. Disabling `learn` also disables the **Learn this skill** action in SkillHub after the local Skill list has refreshed.+- Cindy also includes `/learn`. Bare `/learn` distills the current session; `/learn <description>` learns from a described workflow; `/learn hub:<scope>:<slug> [instructions]` learns from a SkillHub Skill. Cindy gathers evidence, runs the distillation in a separate session, and shows a diff for review before saving the proposed Skill (see also the Slash commands topic).+- The built-in `cindy-skill-creator` and `learn` entries appear in the local Skill list with an Official badge. You can disable or enable either entry there; because they ship with Cindy, they cannot be uninstalled. The setting applies to new or restarted sessions. Disabling `learn` also disables the **Learn this skill** action in SkillHub after the local Skill list has refreshed.**Disabling or uninstalling a local skill (Desktop):**-- Open a skill's detail page to use the switch in the top action bar. For a market skill installed in multiple locations, select the local copy from the location dropdown beside the switch. Turning it off keeps the files and removes the skill from new-task command previews. Active tasks retain the skill preference snapshot applied when their Agent started, including their command palette. New or restarted Claude Code, Codex, and Pi agents apply that choice; an already-running agent keeps its existing context.+- Open a skill's detail page to use the switch in the top action bar. For a market skill installed in multiple locations, select the local copy from the location dropdown beside the switch. Turning it off keeps the files and removes the skill from the `/` palette of new sessions. Running sessions keep the skills (and command palette) they started with; new or restarted Claude Code, Codex, and Pi sessions apply the change.- Renaming a local skill while publishing keeps its enabled or disabled setting. If renaming fails, Cindy restores the original folder and content.- For a Skill inside a namespace/container folder, renaming changes only the Skill's own folder name; the namespace stays the same.- The switch is local to this device and Cindy profile. It does not edit external CLI settings, sync to other devices, or override a native engine's own disabled state. Enable it again to let the engine discover it normally.-- In the skill's details, use **… → Uninstall skill**. Confirm the location and shared-file impact. Standalone skills, including locally written skills without a market installation record, move to the system trash. Recovery is through the operating system's trash.-- External source imports remove only their discovery links and keep the external source files. Package-owned Skills must be removed through their owning package. Cindy's built-in Skills cannot be uninstalled, but they can be disabled. Skills provided by Cindy plugins are managed on the owning plugin's page and loaded through Cindy's private skill entries.+- In the skill's details, use the **⋯** button next to the switch → **Uninstall Skill**. Confirm the location and shared-file impact. Standalone skills, including locally written skills without a market installation record, move to the system trash. Recovery is through the operating system's trash.+- External source imports remove only their discovery links and keep the external source files. Package-owned Skills must be removed through their owning package. Cindy's built-in Skills cannot be uninstalled, but they can be disabled. Skills provided by Cindy plugins are managed on the owning plugin's page.- Offline removal of an automatically synced skill is remembered on this device, including after restart or sign-in. Automatic sync skips it until you explicitly install it again.-- Uninstalling a shared copy affects external CLIs that use it. It does not unpublish a skill or delete other users' copies. If file removal fails, Cindy keeps the installation. If cleanup is interrupted, the notice offers **Retry cleanup**. The unfinished operation survives closing, reloading, and restarting Cindy; reopen Skills in the same Cindy profile and account to continue. Cindy pauses conflicting installations until cleanup completes. Retrying never moves files to the trash a second time and preserves externally restored or replaced content.+- Uninstalling a shared copy affects external CLIs that use it. It does not remove the published skill or delete other users' copies. If file removal fails, Cindy keeps the installation. If cleanup is interrupted, the notice offers **Retry cleanup**, even after restarting Cindy (reopen Skills with the same account); installs that would conflict wait until cleanup finishes.++**Viewing and editing a local skill:**++- Click a skill under **Local Skills** to open its detail page: its files and frontmatter, an **Edit** / **Save** action for editing files in place, and a **Usage** panel (last 30 days of local use, with **Diagnose this Skill**, which creates a new session with usage stats filled into the composer; it doesn't send automatically).**Publishing your own skill:**-- Find the publish action on the Skills page and point it at the skill's local folder. It zips the folder and uploads it — reading your directory in place, without copying or moving anything.-- On first publish, choose visibility: personal identities support public or private; organization identities support public or organization sharing. Public publication is subject to review.-- The local registry records what you published, so the app knows it's "yours" for future updates.+- On a local skill's detail page, click **Publish to Market**. Cindy packs the skill folder in place (nothing is copied or moved) and uploads it.+- On first publish you set the display name, description, version, optional tags, and **Visibility**: personal identities choose **Public** or **Only me**; organization identities choose **Public** or **Team** (and can publish as a team). Public publication goes through review; the detail page shows the review status.+- Cindy records what you published, so it knows the skill is yours for future updates.**Updating an already-published skill:**-- On your own skill's detail page you'll see a **发布新版本** (Publish New Version) button.-- Edit the local folder however you want and publish — the version is **auto-incremented server-side**, you don't pick a number.-- **Old versions stay live alongside the new one.** Users who already installed an older version keep it until they choose to update.-- The display name and description are sent on every republish, so editing those metadata fields just means doing a republish.+- On your skill's detail page, click **Publish new version**. The **Version** field is pre-filled with the next patch version (format `x.y.z`); you can change it. Describing what changed is required.+- Publishing a new version doesn't change visibility. Users who installed an older version keep it until they update (**Get update** on the market card, or **Update to v…** on the detail page).**Managing an already-published skill:**-- Change who can see or use your skill from the management menu — public, shared with selected teams / departments, or private to you.-- **Unpublish** makes it private again and returns ownership to your personal scope; it does not remove copies other users already installed.-- **Authorship is fixed.** Only the original author's account can publish new versions of a given skill (enforced server-side as `NOT_AUTHOR`).+- In the SkillHub catalog (the **Managed** filter lists skills you can manage), open the skill's **Manage** menu: **Edit info** (display name and description), **Manage visibility** (**Public** / **Team** / **Only me**; choosing Public submits the current version for review, Only me removes it from the market), **Clone**, and **Delete**.+- **Delete** removes the cloud release from the market and team access and can't be undone. Copies other users already installed, and your local files, are not affected.+- Only the original author — or, for team-published skills, the team's admins and publishers — can publish new versions.**Browsing and installing others' skills:**-- The marketplace lets you install skills into your global skills directory. You can request a specific version on install; the local registry tracks which version you have.+- Search the SkillHub catalog, then click **Clone** on a skill and choose an install location: **Global**, one of your **Projects**, or **Other directory**. If a skill with the same name already exists there, Cindy asks before overwriting and keeps a backup of the old copy.**Notes:**- A skill folder without a `SKILL.md` at its root won't be picked up (lowercase `skill.md` is also accepted).- Project-scoped skills only show up in sessions whose working directory matches — useful for skills tied to a particular repo's conventions.- Editing files inside an installed skill folder takes effect on the next session start; you don't need to reinstall.-- Uninstalling only removes your local copy. For your own published skills, use unpublish or manage visibility to change market availability.--In a new-task draft, the slash hover panel offers detail links for global skills. Project skill detail links are available after creating a task, when SkillHub can include that project in its scan.+- Uninstalling only removes your local copy. For your own published skills, use **Manage visibility** or **Delete** to change market availability.diff --git a/apps/desktop/help-knowledge/storage.md b/apps/desktop/help-knowledge/storage.mdnew file mode 100644index 0000000000..c99c2d1ffa--- /dev/null+++ b/apps/desktop/help-knowledge/storage.md@@ -0,0 +1,44 @@+---+id: storage+title: Storage and cleanup+summary: Settings > Storage shows how much disk Cindy uses and lets you set the Chat workspace folder, clear media caches, run Disk cleanup, clean the database (permanently removes old archived history), and handle Worktree cleanup.+---+**Settings > Storage** is where you see and manage what Cindy keeps on this computer. Your database and media cache stay local, and Cindy never deletes your history automatically — every cleanup here is something you start and confirm.++**Chat workspace:**++- The folder used by new sessions that have no project folder. Cindy creates an account-specific subfolder inside it.+- **Open directory**, **Change directory**, **Restore default location**, and **Full path** (with **Copy path**).+- Changing it only affects new sessions — existing files stay where they are.++**Local storage overview:**++- Shows database size, media files, the image cache, and the message attachment cache. **Refresh storage usage** recalculates.++**Media files:**++- **Image cache directory** and **Message attachment cache directory** — **Open directory** to look inside, or **Clear image cache** / **Clear attachment cache**. Clearing deletes only those cache files, but images or attachments in older messages may stop showing, and it can't be undone.+- **Remote image cache** — images downloaded for reuse. **Clear remote cache** frees the space; images download again when opened, and original files and saved attachments are kept.++**Database:**++- **Database cleanup** — click **Scan database** to see what would be removed. It targets the message history of **soft-deleted and archived** tasks not updated for longer than the period you pick (**7 days** by default, or 1, 3, or 6 months).+ - **Clean Unarchived Tasks** (off by default) also includes old unarchived tasks: their history is removed, their agent context is reset, and they leave the task list (teammates' main tasks stay). Cindy asks you to confirm, because such a task could still be running.+ - **Back up database before cleanup** is on by default (the default backup lives in the database folder and only the latest cleanup backup is kept; **Choose backup folder** to put it elsewhere).+ - **Confirm database cleanup** → **Restart and clean up database**: Cindy restarts, backs up (if enabled), removes the history, compacts, and verifies the database before opening it. If a step fails, Cindy keeps or restores the original database.+ - **This permanently removes that history.** With a backup you can restore task and message data from it (**Open cleanup backup**); without one, it can't be recovered. Records used to review or undo file changes from those messages aren't in the backup. Files and generated media on disk are not deleted.+- **Database reminder threshold** — Cindy reminds you to clean up when the database grows beyond this size (10 G by default, 1–1024). **Do not remind me again** hides these reminders.++**Disk cleanup and storage check** (below the database card):++- **Disk cleanup** — **Scan** finds media no session references anymore, caches over their limit, obsolete leftover folders, and stray temporary files, and shows the totals first. Nothing is deleted until you click **Clean up**. Images in composer drafts and queued messages are always protected.+- **Storage check** — **Check** compares Cindy's media records with the files on disk and reports any mismatch. It never deletes anything.++**Worktree cleanup:**++- Lists worktrees Cindy is still waiting to clean up (for example, still in use or with changed contents), with their state and reason. Use **Retry cleanup**, **Keep directory** to keep it, or **Refresh**. See the worktrees topic for how worktrees are created and removed.++**Notes:**++- Archiving or deleting a session in the sidebar keeps its history in the database; Database cleanup is what actually frees that space (see the sidebar topic).+- Logs live elsewhere: **Settings > About** (see the updates-and-about topic).diff --git a/apps/desktop/help-knowledge/teammates.md b/apps/desktop/help-knowledge/teammates.mdnew file mode 100644index 0000000000..c30705a35f--- /dev/null+++ b/apps/desktop/help-knowledge/teammates.md@@ -0,0 +1,103 @@+---+id: teammates+title: Teammates (Cindy and custom companions)+summary: Teammates are long-lived AI companions (the built-in Cindy plus ones you create) listed at the top of the left sidebar. Chat with them, set their profile, model, memory, capabilities, projects, voice, messaging platforms, and routines; let them run background tasks, follow up on your sessions, and keep todos; put several in a group chat; choose which computer runs them.+---+A **teammate** is a long-lived AI companion with its own name, avatar, personality, memory, skills, and one continuous conversation. **Cindy** is built in and is added automatically the first time you use teammates; you can create more. A teammate belongs to your account, not to one computer: you see the same teammate on every computer and on your phone, but it runs on one **main computer**.++**Where teammates are:**++- The **Teammates** section sits at the top of the left sidebar, above **Pinned** and your task list. It mixes private chats and group chats, newest activity first; pinned ones stay on top. The first 5 show, then **Show all N**.+- Click **Teammates** (the section title) or its arrow to collapse the section into one row of avatars (overflow goes into **+N**). An unread dot marks new replies.+- Hover the section header for the **+** menu: **Discover Teammates**, **New teammate**, **New Group Chat**, and **Enter Invite Link…**. If you have hidden teammates, an eye button there shows or collapses them.+- Right-click a teammate row: **Pin to top** / **Unpin**, **Hide**, **Duplicate teammate** (not for Cindy), **Delete permanently**. Hidden teammates move to **Hidden (N)** at the bottom of the section; click the eye on a row to unhide it.+- The sidebar **Search** also finds teammates (name, introduction, skills) and group chats (group name, member names); they're listed before session results.+- Teammates running on your other computers also show up here, under their main computer.+- When the sidebar is collapsed to icons, the teammates icon (with an unread count) opens the teammate you last viewed.++**Chatting with a teammate:**++- Click a teammate to open its chat and type in the box ("Say something to …"). The conversation is one continuous timeline; it isn't replaced each day. Long histories are compacted automatically, and the teammate keeps the full record and its memory.+- The **permission** chip under the message box works like in a normal task (**Default permissions**, **Auto-review**, **Full access**, and for Claude Code also **Auto accept edits**). There's no model picker in the teammate's message box; set the model in its settings.+- Hover a message for **Reply** (quotes it above the message box) and **Start a thread**. A thread opens in a side panel with its own separate conversation. Work started in a thread reports back in that thread. Threads are desktop-only for now.+- The chat header has the teammate's name/avatar and a gear (both open its settings), a grid button (**… at a glance**, see below), and for teammates on this computer **Add … to a Group Chat** (**New Group Chat…** or **Add to Existing Group**).+- In a normal task you can also @-mention a teammate ("Message or delegate") to send it a message or hand it work.++**Creating a teammate:**++- **+ → New teammate** opens **Create a teammate**. Pick a portrait (or **Upload image**), enter a **Name** (names must be distinct), and check the **Model**. It follows the default teammate model unless you change it ("This selection applies only to this teammate"). If no model is connected, connect a provider first. Click **Create teammate**.+- Cindy then prepares the teammate ("Getting ready to meet"), and it sends a first hello. You can start typing right away. If preparation pauses, everything done so far is saved; use **Change Model** or retry.+- **Import an Agent** (in the same dialog) brings in a teammate from Hermes or OpenClaw on this computer. You can bring its personality, memories, skills, environment and connections, automations, and messaging bots.+- **Discover Teammates** lists ready-made teammates with their skills and routines; click **Invite** to add one. Their routines start turned off.+- You can also ask Cindy to create a teammate for you.++**Teammate settings** (click the name/avatar or gear in the chat header; changes save automatically):++- **Profile** — avatar (**Change avatar**; PNG, JPEG, or WebP under 5 MB), name ("What Should We Call Them?"), and **Self-introduction**.+- **Personality & conversation** — free text describing who the teammate is and how it talks.+- **Memory** — **Use memory** toggle (off: memories aren't read or saved), **What you know about me**, and the teammate's saved memories grouped as About you / Your preferences / Projects / References. Search, edit, or delete them. Each teammate's memory is its own, separate from Settings > Personalization > Memory.+- **Linked Projects** — projects the teammate may read for background when needed (**Choose a project**, **Add**). Linking doesn't assign the project's sessions or start background work.+- **Model** — **Primary Model** with up to 5 candidates (**Add fallback model**; **Backup models (N)**). If one is unavailable, it automatically continues with the next. **Task Model** is the model for new background tasks; by default it uses **Inherit Primary Model**. **Restore default model** goes back to the shared default.+- **Voice** — speech model, voice, speech speed, and **Read by sentence** for voice conversations (see below).+- **Capabilities** — choose which **Built-in tools**, **MCP** servers, and **Skills** this teammate may use (items unavailable on this computer are marked). By default it follows the tools enabled on this computer; turning one off switches to an explicit list. **Input Handoff** lets it ask you to type a password or code on any of your devices; it never sees the value.+- **Messaging platforms** — give the teammate its own bot (see below).+- **Routines** — follow-ups, built-in routines, nightly learning, and your own routines (see below).+- **Search teammate history** — search past messages, background tasks, and archived records.+- **Advanced configuration** — **Open folder** opens the teammate's own folder (SOUL.md, USER.md, system_prompt.md, skills/, memories/). Use it only to fix something by hand.+- Also on the settings home: **Main Computer** (see below), **Restart teammate** (stops the current run for a fresh start; history, memory, and files are kept; send a message to continue), and **Permanently delete teammate**. Deleting removes its profile, memory, working directory, channel connections, and automated tasks. Task history and separate worktrees are kept. A paused teammate shows **Resume teammate**.++**What a teammate can do for you:**++- **Background tasks:** for bigger work, a teammate starts a normal Cindy task. The task shows in your task list as "…'s background task", and a **Background task** card in the chat lets you **Open task** or **Stop task**. The result comes back to the teammate's chat.+- **At a glance** (grid button in the chat header) has up to three blocks:+ - **Computers**: where it runs and other computers it can move to.+ - **Todos**: open todos, due ones first. Click one to quote it in the message box and tell the teammate what to change. Right-click → **Delete**. **Check for me** asks it to look through your connected mail and chats for new todos; **Check every day** does that on weekdays at 9:00 and stays quiet when nothing is new.+ - **Outputs**: finished files from its tasks.+- **Follow Up on My Sessions** (Settings → **Routines**): when on, the teammate reads summaries of sessions you create on this computer, assesses progress, and tells you about important changes. This uses your model allowance. It's off by default, and only one teammate can hold it; Cindy may offer to turn it on in chat (**Start Following Up** / **Not Now**). Tasks you hand to a teammate are followed by that teammate either way (built-in **Follow assigned sessions**). It reacts when they stall, need you, fail, reach a usage limit, or finish. Sessions a teammate is looking after show its avatar in the sidebar ("This task is looked after by …").+- **Routines:** **Add Routine** gives the teammate standing instructions plus triggers (Hourly, Daily, Weekdays, Weekly, Monthly, Interval, Cron, or plugin events; any trigger can start a run). Each routine has **Run Now**, **Run History**, and an optional **Stay quiet when checks find nothing**. Built-in routines can only be paused or resumed. Routines run on the teammate's main computer and are managed on the desktop.+- **Nightly learning** (built-in, on by default): every night at 3:30 (or weekly), it reviews new conversations, tidies memories, and saves approaches that keep coming up as skills. When there's nothing new, it skips without calling a model. Open it under **Routines** for **Review now**, **Frequency**, **History** (with **Undo** per change), and **Approaches being watched** (**Make it a skill** / **Don't learn**). It runs only on the main computer; with memory off, only skills are reviewed.+- Teammates can message each other; those exchanges appear as cards in the chat.++**How permissions work (high level):**++- When you ask in its chat, a teammate acts on your behalf under the chat's permission mode. **Full access** doesn't ask, **Auto-review** reviews automatically, and **Default permissions** asks you. No extra confirmation cards are added.+- When it acts on its own (a follow-up wake-up, a message from another teammate or a stranger), it can only touch its own background tasks and tasks already handed to it. It can keep those going toward the original goal or switch their model. Anything else on your sessions (new requirements, stopping, renaming, archiving, creating or changing routines) becomes a todo or a note to you. Reply to it and the teammate does it on your behalf.+- Routines you set up run as your standing instructions.++**Messaging platforms (the teammate's own bot):**++- On **Messaging platforms**, click **Add platform** and paste the credentials of a bot you created on **Telegram**, **Feishu** / **Lark**, **WeCom**, **DingTalk**, or **Discord**. Credentials stay on this computer. If the same bot is already set up in Settings > IM bots, choose **Use the bot already set up in Cindy** to move it over; Settings > IM bots is then left empty. A bot can be connected in only one place.+- You can also link this computer's **Personal WeChat** to one teammate at a time. Where available, you can assign the official X bot (@askmycindy) to it; X replies are public.+- Anyone who messages the bot is talking to the teammate. Your own messages go into its main conversation with its permissions. Each group, topic, and DM from someone else gets its own hidden task with restricted permissions; risky actions there need your confirmation. If other people can reach the bot, prefer **Auto-review** or **Default permissions** over **Full access**.+- Each connection lists its **… tasks**; **Start over** gives one a clean context while keeping history. For Telegram groups, choose **@ only** or **Always** (reads every message and decides whether to chime in; each one costs tokens). **Remove** stops the bot and deletes its credentials. Memory is shared across all of a teammate's channels.++**Voice conversations (desktop):**++- Set up **Voice** in the teammate's settings (needs a text-to-speech model from a connected provider). Then click **Start voice conversation** in the message box. Speak; it sends after a pause and reads replies aloud. Speak again to interrupt; **Mute** and **End** are on the call bar.+- Work you ask for during a call is handed to the teammate's main conversation ("Work from this call"). A record of the call is saved as a card. Voice mode works for teammates on this computer, one-on-one only; it's not available on the phone.++**Group chats:**++- **+ → New Group Chat**: give it a name and choose teammates (up to 6). Group chats live in the same Teammates section.+- **@** a teammate to have only them reply, or **@Everyone**. **Group Settings** has **When No One Is @Mentioned** (**Anyone Can Reply** / **Only @Mentioned Reply**). A round runs up to 3 passes and 10 messages, then waits for **Continue Discussion**.+- For real work, the **Organizer** can split it into steps. Review the plan card, then **Start Plan** or **Skip Plan**. Each step waits for **Start Next Step** unless **Step handoffs** is on. Set a **Project Folder** for the group; with a git project, each work split gets its own worktree and branch.+- Invite people with **Invite to group** / **Invitation link**; they can bring their own companions (**Invite my companion**). **Access for other members** controls what others may use from your companion: **Owner only**, **Group conversation only**, or **Owner authorized tools and information**.+- Group messages are separate from the teammate's private chat; memories it saves in a group are still its own. Groups can be archived, left (**Leave group**), or dissolved by the owner (**Dissolve group**).++**Main computer and other devices:**++- The computer where you created a teammate is its **Main Computer**. Its working folder, logins, model connections, and routines live there. Other computers and your phone see the same teammate and can chat with it through remote control. This needs "Allow same-account devices to control this machine" on in Settings → Remote Connection on the main computer.+- To move it, use **Main Computer** in its settings or **Computers** in **at a glance**, then **Move to "…"**. Chat history, memory, skills, todos, and routine definitions come along. Working directories, connections, and local files stay where they are. On the new computer it uses that computer's default permissions, and if its model isn't available there, it follows that computer's default model.+- If the main computer is offline, other computers show the cloud copy of the conversation (read only) and offer **Continue on This Computer**. Only visible messages and final replies are saved in the cloud; tool details stay on the main computer.+- If you had a separate Cindy on several computers, Cindy's settings show **Merge Duplicate Cindys**. It merges them into one and keeps everyone's history, memories, skills, todos, and routines.++**On your phone:**++- The phone's home screen starts with the same **Teammates** section (tap **+** to create a teammate or group chat). Creating a teammate needs a computer that's online and signed in to the same account.+- In a chat, tap the teammate's avatar at the top for **at a glance**. Use the settings button for **Teammate settings**: Profile & Personality, Memory, Skills & Connections, Artifacts, Teammate Messages, Models & Permissions, Main Computer, and Messaging platforms.+- Routines can't be managed on the phone (except **Check every day**). Voice conversations and private threads are desktop-only.++**Settings shared by all teammates** (Settings > General > **Teammates**):++- **Default model**: used by teammates that don't choose their own, independent of new-task model choices.+- **Show which message a reply answers**: shows above a teammate's reply which of your messages it answers. It's off by default.diff --git a/apps/desktop/help-knowledge/tools.md b/apps/desktop/help-knowledge/tools.mdnew file mode 100644index 0000000000..11ed3af197--- /dev/null+++ b/apps/desktop/help-knowledge/tools.md@@ -0,0 +1,46 @@+---+id: tools+title: Tools and automation settings+summary: Settings > Tools turns built-in agent tools (Collab Mode, SSH commands, Document Toolkit, Slack, Feishu Bot, Personal WeChat, Browser) on or off per project and adds custom MCP servers; Settings > Automation controls browser, computer, and Android automation.+---+Two Settings tabs decide which tools the agents get: **Tools** (built-in tools and your own MCP servers) and **Automation** (letting agents operate a browser, this computer, or an Android phone). Tool changes only apply to **new or restarted** sessions — a running session keeps the tools it started with.++**Settings > Tools > Built-in Tools:**++- Each row is a tool the agents can call, with an on/off switch:+ - **Collab Mode** — lets a session start a team of worker agents (see the collaboration topic). Turning it off only affects new sessions and doesn't stop workers that are already running.+ - **SSH commands** — run commands on remote computers through the SSH hosts you've configured (see the remote topic).+ - **Document Toolkit** — create and read back PDF, Word, Excel, and PowerPoint files.+ - **Slack** — search, read, and send messages with your connected Slack account.+ - **Feishu Bot** — send files and notifications to Feishu users through your bound bot.+ - **Personal WeChat** — send messages to known WeChat contacts.+ - **Browser** — browse and act on web pages. It's listed here only when the scope is **New session defaults**; per-project control is under **Settings > Automation**.+- Core tools (memory, scheduling, and the helper that tells agents what's available) are always on and aren't listed. Smart Contacts has its own switch under **Settings > Personalization > Smart Contacts**.+- **Scope picker** (top right): **New session defaults** applies to sessions without a project override; choosing a recent project changes the setting for that project only. Project settings take priority. When you open Settings from a project session, that project is selected first.+- A small label on each row shows where the current value comes from: **project**, **user default**, or **product default** (hover for details). After you change a row, it shows **Custom**; **Restore default** clears your change.++**Settings > Tools > External tools (custom MCP):**++- Click **Add MCP server** to connect your own **remote** MCP server. Enter a **Display name**, the **Transport** (http or sse), the **Endpoint URL**, and optionally a **Bearer token** and extra **Headers**.+- HTTP / Streamable HTTP servers work with both Claude and Codex; **SSE works with Claude only**. Codex only supports Bearer-token authentication, so custom headers apply to Claude.+- A saved server is available to new sessions right away; sessions already running aren't affected. Use the row's edit or delete buttons to change or remove it (deleting stops loading it in new sessions).+- Local (command-line) MCP servers can't be added here — only remote endpoints.++**Settings > Automation:**++- **Browser automation** (on by default) — gives agents a dedicated browser, separate from your everyday one.+ - **Automation browser target**: **Standalone external browser** (the default — a separate Chrome window with its own "Cindy" profile, so it never touches your normal tabs, bookmarks, or sign-ins) or **Sidebar embedded browser** (the agent works in browser tabs in the right sidebar). The two keep separate sign-ins. If the embedded browser loses its connection, use **Reconnect Browser** / **Recover Browser**.+ - With the external browser, click **Open agent browser** to sign in to sites once; the agent keeps that session afterward. If no browser is found, **Download Chrome**.+ - **Use my browser logins** (external browser only) — lets the agent open sites with your signed-in Chrome, Edge, or Brave accounts. Cindy asks you to confirm first; on macOS it also needs **Full Disk Access**. Turning it off deletes the copied logins.+ - This switch applies to the project of the session you came from (or to new-session defaults if there isn't one).+- **Computer automation** (off by default) — lets agents see windows, read the UI, click, and type in local apps. It's powered by the open-source **CUA Driver**, which Cindy installs the first time you turn it on.+ - On macOS it needs **Accessibility** and **Screen Recording** for CuaDriver; Cindy walks you through granting them (**Grant**, **Recheck**). macOS may still ask for extra confirmation for a specific app or action later — approve those only when you've just started a computer-use task.+ - Windows needs no extra permission; keep the target app visible on the active desktop. On Linux, the desktop session must allow window inspection and input automation.+ - **Check for updates** / **Update** keeps the driver current (you may need to re-grant permissions after an update).+- **Android automation** (off by default) — lets agents inspect and operate a connected Android phone over ADB. Cindy prepares ADB for you. To connect: use a data-capable USB cable, enable **Developer options** and **USB debugging** on the phone, keep it unlocked, allow the debugging prompt, then click **Recheck**. Pick a default **Device** (or **Auto**) and, if needed, a custom ADB path.+- Computer and Android automation apply to the whole computer (not per project) and only to sessions started afterward.++**Notes:**++- Plugins add more tools on their own pages under **Settings > Plugins** (see the plugins topic).+- If a tool you expect is missing in a session, check its switch here (and the project scope), then start a new session.diff --git a/apps/desktop/help-knowledge/updates-and-about.md b/apps/desktop/help-knowledge/updates-and-about.mdindex 6ff4a2d5ae..ca747b5b0a 100644--- a/apps/desktop/help-knowledge/updates-and-about.md+++ b/apps/desktop/help-knowledge/updates-and-about.md@@ -1,22 +1,35 @@---id: updates-and-abouttitle: App updates, version and logs-summary: How auto-updates apply; where to find the app and agent versions, the debug-log toggle, and the logs folder.+summary: How app updates download and apply (relaunch banner, idle install, Beta channel); where to find the app and agent versions, agent updates, debug logs, log upload, and the logs folder.tab: about----Cindy checks for updates automatically and downloads them in the background. When an update is ready, you're prompted to **relaunch** the app to apply it.+Cindy checks for updates automatically and downloads them in the background. When an update is ready, a banner appears in the sidebar (**Updated to <version> · Relaunch to apply**) with a **Relaunch** button and **See What's New**.++**Applying an update:**++- Click **Relaunch** to install. If a task or other background work is still running, Cindy warns you first (**A task is still running**) and offers **Restart App Anyway** or **Cancel**.+- Click **Later** to hide the banner. The flame button on your account card at the bottom-left of the sidebar then shows **Update ready — click to review** and brings the banner back. Otherwise, the update is applied the next time you start Cindy.+- When no update is pending, the same flame button opens **View release notes**.+- You can also ask a Cindy agent to update Cindy or to turn idle installation on or off; Cindy always shows a confirmation card first.**Settings > About:**-- **App version** — the desktop app's release.-- **Claude Code version** / **Codex version** — the agent CLIs Cindy manages and downloads at runtime into your user data (they update separately from the app).-- **Auto-relaunch when idle** — an optional toggle that lets a downloaded update apply itself during idle time instead of waiting for you to click relaunch.-- **Debug log toggle** — turn on more verbose logging when reproducing a problem. Leave it off for normal use.-- **Open logs folder** — opens the logs directory in your OS file browser, useful when you need to attach logs to a bug report.-- **Storage** — a storage-management card for reviewing / clearing local app data.+- **App version** — shown here in the China edition. In every edition, the app version also appears under your name on the account card at the bottom-left of the sidebar.+- **Install updates while idle** (off by default) — after an update finishes downloading, Cindy restarts to install it once the computer has been idle for 10 minutes and no tasks are running. On Linux this row only explains how updates work there: system `.deb` installs ask for your system password; other installs use their package manager.+- **Usage statistics** — turn launch and usage reporting on or off (on by default once you accept the Privacy Policy).+- **Claude Code version** / **Codex version** / **Pi version** — the agent runtimes Cindy manages. They update separately from the app: open the row's menu to **Check for updates** or update (Cindy restarts to apply it). For Pi you can also **Update from Pi GitHub** or **Restore Cindy release**.+- **Debug logs** — turn on more verbose logging while reproducing a problem. Leave it off for normal use.+- **Logs directory** → **Open** — opens the logs folder in your file manager.+- **Upload logs automatically on crash** (off by default) and **Upload logs** → **Upload** — sends recent run records to Cindy and gives you an upload code to quote when you report a problem.+- **Terms of Service**, **Privacy Policy**, and **Connect with Cindy** community links.++**Related settings:**++- **Beta channel** — **Settings > General > Experimental**. Gets upcoming updates early; takes effect after a restart. Not shown on Linux.+- **Storage** — its own Settings tab (**Settings > Storage**): media and cache usage and cleanup, database cleanup, the Chat workspace folder, and **Worktree cleanup**.**Notes:**-- By default, a downloaded update only triggers a relaunch prompt — it won't restart you unexpectedly. If you enable **Auto-relaunch when idle**, Cindy will restart itself automatically once the idle and busy gates pass, without an additional prompt.-- If you want to skip auto-update temporarily, you can dismiss the relaunch prompt; the update applies next time you start the app.-- The Claude Code / Codex versions you see here are what Cindy's **managed agents** run as — they're independent from any Claude / Codex CLI you have installed globally on your shell.+- A downloaded update never restarts Cindy unexpectedly unless you turn on **Install updates while idle**.+- The Claude Code / Codex / Pi versions shown here are what Cindy's managed agents run. They are independent of any Claude Code, Codex, or Pi CLI installed on your system.diff --git a/apps/desktop/help-knowledge/usage.md b/apps/desktop/help-knowledge/usage.mdnew file mode 100644index 0000000000..6e4c65c60d--- /dev/null+++ b/apps/desktop/help-knowledge/usage.md@@ -0,0 +1,39 @@+---+id: usage+title: Usage, billing, and Activity Monitor+summary: Where to see how much you've used — Settings > Usage history (token usage by day, agent, model, and session), Settings > Usage and billing (plan, balance, top-ups, orders), the macOS menu bar usage icon, and Activity Monitor (CPU / memory of Cindy and agent processes).+---+Cindy has several places that answer "how much have I used?". Pick the one that matches the question.++**Settings > Usage history — tokens used in Cindy:**++- Shows token usage produced inside Cindy. It's available whenever you're signed in, including local mode. It counts tokens, not money.+- **Device**: **All devices** (the default, combined across your computers) or one device such as **This device**. Other computers are read only while they're online and allow remote control; if one can't be reached, its last-read data is used and the page warns that totals may be low.+- **Time range**: **All time**, **Last 30 days** (default), **Last 7 days**, or **Today**. Use **Date**, or click a day in the heatmap, to look at a single day.+- Cards: **Overview** (tokens, **Active streak**, cache hit rate, **Models used**), **Activity heatmap**, **Daily tokens (30d)** stacked by model, **By Agent / harness**, **By model** (input, output, cache read, cache write, cache hit rate), and **Top sessions by tokens** — click a session there to open it (when it's reachable).+- Recording started when this page shipped; earlier history isn't backfilled.++**Settings > Usage and billing — your Cindy plan and balance:**++- Shown only when you're signed in with a personal Cindy account (not local mode or an organization account).+- **Current plan** — your subscription, renewal or end date, and **View plans** / **Change plan**. You can **Cancel Subscription** (no refund; the plan stays available until the period ends, and you can't buy a new plan until then) or **Resume Subscription**. Stripe-billed plans also offer **Manage in Stripe**.+- **Available balance** — **Plan Balance**, **Top-up Balance**, and **Promotional Balance**. **Add Funds** buys more.+- **Balance Usage** — how much of each balance is used and remaining, plus **Promotional Balance Details** (each grant's amount and expiry).+- **Order history** — top-up and subscription orders started in Cindy, with status. Unpaid orders are canceled automatically when they time out. **Request Invoice** on a completed order shows the details to email to support.+- Provider pages link here too: **View usage** / **Buy a plan** on **Cindy AI** in **Settings > Providers** (see the providers topic).++**Menu bar (macOS):**++- **Show remaining usage in the menu bar** (on by default) is in **Settings > General > App Behavior**. It adds a Cindy icon to the macOS menu bar; click it to see what's left on your subscriptions, your Cindy balance, and today's usage.++**Activity Monitor — what's using CPU and memory:**++- Open the **Menu** button at the top left and choose **Activity Monitor**. It opens in its own window.+- **Agent Processes** (the agent runtimes behind your tasks, with the task they belong to) and **App Processes** (Cindy's own processes), each with **CPU**, **Memory**, and **PID**. Click a column to sort.+- Select an agent process and click **Stop Process** to force-stop it. Every task using that process is interrupted; send another message in the task to start its agent again. Cindy's own app processes can't be stopped here.+- It shows processes on this computer only, not on remote computers you control.++**Other usage views:**++- Voice input usage (duration, count, estimated cost): **Settings > Voice Input** (see the voice input topic).+- A plugin's own calls: **Your calls** on its page under **Settings > Plugins**; a skill's recent use: its **Usage** panel on the skill detail page.diff --git a/apps/desktop/help-knowledge/voice-input.md b/apps/desktop/help-knowledge/voice-input.mdindex edf6f2b7ba..9d017ec30a 100644--- a/apps/desktop/help-knowledge/voice-input.md+++ b/apps/desktop/help-knowledge/voice-input.md@@ -1,41 +1,47 @@---id: voice-inputtitle: Using voice input (dictation)-summary: Push-to-talk dictation that inserts ASR text into the focused composer; configure shortcut, mic, language, refinement, and dictionary in Settings > Voice Input.+summary: Dictate with the global shortcut, the composer mic button, or by holding the composer; configure voice service, language, microphone, shortcut, refinement, dictionary, voiceprint filter, and sounds in Settings > Voice Input.tab: voice-input----Voice input lets you dictate instead of typing. Hold a shortcut, talk, release — the recognized text is inserted into whichever composer has focus.--**Activation:**--- Default shortcut: **Alt+Space** on macOS, **Ctrl+Shift+Space** on Windows. The shortcut is customizable on macOS and Windows. On Linux, global voice shortcuts are not supported yet (the setting is hidden).-- The shortcut is **global** — it works even when Cindy isn't the focused app, so you can dictate into the composer from anywhere.-- It's **push-and-hold**: press to start listening, release to stop and submit.-- Pressing the microphone immediately shows the recording controls and starts the timer. The start cue acknowledges the action and plays independently while capture starts in the background. Once audio becomes available, it is retained even while the recognition service is connecting; up to about 60 seconds is buffered locally and sent in order when it connects. You can stop before connection completes without discarding that buffered audio.--**Settings:**--- **Microphone** — pick a specific device or leave on "auto" (system default).-- **Language** — auto / 中文 (zh-CN) / 繁體中文 (zh-TW) / English / 日本語 / 한국어. Those are the supported recognition locales — there's no free-form locale field.-- **Refinement** — optional LLM post-process that cleans up the raw ASR text (fixes punctuation, removes ums, joins fragments). You can write a custom refinement prompt to bias it toward your preferences.-- **Mute system audio while recording** — toggle; reduces feedback / echo from your speakers picking up onto the mic. On by default.-- When enabled, the start cue finishes before system audio is muted. Capture continues throughout. Audio captured before the mute finishes is retained too, so the beginning of your speech is not cut off; a brief amount of sound already playing through the speakers may also be captured.-- **Fast activation** — toggle; keeps the mic warm so press-to-talk has lower latency at the cost of slightly more background mic usage. Off by default.-- **Interaction sound** — a short sound cue when dictation starts / stops. On by default.--**Dictionary (improves recognition of your terms):**--- **Manual entries** — type terms you want recognized (people's names, product names, jargon).-- **Learned entries** — when you edit the text shortly after dictation lands in the composer, the external-edit inspector picks up your corrections and feeds them back as automatic dictionary entries. Both manual and learned entries feed the refinement pass.+Voice input lets you dictate instead of typing. The recognized text goes into the Cindy composer you're typing in, or, when another app is in front, is typed into that app's focused text field.++**Starting and stopping:**++- **Global shortcut** — default **Option+Space** (shown as Alt+Space) on macOS and **Ctrl+Shift+Space** on Windows. **Hold** it, speak, and release to finish. Or **tap** it once to keep recording hands-free, then press it again to finish. It works even when Cindy isn't the active app. On Linux the global shortcut isn't supported yet; use the in-app voice input button.+- **Composer mic button** — click **Start voice input** in the message composer; click again to stop.+- **Hold the composer** — press and hold the left mouse button inside the composer for a moment to start; release to stop and insert the text.+- Recording starts right away when you trigger it. Audio captured while the recognition service is still connecting is kept (about 60 seconds is buffered) and sent once it connects, so the beginning of your speech isn't lost.+- Outside Cindy, a small overlay shows the live transcript. **Finish and input** inserts it; **Close and discard** cancels. If Cindy can't type into the current app, the overlay keeps the text with a **Copy** button.++**Settings > Voice Input:**++- **Service source → Voice service** — **Cindy official service** (default; uses your Cindy sign-in) or **Custom voice service** (choose a recognition model that uses its own credentials, or **Custom realtime ASR** with your own WebSocket URL, model, and API key; **Test connection** checks it). When you're not signed in, use a custom service.+- **Basics:**+ - **Language** — **Auto** (default) / Simplified Chinese / Traditional Chinese / English / Japanese / Korean. The China edition defaults to Simplified Chinese. There's no free-form locale field.+ - **Microphone** — **Auto** (system default) or a specific device.+ - **Shortcut** — click to record a new one: a key combination, a standalone F1–F24 key, or (macOS) holding Left/Right Command, Option, Control, or Fn. **Clear** removes it. Cindy rejects shortcuts that conflict with system or common editing shortcuts.+- **Refinement:**+ - **Voice refinement** (on by default) — cleans up punctuation, terminology, and spoken phrasing. Turn it off to use the raw recognition result.+ - **Refinement rules** — editable instructions for your preferred style (they guide cleanup only; the model won't answer or expand content).+ - **Custom dictionary** — add names, product terms, and jargon, optionally with **Often misheard as** variants; **Import CSV** (one term per row) adds many at once. Filter by **All** / **Auto-added** / **Manual**.+ - **Auto-learning** (on by default) — when you correct dictated words by hand, Cindy records candidates and adds them to the dictionary after repeated evidence. Learning from edits in other apps is macOS-only.+ - **Sync across my devices** (on by default) — merges your dictionary between your computers that are online at the same time, device to device, without storing it on a server.+- **Preferences:**+ - **Mute during voice input** (on by default) — temporarily mutes system audio and restores it afterwards. The start cue finishes before muting, and capture isn't interrupted. Not supported on Linux.+ - **Fast voice activation** (off by default) — keeps the microphone ready briefly after use so recording starts about 100 ms sooner; the mic is released after 30 minutes without use.+ - **Sound cues** (on by default) — short sounds when voice input starts and ends.+- **Voiceprint filter → Recognize my voice** (off by default) — record your voiceprint (read three short passages) so only your voice is transcribed. The first use downloads a recognition model. **Strictness**: Relaxed / Standard / Strict. The voiceprint stays encrypted on this computer.+- **Usage** — duration, count, and estimated cost, plus **Recent transcripts** you can copy or delete.**Permissions (macOS):**-- **Microphone** — required; the page prompts the OS dialog if you haven't granted it.-- **Input Monitoring** and **Accessibility** — may be required depending on how the global shortcut is captured. The page shows each permission's status (granted / denied / unknown) so you know what to grant in System Settings.+- **Microphone** — required. The page shows **Granted** or a **Grant** button.+- **Shortcut listener** (Input Monitoring) — needed only for Fn, standalone Command / Option / Control, and standalone F1–F24 shortcuts. Regular key combinations don't need it.+- **Automatic input** (Accessibility) — lets voice input type into other apps, read nearby text for refinement context, and learn from your edits there.**Notes:**-- If Cindy reports "Too many voice requests", wait a moment before starting another recording. Cindy voice providers share this account limit, so automatic provider switching stops for that attempt. The limit does not mark the providers as faulty or impose a provider cooldown.-- Push-and-hold is the primary activation mode; **Fast activation** changes the warm-up behavior, not the trigger itself.-- If the global shortcut fails to register (e.g. it's already taken by another app like an IME), you'll see a warning in the page; pick a different shortcut.-- If the connection drops mid-dictation, the text recognized up to that point is kept — it lands in the composer (or stays in the overlay with a copy button) alongside the error, instead of being discarded. That salvaged text is raw ASR output: refinement needs the connection that just failed, so it is skipped.+- If Cindy reports "Too many voice requests", wait a moment before trying again.+- If the shortcut can't be registered (for example, another app or input method already uses it), the page shows a warning; pick a different shortcut.+- If the connection drops mid-dictation, the text recognized so far is kept — in the composer, or in the overlay with a copy button — instead of being discarded. That text isn't refined.diff --git a/apps/desktop/help-knowledge/workspace-panels.md b/apps/desktop/help-knowledge/workspace-panels.mdnew file mode 100644index 0000000000..1e2950dd12--- /dev/null+++ b/apps/desktop/help-knowledge/workspace-panels.md@@ -0,0 +1,74 @@+---+id: workspace-panels+title: Right sidebar panels (files, terminal, browser, side chat, and more)+summary: Open the right sidebar next to a task and add panels with + (Add tab) — File browser, Review, Subagent, Background tasks, New side chat, Browser, Terminal, plugin panels — then reorder, move tabs to another edge, or pop them out into their own window.+---+The **right sidebar** sits next to an open task. It holds tabs ("panels") that help you work alongside the conversation. Tabs belong to the task: switching tasks switches to that task's tabs, and they come back when you return. Plugin panels are the exception (see below).++**Opening it:**++- Click the panel icon at the top-right of the window (**Show sidebar**). The same area has **Maximize sidebar** / **Restore sidebar**, **Close sidebar**, and **Open the current tab in a new window**.+- With no tabs yet, the sidebar shows **Add a panel** with the same choices as the **+** menu.+- The right sidebar is only available while a task is open.++**The + (Add tab) menu:**++- **File browser** — browse the task's working folder (details below).+- **Review** — the task's git changes; see the Reviewing changes topic.+- **Subagent** — progress, results, and details of subagents created in this task. It's shown for local Pi tasks (and for tasks that still have Pi subagent records).+- **Background tasks** — background agents, shell commands, and workflows started in this task, split into **Running** and **Completed**; open one to see its progress or output, or **Stop** it. Clicking a background-task card in the chat also opens this tab.+- **New side chat** — see Side chat below.+- **Browser** — a built-in web browser (details below).+- **Terminal** — a shell in the task's working folder (details below).+- **Plugins** — installed plugins that provide a panel are listed under their own heading.+- Review, Subagent, and Background tasks can be open only once per task; the menu then shows **Open** and switches to the existing tab. You can open several File browser, Browser, Terminal, and side chat tabs.+- Other tabs open automatically from elsewhere: the **Team** tab when you use Collaboration (see the Collaboration topic), plus some plugin and mail pages.+- In IM bot chats, the menu is shorter (no Review, Subagent, Background tasks, Terminal, or side chat).++**File browser:**++- A file tree of the working folder next to a preview of the selected file. Drag the divider to resize the tree.+- **Filter files…** narrows the tree by file name; the search button switches to **Search file contents** (with **Match case**). The tree also has **Collapse all**, **Refresh**, and **Show ignored directories**.+- **Edit** a text file right in the panel, then **Save** (Cmd/Ctrl+S). If you switch files with unsaved changes, Cindy asks first.+- Right-click a file for **Open in file browser** (opens it in another File browser tab), **Copy File Path**, and **Show In Folder** (these two for local tasks); web pages also offer **Open in sidebar** (the built-in Browser) and **Open in browser**. Click an image to preview it. Creating, renaming, or deleting files isn't done from this panel.+- Drop a file onto the panel to preview it (one file at a time).+- Also works for tasks on another computer or an SSH host (file-name filtering and search there need ripgrep on the remote machine). Tasks without a working folder can't use it.++**Browser:**++- Address bar (**Enter URL or search**), **Back**, **Forward**, **Reload**, and **New tab**.+- **Screenshot** copies the page to the clipboard.+- **Comment on the page**: click an element (Shift-drag to select an area), describe the change you want, and **Add comment** — the comment goes into the chat input so you can send it to the agent. Press Esc to leave comment mode.+- **More** has **Open in system browser** and **Copy link**.+- Sign-ins and cookies are kept between sessions. You can reference this task's browser tabs in the composer with **@** (see the Attachments topic).++**Terminal:**++- Opens a shell in the task's working folder. Choose the shell in **Settings > Personalization > Default terminal shell** (new tabs only).+- Shortcut: **Ctrl+`** (**Open terminal**) focuses the existing terminal tab or creates one.+- If the shell exits, click **Restart**. Terminals don't survive quitting Cindy.+- Available only for tasks running on this computer — not for tasks on another computer or an SSH host.++**Side chat:**++- Ask a separate question with the main task's context without interrupting it. Open one with **New side chat** in the + menu, with `/side` (text after it becomes the draft question), or by selecting text in a reply and clicking **Side chat**.+- It uses the context completed when you opened it; the main task keeps running. **Open full task page** shows it full-size; **Back to main task** returns.+- Wait until the main task has its first reply. Not available in shared tasks you joined, in IM bot chats, or in tasks whose agent runs in another environment.+- Closing the tab asks **Close side chat?** — it stops the side chat and deletes its history (**Close and delete**).++**Plugin panels:**++- Shown as tabs marked "Global tab, kept across tasks": they stay put when you switch tasks.+- Reopen a closed plugin panel from the **+** menu or with **Open in the sidebar** (see the Plugins topic).++**Managing tabs:**++- Drag tabs to reorder them. Right-click a tab for **Close**, **Close others**, **Close all**, **Move to …**, and **Open in a new window**.+- **Move to left side / top / bottom / right sidebar** places the tab on another edge of the chat area. You can also drag a tab out of the tab strip to an edge, or onto another tab strip.+- **Pop out:** **Open in a new window** (or drag a tab outside the Cindy window) opens just that tab in its own window. Use **Move to window:** to add a tab to a window you already popped out, or **Merge back into the sidebar** to return it.+- Shortcuts: **Previous / Next right sidebar tab** — Cmd+Shift+[ / ] on macOS, Ctrl+PageUp / PageDown on Windows, or Ctrl+Shift+Tab / Ctrl+Tab. On macOS, Cmd+W closes the active tab while the sidebar has focus. Tab-switching and **Open terminal** shortcuts can be changed in **Settings > Keyboard Shortcuts**.+- Panel positions and widths are saved; **Settings > Appearance > Restore default layout** resets them.++**Activity Monitor:**++- Not a sidebar tab: open the menu button at the top-left of the window and choose **Activity Monitor**. It lists **Agent Processes** and **App Processes** with CPU and memory, and can **Stop Process** for an agent process (tasks using it are interrupted; send another message to restart).diff --git a/apps/desktop/help-knowledge/worktrees.md b/apps/desktop/help-knowledge/worktrees.mdindex 61170dd95d..815814d0a6 100644--- a/apps/desktop/help-knowledge/worktrees.md+++ b/apps/desktop/help-knowledge/worktrees.md@@ -1,40 +1,49 @@---id: worktreestitle: Working in an isolated git worktree-summary: Opt a session into an isolated git worktree so its changes stay on their own branch and folder.+summary: Start a session in an isolated git worktree (Branch | worktree control in the new-session view) so its changes stay on their own branch and folder; what happens on archive/delete and how to restore.---A session can run inside its own **git worktree** — a separate working copy of the same repository on its own branch — so the agent's edits don't mix with your main checkout.**Opting in:**-- In the new-chat view, click the working-directory chip → the **gear** icon to open the Advanced popover → tick the **worktree** checkbox.-- The worktree is created **on first message send**, not on session create. An overlay shows the status (creating / ready / failed).+- In the new-session view, pick a project that is a git repository. A combined control appears next to the project picker: the left half shows the **Branch** (the source branch, defaulting to the repo's current branch; click it to search and choose another), and the right half is a **worktree** checkbox. Tick **worktree** to use isolation.+- The checkbox is off by default. Cindy remembers your last explicit choice for new sessions on that computer.+- The worktree is created when you send the first message. While it's being created, an overlay shows **Creating worktree · input disabled** and the composer unlocks automatically when it's ready.+- Next to the branch, a refresh button can pull the latest commits for the selected branch from its remote before you start.**Where it goes:**-- Folder: `<repo>/.cindy-worktrees/<auto-generated-name>/` — kept inside the same repo so git tooling can find it. Worktrees created by older versions under `.xdt-worktrees/` remain supported.-- Branch: `cindy/<auto-generated-name>` — auto-created off the source branch you pick, which defaults to the repo's current branch (HEAD). Branches created by older versions under `xdt/` remain supported and do not need to be renamed.+- Folder: `<repo>/.cindy-worktrees/<auto-generated-name>/`. Worktrees created by older versions under `.xdt-worktrees/` remain supported.+- Branch: `cindy/<auto-generated-name>`, created from the source branch you picked. Older branches under `xdt/` remain supported.+- Untracked local files (such as `.env`) aren't in a new worktree by default. List them in a `.worktreeinclude` file (`.gitignore` syntax) at the repo root and Cindy copies matching untracked files into each new worktree.**Lifecycle:**-- Created on first agent turn that needs files.-- Survives across app restarts as long as the session is around. Transient closes (`/clear`, reconnecting, quitting the app) do **not** remove it.-- The folder is removed only when you explicitly **delete or archive** the session; the **branch is preserved** (no force-delete), so you can keep the work or merge / cherry-pick from it later.-- Uncommitted changes and new files inside the worktree are **saved in Git** before the folder is removed, and come back when the session is restored. Files ignored by `.gitignore` (dependencies such as `node_modules`, build output, caches) and Git repositories cloned or initialised inside the worktree are **not kept**; reinstall, rebuild or clone them again after restoring. Restored files go through Git's usual line-ending and symlink handling, like a normal checkout. Removal is skipped if you leave a `.worktree-keep` file in it or another live session is still using it.+- The worktree survives app restarts, `/clear`, and reconnecting.+- When you **archive** or **delete** the session, Cindy removes the worktree folder but **keeps the branch**, so you can still merge or cherry-pick from it. Uncommitted changes and new files are first saved as a git snapshot (the confirmation dialog warns you when there are any). Files ignored by `.gitignore` (dependencies like `node_modules`, build output, caches) and git repositories cloned inside the worktree are **not kept**.+- Removal is skipped if the folder contains a `.worktree-keep` file or another live session is still using it.+- **Restoring:** when you unarchive the session and send a message, Cindy recreates the worktree from the kept branch and re-applies the snapshot automatically. If you open a session whose worktree is missing, a banner offers **Restore worktree** (or **Apply changes** if only the snapshot is pending).+- **Settings > Storage > Worktree cleanup** lists worktrees still awaiting cleanup, with **Retry cleanup** and **Keep directory**.-**When the option is disabled:**+**When the control doesn't appear:**-- Git is not installed on your machine.-- The chosen working directory isn't a git repository.-- The chosen directory is already inside an existing worktree (no nesting).+- The control is hidden when Git isn't installed, the chosen folder isn't a git repository, or it's already inside a worktree (no nesting). It's also hidden for projectless chats.+- While Cindy is still detecting the folder, or detection failed, the checkbox is disabled and its tooltip explains why.-**Error cases (shown on the overlay):**+**Errors (shown if creation fails):**-- `permission-denied` (can't write into the repo dir), `git-crypt-locked` (repo uses git-crypt and isn't unlocked), `dubious-ownership` (git's safe.directory check tripped), `lfs-error`, `not-a-git-repo`, `git-not-installed`.+- Cannot access worktree directory (permissions), git-crypt repository is not unlocked, Dubious git repository ownership (`safe.directory`), git-lfs sync failed, Current directory is not a git repository, git not detected. Each comes with a suggested fix.++**Other places that use worktrees:**++- Automations can run in a **New worktree session** (see the Scheduling automations topic), collaboration Workers can get their own worktree (**Create a worktree for this Worker**), and importing a shared session offers **Create in a worktree**.+- In **Sidebar settings → Session info**, tick **worktree** to show each session's worktree name; click it to open the folder.+- A session whose working directory is a worktree can't be moved to another project yet.**Notes:**-- The agent's **Read / Write / Bash** tools automatically operate inside the worktree path — you don't need to switch directories manually.-- Because the branch survives the worktree, "did the agent's changes go anywhere?" is always answerable with `git branch --list 'cindy/*' 'xdt/*'`.-- Worktrees are a per-session opt-in — they're not on by default and they don't change non-worktree sessions in any way.+- The agent works inside the worktree folder automatically; you don't need to switch directories.+- Because the branch outlives the worktree, you can always find the agent's work with `git branch --list 'cindy/*' 'xdt/*'`.+- Worktrees don't affect sessions that don't use them.diff --git a/apps/desktop/src/main/__tests__/helpRouting.test.ts b/apps/desktop/src/main/__tests__/helpRouting.test.tsindex fb653aa7ce..7dde4e6289 100644--- a/apps/desktop/src/main/__tests__/helpRouting.test.ts+++ b/apps/desktop/src/main/__tests__/helpRouting.test.ts@@ -116,7 +116,7 @@ describe('HELP_KNOWLEDGE invariants', () => {expect(providers?.tab).toBe('providers');expect(providers?.title).toContain('Cindy AI');- expect(providers?.summary).toContain('Model Providers');+ expect(providers?.summary).toContain('Settings > Providers');// 「工具密钥」(api-keys) 面板已于 2026-07-13 下架:Mivo / 搜索 key 改由// 对应意识详情页收单,帮助文档随面板一并退役。@@ -126,7 +126,7 @@ describe('HELP_KNOWLEDGE invariants', () => {it('collaboration guidance includes Pi as a lead and worker, locally or over SSH (round 42)', () => {const collaboration = HELP_KNOWLEDGE.find((d) => d.id === 'collaboration');- expect(collaboration?.content).toContain('Claude Code, Codex, and Pi can lead locally or over SSH');+ expect(collaboration?.content).toContain('Claude Code, Codex, and Pi sessions can lead, locally or over SSH');expect(collaboration?.content).toContain('Claude Code, Codex, or Pi');// 轮 42:Pi SSH remote 能力落地后「local-only」表述已过时,不得再出现。expect(collaboration?.content).not.toContain('Pi sessions are local-only');diff --git a/apps/desktop/src/main/maker-ipc/helpKnowledge.generated.ts b/apps/desktop/src/main/maker-ipc/helpKnowledge.generated.tsindex 5c45e94271..474e760508 100644--- a/apps/desktop/src/main/maker-ipc/helpKnowledge.generated.ts+++ b/apps/desktop/src/main/maker-ipc/helpKnowledge.generated.ts@@ -14,163 +14,212 @@ export const HELP_KNOWLEDGE: HelpKnowledgeDoc[] = [{id: "appearance",title: "Appearance, theme, wallpaper and fonts",- summary: "Theme, theme family, wallpaper, fonts, sidebar styles, panel layout, and reply animation — all in Settings > Appearance.",+ summary: "Theme, appearance mode, local and imported themes, wallpaper, UI / code / terminal fonts, sidebar styles, panel layout, and reply animation — all in Settings > Appearance.",tab: "appearance",- content: "Settings > Appearance controls how the app looks. It starts with the theme and appearance mode, followed by Wallpaper, Fonts, Sidebar & panels, and Reply Animation. Display language and notifications stay in Settings > General (see the language and notifications topic).\n\n- **Theme**: pick from the registered theme families; each family supplies its own colors for light and dark. The app uses a VSCode-style token system (see docs/design-rules/cindy-design-system.md) so themes only override what they need. Families with only one variant are tagged Light only / Dark only; while one is selected the appearance mode is fixed to that variant, and your saved mode returns when you pick a two-variant theme.\n- **Appearance mode**: light, dark, or system (follows your OS).\n- **Wallpaper**: choose a Cindy scene, your own image, or a video for the whole app. The wallpaper is independent of the theme.\n- **Fonts**: choose fonts and sizes for interface text and code. Code font size changes code blocks, change comparisons, and editors; inline code, status numbers, badges, and path labels keep their size.\n- **Pinned Style**: show pinned sessions as Text, Cards, or List. **List Style** controls unpinned sessions: Text uses one line; List adds a preview.\n- **Local theme brand identity**: this is an optional power-user feature configured in the local theme JSON. The settings page intentionally keeps only create-copy, open-folder, and refresh actions. New copies include self-explanatory example paths directly in the JSON.\n- **Export / open local theme files**: export the current theme's tokens to a file or open a local theme JSON for inspection / sharing. To replace the icon and logo used on the new-chat page:\n\n ```json\n {\n \"brand\": {\n \"icon\": \"/absolute/path/to/your-image-folder/icon-square-50x50px.png\",\n \"logo\": \"/absolute/path/to/your-image-folder/logo-horizontal-110x37.5px.png\"\n }\n }\n ```\n\n The example filenames describe the final display areas; source images may be exported at 2x/3x while keeping the same aspect ratios. Paths must be absolute JSON strings; spaces do not need shell quotes. PNG, JPEG, and WebP are supported. Transparent padding is trimmed at render time without changing the source file. Only `brand.icon` and `brand.logo` are recognized; the previous single-image fields are not compatible with the redesigned two-asset layout.\n\n- **Panel layout**: panel positions and widths are saved automatically; **Restore default layout** puts them back.\n- **Reply Animation**: new reply text fades in while it streams. Turn it off to show text immediately. It stays off when your system reduces motion.\n\n**Notes:**\n\n- Theme changes apply immediately; no restart needed.",+ content: "Settings > Appearance controls how the app looks. It starts with **Theme** and **Appearance Mode**, followed by **Wallpaper**, **Fonts**, **Sidebar & panels**, and **Reply Animation**. Display language and notifications stay in Settings > General (see the language and notifications topic).\n\n**Theme:**\n\n- **Theme**: pick a theme family; each family supplies its own light and dark colors. Local themes are marked **(Local)**. Families with only one variant are tagged **Light only** / **Dark only**; while one is selected the appearance mode is fixed to that variant, and your saved mode returns when you switch to a two-variant theme.\n- **Appearance Mode**: **Light**, **Dark**, or **Auto** (follows your system).\n- **Local themes** row:\n - **Create local copy** — writes an editable JSON copy of the current theme into your themes folder.\n - **Import VSCode / Obsidian theme** — pick a VSCode or Obsidian theme file and turn it into a local theme, then select it in the theme list. VSCode themes that inherit colors via `include` can't be imported on their own.\n - **Open themes folder** — opens the folder that holds local theme JSON files.\n - **Refresh** — reloads local themes after you edit them.\n- **Local theme brand identity** (optional, for power users): in a local theme JSON you can replace the icon and logo shown on the new-chat page. New copies already contain example paths to replace:\n\n ```json\n {\n \"brand\": {\n \"icon\": \"/absolute/path/to/your-image-folder/icon-square-50x50px.png\",\n \"logo\": \"/absolute/path/to/your-image-folder/logo-horizontal-110x37.5px.png\"\n }\n }\n ```\n\n The example filenames describe the final display areas; source images may be exported at 2x/3x while keeping the same aspect ratios. Paths must be absolute JSON strings; spaces do not need shell quotes. Transparent padding is trimmed at render time without changing the source file. Only `brand.icon` and `brand.logo` are recognized. If a path doesn't exist, the default artwork is shown. Save the file, then click **Refresh**.\n\n**Wallpaper:**\n\n- Choose **None**, one of the built-in Cindy scenes / backgrounds, or **Custom wallpaper** for the whole app. The wallpaper is independent of the theme; **Reset** restores the default wallpaper, visibility, blur, and display settings.\n- **Choose Wallpaper** / **Replace Wallpaper** / **Remove Wallpaper**: PNG, JPEG or WebP up to 20 MB, or an MP4 (H.264) video up to 100 MB. It's stored on this device.\n- **Display**: **Static** (default) or **Dynamic** for animated scenes and custom videos. Motion pauses when the window is hidden and follows your system's reduced-motion setting; videos loop silently.\n- **Wallpaper Visibility** (0% hides it, 100% shows it without an overlay) and **Wallpaper Blur** (0 turns it off; blurring animated wallpaper uses more GPU).\n\n**Fonts:**\n\n- **UI font** / **UI font size** — menus, settings, and chat text (12–24).\n- **Code font** / **Code font size** — code blocks, change comparisons, and editors (10–24). Inline code, status numbers, badges, and path labels keep their size.\n- **Terminal font** / **Terminal font size** — only the built-in terminal (10–24, default 13). A monospace font is recommended.\n- Each font picker offers presets or a **Custom** font name; each size has a **Reset to default** button.\n\n**Sidebar & panels:**\n\n- **Pinned Style**: show pinned sessions as **Text**, **Cards**, or **List** (same as the view-style button on the sidebar's **Pinned** header).\n- **List Style**: for unpinned sessions — **Text** uses one line; **List** adds a preview (same as **Display** in the sidebar's **Sidebar settings**; see the Sidebar topic).\n- **Panel layout**: panel positions and widths are saved automatically; **Restore default layout** puts them back.\n\n**Reply Animation:**\n\n- **Fade In Text**: new reply text fades in while it streams. Turn it off to show text immediately. It stays off when your system reduces motion.\n\n**Notes:**\n\n- Theme changes apply immediately; no restart needed.\n- To zoom the whole window, use **Cmd/Ctrl + =** / **Cmd/Ctrl + -**, and **Cmd/Ctrl + 0** to reset (on macOS also under the **View** menu). The zoom level is remembered.",},{id: "attachments",title: "Attaching files and images, and referencing code",- summary: "Paste / drag images, @-mention files, and add extra reference directories from the composer.",- content: "You can give the agent more than plain text.\n\n**Images:**\n\n- **Paste** an image from the clipboard, or **drag** image files onto the composer.\n- Accepted formats: **PNG, JPEG, WEBP, and GIF**.\n- Thumbnails appear above the input; click one to open a full-screen lightbox preview (remove from there, or ESC to close).\n\n**`@`-mention to reference files / directories / agents:**\n\n- Type `@` to open the picker. It shows files, directories, and registered agents in one flat list.\n- Scope: the picker scans your session's **working directory** (extra reference directories are given to the agent as context but aren't part of the `@` picker's file scan).\n- Matching is fuzzy, with a boost for prefix matches on the filename.\n\n**Extra reference directories (read-only context for the agent):**\n\n- Open the composer's **+** menu (to the left of the permission selector — it also hosts New Goal, Plan Mode, and Plugins) and add a directory from the OS folder dialog.\n- Per-session; up to **10** extra directories. This section shows for **Claude Code** sessions (Codex ignores extra directories).\n- Directories that are subdirectories of the current working directory are silently skipped (already covered). Parent / ancestor directories are flagged with a warning.\n- The agent gets read-only context for these directories — it can grep / read them, but writes still go to the working directory.\n\n**Notes:**\n\n- Image attachments are sent with that user message only; they're not implicitly re-included on subsequent turns.\n- The `@`-mention just inserts a reference string into your prompt — it's the agent's job to actually open / read the referenced item.",+ summary: "Paste / drag images and files, use the composer's + / @ panel to attach files, reference files, tasks, browser tabs and more, and add extra directories the agent can work in.",+ content: "You can give the agent more than plain text.\n\n**Images and files:**\n\n- **Paste** an image from the clipboard (e.g. a screenshot), or **drag** files onto the composer / task window. You can also open the composer's **+** menu and choose **Add files, images, or videos…** to pick files from the OS dialog.\n- Images (**PNG, JPEG, WEBP, GIF**) are sent to the model as images. Other files (PDF, Office documents, code, text, videos, etc.) are attached by path, and the agent opens or reads them as needed.\n- Attachments appear above the input. Click an image to open a full-screen preview; click a text / code file to preview its contents. Hover a thumbnail and click **×** to remove it.\n- Pasting a copied folder inserts it as a folder reference instead of an attachment.\n- Pasting a very long block of text (about 24+ lines or 4,000+ characters) folds it into a **Pasted text** chip; click the chip to view or edit it.\n- Local file attachments aren't available in tasks running on an SSH remote host — paste the content directly instead.\n\n**The + menu and `@` mentions:**\n\n- Clicking **+** (left of the composer toolbar) and typing `@` open the same panel. It lists actions — **Add files, images, or videos…**, **New goal**, **Plan mode**, **Collaboration**, and **Add directory…** — followed by your installed plugins and the task's extra directories. Type to filter; **↑ / ↓** and **Enter** select.\n- Things you can reference with `@`:\n - **Files and folders** in the task's **working directory** (extra directories aren't scanned by the picker).\n - Agents defined in the project.\n - **Tasks** — type a keyword to find past tasks and reference them.\n - **Teammates** — \"Message or delegate\" to one of your teammates.\n - The current task's built-in **browser tabs** and open **app windows** on this computer (local tasks only).\n - **Plugin** commands and plugin resources.\n- Matching favors names that start with what you typed, then falls back to in-order fuzzy matching on names and paths.\n\n**Extra directories:**\n\n- Choose **Add directory…** in the **+** menu to pick a folder from the OS dialog. The agent can **read and write** in it, in addition to the working directory. Added directories are listed under **Directories** in the same panel; hover one to remove it.\n- Per task; up to **10** extra directories (**Limit reached (10)** when full). Available for local Claude Code, Codex, and Pi tasks; not available for SSH remote or linked-device tasks.\n- Choosing a subdirectory of the working directory is silently skipped (already covered). Choosing a parent / ancestor directory asks you to confirm first, because the agent could then modify everything inside it.\n\n**Notes:**\n\n- Attachments are sent with that one message only; they're not implicitly re-included on later turns.\n- An `@` reference just inserts a reference into your prompt — it's the agent's job to actually open / read the referenced item.",},{id: "collaboration",title: "Collaboration mode (lead + workers)",- summary: "Turn on Collab to bring one or more worker agents in beside your session and delegate tasks to them.",- content: "Collaboration (\"协同\" / Collab) lets one session — the **lead** — bring in one or more **worker** agents to run alongside it (a \"team\"). The lead delegates tasks to workers through a built-in MCP, and you can watch and steer each worker.\n\n**Turning it on:**\n\n- In any regular project or dialogue session, open the composer's **+** menu and choose **Collaboration mode**. You can also choose it before sending the first message of a new task. Claude Code, Codex, and Pi can lead locally or over SSH.\n- Pick the first worker's agent (Claude Code, Codex, or Pi), and optionally its role / model / effort / an initial task. If you have not chosen a Worker permission before, the initial default is **Full access** — it is a selectable default, not a fixed mode. You can choose **Auto-review** instead. Once you create a Worker with either mode, Cindy remembers that choice for the next Worker instead of replacing it with the product default.\n- Start. This creates the team plus that first worker.\n\n**Adding more workers:**\n\n- From the collaboration tab's worker list, use **Create new worker** (+) to add more.\n- Default limits are a **soft cap of 5** and a **hard cap of 8** workers per lead — you'll see a warning past the soft cap, and creation is blocked at the hard cap. Both are adjustable (1–20) in the Collaboration section under **Settings > General**.\n\n**While collaborating:**\n\n- Each worker is a full session with its own history, tools, working directory, and model. **Full access** skips routine approvals; **Auto-review** lets safe work proceed automatically while higher-risk actions may still be denied or require confirmation. The Worker creation panel remembers your last permission choice alongside its Agent and model preferences. The same remembered choice is used when an agent tool creates a Worker; an explicit `start_team` permission updates that shared choice. Raising it from Auto-review to Full access waits for your confirmation first; cancelling or letting the confirmation time out leaves the preference and team unchanged. Changing it does not affect the lead or workers that already exist. Workers inherit the lead's working directory by default (or share the worktree when the lead uses one).\n- A worker's model defaults to your last new-session choice for that vendor, falling back to the lead's model.\n- Only one worker is **focused** (shown) at a time — switch between them with the worker tabs / dropdown. You can intervene in any worker directly, and archive workers individually.\n\n**Turning it off:**\n\n- Closing collab **confirms with you first** (in-progress work is lost), then ends the workflow and archives **all** the team's workers. The conversation history stays searchable.\n\n**Notes:**\n\n- A lead can be a Claude Code, Codex, or Pi **project or dialogue** session, locally or over SSH. Remote workers are spawned on the **same remote host**, and the collaboration MCP reaches the remote side over an SSH remote-forward to the local bridge (with a persistent bearer token). A worker can't start its own sub-collab (no nesting).\n- A lead can have only **one active collaboration workflow** at a time. If you have two clients open on the same account (e.g. dev + release) and start collab on the same session in both, the second fails with \"开启协同失败\" because a workflow is already active — close one client or end the existing collab.",+ summary: "Turn on Collaboration from the composer's + menu to bring one or more worker agents in beside your session — locally, in their own worktree, or on another computer — and delegate tasks to them.",+ content: "Collaboration (\"Collab\") lets one session — the **lead** — bring in one or more **worker** agents to run alongside it (a \"team\"). The lead delegates tasks to workers through built-in collaboration tools, and you can watch and steer each worker.\n\n**Turning it on:**\n\n- In a regular project or dialogue session, open the composer's **+** menu and choose **Collaboration**. You can also do this before sending the first message of a new session. Claude Code, Codex, and Pi sessions can lead, locally or over SSH. It isn't offered inside a worker or in a teammate chat.\n- The **Start Collaboration** panel (\"Bring in a Worker\") sets up the first worker:\n - **Agent** (Claude Code, Codex, or Pi), **Model**, **Reasoning Effort**, an optional **Role Name** (just a label to tell workers apart), and an optional **Initial Task**.\n - **Permission Mode**: if you haven't chosen one before, it starts at **Full access** (skips routine approvals). You can choose **Auto-review** instead (safe work proceeds automatically; higher-risk actions may still be denied or need confirmation). Switching to Full access asks you to confirm first.\n - **Project** (local project sessions): work directly in the lead's project and branch, or **Create a worktree for this Worker** branching from a branch you choose. If the directory has uncommitted changes, the worker continues with them.\n - **Run on**: keep the worker **Local**, or run it on another computer on your account that's online and has remote control enabled. Its session, folder, and commands then stay on that computer (choose **Chat** or **Specify folder** for where it works there).\n- The panel remembers your last Agent, model, effort, and permission choice for the next worker. Changing it doesn't affect the lead or workers that already exist.\n- If the + menu says collaboration is disabled, turn **Collab Mode** back on under **Settings > Tools** (it can also be overridden per project).\n\n**Adding and managing workers:**\n\n- Workers appear in the right sidebar's **Team** tab. The worker header lists all workers (choose **Tabs** or **Dropdown** under **Worker list layout**); use **Create new worker** there to add more.\n- Default limits are a **Worker Soft Limit of 5** and a **Worker Hard Limit of 8** per lead — you'll see a warning past the soft limit, and creation is blocked at the hard limit. Both are adjustable (up to 20) in **Settings > General > Collaboration**, which also has **Idle Release (minutes)** to automatically release workers after they sit idle (0, the default, turns it off).\n- Only one worker is shown at a time — switch between them in the worker header. You can message any worker directly, **edit** a worker's role name / label, or **archive** a worker individually (archived workers can't be restored; create a new one instead).\n\n**While collaborating:**\n\n- Each worker is a full session with its own history, tools, and model. Workers use the lead's working directory by default (or the shared / new worktree chosen above).\n\n**Turning it off:**\n\n- Choose **Collaboration** again in the **+** menu, or close the **Team** tab. Cindy **asks you to confirm first** — running workers stop immediately and in-progress results are lost — then ends the team and archives **all** of its workers. The conversation history stays searchable.\n\n**Notes:**\n\n- With an SSH lead, workers are created on the **same remote host**. A worker can't start its own collaboration (no nesting).\n- A lead can have only **one active collaboration** at a time. If you have two Cindy clients open on the same account and start collaboration on the same session in both, the second fails with \"Failed to start collaboration\" because one is already active — close one client or stop the existing collaboration.",},{id: "commands",title: "Slash commands",- summary: "Type \"/\" in the composer to open a palette merging your skills, built-in app commands, and the agent's own commands.",- content: "Type `/` at the start of the composer to open the command palette.\n\n**What's in the palette (priority order on collisions):**\n\n1. **Your installed skills and Cindy's built-in Skills** (highest priority) — anything in your global or project skills directory, including `/cindy-skill-creator` and `/learn` when enabled (see the Skills topic).\n2. **Built-in app commands** — `/help` (show available commands), `/clear` (clear the current session context — resets the conversation in place without creating a new session), `/cmd` (run a shell command in the working directory), `/issue` (file an issue), `/goal` (start a goal-driven run), `/workflows`, and `/jump-session`.\n3. **The agent's own commands** — e.g. `/compact`, `/agents`, `/memory` (the exact list depends on the agent; Claude Code and Codex each contribute their own).\n\n`/issue` does not require a GitHub plugin or GitHub account. The confirmation card uses Cindy's official bot by default; if a working account is already configured under **Plugins > Cindy GitHub**, that account appears as an optional submission identity.\n\nBefore submitting, Cindy helps clarify missing details so the issue is actionable. For bugs, it can collect reproduction steps and user-approved, redacted diagnostic summaries; the confirmation card always shows the final public content. After creation, Cindy returns the issue link and can help interested users reproduce or fix the problem from source because Cindy is open source.\n\n**Using the palette:**\n\n- Start typing to filter. Matching is **contains-based** (case-insensitive substring matching on the command name) — not fuzzy.\n- Up to 25 matches shown at a time.\n- **↑ / ↓** move focus, **Enter** runs the focused command, **Esc** closes the palette.\n\n**Notes:**\n\n- If a name collides across sources (same command in a skill and in a built-in), the higher-priority source wins; the lower-priority one is silently skipped (logged as a warning).\n- Most built-in app commands run inside the desktop app (a few, like `/workflows` and `/jump-session`, are handled directly in the UI); agent commands are sent as prompt prefixes to the agent.",+ summary: "Type \"/\" in the composer to open a palette merging your skills, built-in app commands (/help, /clear, /side, /cmd, /issue, /review, /goal, /plan, /learn …), and the agent's own commands; type \"$\" for installed plugin commands.",+ content: "Type `/` at the start of a line (or after a space) in the composer to open the command palette.\n\n**What's in the palette (priority order on collisions):**\n\n1. **Your installed skills and Cindy's built-in Skills** (highest priority) — anything in your global or project skills directory, including the built-in `/cindy-skill-creator` (see the Skills topic).\n2. **Built-in app commands:**\n - `/help` — show a help card listing every available command with usage examples.\n - `/clear` — clear the current session context in place (wipes its messages and state without creating a new session; the session stays in the sidebar).\n - `/side` — open a side chat with this session's context; any text after it becomes the draft question.\n - `/cmd <command>` — run a shell command in the working directory and show the output in the chat (for a session on another of your computers, it runs on that computer).\n - `/issue [description]` — send feedback or a bug report to the Cindy team (see below).\n - `/review [focus or path]` — review the current session's work in a fresh, read-only reviewer session (local sessions only; see the Reviewing changes topic).\n - `/goal <condition>` — set a goal the agent keeps working toward across turns; `/goal` alone opens the **New goal** dialog, and `/goal clear` clears it.\n - `/plan` — toggle **Plan mode** on or off (when available for the session; see the Plan mode topic).\n - `/learn [description]` — distill a reusable Skill from this session or a described workflow, reviewed as a diff before saving (when enabled; see the Skills topic).\n - `/workflows` — show the progress of workflows running in this session.\n - `/jump-session` — jump straight to a session by its ID.\n3. **The agent's own commands** — e.g. `/compact` and `/context` for Claude Code. The exact list depends on the engine (Claude Code, Codex, or Pi).\n\n**Plugin commands (`$`):**\n\n- Type `$` (a full-width `$` or `¥` also works) to open a separate palette listing commands from your installed, enabled plugins.\n\n**Using the palette:**\n\n- Start typing to filter. Matching is substring-based (case-insensitive) on the command name — exact matches rank first, then prefix matches, then other matches; Cindy's official commands rank ahead on ties. It is not fuzzy.\n- Up to 25 matches are shown at a time.\n- **↑ / ↓** move focus, **Enter** or **Tab** inserts the focused command into the composer (add any arguments, then send), **Esc** closes the palette.\n\n**`/issue` feedback:**\n\n- `/issue` doesn't require a GitHub plugin or GitHub account. Before submitting, the agent helps clarify missing details so the issue is actionable; for bugs it can collect reproduction steps and user-approved, redacted diagnostic summaries.\n- The confirmation card always shows the final public **Title** and **Body**, lets you choose the **Submission Identity** — **Cindy official bot (default)**, or **My GitHub account** if a working account is connected on the **GitHub** plugin's detail page under **Plugins** — and set a **Public Attribution** (or **Anonymous**). You must confirm the content contains nothing private before **Submit Feedback**.\n- After creation, Cindy returns the issue link and can help you reproduce or fix the problem from source, because Cindy is open source. To see issues you've submitted, open **Issues** from the app menu (top-left **Menu**, or **Help > Issues** in the macOS menu bar).\n\n**Notes:**\n\n- If a name collides across sources (same command in a skill and in a built-in), the higher-priority source wins; the lower-priority one is silently skipped.\n- Built-in app commands run inside the desktop app; agent commands and skills are sent to the agent as the start of your prompt.",},{id: "connections",title: "Connecting external accounts",- summary: "External accounts (Google, Jira / Confluence, GitHub, GitLab) are connected on their plugin detail pages under Plugins; Slack and Telegram are bound under Settings > IM bots; Codex sign-in lives in Settings > Model Providers.",- content: "External account connections have moved out of the old \"Third-Party Platforms\" section — each integration now lives on its own plugin detail page under **Plugins**, and the plugin's tools use the account you connect there.\n\n**Where to connect what:**\n\n- **Slack** — connect under **Settings > IM bots** (the \"Cindy\" tab). Each workspace has **Slack Communications on This Device** (本机 Slack 通讯) when the server supports it. Multiple authorized devices can search, read, and post as the Slack user, while only one device receives Bot chats and tasks. Turning off local communications does not revoke the account or affect other devices. Enabling communications on a previously displaced device may open Slack authorization, but never takes over the Bot. Old servers retain the original binding-based behavior.\n- **Telegram** — under **Settings > IM bots** (the \"Cindy\" tab), turn on Telegram and choose **Link Telegram**. Open the one-time link, start the official Cindy bot, and complete confirmation if prompted. The same card lets you cancel, retry, open the bot, add it to a group, or unlink; no personal bot token is required.\n- **Google** (Gmail / Calendar / Drive / Sheets) — handled by the Filo Google plugin (Plugins > Filo Google).\n- **GitHub** — handled by the Cindy GitHub plugin (Plugins > Cindy GitHub; existing PAT connections are migrated automatically). This connection is optional for `/issue`: feedback uses Cindy's official bot by default, and a working GitHub connection only adds the option to submit as yourself.\n- **GitLab** — handled by the Cindy GitLab plugin (Plugins > Cindy GitLab; existing PAT connections are migrated automatically).\n- **Codex** — sign in under Settings > Model Providers (separate from the Codex CLI binary the app ships with).\n- **FeiShu** — bot only (App ID / Secret under Settings > IM bots; see the FeiShu bot topic). FeiShu is no longer a sign-in method for Cindy itself.\n- **Cindy AI** — Settings > Model Providers. Cindy Web Search uses this account by default and needs no key; optional Brave / Tavily / DeepSeek search keys are configured on that plugin's detail page.\n\n**Managing connections:**\n\n- Each plugin detail page shows the connected accounts (with status) and lets you reconnect, disconnect, or switch the default account.\n- Disconnect removes the local token only; you may also want to revoke access on the provider's side.\n- A connection error usually means the token expired or was revoked on the provider's side; reconnect to refresh.",+ summary: "External accounts (the Gmail / Google Drive / Google Calendar / Google Sheets plugins, GitHub, GitLab, and Jira / Confluence via XD Atlassian for organizations that provide it) are connected on each plugin's detail page under Plugins; Slack, Telegram, X and personal chat bots are set up under Settings > IM bots; ChatGPT / Claude / Grok subscriptions and Cindy AI live in Settings > Providers.",+ content: "There is no separate \"Third-Party Platforms\" settings page any more — each integration lives on its own plugin detail page under **Plugins**, and the plugin's tools use the account you connect there. Chat apps are set up under **Settings > IM bots**, and model accounts under **Settings > Providers**.\n\n**Where to connect what:**\n\n- **Slack** — **Settings > IM bots**, **Official** section (requires a Cindy account). Turn on the Slack card to authorize; use **Add Slack workspace** to bind more workspaces. Each workspace has **Slack Communications on This Device** when the server supports it: multiple authorized devices can search, read, and post as the Slack user, while only one device receives Bot chats and tasks. Turning off local communications does not revoke the account or affect other devices. Enabling communications on a previously displaced device may open Slack authorization, but never takes over the Bot.\n- **Telegram (official bot)** — **Settings > IM bots**, **Official** section: turn on Telegram and choose **Link Telegram**. Open the one-time link, start the official Cindy bot, and confirm if prompted. The same card lets you cancel, retry, open the bot, add it to a group, or unlink; no personal bot token is required.\n- **X (Twitter)** — **Settings > IM bots**, **Official** section: authorize your X account, then mention the Cindy bot in a post to send a task to this device. Replies are posted publicly, so use it only for public questions.\n- **Your own chat bots** — **Settings > IM bots**, **Personal** section: **Personal WeChat**, **WeCom Intelligent Bot**, **Feishu / Lark Bot** (App ID / App Secret; see the Feishu bot topic), **DingTalk Bot**, **Discord Bot**, and **Telegram Bot** (your own bot token). You manage these credentials yourself.\n- **Google** — the separate **Gmail**, **Google Drive**, **Google Calendar**, and **Google Sheets** plugins (**Plugins > Gmail**, and so on). Each one is authorized separately on its own detail page. Some organization accounts get a single combined **Filo Google** plugin instead.\n- **Jira / Confluence** — the **XD Atlassian** enterprise plugin (**Plugins > XD Atlassian**), only for organizations that provide it. Earlier Jira / Confluence connections carry over automatically.\n- **GitHub** — the **GitHub** plugin (**Plugins > GitHub**; earlier token connections carry over automatically). This connection is optional for `/issue`: feedback uses Cindy's official bot by default, and a working GitHub connection only adds the option to submit as yourself.\n- **GitLab** — the **GitLab** plugin (**Plugins > GitLab**; earlier token connections carry over automatically).\n- **ChatGPT (Codex), Claude, and Grok subscriptions** — **Settings > Providers**: **OpenAI** (use your local Codex account or sign in to a separate ChatGPT account), **Anthropic** (**Sign in with Claude Code**), and **xAI** (SuperGrok). This is separate from the Codex / Claude Code programs the app ships with.\n- **Cindy AI** — **Settings > Providers**, provisioned automatically after you sign in to Cindy. The **Web Search** plugin uses this account by default and needs no key; optional Brave / Tavily / DeepSeek search keys are configured on that plugin's detail page.\n- **Feishu / Lark** is bot-only (see above); it is not a sign-in method for Cindy itself.\n\n**Managing connections:**\n\n- Each plugin detail page shows the connected account(s) and their status, and lets you reconnect or disconnect. If a plugin later needs new permissions, it asks you to reconnect.\n- Disconnect removes the local token only; you may also want to revoke access on the provider's side.\n- A connection error usually means the token expired or was revoked on the provider's side; reconnect to refresh.",},{id: "feishu-bot",- title: "Setting up the FeiShu (Lark) bot",- summary: "Configure the FeiShu bot (App ID / App Secret) in Settings > IM bots to get DM notifications and bot interaction.",+ title: "Setting up the Feishu / Lark bot",+ summary: "Connect your own Feishu or Lark bot (App ID / App Secret) under Settings > IM bots > Personal to chat with Cindy, get session DMs, and use it in group chats.",tab: "im-bot",- content: "Cindy can send DMs and take commands through a FeiShu (Lark) bot. (This is unrelated to signing into Cindy — you sign in with your Cindy account; FeiShu is no longer a login method.)\n\n**Setting up the bot:**\n\n- Open Settings > IM bots, then the FeiShu Bot section.\n- Paste the bot's **App ID** and **App Secret**, then click **Bind**.\n- The status badge updates: **needs-config** → **connecting** → **connected** (or **conflict** / **error**).\n\n**Binding the bot to you:**\n\n- After the badge shows **connected**, the page instructs you to DM the bot a code (or click an action) to complete the binding. Until you bind, the bot won't know which FeiShu user to DM.\n\n**Notifications:**\n\n- Toggle **message notifications** on the same page to get bot DMs when sessions finish, or when the bot has updates for you.\n- **System Notifications** is a separate toggle in **Settings > General > Notifications**. It covers sessions finishing, encountering an error, or needing a reply. See the language and notifications topic for background-window and Agent Island behavior.\n\n**Group chats:**\n\n- In a group, @ the bot to ask a question. The bot opens a new thread (话题) for the reply — the main stream stays clean and each thread is its own session, like Slack threads.\n- The bot pages back through recent chat history (newest first, until the messages stop being relevant to your question) and answers with that history as context — including images and files shared in the chat. For a fresh thread opened from the main stream, the context still comes from the main stream history.\n- @-ing the bot inside an existing thread uses that thread's own history as context.\n- Thread sessions are titled `[飞书·{group name}·{topic summary}] {id suffix}` in the sidebar.\n\n**Permissions for group features:**\n\nGroup chat context needs an extra app permission that is not part of the default setup:\n\n1. Open the **FeiShu (Lark) Open Platform** → Developer Console → your app → **Permissions & Scopes** (权限管理).\n2. Search for and enable **both** of the following (searching by the scope id works regardless of the console language):\n - **\"Read all messages in associated group chat\" (获取群组中所有消息, `im:message.group_msg`)** — pulls the chat history via API.\n - **\"Read user and bot messages in group chats\" (获取群组中用户和机器人消息, `im:message.group_msg.include_bot:read`)** — needed for group message events, so the history covers messages from other bots too.\n These are sensitive permissions: on FeiShu they are only available to self-built apps (自建应用), not store apps.\n3. Create and publish a new app **version** (self-built apps need admin approval).\n\nWithout it the bot still answers when @-ed, but with no chat context, and it DMs the owner a one-line heads-up pointing here. Optionally, enabling **\"Obtain basic group information\" (获取群基本信息, `im:chat:readonly`)** lets session titles show the real group name instead of an id suffix.\n\n**Notes:**\n\n- The \"conflict\" badge state means another Cindy instance is already bound to the same bot for this user — typical when running dev + release simultaneously. Disconnect on one side to clear it.\n- Settings > IM bots also hosts the Slack and Telegram bindings (the \"Cindy\" tab) and a bring-your-own Discord bot, alongside this FeiShu bot.",+ content: "Cindy can take requests and send you DMs through a Feishu or Lark bot that you create yourself. (This is unrelated to signing into Cindy — Feishu is not a sign-in method.)\n\n**Setting up the bot:**\n\n- Open **Settings > IM bots**. In the **Personal** section, expand **Feishu / Lark Bot**.\n- Under **Service**, choose **Feishu** or **Lark**.\n- Don't have a bot yet? Click **Create on Feishu Open Platform ↗** (or **Lark Open Platform**) to create one, then come back with its credentials.\n- Paste the bot's **App ID** and **App Secret**, then click **Bind**.\n- The status badge shows **Needs configuration** → **Connecting…** → **Connected**. Other states: **Reconnecting…**, **Used by another device**, **Connection failed**.\n\n**Binding the bot to you:**\n\n- After it connects, send the bot any private message once. The first person to DM the bot becomes its **Owner**; until then the card shows \"Waiting for first private-message bind\".\n- DMs from anyone other than the owner are silently ignored.\n- The bound card shows the service, App ID, and owner, with **Reconnect Bot** and **Unbind**. Unbinding disconnects the bot and removes the saved App ID / App Secret / owner from this computer.\n\n**New conversation settings:**\n\n- **New Conversation Configuration** in the same card picks the Agent and model for new Feishu conversations (or **Automatic**, which uses an available recommended model). Existing chats switch on their next message unless their model was changed manually.\n- **Permission mode for new group tasks** sets the permission mode for every session created in a group chat (@-ing the bot, `/new`, `/ctr`). It defaults to **Auto-review**. Group messages can come from other members, so choose **Full access** only for groups you trust. Use `/permission` in the group to change an existing one.\n\n**Notifications (shown after the bot is bound):**\n\n- **Online/offline announcement → Message notification** — the bot posts a message when this computer connects or disconnects.\n- **Session notifications → Send a Feishu / Lark DM when a session finishes or needs a reply** — DMs you only while the Cindy window is unfocused. You can turn it on only after the owner is bound.\n- **System Notifications** (desktop pop-ups) is a separate toggle in **Settings > General > Notifications** — see the language and notifications topic.\n\n**Chat commands:**\n\n- `/new` starts a fresh context, `/model` switches the model, `/permission` changes the permission mode, `/ctr` takes over a desktop session (`/exctr` ends the takeover), and `/help` lists the commands.\n- Send `!stop` to stop the current run and clear queued messages. The session is kept, so you can continue it.\n\n**Group chats:**\n\n- In a group, @ the bot to ask a question. The bot opens a new thread (话题) for the reply, so the main chat stays clean and each thread is its own session.\n- When @-ed in the main chat, the bot reads back through recent group messages (newest first, up to about 250) and stops once they're no longer relevant to your question. It answers with that history as context, including images and files shared in the chat.\n- @-ing the bot inside an existing thread uses that thread's own history as context.\n- These sessions appear in the sidebar's Chat group: thread sessions are titled like `[飞书·{group name}·{topic summary}] {id suffix}`, group sessions `[飞书·群] {group name}`, and DMs `[飞书·DM] {summary}`. Lark bots use \"Lark\" instead of \"飞书\".\n\n**Permissions for group features:**\n\nGroup chat context needs extra app permissions that are not part of the default setup:\n\n1. Open the **Feishu (or Lark) Open Platform** → Developer Console → your app → **Permissions & Scopes** (权限管理).\n2. Search for and enable **both** of these (searching by the scope ID works in any console language):\n - **\"Read all messages in associated group chat\" (获取群组中所有消息, `im:message.group_msg`)** — lets the bot read chat history.\n - **\"Read user and bot messages in group chats\" (获取群组中用户和机器人消息, `im:message.group_msg.include_bot:read`)** — includes messages from other bots in that history.\n These are sensitive permissions: on Feishu they are only available to self-built apps (自建应用), not store apps.\n3. Create and publish a new app **version** (self-built apps need admin approval).\n\nWithout these permissions the bot still answers when @-ed, but without chat context, and it DMs the owner a short notice with a link to the app's permission page. Optionally, enabling **\"Obtain basic group information\" (获取群基本信息, `im:chat:readonly`)** lets session titles show the real group name instead of an ID suffix.\n\n**Notes:**\n\n- **Used by another device** means this App ID is already connected on another device or Cindy instance (for example, running two Cindy builds at once). Unbind it on one side to clear the conflict.\n- A bot can live in only one place. You can also give a bot to a teammate from that teammate's **Messaging platforms** page. Choosing the bot already set up in Cindy moves it out of Settings > IM bots, and its chats continue with the teammate.\n- Settings > IM bots also has an **Official** section (Slack, Telegram, and X through Cindy's official bots, no credentials needed). The **Personal** section also has Personal WeChat, WeCom, DingTalk, Discord, and Telegram bots. See the integrations topic.",},{id: "getting-started",title: "Getting started with Cindy",- summary: "Create a session, choose an agent and working directory, send your first prompt, and iterate.",- content: "Cindy is a desktop app that runs AI coding agents (Claude Code and Codex) for you.\n\n**Your first session:**\n\n- Click **+ New** in the top of the left sidebar.\n- Choose the **agent** (Claude Code or Codex).\n- For code work, choose a **working directory** — the agent runs its file ops there. (You can skip this for plain-Q&A sessions.)\n- Optionally pick a **model** in the composer toolbar.\n- Type your goal and send.\n\n**Iterating:**\n\n- A session is a persistent chat thread: ask, review the result, then ask the next thing.\n- While the agent is replying, the **Send** button becomes **Stop** — click it to interrupt. You can keep typing in the meantime; follow-ups queue up and send in order.\n- Type `/` in the composer to open the slash-command palette (see the Commands topic).\n\n**Organizing:**\n\n- All sessions are listed in the left sidebar (see the Sidebar topic). Switch between them by clicking; each session remembers its agent, model, and working directory.\n- Keep unrelated tasks in **separate** sessions — sharing one session for everything dilutes context.\n\n**Tone tip:**\n\n- Describe the goal and context first, then what result you want. That works better than a single short command. The agents respond well to \"here's what I'm trying to do, here are the constraints, here's what 'done' looks like\".",+ summary: "Start a new session with New (⌘N / Ctrl+N), choose a project and a model (the engine — Claude Code, Codex, or Pi — follows the model), send your first prompt, and iterate.",+ content: "Cindy is a desktop app that runs AI agents (Claude Code, Codex, and Pi) for you.\n\n**Before your first message:**\n\n- You need at least one model. If none is available, Cindy prompts you to sign in to Cindy (Cindy AI, the official service) or connect a provider with your own API key. You can do this later in **Settings > Providers**.\n\n**Your first session:**\n\n- Click **New** (the ⊕ icon) at the top of the left sidebar, or press ⌘N / Ctrl+N.\n- Use the folder picker (**Chat or select project**) to choose a project folder for code work, where the agent reads and edits files. Choose **Chat** for plain Q&A that doesn't need a project.\n- Pick a **model** in the composer's model selector. There's no separate agent picker: the engine (Claude Code, Codex, or Pi) follows the model you choose and is shown next to it (see the Choosing a model topic). A new session starts with the model you last used.\n- Optionally set a **permission mode** in the same toolbar. If you have remote devices connected, you can also choose which device runs the session.\n- Type your goal and send. By default **Enter** sends and **Shift+Enter** adds a new line. You can switch to ⌘↵ / Ctrl+Enter under **Settings > General > Message input**.\n\n**Iterating:**\n\n- A session is a persistent chat thread: ask, review the result, then ask the next thing.\n- While the agent is working, the composer shows a separate **Stop** button; click it to stop the current run.\n- You can keep typing while the agent is busy: the send button becomes **Queue message**, and queued messages go out in order after the current turn. With the default Enter-to-send setting, press ⌘↵ (Ctrl+Enter on Windows/Linux) instead to steer the running turn without stopping it (see the Sessions and the chat thread topic).\n- Type `/` in the composer to open the slash-command palette (see the Commands topic).\n\n**Organizing:**\n\n- All sessions are listed in the left sidebar (see the Sidebar topic). Click one to switch to it. Each session remembers its engine, model, and working directory.\n- Keep unrelated work in **separate** sessions. Putting everything in one session dilutes the context.\n\n**Tone tip:**\n\n- Describe the goal and context first, then the result you want. That works better than a single short command. The agents respond well to \"here's what I'm trying to do, here are the constraints, here's what 'done' looks like\".",},{id: "import-sessions",title: "Importing Codex and Claude conversations",- summary: "Scan and import existing Codex or Claude Code chat history from Settings > Session Import.",+ summary: "Import existing local Codex or Claude Code history, or a shared .cshare session file, from Settings > Session Import.",tab: "import",- content: "If you already have Codex or Claude Code conversations on disk (from the CLIs), Import brings them into Cindy so they show up in the sidebar alongside your other sessions.\n\n**Importing:**\n\n- Open Settings > Session Import.\n- Click **Scan** to discover sessions in the standard Codex / Claude Code locations on your machine.\n- Filter the results:\n - by **source** — Codex or Claude Code;\n - by **placement** — by project (sessions inside a specific working directory) or by chat (free-form sessions not tied to a project).\n- Tick the sessions or projects you want, then click **Import**.\n- They appear in the left sidebar.\n\n**Notes:**\n\n- Importing is **one-way and non-destructive** — the originals on disk are left in place; nothing is synced back. Re-importing the same session is a no-op.\n- Imported threads are read-only history at first; you can continue them just like any other Cindy session.\n- This is for bringing in past conversations — it's unrelated to archiving / deleting sessions (see the Sidebar topic for those).",+ content: "If you already have Codex or Claude Code conversations on this computer (from the CLIs or the Codex app), Session Import brings them into Cindy so they show up in the sidebar with your other sessions.\n\n**Importing local history:**\n\n- Open **Settings > Session Import** (the page is titled **Local Session Import**).\n- Cindy scans the standard Codex and Claude Code history locations when you open the page. Click **Scan** to rescan. Scanning is read-only: it doesn't change your database or the sidebar.\n- Filter the results:\n - **Source** — All / Codex / Claude;\n - **Placement** — All / **Existing projects** (sessions that will go under a project group) / **Chat** (Codex sessions without a project).\n- Tick individual sessions, a whole project group, or **Select all in current filter**, then click **Import selected**.\n- A toast reports how many sessions were inserted or updated. They appear in the left sidebar.\n\n**Importing a shared session (.cshare):**\n\n- Click **Import from .cshare…** on the same page, or drag a `.cshare` file onto the Cindy window.\n- If the file is encrypted, enter the password from the sender. Then review the preview, choose the matching working directory on this computer, and optionally create it in a worktree.\n- If the same session already exists here, you can **Overwrite & import** or cancel.\n- To create a `.cshare` file yourself, open a session's menu (right-click it in the sidebar, or use its ⋮ menu) and choose **Export session…**. You can protect the file with a password. Exporting a collaboration lead includes its workers. Sessions that run on a remote machine or another device can't be exported.\n\n**Notes:**\n\n- Importing local history is **one-way and non-destructive**: the original files stay where they are, and nothing is synced back.\n- Sessions that are already in Cindy, plus internal sessions, are skipped and don't appear in the list (the summary counts them as **Skipped**).\n- You can continue imported sessions just like any other Cindy session.\n- Only import shared files from people you trust. The imported history can influence the agent, and a continued session can still read and write local files and run commands.\n- This is for bringing in past conversations. It's unrelated to archiving or deleting sessions (see the Sidebar topic for those).",},{id: "integrations",title: "What Cindy integrates with",- summary: "Supported external services — FeiShu, Slack, Telegram, Discord, Google, Jira / Confluence, GitHub, GitLab — and what's not supported.",- content: "At a glance, Cindy integrates with:\n\n- **FeiShu (Lark)** — DM-style bot notifications and interaction (Settings > IM bots). See the FeiShu bot topic. (FeiShu is no longer a sign-in method for Cindy itself.)\n- **Slack** — Settings > IM bots (the \"Cindy\" tab) manages workspace connections. With a compatible server, **Slack Communications on This Device** enables agent tools (search / read history / post as the authorized user) independently on multiple devices; Bot chats and tasks still go to a single receiving device. Enabling communications does not change that receiver.\n- **Telegram** — link your Telegram account to the official Cindy bot under Settings > IM bots (the \"Cindy\" tab). No bot token is required. Telegram tasks use the same local projects and Cindy sessions as Slack, while keeping channel-specific project and model preferences separate.\n- **Discord** — a Discord bot you bring yourself (your own Discord App + bot token) under Settings > IM bots.\n- **Google** — Gmail / Calendar / Drive / Sheets, via the Filo Google plugin (Plugins > Filo Google).\n- **GitHub / GitLab** — via the Cindy GitHub and Cindy GitLab plugins (Plugins).\n- **The coding agents** — Claude Code and Codex can reach pay-per-token models through **Cindy AI** in Settings > Model Providers.\n\n**Not supported:**\n\n- There is no built-in Microsoft Teams integration. For chat notifications, use the FeiShu, Slack, Telegram, or Discord bot.\n- There's no native Linear integration as a first-party feature — those would have to come via an MCP a user installs themselves.\n\n**Notes:**\n\n- Most of these connections are accessed by the agents through their tools (plugins / MCPs), not directly by the desktop app. Connecting on the plugin detail page is what gives its tools access to your account.",+ summary: "Supported external services — chat apps (Slack, Telegram, X, Feishu / Lark, WeChat, WeCom, DingTalk, Discord), Google (Gmail / Drive / Calendar / Sheets plugins), Jira / Confluence (XD Atlassian, some organizations), GitHub, GitLab, custom MCP servers — and what's not supported.",+ content: "At a glance, Cindy integrates with the services below.\n\n**Chat apps (Settings > IM bots):**\n\nThe IM bots page has two sections:\n\n- **Official** — Cindy's own bots. You don't create a bot or enter credentials; just turn on the channel's switch and follow the linking steps. This section isn't available in local mode.\n - **Slack** — bind one or more Slack workspaces and send tasks to Cindy from Slack. Each workspace can also turn on **Slack Communications on This Device** where supported (see the Connections topic).\n - **Telegram** — link your Telegram account to the official Cindy bot with **Link Telegram**. No bot token is required.\n - **X** — link your X (Twitter) account and send tasks by mentioning the Cindy bot on X (shown when the server offers it).\n - Official channels share this computer's project list and sessions, while each channel keeps its own account and project / model preferences.\n- **Personal** — bots you create and manage yourself, each with its own credentials and settings:\n - **Personal WeChat** — scan to connect and use Cindy from a private chat.\n - **WeCom Intelligent Bot** — Bot ID + Secret. Also hosts **WeCom group notifications** (a group-bot webhook for automation completion alerts).\n - **Feishu / Lark Bot** — App ID + App Secret. See the Feishu bot topic.\n - **DingTalk Bot** — Client ID + Client Secret (Stream mode), for DMs and group chats.\n - **Discord Bot** — your own bot token + your Discord user ID, for Discord DMs.\n - **Telegram Bot** — your own BotFather token + your Telegram user ID, for DMs and group chats.\n- Each channel has a **New Conversation Configuration** that picks the Agent and model for new conversations from that channel.\n- Teammates can also have their own bots on their **Messaging platforms** page. A bot can be connected in only one place.\n\n**Accounts and tools (Plugins):**\n\n- **Google** — four separate built-in plugins: **Gmail**, **Google Drive**, **Google Calendar**, and **Google Sheets**, each authorized separately. Some organization accounts get a single combined **Filo Google** plugin instead.\n- **GitHub / GitLab** — the built-in **GitHub** and **GitLab** plugins.\n- **Jira / Confluence** — the **XD Atlassian** enterprise plugin, only for organizations that provide it.\n- Other built-in plugins include **Art**, **Web Search**, **Mermaid**, and **Notion** (see the Plugins topic).\n- Connect accounts on each plugin's detail page. That connection is what gives the plugin's tools access to your account (see the Connections topic).\n\n**Models:**\n\n- The agents can use pay-per-token models through **Cindy AI** (the official service) or through your own providers and API keys, all under **Settings > Providers** (see the Providers topic).\n\n**Custom MCP servers:**\n\n- Add any remote MCP server (http / sse) under **Settings > Tools > External tools > Add MCP server**. Use this for services without a built-in integration.\n\n**Not supported:**\n\n- There is no built-in Microsoft Teams integration. For chat, use one of the IM bots above.\n- There's no first-party Linear integration. Add a Linear MCP server yourself under **Settings > Tools > External tools** if you need it.\n\n**Notes:**\n\n- The agents mostly reach these services through their tools (plugins / MCPs), not through the desktop app directly.",+ },+ {+ id: "keyboard-shortcuts",+ title: "Keyboard shortcuts and accessories",+ summary: "Settings > Keyboard Shortcuts lists Cindy's default shortcuts and lets you rebind, clear, or reset them; Accessories sets up Work Louder keyboards and Xbox, PlayStation, Switch, or generic game controllers.",+ content: "Open **Settings > Keyboard Shortcuts** to see and change Cindy's app shortcuts. Below, **Mod** means ⌘ on macOS and Ctrl on Windows / Linux.\n\n**Default shortcuts:**\n\n- **New session** — Mod+N\n- **Toggle sidebar** — Mod+B\n- **Cycle permission mode** — Shift+Tab (while the message box is focused)\n- **Previous / Next right sidebar tab** — macOS: ⌘⇧[ / ⌘⇧] or Ctrl+Shift+Tab / Ctrl+Tab; Windows / Linux: Ctrl+PageUp / Ctrl+PageDown or Ctrl+Shift+Tab / Ctrl+Tab\n- **Open terminal** — Ctrl+` (opens a terminal in the right sidebar, or focuses the existing one)\n- **Find in page** — Mod+F\n- **Search in project** — Mod+Shift+F (while browsing documents)\n- **Zoom in / Zoom out / Reset zoom** — Windows / Linux: Ctrl+= / Ctrl+- / Ctrl+0. On macOS, zoom uses the standard View menu shortcuts and isn't listed here.\n- Built-in browser: **Focus address bar** Mod+L, **Browser back** / **Browser forward** Alt+← / Alt+→ (Option on macOS), **Reload page** Mod+R or F5\n\nThese also work but aren't shown on the page: Mod+1 … Mod+9 jump to the first nine visible sessions in the sidebar; Mod+S saves the document you're editing; Mod+W closes the active right sidebar tab when focus is there (on macOS it otherwise closes the window); ⌘, opens Settings on macOS.\n\n**Changing a shortcut:**\n\n- Click the pencil on a row, press the new key combination, or click **Cancel**. The trash button removes the binding (the row shows **Not set**).\n- Cindy rejects combinations that can't be used, are **Reserved by the operating system**, conflict with another Cindy shortcut (it names which one), or conflict with the voice input shortcut.\n- A changed row gets **Reset to default**; **Reset all to defaults** appears at the top once anything is customized.\n- Other shortcuts live elsewhere: the voice input shortcut is in **Settings > Voice Input** (see the voice input topic), and the send key (**Use ⌘↵ to send**) is under **Settings > General > Message input**.\n\n**Accessories (keyboards and controllers):**\n\n- Open **Accessories** at the top of the page. A device also gets its own entry on the main page once it's connected or turned on.\n- **Work Louder Codex Micro** and **Work Louder Creator Micro 2** — open tasks from the keys and show their status with key lighting.\n - Connect with a data-capable USB cable and turn on **Enable**. Cindy uses the official Work Louder software that comes with a supported ChatGPT or Codex app, so keep one installed; close ChatGPT if it's holding the device. On macOS, allow Cindy under **Input Monitoring**.\n - **Lighting**: **Brightness** and **Auto-dim**.\n - **Agent keys**: **Task order** (**Sidebar order**, **Recent tasks**, **Priority**, or **Custom**), **Focus Cindy with a single tap**, and each key's role — a **Task key** switches tasks and shows status (only the first 6 task keys can light independently), an **Action key** runs a shortcut.\n - Also **Microphone** (separate mic keys), **Joystick** directions, **Encoder** mode (switch tasks, navigate the composer, reasoning effort, scroll, or custom), the keycap **Layout** editor, and **Reset layout**.\n - **Codex compatibility** (macOS only): **Prevent Codex from occupying the device** while Cindy runs; quit and reopen Codex to apply.\n- **Xbox controller**, **PlayStation controller** (DualSense / DualShock), **Switch controller** (Switch Pro / Joy-Con), and **Generic controller** — scroll, talk, send, and switch tasks with a gamepad over USB or Bluetooth.\n - Turn on **Enable**; Cindy only listens while it's in the foreground.\n - Click a control to assign an action (a command, a skill, or **Voice input** — hold to talk, release to stop). A stick can **Scroll conversation** or use **Custom directions**. **Reset mapping** restores the defaults.",},{id: "language-and-notifications",title: "Display language and notifications",- summary: "Display language and desktop / FeiShu notifications — in Settings > General.",+ summary: "Display language and desktop notifications in Settings > General, plus where IM-bot notifications, Agent Island, and other General settings (app behavior, send shortcut) live.",tab: "general",- content: "Settings > General controls the display language and how the app notifies you. Theme, wallpaper, and fonts live in Settings > Appearance (see the appearance topic).\n\n**Language:**\n\n- Display language: System / English / 简体中文 / 繁體中文 / 日本語 / 한국어. Used for the interface and AI error reports. Other replies follow your message or language request. On mobile, the phone's display language is used for AI error reports for messages sent from that phone.\n\n**Notifications:**\n\n- **System Notifications** — notify when a session finishes, encounters an error, or needs a reply. Regular sessions notify only while the window is in the background. When Agent Island is enabled, it handles these alerts instead.\n- **FeiShu DM on session finished** — requires the FeiShu bot to be configured (see the FeiShu bot topic).\n\n**Notes:**\n\n- Language changes apply immediately, but a few static strings may need a window reload to update.",+ content: "Settings > General controls the display language, desktop notifications, and a few app-wide behaviors. Theme, wallpaper, and fonts live in Settings > Appearance (see the appearance topic).\n\n**Language:**\n\n- **Display language**: Auto (follow system) / English / 简体中文 / Traditional Chinese (繁體中文) / 日本語 / 한국어. It's used for the interface and AI error reports. Other replies follow your message or language request.\n- Changes apply immediately. With **Auto (follow system)**, Cindy reads the system language at startup, so restart Cindy after changing your computer's language.\n- On mobile, the phone app's display language is used for AI error reports for messages sent from that phone.\n\n**Notifications:**\n\n- **System Notifications** — notify when a session finishes, encounters an error, or needs a reply. Regular sessions notify only while the window is in the background.\n- **Agent Island** (macOS 14 or later, in its own **Settings > Agent Island** tab) shows running, needs-reply, and unread completed tasks at the top of the display. When it's on, it handles these alerts instead of system notifications.\n- **Feishu / Lark DMs** are no longer set here. Turn on **Send a Feishu / Lark DM when a session finishes or needs a reply** in the Feishu / Lark Bot card under **Settings > IM bots** (see the Feishu bot topic). It needs a bound bot and only sends while the window is unfocused.\n- **WeCom group notifications** (automation completion alerts via a WeCom group-bot webhook) are in the WeCom Intelligent Bot card under Settings > IM bots.\n\n**Other settings in General:**\n\n- **App Behavior** — **Keep computer awake** (so background agents and scheduled tasks keep running), **Start at Login** (off by default), **Show remaining usage in the menu bar** (macOS), **When closing the main window** (Windows: Minimize to tray / Quit Cindy; Linux: Minimize window / Quit Cindy), and **Prevent Accidental Clicks When Returning** (macOS / Windows).\n- **Message input** — **Use ⌘↵ to send** (Ctrl+Enter on Windows / Linux). When it's off (the default), Enter sends and Shift+Enter inserts a new line.\n- General also holds your account info, Teammates, Collaboration limits, Agent Resource Usage, Git Safety, Pi extensions, Experimental, and Logout.",},{id: "memory",title: "Memory across conversations",- summary: "The \"Cindy\" master switch turns on shared cross-agent memory (default on); when off, each agent uses its own native memory — all in Settings > Personalization > Memory.",+ summary: "Settings > Personalization > Memory — the \"Cindy\" switch (on by default) turns on shared memory per working directory; when it's off, Claude Code and Codex use their own native memory. Teammates such as Cindy keep their own memory, viewable in the teammate's settings.",tab: "personalization",- content: "Memory lets the agents carry context across your conversations — facts about you, your projects, and how you like to work. There are two kinds, and one master switch chooses between them.\n\n**Where it lives:**\n\n- Settings > Personalization > Memory.\n\n**The \"Cindy\" master switch (default ON):**\n\n- Turning **Cindy** on enables a **shared, cross-agent memory** scoped to each working directory — Claude Code and Codex sessions in the same directory draw on the same memory.\n- While it's on, the per-agent native toggles below are **disabled / grayed out**, because the shared memory supersedes them.\n- It's **on by default** for new installs.\n\n**Per-agent native memory (used when the master switch is OFF):**\n\n- When Cindy is off, each agent falls back to **its own native memory** — so turning the master off does not mean \"no memory,\" it means Claude Code and Codex each remember things their own way.\n- Separate toggles let you enable / disable native memory for **Claude Code** and **Codex** independently (these are only interactive while the master switch is off).\n\n**Resetting:**\n\n- **Per-agent reset** clears a single agent's native memory.\n- **Reset all Cindy memory** clears the shared cross-agent memory for every working directory.\n- A section-level **reset to defaults** restores the switches themselves.\n- Resets ask for confirmation and aren't undoable.\n\n**What memory is, what it isn't:**\n\n- It's the agents' **long-term memory across sessions** — notes about you and your projects.\n- It's **not** the in-session chat history (that's always persisted to the session itself).\n- It's **not** your Personalization custom instructions (those are standing instructions you write; memory is what the agents record about you).\n\n**Notes:**\n\n- Changes take effect on the **next** chat / agent turn — toggling mid-conversation doesn't retroactively scrub current context.\n- Memory is stored locally and does not migrate if you change machines or wipe the app's user data.",+ content: "Memory lets the agents carry context across your conversations — facts about you, your projects, and how you like to work. There are two kinds, and one master switch chooses between them.\n\n**Where it lives:**\n\n- Settings > Personalization > **Memory**.\n\n**The \"Cindy\" master switch (default ON):**\n\n- Turning **Cindy** on enables a **shared memory** scoped to each working directory — Claude Code, Codex, and Pi tasks in the same working directory draw on the same memory.\n- While it's on, the **Claude Code** and **Codex** native-memory rows below are **disabled / grayed out**, because the shared memory replaces them.\n- The **Pi** row is the opposite: it's only available while Cindy is on. It controls whether Pi saves its long-session compression summaries to Cindy memory for future Pi tasks.\n\n**Per-agent native memory (used when the master switch is OFF):**\n\n- When Cindy is off, Claude Code and Codex each fall back to **their own native memory** — so turning the master off doesn't mean \"no memory\"; it means each agent remembers things its own way.\n- The **Claude Code** and **Codex** rows each have their own switch (only usable while Cindy is off).\n\n**Resetting:**\n\n- The reset button on the **Cindy** row deletes all Cindy memories for every project.\n- The reset button on an agent row deletes all memories that agent has accumulated across every project.\n- When you've changed the switches, a **Custom** badge and a **Restore default** button appear next to the Memory title; it restores the switches (not the memories).\n- Resets ask for confirmation and can't be undone.\n\n**Teammates (such as Cindy) have their own memory:**\n\n- A teammate keeps its own memory, separate from the task memory above. To see or edit it, open the teammate's chat, click its name / avatar at the top to open its settings, and choose **Memory**. There you can search memories, open one to read or edit it, delete it, or turn **Use memory** off.\n\n**What memory is, what it isn't:**\n\n- It's the agents' **long-term memory across tasks** — notes about you and your projects.\n- It's **not** the chat history of a task (that is always saved with the task).\n- It's **not** your custom instructions in Personalization (those are standing instructions you write; memory is what the agents record).\n- To make the agent remember something specific, just tell it in the chat (for example \"remember that we use pnpm in this repo\").\n\n**Notes:**\n\n- After changing a memory switch you usually need to **start a new task** for it to take effect; running tasks keep the setting they started with (running Codex tasks switch after their current turn finishes).\n- Memory and these switches belong to your signed-in account on this computer. Memory is stored locally and doesn't move to another computer.",+ },+ {+ id: "mobile-app",+ title: "Using Cindy on your phone (mobile app, notifications, remote desktop)",+ summary: "Control your computer's Cindy tasks from the Cindy mobile app — no pairing, just sign in with the same account and allow device control; get push notifications when a task finishes, fails, or needs your reply; and view or operate your computer with Remote desktop.",+ tab: "remote-control",+ content: "The Cindy mobile app is a remote control for Cindy on your computer. The agents still run on the computer; the phone shows the same tasks and lets you work with them.\n\n**Connecting your phone (no pairing code):**\n\n1. Install the mobile app. On the computer, click the phone icon (**Download Cindy Mobile**) in the account area of the left sidebar to get a download QR code.\n2. Sign in on the phone with the **same Cindy account** as the computer.\n3. On the computer, turn on **Allow same-account devices to control this machine** (**Settings > Remote Connection**, or the switch in the download dialog).\n\nYour computer then shows up on the phone automatically. If the phone says the computer is offline, open Cindy on the computer and stay signed in; the phone reconnects on its own. Revoked phones can't reconnect until you restore their access in **Settings > Remote Connection** (see the Remote topic).\n\n**What you can do from the phone:**\n\n- Browse tasks by computer, open them, and start a **new task** (choose the computer, then a project or **Chat (no project)**, the agent, and optionally a worktree).\n- Send messages (messages sent while the agent is working are queued), **Stop**, attach photos, files, or screenshots, use voice input, and switch model, reasoning effort, and permission mode. **Plan Mode** and **Goal Mode** are available too.\n- Approve permission prompts, answer the agent's questions, and review plans.\n- Rewind (**Back to here**), fork (**Start new session from here**), ask in a side chat, and manage Collaboration workers.\n- Browse a task's files (read-only), use plugins, and chat with Teammates.\n- Share a task with other accounts, or join one (**Settings > Share Task > Manage Sharing** on the phone; see the Sharing tasks topic).\n- On iPhone, share content from other apps into Cindy; it's placed in the composer, not sent automatically.\n- The phone can't change the computer's global settings, sign-in, or API keys, and it can't be controlled itself.\n\n**Push notifications:**\n\n- On the phone, open **Settings** and turn on **Message and task notifications** (off by default; the system asks for permission). You're notified about chat messages and tasks that **finish, fail, or need your reply** (a question, permission prompt, or plan review).\n- **Show chat previews** controls whether notifications include message text.\n- No push is sent while you're watching that task live on another device, or while the Cindy window on the computer is focused. While the phone app is open, task notifications don't pop up as banners.\n- An automation's runs notify the phone only if that automation has desktop or Feishu notifications turned on.\n- There's no separate setting on the computer: the phone's switch is the only control.\n- On Android phones without Google push, turn on **Keep chat connected in background** to keep receiving chat messages (it shows an ongoing notification you can stop).\n\n**Remote desktop (see and operate the computer's screen):**\n\n- On the computer, turn on **Allow remote desktop** in **Settings > Remote Connection** (device control must also be on). On macOS, grant **Screen recording** (to view) and **Accessibility** (to control) when prompted; the phone can open the permission guide on the computer.\n- On the phone, tap **Remote desktop** in a task's header, on the computer's task list, or in its device details. Control is requested when you connect; switch on **View only** to just watch.\n- Phone tools include **Keyboard**, **Trackpad** / **Touch** input, **All windows** / **Desktop**, clipboard copy and paste between phone and computer, **Sound**, display quality and resolution, and a Security section (**Lock Computer on Exit**, **Computer privacy screen**, **Clipboard sync**, and on iPhone with a Mac, **Automatically unlock computer**).\n- From another computer on your account, open **Remote desktop** from the sidebar's device scope menu, the button next to a task that runs on that computer, or **Settings > Remote Connection**. It opens in its own window with **Controls** for display, sound, clipboard, and security.\n- Only one device can use a computer's remote desktop at a time; **Take over connection** disconnects the other device. The controlled computer shows who is viewing or controlling it, with **Disconnect**.\n\n**Notes:**\n\n- The phone has no agent of its own: if the computer is asleep, offline, or signed out, its tasks can't run. Turn on **Keep computer awake** in **Settings > General > App Behavior** to keep it reachable.\n- The Mainland China edition is a separate app with its own accounts (it signs in with a phone number). Use the same edition and the same account on the phone and the computer.",},{id: "models-and-compat",title: "Choosing a model",- summary: "Pick a session's model from the composer dropdown; non-Anthropic models work automatically (no compat toggle to flip).",+ summary: "Pick a task's model from the model selector in the composer; Cindy picks the engine (Claude Code / Codex / Pi) for you, and Customize lets you change engine, reasoning effort, and Fast. Compatibility-mode engines are off by default and are turned on per model in Settings > Providers.",tab: "providers",- content: "You pick the model a session uses from the **model dropdown in the composer toolbar**. In the dropdown, models are organized under the source providers you've connected. The brand groups they belong to are **Anthropic**, **OpenAI**, **Budget GPT**, **xAI**, **Google**, and **Domestic** (Kimi / GLM / Qwen / DeepSeek), plus non-chat model kinds (image / audio / video / …).\n\nThe default depends on the agent: Claude Code defaults to **Opus 4.8**, Codex defaults to a **GPT** model.\n\n**Switching mid-session:**\n\n- Open the dropdown and pick a different model. It takes effect on the **next turn**.\n\n**Using non-Anthropic models:**\n\n- You don't need to flip anything. Claude Code's SDK emits Anthropic-specific request fields (`thinking`, `cache_control`, etc.) that non-Anthropic backends would reject, so the app **always** routes requests through a local proxy that scrubs those fields per-model before forwarding. Claude / Anthropic models pass through untouched.\n- (This used to be a manual \"Compat Mode\" toggle in Settings. That toggle has been retired — the handling is now automatic and always on, so there's nothing to configure.)\n\n**Notes:**\n\n- Which models you can pick depends on the providers connected in **Settings > Model Providers** (see the Providers topic).\n- \"Budget GPT\" is a lower-cost routing tier for GPT models; \"Domestic\" groups the China-hosted models.",+ content: "You pick the model a task uses from the **model selector in the composer toolbar** (the pill showing the current model). It works the same for a new task and inside an existing one.\n\n**The model panel:**\n\n- A search box at the top; a rail on the left with **Recent**, **Favorites**, **All**, and one icon per connected provider (source); the model list on the right; and **Configure models** at the bottom to connect more sources.\n- You choose a **model**; Cindy automatically picks the recommended **engine** (Claude Code, Codex, or Pi) for it and shows it in the pill.\n- Hover a model row and click **Customize** (or right-click the row) to change its engine, **Reasoning Effort**, and **Speed** (Fast, where supported). **Reset to recommended** undoes your changes.\n- Click ☆ on a row to **Save to favorites** — a favorite keeps that exact setup (source, model, engine, effort, Fast). **Recent** lists the last few setups you used.\n\n**Default model for a new task:**\n\n- It depends on what you've connected: with a ChatGPT (OpenAI) subscription it's **GPT-6.1 Sol** on Codex; with a Claude (Anthropic) subscription it's **Opus 5.5** on Claude Code; otherwise it's Cindy AI's **Auto** on Pi. After that, a new task starts with the model you last picked.\n\n**Switching mid-task:**\n\n- Open the selector and pick a different model. It applies from your **next message**; while a switch is pending, the pill's tooltip shows \"Current\" and \"Next message\".\n- If the new model runs on a **different engine**, Cindy asks you to confirm (**Switch Agent?**). The new engine continues from a handoff summary of the conversation, so some reasoning or details may not carry over perfectly.\n\n**Using non-Anthropic models:**\n\n- There's no global compatibility toggle to flip. Each model lists the engines that support it natively, and the recommended one is chosen automatically.\n- Some engines can only run a model in **Compatibility mode** (requests are converted). These are **off by default**, because they can be unstable. To turn one on, open **Settings > Providers**, open the model's **Advanced settings**, and enable that engine under **Engine support**. It is then shown with a \"Compatibility mode\" label in the model's Customize options.\n\n**Notes:**\n\n- Which models you can pick depends on the sources connected in **Settings > Providers** (see the Providers topic). Use **Configure models** at the bottom of the panel to get there.\n- Models from your other computers or shared with you can also appear in the panel, labeled with where they run.",},{id: "personalization",- title: "Custom instructions (personalization)",- summary: "Standing instructions the agents follow across every session — in Settings > Personalization.",+ title: "Custom instructions and the Personalization page",+ summary: "Settings > Personalization — custom instructions the agents follow in every new task, plus Memory, Subagents, Auxiliary models, Vision Bridge, Smart Contacts, context compaction, default terminal shell, where links open, and Tips toggles.",tab: "personalization",- content: "Personalization is where you give the agents standing instructions that apply to **every** session — tone, language, conventions, things you want the agent to always keep in mind.\n\n**Setting it up:**\n\n- Open Settings > Personalization.\n- Write your instructions in the text area, then save.\n- Useful examples: \"respond in Chinese\", \"default to TypeScript\", \"prefer pnpm over npm\", \"always check tests before reporting done\".\n\n**Other things on this page:**\n\n- **Memory** — long-term memory across sessions (see the Memory topic).\n- **Subagent models** — choose which models the agents' subagents use.\n- **Tips** — a few functional toggles (e.g. silent encrypted retry, chat embedding).\n\n**Notes:**\n\n- Changes apply to **new turns**; they don't rewrite previous messages in current sessions.\n- Per-session overrides (a stricter rule for one specific session) are best done in that session's first message, not here.\n- Don't put secrets here — instructions are sent with every prompt.",+ content: "Personalization is where you give the agents standing instructions that apply to **every** new task — tone, language, conventions, things you want the agent to always keep in mind. The same page also holds several agent-behavior settings.\n\n**Custom instructions:**\n\n- Open **Settings > Personalization**. The first section (**Personalization · Custom Prompts**) has a text box for your instructions.\n- Write your instructions, then click **Save**. The limit is 8,000 characters; a counter shows how much you've used.\n- They're appended to the end of the agent's system prompt in Cindy's own tasks.\n- Useful examples: \"respond in Chinese\", \"default to TypeScript\", \"prefer pnpm over npm\", \"always check tests before reporting done\".\n\n**Other sections on this page:**\n\n- **Memory** — long-term memory across tasks (see the Memory topic).\n- **Subagents** — the default model Claude Code uses for subagents, and **Codex smart routing** (lets Codex pick from more connected models per subtask; off uses Codex's native assignment).\n- **Auxiliary models** — the small models used for background chores such as automatic task naming, composer input suggestions, the Help assistant, pinned-task summaries, and IM titles. Keep **Automatic recommendation**, or choose **Custom** to set a **Preferred** model and up to two fallbacks.\n- **Vision Bridge** — lets text-only models (e.g. DeepSeek) read images: images are first described by a multimodal model you choose (a **Primary** plus an optional **Backup**). Turn on **Enable Vision Bridge**, then pick which models route through it.\n- **Smart Contacts** — a people / organization directory the agents can look up and add to as they work (uncertain entries wait in **Pending** for your review). You can browse, search, and edit entries here, and optionally turn on **Sync across my devices**.\n- **Context compaction** — the context-usage percentage at which Claude Code and Pi automatically compact a long conversation (Codex sets its own threshold).\n- **Default terminal shell** — the shell used for new terminal tabs in the right sidebar (existing tabs aren't affected).\n- **Open Links** — whether links in messages open in Cindy's **Built-In Browser** or your **System Default Browser**, set separately for web pages and local pages. Right-click a link to pick another option just for that link.\n- **Tips** — a few functional toggles: **Input Suggestions** (suggests your next message after a reply; press Tab to insert it), **Silently drop protocol-encrypted content and retry** (helps when switching providers mid-task; on by default), **Enable chat semantic indexing** (not available in local mode), and **Show question navigation rail** (tick marks beside the chat to jump between your questions).\n\n**Notes:**\n\n- Custom instructions apply to **new tasks**; tasks that are already running keep the instructions they started with and don't rewrite earlier messages.\n- They apply only to Cindy's built-in tasks, not to Claude Code, Codex, or Pi you run yourself in a terminal or another IDE.\n- Custom instructions are saved on this computer for your signed-in account; they don't sync to other devices.\n- For a stricter rule in one specific task, put it in that task's first message instead.\n- Don't put secrets here — the instructions are sent to the model with every task.",},{id: "plan-mode",- title: "Reviewing and approving plans",- summary: "When the agent proposes a plan, approve it, edit it, or send feedback before it proceeds.",- content: "For larger tasks the agent may first propose a **plan**, shown as a card in the chat. You decide whether to approve, edit, or send feedback before any code runs.\n\n**Card controls:**\n\n- **Approve** — the agent receives the (possibly edited) plan and starts executing.\n- **Feedback** — opens a text row; what you write is sent back to the agent as a denial reason. The agent can then propose a revised plan or take a different path.\n- **Edit** — open the plan markdown in the card and edit it directly. Edits are auto-saved (debounced) to a plan file on disk; on Approve the edited version is what the agent uses.\n- **Cancel** (the toolbar **✕** or **Esc**) — dismiss the review and end the plan loop without approving or sending feedback.\n- **Expand / Half / Minimize** — change the card's display size to read or skim it.\n\n**How it appears:**\n\n- The agent emits a `plan_review` event with the plan markdown; the app renders the card in the chat thread and pauses the agent until you respond.\n\n**Notes:**\n\n- For Claude Code, this card is exactly how the SDK's `ExitPlanMode` tool surfaces: the app intercepts that tool call into a `plan_review` interaction and, on Approve, auto-exits plan mode. Cindy's renderer ↔ main `resolveInteraction` layer is what renders the card and returns your decision.\n- An approved plan card stays in your chat history with an \"approved\" state; you can scroll back to it.\n- If you close the window while a plan card is open, the agent is still paused waiting for your decision — reopen the session to resolve it.",+ title: "Plan mode and reviewing plans",+ summary: "Turn on Plan mode from the composer's \"+\" menu (or type /plan) so the agent proposes a plan first; then approve it, edit it, send feedback, or dismiss it before any work starts.",+ content: "In **Plan mode** the agent first proposes a **plan** and waits for your approval before it starts doing the work. It's available for Claude Code, Codex, and Pi tasks.\n\n**Turning Plan mode on:**\n\n- Click the **+** button in the composer (or type `@`) and choose **Plan mode**, or type `/plan` and pick it.\n- While it's on, a **Plan mode** chip appears above the composer. Click the ✕ on the chip (**Exit plan mode**) to turn it off.\n- After you approve a plan, Plan mode turns off automatically and the agent starts executing. To plan again, turn it back on.\n\n**Reviewing the plan:**\n\nWhen the plan is ready, a **Review the agent's plan** card replaces the composer at the bottom of the task, with the plan and a **Content** outline beside it.\n\n- **Approve the plan and start coding** — click it or press **Enter**. The agent receives the plan (including your edits) and starts working.\n- **Tell the agent what to do instead** — click this row, type your feedback, and press Enter or click send. The agent revises the plan and shows it again.\n- **Edit plan** (toolbar) — edit the plan's Markdown directly. Edits are saved automatically, and Approve uses the edited version. **Exit edit mode** goes back to the formatted view.\n- **Dismiss this plan** (the toolbar **✕**, or **Esc**) — closes the review without approving or sending feedback; the agent doesn't start the work.\n- **Expand / Collapse to half / Minimize** — change the card's size so you can read the chat above it; a minimized card can be restored from its bar.\n\n**In the chat history:**\n\n- Each plan stays in the task as a **Plan review** entry marked **Approved**, **Revision requested** (with your feedback), **Cancelled**, or **Expired**. Use **Show full plan** to read an approved plan again.\n- If the agent's turn is stopped or the task closes before you decide, the review is marked **Expired** and can no longer be answered — send a new message to continue.\n\n**Notes:**\n\n- You can also review and approve plans from the Cindy mobile app when you control this computer from your phone.",},{id: "plugins",title: "Plugins (extending what the agents can do)",- summary: "What plugins are, how to browse / install / enable / disable them, per-working-directory control, the permission model, and invoking one with a $command.",+ summary: "Where the Plugins page is, the built-in plugins, installing (marketplace, .cindy file, custom marketplaces), enabling / disabling globally or per project, updates, uninstalling, account / key setup, $commands, panels, and the permission model.",tab: "ghosts",- content: "Plugins (internally \"意识\" / Ghost) extend what the agents can do — they add AI-callable tools, panels that live as pinned tabs in the right sidebar, full-page app views, and \"ask Cindy to make it\" media generation. Each plugin is a self-contained `.cindy` capability pack that runs in its own isolated sandbox.\n\n**The Plugins page:**\n\n- Open it from the **Plugins** entry in the left sidebar.\n- Browse by origin: **all / built-in / team-shared / external**. The team-shared (enterprise) group only shows if your account belongs to an organization.\n- Each plugin has a detail page — that's where you enable it, connect an account / enter an API key, and read what it can do.\n\n**Built-in plugins that ship with the app:**\n\n- **Cindy Art** (`$cindy-art`) — generate and edit images and short videos.\n- **Cindy Web Search** (`$cindy-web-search`) — public web search. **No key needed**: by default it searches through Cindy AI and uses your signed-in Cindy account's balance (unavailable in local mode). Results come with a title, link, and short summary. To search through your own service instead, add a Brave, Tavily, or DeepSeek key on its detail page. A failed search is reported as-is; it never silently switches to another source.\n- **Cindy Mermaid** (`$cindy-mermaid`) — build / repair Mermaid diagrams.\n- **Cindy GitHub** (`$cindy-github`) / **Cindy GitLab** (`$cindy-gitlab`) — read / write issues, PRs / MRs, reviews, CI, releases.\n- **Filo Google** (`$filo-google`) — Gmail / Calendar / Drive / Sheets.\n\n**Enabling and disabling:**\n\n- **Enable / disable (globally)** from the plugin's detail page. Disabling puts its tools and panel to sleep but keeps the plugin installed — its right-sidebar tab is kept and comes back when you re-enable it. A plugin's panel shows up as a pinned tab in the right sidebar that stays across tasks; close it from the tab, and reopen it later by right-clicking the plugin's entry in the left sidebar and choosing \"Open in the sidebar\".\n- **Disable for one working directory only** — a plugin can be turned off just for a specific project. In that project's sessions it behaves as if it isn't installed (its tools and `$command` are unavailable there); everywhere else it stays on.\n\n**Installing:**\n\n- Install from the marketplace, or add a `.cindy` file manually (drag it in / pick it from disk).\n- Choosing Install, dragging a package, or picking one from disk first shows a **confirmation dialog** listing everything the plugin declares it can do (network domains, credentials, background tasks, skills, bundled code, panels…). Nothing is installed until you confirm; after that the plugin is enabled right away.\n- When you ask an agent to install a plugin (`ghost_market_install` / `ghost_forge_install`), the same confirmation appears as a card in that task — whatever the task's permission mode, including Full Access. It also works from the phone when you control this computer remotely. If you deny it, nothing is installed.\n- Plugins pushed by your organization (server default installs) are installed and updated silently without a confirmation.\n- The server selects the release appropriate for this Cindy version before marketplace delivery. Desktop does not apply a second `minCindyVersion` gate; custom marketplaces and local `.cindy` packages follow the same install policy.\n- Updates from stable marketplace sources are applied automatically when the plugin is idle — **as long as the new version doesn't need more permissions**. If it does, the automatic update pauses, the old version keeps working, and the plugin card shows \"New permissions need confirmation\"; clicking Update shows the added permissions for you to confirm. Failed attempts retry later; you can still click Update to retry immediately.\n\n**Creating or updating your own plugin:**\n\n- Choose **Create with Cindy** from the Plugins page's add menu. Cindy opens a new-task draft with the built-in Plugin Creator's actual command, selecting the built-in version even if a user Skill has the same name. Describe what you want and send it using your chosen Agent and model.\n- The built-in `cindy-plugin-creator` Skill supports creating, editing, debugging, packaging, and publishing plugins. Its bundled references contain the authoring documentation for the current Cindy version. If it is unavailable, the page reports that and keeps your existing draft.\n- Creating and packaging a plugin produces a `.cindy` file. Installation and publication are separate actions that you can request when ready.\n\n**Using a plugin with `$command`:**\n\n- Type its command at the **start** of a message — e.g. `$cindy-art a poster for…`. Only enabled plugins that declare that command match.\n- You don't have to use `$command`: the agents can also pick an enabled plugin's tools on their own when a task calls for it.\n\n**The permission / sandbox model (why plugins are safe):**\n\n- A plugin's sandbox has **no direct filesystem, network, or system access**. The host mediates every operation and enforces its actual execution context.\n- Autonomous/background plugin behavior must stay within the capabilities and scopes declared directly in its manifest. Whether an in-flight plugin tool runs is decided by the existing Agent authorization flow; network and working-directory operations bind to that call, while bundled code and CLIs continue to use the existing Node worker. Plugin authors do not need a client change or a manifest Slot to pre-register a specific command, host, or path.\n- **Autonomous network** access is limited to a domain allowlist the plugin declares; the host proxies every request. Credentials never enter the sandbox — the host injects them only into declared requests whose target matches that declaration.\n- Because of this, a plugin **can't see your files** on its own. A file reaches a plugin only when you or the agent explicitly hand one over (an image attachment, or a directory the host passes through after an ownership check).\n\n**Configuring a plugin's account / API key:**\n\n- Do it on the plugin's **detail page**. Plugins that use a key (optional search-service keys for Web Search, GitHub/GitLab tokens) have a field there; the value goes straight into the host's encrypted store and can't be read back.\n- Plugins that use OAuth (for example Filo Google) show a **Connect account** button instead.\n\n**Notes:**\n\n- Enabling or disabling a plugin takes effect on the **next** agent query, not retroactively in the current turn.\n- Skills (`/` slash commands) and Plugins (`$` commands) are two tabs of the same management surface — see the Skills topic for skills.",+ content: "Plugins extend what the agents can do — they add tools the agent can call, panels that live as tabs in the right sidebar, full-page app views, and integrations with outside services. Each plugin is a self-contained `.cindy` package that runs in its own isolated sandbox.\n\n**The Plugins page:**\n\n- Open it from **Plugins** in the left sidebar, or from **Settings > Plugins**. **Plugins** and **Skills** are the two tabs at the top of the same page (see the Skills topic for skills).\n- **Installed** lists your plugins; **Recommended** lists plugins from the marketplace, filtered by source: **All / Public / Team shared / Custom**. **Team shared** only appears if your account belongs to an organization.\n- Use **Search plugins** to find one. Each card has a quick action — **Use** opens its panel, **Chat** starts a new task with the plugin, **Manage** (or clicking the card) opens its detail page.\n- The detail page is where you enable / disable it, connect an account or enter an API key, see what it contains and its version, view **Your calls** (usage), and uninstall it.\n\n**Built-in plugins:**\n\n- **Art** (`$cindy-art`) — create and edit images or generate short videos.\n- **Web Search** (`$cindy-web-search`) — public web search. **No key needed**: by default it searches through Cindy AI using your signed-in account's balance (not available in local mode). To use your own service instead, add a Brave, Tavily, or DeepSeek key on its detail page. A failed search is reported as-is; it never silently switches to another source.\n- **Mermaid** (`$cindy-mermaid`) — fix and normalize Mermaid diagrams written by the agent.\n- **GitHub** (`$cindy-github`) / **GitLab** (`$cindy-gitlab`) — repositories, issues, pull / merge requests, reviews, CI. GitLab also supports self-hosted instances.\n- **Gmail**, **Google Drive**, **Google Calendar**, **Google Sheets** — each is a separate plugin and connects its own Google account authorization. Some organization accounts get a single combined **Filo Google** plugin instead.\n- **XD Atlassian** (Jira / Confluence) — an enterprise plugin, only for organizations that provide it.\n- **Notion**, plus mail plugins such as **iCloud Mail**, **QQ Mail**, and **163 Mail**.\n- The exact list can vary by account and organization.\n\n**Installing:**\n\n- Click **Install** on a Recommended plugin, or open **Add plugin** and choose **Install from file…** to pick a `.cindy` file. You can also drag a `.cindy` file anywhere onto the Cindy window, or double-click it in your file manager.\n- **Add plugin > Add Plugin Marketplace…** adds a custom marketplace from a GitHub repository, Git URL, or local folder; its plugins then show under **Custom**.\n- Before anything is installed, a **confirmation dialog** lists everything the plugin will be able to do (network addresses, credentials, background tasks, skills, bundled code, panels…). Nothing is installed until you confirm; after that the plugin is enabled right away.\n- When you ask an agent to install a plugin, the same confirmation appears as a card in that task — whatever the task's permission mode, including Full Access. This also works from your phone when you control this computer remotely. If you deny it, nothing is installed.\n- Plugins installed for you by default (for example, ones your organization rolls out) are installed and updated without a confirmation.\n\n**Updates:**\n\n- Updates from the marketplace are applied automatically when the plugin is idle — **as long as the new version doesn't need more permissions**. If it does, the old version keeps working and the card shows \"New permissions need confirmation\"; click Update to review and confirm the added permissions.\n- When updates are waiting, the page shows \"N plugin updates available\" with **Update all** (or **Ignore this round**). Failed updates retry later; you can retry right away from there.\n\n**Enabling, disabling, and uninstalling:**\n\n- **Enable / disable** with the switch on the plugin's detail page. Disabling puts its tools and panel to sleep but keeps the plugin installed (its card shows **Disabled**).\n- **Off for one project only** — use the scope picker at the top of the Plugins page to choose a project (**Project Scope**), then turn the plugin off there. In that project's tasks it behaves as if it isn't installed (its tools and `$command` are unavailable); everywhere else it stays on. Its card shows **Off in this project**. A plugin disabled globally must be enabled in the **Global** view first.\n- **Uninstall** from the detail page (**Uninstall plugin**). Its panel and capabilities are removed; its layout spot is restored if you install it again.\n- Enabling or disabling takes effect on the agent's **next** query, not in the current turn.\n\n**Panels and app views:**\n\n- A plugin's panel appears as a tab in the right sidebar and stays across tasks. If you close it, reopen it with **Open in the sidebar** on the plugin's detail page, or from the right sidebar's **+** menu.\n- Plugins with a full-page view appear in the left sidebar (under **More** by default; use **Customize** to move one to the top). Right-click one to **Open as page**, **Open in the sidebar**, or **Open in a separate window**.\n\n**Configuring a plugin's account / API key:**\n\n- Do it on the plugin's **detail page**. Plugins that use a key or token (optional search keys for Web Search, GitHub / GitLab / Notion tokens, mail app passwords) have a field there; the value is stored encrypted on this computer and can't be read back.\n- Plugins that use account authorization (for example the Google plugins) show a button to connect your account instead.\n- If you try to use a plugin that isn't set up yet, Cindy asks you to open its settings first.\n\n**Using a plugin with `$command`:**\n\n- Type its command at the **start** of a message — e.g. `$cindy-art a poster for…`. Only enabled plugins that declare that command match; a matched command turns into a pill with the plugin's icon.\n- You can also pick a plugin from the composer's **+** menu, which puts its command at the start for you.\n- You don't have to use `$command`: the agents can also pick an enabled plugin's tools on their own when a task calls for it.\n\n**Creating your own plugin:**\n\n- Choose **Add plugin > Create with Cindy**. Cindy opens a new-task draft that uses the built-in plugin creator Skill (`cindy-plugin-creator`); describe what you want and send it with your chosen agent and model.\n- That Skill can create, edit, debug, package, and publish plugins. Packaging produces a `.cindy` file; installing and publishing are separate steps you ask for when ready.\n\n**The permission / sandbox model (why plugins are safe):**\n\n- A plugin's sandbox has **no direct file, network, or system access**. Cindy carries out every operation on its behalf.\n- Background behavior is limited to what the plugin declared and you approved at install. Whether a plugin tool call in a task runs follows that task's normal permission mode.\n- Network access on its own is limited to the addresses the plugin declares. Credentials never enter the sandbox — Cindy adds them only to requests going to the declared address.\n- So a plugin **can't see your files** on its own. A file reaches a plugin only when you or the agent explicitly hand one over (for example an image attachment, or a folder you approve).\n\n**Notes:**\n\n- You can manage the plugins installed on your other computers from the same page by switching the computer at the top (the target computer must be online and on a recent Cindy version).",},{id: "providers",- title: "Model providers and Cindy AI",- summary: "Cindy AI is auto-provisioned after sign-in for pay-per-token model access; manage it (and custom providers) in Settings > Model Providers.",+ title: "Providers and Cindy AI",+ summary: "Model sources live in Settings > Providers — Cindy AI is provisioned automatically after sign-in for pay-per-token model access; add, reorder, disable, and share other providers (subscription accounts, API keys, local models, custom endpoints) there.",tab: "providers",- content: "Claude Code, Codex, and lightweight utility models can use **Cindy AI** for pay-per-token model access. Voice input is included for signed-in Cindy users and uses the separate voice service; it does not require or receive a Cindy AI key.\n\n**Cindy AI:**\n\n- Open **Settings > Model Providers** (the tab is labeled \"Providers\"). Cindy AI is pinned at the top.\n- You don't paste a key — it's **provisioned automatically after you sign in**. The page fetches its credentials for you (\"Fetching credentials…\" while it does).\n- Controls: **Refresh credentials**, **Rotate key**, and **Disconnect** (which clears the Cindy AI key saved locally). Because it's tied to your account, it's available on each device you sign into.\n\n**Status badges:**\n\n- **Connected** — this device has usable Cindy AI credentials.\n- If it's still fetching or disconnected, use Refresh credentials / sign in.\n\n**Custom providers:**\n\n- You can also add your own provider (a base URL + key) — that's the flow with a paste-a-key dialog. Cindy AI itself never uses that dialog.\n\n**Notes:**\n\n- Direct service keys such as Brave Search, Tavily, and DeepSeek search are **not** managed here — they live on the matching plugin detail pages under Plugins (Cindy Web Search). They are optional; web search works through Cindy AI without them.\n- Rotating or replacing credentials takes effect for new turns; in-flight requests finish on the previous key.",+ content: "A provider is a source of models (with its own billing). Every model you can pick in the composer comes from a provider connected in **Settings > Providers**. Voice input is included for signed-in Cindy users and uses the separate voice service; it does not require or receive a Cindy AI key.\n\n**The Providers page:**\n\n- The left column lists your connected providers; click one to see its details and models on the right. **Drag to reorder** the list.\n- **Add provider** is at the bottom of the list. **Enter Share Link…** (top right) is where you paste a provider share link someone sent you.\n\n**Cindy AI:**\n\n- Cindy AI is the official service: all models, for Claude Code, Codex, and Pi, billed per token from your balance — no keys to manage.\n- If you're not signed in, a **Cindy AI** row marked **Recommended** appears at the top of the list with a **Sign in to Cindy** button.\n- You don't paste a key — it's **provisioned automatically after you sign in** (\"Fetching credentials…\" while that happens, then **Connected**). Because it's tied to your account, it works on every computer you sign into. Organization accounts may show \"Not enabled for your organization\".\n- The detail pane shows your remaining balance or plan allowance, with **View usage** and **Buy a plan** / **Upgrade plan** (these open **Settings > Usage and billing**).\n- Credential actions are in the **···** (More Actions) menu: **Refresh credentials**, **Rotate key**, **Disconnect** (clears the locally saved Cindy AI key), plus the masked key (click to copy).\n- **Rotate key** revokes the old key immediately: running sessions, remote sessions, and mobile voice that use it need to be restarted.\n\n**Adding other providers (Add provider):**\n\n- The wizard has three steps: **Choose provider → Connect → Choose models**. Options are grouped as **Recommended**, **Detected on this computer**, **Subscription account** (sign in through your browser, e.g. Claude Code, ChatGPT, SuperGrok), **Local models** (run models on this computer — Cindy can install llama.cpp or Ollama and download models — or connect a local server that's already running), **More providers** (API key), and **Custom**.\n- CLIs already signed in on this computer appear under **Detected on this device** in the list, with **Use local account**.\n- **Custom endpoint** connects any OpenAI- or Anthropic-compatible service: display name, which runtime(s) use it (Claude Code / Codex / Pi), base URL, authentication (API key, subscription account sign-in, or none), and models. **Fetch model list** pulls available models; **Test connection** sends short requests (may cost a little); **Quick fill from a template** pre-fills known services.\n\n**Managing a provider:**\n\n- In the detail pane, turn individual models on or off (**Disable This Model** / **Enable All**); only enabled models appear in the model picker.\n- The **···** menu has **Rename**, **Edit connection settings** (custom providers), **Disable This Provider** (keeps credentials; enable it again later), and **Delete** (message history is kept).\n\n**Sharing a provider (Remote & Sharing):**\n\n- A connected provider shows **Allow remote use**: when on, your other computers and people you share with can use its models; the agent runs on this computer using its sign-in and quota. This needs remote control allowed on this computer (**Settings > Remote Connection**).\n- **Remote & Sharing** opens sharing management: **Create Share Link** (one-time link; you approve the request, optionally checking a pairing code), see **People You Share With** with usage, and **Pause Sharing** / **Remove** people.\n- Providers from your other computers appear in the list with a remote badge; providers others shared with you appear under **Providers Shared with Me**.\n\n**Notes:**\n\n- Direct service keys such as Brave Search, Tavily, and DeepSeek search are **not** managed here — they live on the **Web Search** plugin's detail page under **Plugins**. They are optional; web search works through Cindy AI without them.",},{id: "remote",title: "Remote workspaces and device control",- summary: "Run agents on a remote machine over SSH, and let another signed-in device (e.g. your phone) control this desktop — both in Settings > Remote & device control.",+ summary: "Run agents on a remote machine over SSH, let your other signed-in devices (e.g. your phone) control this computer or view its desktop, and manage local app connections — all in Settings > Remote Connection.",tab: "remote-control",- content: "Settings > **Remote & device control** has two parts: **SSH remote workspaces** (run an agent on another machine) and **My Devices** (let other devices you're signed into control this one).\n\n**SSH remote workspaces:**\n\n- **Add a host** with the **+** button: give it an alias, hostname (IP or domain), user, port (default 22), and an auth method (SSH agent or a key file). Hosts from your `~/.ssh/config` also show up automatically; you can re-read that file at any time.\n- Once a host connects (status dot turns green), expand it to:\n - **Install the agents** (Claude Code / Codex) on the remote machine, and run a quick test prompt against them.\n - **Start a remote session** — enter a working directory on the remote (defaults to `~`); the app creates the directory if needed and opens a normal session whose agent runs **on the remote machine** with that remote working directory. It appears in your sidebar like any other session.\n- The agent-facing SSH tools (via the `cindy-ssh` plugin) reuse these **same** configured hosts — if a tool reports the host isn't found, add it here first.\n\n**My Devices (controlling this desktop from another device):**\n\n- **No manual pairing** — any device signed into the **same Cindy account** is discovered automatically and appears in the list (this is how the phone app connects to your desktop).\n- The **top card is this machine**: rename it, see its relay-connection status, and toggle the master switch **\"allow other devices to control this machine\"**.\n- Each **other device** has two independent switches: **you control it** (outbound) and **allow it to control this machine** (inbound). You can rename devices, and delete ones that are offline.\n\n**What a controlling device can and can't do:**\n\n- A controlling device (like the phone app, which is control-only — it has no local agent or database of its own) can drive sessions: create / close / fork them, send and steer input, approve interaction prompts, switch model / effort / permission mode, rewind, manage collaboration workers and scheduled tasks, and read session / message lists.\n- It can **not** touch auth or API keys, change global settings, write the database directly, trigger updates, or run local shell / file dialogs on the controlled machine. A new remote session's working directory must be currently accessible on the controlled desktop; disconnected network drives are rejected.\n\n**Notes:**\n\n- Remote project sessions **can** be the lead in a collaboration workflow — workers then run on the same remote host (see the Collaboration topic).\n- Device control works only between devices on the **same account**; the relay just routes between them — your session data isn't stored on it.",+ content: "**Settings > Remote Connection** has three parts: **My devices** (control your other signed-in devices, or let them control this one), **SSH remote hosts** (run an agent on another machine), and **Local App Connections**.\n\n**SSH remote hosts:**\n\n- **Add** opens \"Add a remote machine\": alias, optional display name, hostname (IP or domain), user, port (default 22), and authentication — **SSH agent (recommended)**, optionally pinned to one key, or an **Identity file** (encrypted keys aren't supported; use ssh-agent for those). Optionally turn on **Route agent traffic via proxy** (a Cindy-managed tunnel back to your local proxy, or your own proxy URL reachable from the remote).\n- Hosts from your `~/.ssh/config` show up automatically (their connection details are read-only in Cindy); the refresh icon (**Reload ~/.ssh/config**) re-reads the file. **Manage keys** lets you browse, generate, or unlock SSH keys; **Setup key** helps install a key on a host.\n- Each host has **Connect** / **Disconnect**, **Auto-connect** (connect when Cindy starts), **Edit**, and **Remove**. Status reads Disconnected / Connecting… / Authenticating… / **Ready**.\n- Once a host is Ready, **Show agent setup** lets you **Install** / **Reinstall** / **Uninstall** Claude Code, Codex, and Pi on it, **Sync auth** (copy your local Codex sign-in to the remote), and run a **Quick test** prompt.\n- To work on the remote: on the new-session page, open the project picker and choose **Add remote project**, pick the host under **Remote hosts (SSH)**, then choose (or create) a folder. Claude Code, Codex, and Pi can all run over SSH; if the agent isn't installed on the host yet, Cindy installs it automatically when you send the first message. Agent setup also has **Start a remote Codex session** (working directory defaults to `~`; you're asked before a missing directory is created). The session appears in your sidebar like any other, but its agent runs **on the remote machine**.\n- Models that only work locally (e.g. Ollama) or subscription-direct models may not be available in SSH sessions; Cindy AI works.\n- The agent's built-in **SSH commands** tool (Settings > Tools) reuses these **same** hosts — if it reports a host isn't found, add it here first.\n\n**My devices (shown when you're signed in):**\n\n- **No manual pairing** — every device signed into the **same Cindy account** appears automatically (this is how the phone app connects to your computer). If none appear, sign in with the same account on the other device.\n- The **top card is this device**: rename it, see its connection status, and the master switch **\"Allow same-account devices to control this machine\"** (no per-connection approval; a visible indicator shows while it's being controlled, and you can disconnect at any time).\n- Below it, **Allow remote desktop** lets your signed-in devices view and operate all apps on this computer (device control must also be on). On macOS this needs Screen Recording and Accessibility permission; on Windows, **Lock screen control** additionally needs administrator authorization.\n- Each **other device** has two independent switches: **I control it** and **Let it control this device**, plus rename and delete. Desktops also have a **Remote desktop** button to open their screen. Phones can control this computer but can't be controlled.\n- While another device is connected, a \"<device> is connected to Cindy\" notice appears on this computer with a **Revoke** button (with several devices, **View** opens this page). A revoked device can't reconnect until you restore its access here.\n\n**Using another computer:**\n\n- On the new-session page, the **Device** picker chooses which computer the session runs on. The sidebar's device scope menu shows sessions from your other computers.\n\n**What a controlling device can and can't do:**\n\n- A controlling device (like the phone app, which is control-only — it has no local agent of its own) can drive sessions: create / close / fork them, send and steer input, approve prompts, switch model / effort / permission mode, rewind, manage collaboration workers and Automations, and read session / message lists.\n- It can **not** touch sign-in or API keys, change the controlled computer's global settings, trigger updates, or open local shell / file dialogs there. A new remote session's working directory must be currently accessible on the controlled computer; disconnected network drives are rejected.\n\n**Local App Connections:**\n\n- Lists other local apps you've allowed to use Cindy's models and agent capabilities (Cindy keeps the provider credentials). Revoke a single connection, or all of them, here.\n\n**Notes:**\n\n- Remote project sessions **can** be the lead in a collaboration workflow — workers then run on the same remote host (see the Collaboration topic).\n- To keep a computer reachable while you're away, turn on **Keep computer awake** in **Settings > General > App Behavior**.\n- Device control works only between devices on the **same account**; the relay just routes between them — your session data isn't stored on it.",},{id: "reviewing-changes",title: "Reviewing the agent's file changes",- summary: "Open the Review tab in the right sidebar to see what changed as git-style diffs, and stage / revert hunks.",- content: "When you want to see what the agent changed, use the **Review** tab in the right sidebar — it shows the working directory's changes as git diffs.\n\n**Opening it:**\n\n- In the right sidebar, open the **+** menu and pick **Review** to add the Review tab.\n- The diffs come from a real **git diff** of your working directory against a base ref — not from replaying the agent's individual Edit / Write calls. You can change the base branch it compares against.\n\n**Inside the tab:**\n\n- Changed files are shown as diffs, **expanded by default**; collapse the ones you don't care about.\n- Toggle between **unified** and **split** views. There's also inline word-diff, word-wrap, hide-whitespace, a file tree, and a rich Markdown preview for `.md` files.\n- Colors follow git convention: **red removed / green added**.\n\n**Acting on changes:**\n\n- Each hunk has action pills to **stage / unstage / revert** it — so you can review the agent's edits and selectively keep or roll back individual hunks right from the tab.\n\n**Notes:**\n\n- Because the tab reflects real git state, it shows everything in your working tree — including edits you made yourself — not only what the agent touched this turn.\n- This is separate from **Plan mode** (approving a plan before work starts) and from the **collaboration** worker panes.",+ summary: "See what changed from a message's \"Changed N files\" card (Review / Undo / Reapply) or in the right sidebar's Review tab (git diffs by source, stage / revert, commit and push); use /review for an independent read-only review.",+ content: "There are two places to see what the agent changed: the **changes card** under each reply, and the **Review** tab in the right sidebar.\n\n**Per-message changes card:**\n\n- When a reply edits files, a **\"Changed N files\"** card appears under it, listing the files.\n- **Review** opens those exact changes (\"Changes from this message\") in the Review tab.\n- **Undo** reverts that message's file changes; **Reapply** puts them back. These are blocked while the agent is still replying, and if the files have changed since then, Cindy asks you to review the current changes first. Undo needs Git installed.\n- If some writes (e.g. from shell commands) couldn't be captured exactly, the card says so and only the captured part can be reviewed or undone. Exact per-message changes aren't available for remote sessions yet.\n\n**Opening the Review tab:**\n\n- Show the right sidebar with the panel icon at the top-right of the window, click **+** (Add tab), and pick **Review**.\n- The source dropdown picks what to compare: **Unstaged** (default), **Staged**, **Commit** (pick a commit on this branch), **Branch** (everything since the merge-base with a **Compare base** branch you choose), or **Last turn** (only files the agent touched in its latest turn).\n- Diffs are read from real git state, so the working-tree views include edits you made yourself, not only the agent's.\n\n**Inside the tab:**\n\n- Toggle **Unified** / **Split** views. Diffs are expanded by default; **Expand all diffs** / **Collapse all diffs** and per-file collapse are available.\n- **Show files** opens a file tree (with a filter); **Jump to file** filters and jumps.\n- The view-options menu has word wrap, word diffs, hide whitespace, rich text preview for Markdown files (on by default), and **Copy git apply command**. Changed images show a Before / After preview.\n- Colors follow git convention: **red removed / green added**. Each file can be opened or revealed in its folder.\n\n**Acting on changes:**\n\n- **Stage** / **Unstage** / **Revert** whole files or single hunks, or use **Stage all** / **Unstage all** / **Revert all**. Revert asks for confirmation (untracked files are deleted).\n- **Commit or push** (top of the tab): write a commit message, then **Commit** or **Commit and push** (optionally **Include unstaged changes**), or **Push** local commits. Cindy asks before a force push.\n- Writes are paused while the agent is running, and during detached HEAD, merge conflicts, or an in-progress git operation. Sessions on SSH hosts or another computer are view-only.\n\n**Independent review (`/review`):**\n\n- Type `/review` (optionally with a focus or path) to start a fresh, read-only reviewer session with no memory of the work; it checks code changes, files, documents, or images against the task. It currently works for local sessions only.\n\n**Notes:**\n\n- This is separate from **Plan mode** (approving a plan before work starts) and from the **collaboration** worker panes.",},{id: "scheduler",- title: "Scheduling automated tasks",- summary: "Run a prompt automatically on a schedule (recurring or one-off) against a chosen working directory.",- content: "The Scheduler runs prompts for you on a schedule — recurring (cron-style) or one-off (a specific time).\n\n**Creating a task:**\n\n- Open the Scheduler.\n- Give the task a **prompt** (what the agent should do), a **working directory** (where it runs), and a **schedule** — recurring (interval / cron) or one-time (specific datetime).\n- Pick which **agent** runs it (Claude Code or Codex) and the model / mode you want.\n- Save. The task fires according to its schedule.\n\n**Managing existing tasks:**\n\n- Edit (prompt, schedule, working directory).\n- **Pause / resume** without losing the task definition.\n- Delete.\n\n**Run history:**\n\n- Each task tracks its past runs — when it fired, whether it succeeded, the resulting session (you can open it to see what the agent did).\n\n**Notes:**\n\n- The task only fires while the desktop app is running on your machine — it's a local scheduler, not a cloud cron. Closing the app means tasks won't fire until you reopen it.\n- Tasks run as normal sessions under the hood; you can find their resulting sessions in the sidebar (often grouped by the scheduler).\n- If you have two instances open on the same account (e.g. dev + release), they coordinate so a due task fires only once — you won't get duplicate runs.\n- A failing task shows the failure in run history but doesn't auto-pause itself — you may want to disable a chronically failing task manually.",+ title: "Scheduling automations",+ summary: "Automations run a prompt (or a script) for you on a schedule, once, or on demand — open Automations in the left sidebar to create, pause, run, and review them.",+ content: "**Automations** run a prompt for you on a schedule — recurring, once, or only when you trigger it. Open **Automations** from the top of the left sidebar (if you've customized the sidebar, it may be under **More**).\n\n**Creating an automation:**\n\n- Click **New Automation**. Start from a template (**Start from a template** — e.g. nightly test self-healing, PR gatekeeper, domain radar) or choose **Blank**.\n- Give it a title and a **prompt**, and choose where it runs: a **project** folder (optionally an isolated **worktree**, under **Advanced**) or a plain chat with no project.\n- Pick the **model** (and reasoning effort); the engine — Claude Code, Codex, or Pi — follows the model.\n- **Schedule**: **Automatic** or **Manually** (runs only when you click Run now). For Automatic, choose a preset — every N minutes, hourly, daily, weekdays, weekly, or monthly — in **Cron** mode (fixed clock times) or **Relative interval** mode (waits that long after the previous run finishes). Tick **Once** to run a single time at the next scheduled slot.\n- **Run session**: **New each run**, **Continuous session** (every run continues the same session), or **Bind existing session**.\n- Optional: **Stay quiet when checks find nothing** (only report changes or problems), send the result to a Feishu bot or WeCom group, and a **Pre-run check** — a command that decides whether a round runs or is skipped (**Generate with AI** can write it for you).\n- **Execution** can also be **Script only**: run a command in a project directory without an AI turn.\n- You can also ask an agent in any session to set up an automation for you.\n\n**Managing automations:**\n\n- The list groups automations by project and filters by status (All / Active / Paused).\n- **Run now**, **Edit**, **Rename**, **Pause** / **Resume**, and **Delete Automation**. Deleting asks what to do with the sessions it created: keep, archive, or delete them. Pausing or deleting stops any run in progress.\n- **Promote to project automation** saves it into the project's repository so collaborators share it; project automations are edited through their config file.\n\n**Run history:**\n\n- Select an automation to see **Recent runs**: when each ran, its status (Success, Skipped, Aborted, Interrupted, …), duration, cost, and pre-run check output. **Open session** shows what the agent did; **Restart** re-runs, and a run record can be deleted.\n\n**Notes:**\n\n- Automations run locally: they only fire while Cindy is running on this computer (closing the window to the tray is fine). Runs missed while the app was closed are not made up — the next one fires at the next scheduled time. To stop the computer from sleeping, turn on **Keep computer awake** in **Settings > General > App Behavior**.\n- Automation runs are normal sessions; in the sidebar they're grouped under the automation's name (marked \"Created by automation\"), with Run now / Edit / Pause there too.\n- A limited number of runs execute at the same time; others show \"Waiting for a slot\" and start as soon as one frees up.\n- If several Cindy instances share the same account data on this computer, they coordinate so a due run fires only once.\n- A failing automation shows the failure in its run history but doesn't pause itself — disable a chronically failing one manually.",},{id: "sessions-and-chat",title: "Sessions and the chat thread",- summary: "Create sessions, send / stop / queue messages, clear context, and switch between sessions in the sidebar.",- content: "Each **session** is a persistent chat thread driven by **one agent** (Claude Code or Codex). All your chats live as sessions in the left sidebar.\n\n**Starting a session:**\n\n- Click **+ New** at the top of the sidebar.\n- Pick the agent (Claude Code or Codex) and — for code work — a working directory.\n- Pick a model in the composer toolbar if you want something other than the default.\n- Type your first message and send.\n\n**During a turn:**\n\n- While the agent is replying, **Send** becomes **Stop** — click it to interrupt the current turn cleanly.\n- You can keep typing while the agent is busy — follow-ups are **queued** in order and sent one after another when the previous turn finishes.\n\n**Chat controls:**\n\n- **`/clear`** — clears the current session's conversation context in place (resets messages and state without creating or switching to a new session). The session itself stays in the sidebar; only its content is wiped.\n- **`/compact`** — ask the agent to summarize and compress earlier turns into a shorter context (Claude Code).\n- **Edit** — you can edit your **last** user message. Doing so **rewinds** the conversation: it drops that message and everything after it, then resends your edited message. In sessions with file-rewind support (local git working directories with savepoints), file changes are also rolled back. In sessions without savepoints (remote sessions, non-git directories, or Codex sessions that haven't created savepoints), only the conversation is rewound — filesystem changes remain.\n\n**Switching between sessions:**\n\n- Click any session in the sidebar to switch to it. State is persisted — each session remembers its agent, model, working directory, attachments, etc.\n- Keep **unrelated tasks in separate sessions** — sharing one session for everything dilutes the agent's context and confuses it.\n\n**Notes:**\n\n- Sessions are **per local user** — they live in the local SQLite DB at `<userData>/cindy-<userId>.db` (installs migrated from the legacy app get their sessions copied from the old `xdt-maker-<userId>.db` on first login). They don't sync to other machines automatically.\n- If you have two desktop instances open (e.g. dev + release) on the same user account, both read/write the same DB and will see each other's sessions.",+ summary: "Create sessions, pick model / engine / project, send / stop / queue / steer messages, compact or clear context, rewind, edit, fork, and switch between sessions.",+ content: "Each **session** is a persistent chat thread run by an agent engine (Claude Code, Codex, or Pi). All your sessions live in the left sidebar.\n\n**Starting a session:**\n\n- Click **New** at the top of the left sidebar (or press ⌘N / Ctrl+N).\n- Choose a project folder for code work, or stay in **Chat** for plain Q&A. If you have other computers or SSH hosts, the **Device** picker and **Add remote project** choose where it runs (see the Remote topic).\n- Pick a model in the composer's model selector. There's no separate agent picker: the engine (Claude Code, Codex, or Pi) follows the model and is shown next to it; to run a model on a different engine, use that model's **Customize** options (see the Choosing a model topic). You can also set the permission mode in the composer toolbar.\n- Type your first message and send. **Enter** sends and **Shift+Enter** adds a new line; to send with ⌘↵ / Ctrl+Enter instead, use **Settings > General > Message input**.\n\n**During a turn:**\n\n- While the agent is replying, the composer shows a separate **Stop** button (labeled **Stop generating**) — click it to interrupt the current turn.\n- You can keep typing: the send button becomes **Queue message**, and follow-ups are **queued** and sent one after another when each turn finishes. With the default Enter-to-send setting, press ⌘↵ / Ctrl+Enter instead to **steer** — deliver the message into the running turn without interrupting it.\n- Queued messages appear above the composer, where you can **Steer**, **Edit**, **Delete**, or drag to reorder them. If you stop the turn, the queue pauses until you click **Continue sending**.\n\n**Context:**\n\n- The context ring in the composer shows how full the context is; click it to compact (Claude Code and Pi; Codex manages its own context, so the ring is read-only there). **`/compact`** also asks the agent to summarize earlier turns.\n- **`/clear`** — clears the current session's conversation context in place (resets messages and state without creating or switching to a new session). The session stays in the sidebar; only its content is wiped.\n\n**Message actions (hover a message):**\n\n- **Copy message**; the **More message actions** menu also has **Copy current message link**.\n- **Edit** — available on your **last** message. Editing **rewinds** the conversation to that message and resends your edited text.\n- **Back to here** — rewind to an earlier message of yours: later messages are removed and, when a Git file recovery point exists, project files are rolled back too (the dialog shows how many files). Without one (non-git folders, recovery points turned off in **Settings > General > Git Safety**), only the conversation is rewound. Rewind isn't supported yet for remote Claude Code sessions.\n- **Start a new session** — fork a separate session from that point; the original stays unchanged.\n- **Delete this message** — removes it from the history; the agent's context is rebuilt from what remains on the next send.\n- Replies that changed files show a **Changed N files** card with Review / Undo / Reapply (see the Reviewing changes topic).\n\n**Switching between sessions:**\n\n- Click any session in the sidebar, or press ⌘1–⌘9 / Ctrl+1–Ctrl+9 to jump to the first nine visible sessions. Each session remembers its engine, model, working directory, attachments, and draft.\n- Picking a model that runs on a different engine mid-session switches the engine after you confirm **Switch Agent?** (see the Choosing a model topic).\n- Keep **unrelated tasks in separate sessions** — sharing one session for everything dilutes the agent's context.\n\n**Notes:**\n\n- Sessions are stored locally on this computer, separately for each signed-in account. They aren't copied to your other computers, but with **My devices** (Settings > Remote Connection) you can view and drive sessions on your other signed-in computers from the sidebar.",+ },+ {+ id: "shared-tasks",+ title: "Sharing a task with other people",+ summary: "Share a task with other Cindy accounts through an invitation link (Share Task in the task menu); guests see the full history and message the same Agent. Manage sharing in Settings > Share Task; to send a copy instead, export a .cshare file.",+ content: "**Share Task** lets people on **other Cindy accounts** join one of your tasks. They see its full history, tool results, and attachments, and can message the same Agent. The task keeps running on your computer (the \"host\"), using your models and connections. To use your own phone or other computers on the **same account**, you don't need sharing; see the Remote topic.\n\n**Starting to share (host):**\n\n- Open the task's menu (the **⋯** next to the task title, or right-click it in the sidebar) and choose **Share Task**, then **Start Sharing**. You must be signed in, and the task must be active (not archived). Tasks running on your other computers can be shared too.\n- **Copy invitation** copies a message with the invitation link. Send it to your teammate. Anyone holding the link can join after signing in, with the same Agent access as you, and without needing your approval.\n- A task can have up to **2 guests**. There's also a limit on how many tasks you can host or join at once.\n- Sharing ends automatically if the host computer stays offline. To resume, start sharing again and send a new invitation link.\n\n**Managing a shared task (host):**\n\n- In the task menu, **Manage Sharing** has **Copy invitation**, **Manage Members…** (see members and **Remove** a guest), and **Stop Sharing**.\n- Removing a guest or stopping sharing cuts off access right away and makes old invitation links invalid. The task history and the Agent's work are kept.\n- Stop sharing before you archive or delete a shared task.\n\n**Joining (guest):**\n\n- Open the invitation link: it opens Cindy and joins you right away, then you **Enter Task**. If Cindy doesn't open, choose **Join Shared Task** and paste the link (an older invitation code also works).\n- Where to find **Join Shared Task**: the **+** next to the **Share Task** group in the left sidebar (the group appears once you host or join something), or **Settings > Share Task**.\n- As a guest you can send messages, add attachments, use voice input, stop the Agent, and answer the Agent's permission prompts, questions, and plan reviews. The task's Agent, model, and reasoning settings can only be changed by the host. Side chats and forking aren't available to guests.\n- The task's menu only shows **Leave** for guests. To rejoin later, you need a new invitation link from the host.\n- An invitation only works in a Cindy app that uses the same service — for example, an invitation from the Mainland China edition can't be used in the other edition.\n\n**Settings > Share Task:**\n\n- **Shared by Me**: every task you share (including from your other computers) — click a title to open it, **Manage** to see members and the invitation, or **Stop All Sharing**.\n- **Joined by Me**: tasks you joined — click a title to open it, or **Leave**.\n- **Join Shared Task** button: paste an invitation link or code.\n\n**In the sidebar:**\n\n- Shared tasks you host or joined appear together in the **Share Task** group. Tasks you host show a gold crown.\n\n**Sending a copy instead of live sharing:**\n\n- To give someone a snapshot of a task, use **Export session…** in the task menu to create a `.cshare` file (optionally password-protected). The other person imports it on their own computer and continues independently. See the Importing sessions topic.\n\n**Notes:**\n\n- Only invite people you trust: guests use the Agent with the task's existing permissions, tools, and plugins, and can read and change files in the task's working folder.\n- Ending sharing doesn't take back anything guests already saw or downloaded.",},{id: "sidebar",title: "Finding and organizing sessions",- summary: "Browse, group, filter, sort, search, and pin sessions or projects in the left sidebar.",- content: "The left sidebar lists your sessions. It's the main way to navigate and organize. The **Sidebar display settings** menu holds grouping, sorting, filters, and display options — open it from the sliders icon on the main list header, from **Sidebar display settings** in the device-scope menu, or by right-clicking any blank area of the sidebar.\n\n**Device scope (the list header itself):**\n\n- The main list header doubles as the device scope switcher. Its text names what you're looking at — **All sessions**, **This device's sessions**, a remote device's name, or **N machines** — and clicking it opens the scope menu: all / this device / each connected remote device (single-click switches; checkboxes multi-select), plus **Remote connection settings** and **Sidebar display settings**.\n- The scope applies to the whole sidebar, including the Pinned section.\n- With no remote devices, the header still shows **All sessions ▾**. Clicking it opens a short menu with only **Remote connection settings** and **Sidebar display settings** — no device list. The header stays visible even when the list is empty, only pinned items remain, remote devices/tasks are loading or failed, or the selected remote device is still connecting. There is no separate device row in the top navigation.\n\n**Grouping (independent checkboxes, freely combinable):**\n\n- **Group by project** (default on) — sessions with a project fold into project rows; project rows and stray chats compete for position on the same timeline (a chat active two minutes ago sorts above a project last touched yesterday). Turn it off for a flat list.\n- **Group by device** (default on) — only appears when a remote device is connected and the scope isn't narrowed to a single machine; splits the list into device sections (this device first), each with a collapsible header showing the device name and online state. When the last remote device disconnects — or the scope is a single machine — the option disappears and the effect is lifted; the saved preference is untouched and comes back automatically.\n- **Group chats together** (default on) — collects projectless sessions into one collapsible **Chat** group instead of interleaving them. With device grouping on, each device section gets its own Chat group with independent collapse state.\n- Grouping **by date** has been removed.\n\n**Sorting:**\n\n- **By time** (default) — one activity timeline for the whole list, most recent first. There is no oldest-first option.\n- **Priority** — four tiers, same as Codex: sessions waiting on you (pending reply / approval / error) first, then finished-unread sessions, then running sessions, then the rest by time. A project row inherits the highest priority among its sessions. Unread state is in-memory: after a restart most sessions fall back to time order.\n- **Manual** — drag project rows to set a persistent custom order. Manual order covers **project rows only**; stray chats and the Chat group always follow by recency. Switch to Manual first to drag; other sort modes do not start a drag.\n- Sorting **alphabetically** and **oldest first** have been removed.\n\n**Filters (one row that opens a submenu):**\n\n- **Status** — active / archived / all.\n- **Project** — multi-select working directories plus **Chat** (projectless sessions), or \"all\". Selecting specific projects hides chats unless **Chat** is also checked; check only **Chat** to see projectless sessions alone.\n- **Agent** — Claude Code / Codex / all.\n- **Recent activity** — last 1 day / 3 days / 7 days / 30 days / all.\n- A **Reset filters** action at the bottom clears all four at once. Filters do **not** apply to the Pinned section, except Status (archived pins stay hidden while Status is Active). The list header does not show a standing “filtered” mark; open Sidebar display settings to see which filters are on.\n\n**Display:**\n\n- The main list has **Text** (compact single row) and **List** (two-line row with a preview) styles. Fresh installs default to List; installs upgraded from an older version keep Text so the sidebar looks the way it did before the upgrade. This is separate from the Pinned section's style.\n- **Session info** — choose what shows at the right edge of each row: **Time** (default), **PR status** (a status icon plus a monospace `#number`; only the icon is colored — green open / gray draft / red closed / accent merged; the open green follows the row surface so it stays readable on both the rest background and the inverse selected pill; a small orange dot on the icon means the PR still has unresolved review threads; unknown state shows a gray icon; sessions without a PR show nothing), **Tokens** (compact total like `1.4M`), and **Cost** (`$`/`¥` by currency). Multi-select; items render in the order you selected them.\n\n**Pinning:**\n\n- Right-click a session and pick **Pin** to keep it at the top. There's no cap on the number of pinned sessions; a newly pinned session goes to the **front** of the pin order (most recent pin first).\n- Right-click a pinned session and pick **Unpin** to remove the pin.\n- Open a project's overflow menu and pick **Pin project** to move the whole project into **Pinned**. Project pins are independent from session pins, remain available after a restart or project rename, and can be expanded or collapsed inside **Pinned**.\n- Pinned projects and sessions share one draggable order. Newly pinned items move to the front, while the other pinned items keep their relative order. Use **Unpin project** to return a project to its position under the current project sorting mode.\n- The **Pinned** section keeps its own display style: **Text**, **List**, and **Card** — Card mode is exclusive to Pinned.\n\n**Collapse / expand all groups:**\n\n- The fold button next to the display-settings button collapses or expands every group at once. With both device and project grouping on, it cycles through levels: collapse project groups → collapse device groups → expand all. The tooltip always names the next step.\n\n**Searching:**\n\n- The search box matches against the **session title only** — not message content and not the working directory path. If you can't remember a session's title, browse via project / agent filters instead.\n\n**Right-click actions:**\n\n- On a regular session: **Pin / Unpin**, **Rename**, **Move to project** (submenu), **Copy session link** (a `cindy://session/<id>` deep link), **Open in new window**, **Archive**, **Delete**.\n- On an archived session: **Rename**, **Unarchive**, **Copy session link**, **Delete**.\n\n**Removing a project from the sidebar:**\n\n- Open a local project's overflow menu and choose **Remove Project from Sidebar**. Cindy asks for confirmation, removes the project from the sidebar, and keeps its existing sessions available as projectless chats in the main list. Individually pinned sessions remain in **Pinned**. Those sessions are not archived or stopped, and files on your computer are not deleted.\n- To restore the project, choose **Add Project** and select the same directory again. Cindy restores the existing sessions under their original project grouping instead of creating an empty session.\n\n**Session statuses:**\n\n- Sessions are **active**, **archived**, or **deleted** — there are no other states. Archiving hides a session from the default view without removing it; delete is unrecoverable from the UI.\n\n**Notes:**\n\n- Click the sidebar collapse arrow to shrink it to an icon-only strip — that's purely visual, you don't lose any features.\n- **Pin state, order, hidden projects, project filters, and identity-based collapse/selection state are isolated per Cindy account.** Pin and hidden-project changes sync across open windows for the same account. Display-only preferences (list style, grouping toggles, sorting, session info) remain local to each window, so they don't sync between, say, a dev window and the installed app.",+ summary: "Browse, group, sort, filter, search, tag, and pin sessions or projects in the left sidebar, and find and restore archived sessions.",+ content: "The left sidebar lists your sessions. It's the main way to navigate and organize. The top of the sidebar has navigation rows (**New** — ⌘N / Ctrl+N starts a new session — then **Automations**, **Plugins**, **Search**; use **More → Customize** to show, hide, or reorder them). **Teammates** have their own section above **Pinned** (see the Teammates topic). Below them is the main session list.\n\n**Finding and restoring an archived session:**\n\n- Archived sessions are hidden from the default list, but they are not deleted. Two ways to find one:\n 1. **Search** (top of the sidebar): search covers archived sessions by default, because its **Status** filter starts at **All**. Click a result to open it.\n 2. **Sidebar settings → Session status → Archived** (or **All**): the main list then shows archived sessions. Open **Sidebar settings** from the sliders button at the right end of the list header (it appears when you hover over the header), from **Sidebar settings** in the list-header dropdown, or by right-clicking a blank area of the sidebar.\n- While **Archived** is selected, the sliders button stays visible and shows a small dot so you remember the list isn't the normal view. **All** shows no dot. Switch back to **Active** to return to the normal list.\n- To restore: hover over an archived row and click the **Unarchive** (undo arrow) button, or right-click the row (or use its **⋮** menu, or the **⋯** menu in the open session's header) and choose **Unarchive**. Archived rows show an archive icon in place of the usual status icon.\n- Sending a new message in an archived session restores it to **Active** automatically. An automation linked to an archived session also restores it the next time it runs.\n- To restore many at once: with **Session status** set to **Archived**, a project's **⋮** menu shows **Restore all**, and the Chat group's menu shows **Restore All**.\n\n**Archiving and deleting:**\n\n- Hover over a session row and click the archive button; it turns into a red **Archived** button — click that to confirm (it reverts after about 4 seconds otherwise). You can also choose **Archived** from the right-click menu. A running session can't be archived until you stop it; a session being remote-controlled must be taken back first.\n- Cindy asks for confirmation when the session's worktree has uncommitted changes (they're saved as a git snapshot first) or when a linked automation will keep running and restore the session.\n- Archiving removes the session's pin.\n- **Delete** asks for confirmation and can't be undone from the UI.\n- A project's **⋮** menu has **Archive all** (pinned and running sessions are skipped); the Chat group's menu has the same action.\n- **Multi-select:** Cmd-click (macOS) / Ctrl-click (Windows) toggles rows; Shift-click selects a range. A bar above the list then offers **Archive selected sessions**, **Delete selected sessions**, and **Clear selection**.\n- Sessions are **active**, **archived**, or **deleted**. **Settings > Storage > Database cleanup** can permanently clear the history of old archived (and optionally unarchived) sessions; scan first, and Cindy asks before cleaning.\n\n**Device scope (the list header title):**\n\n- The main list header title is a dropdown that names what you're looking at: **All sessions**, **Local**, a remote device's name, or **N devices**. Click it to switch scope: **All**, **Local**, or a connected remote device (click a row to switch to that device alone; click the checkbox at the right of a row, or Cmd/Ctrl-click, to select several). The menu also has **Remote desktop** (when another desktop computer can be controlled), **Remote connection settings**, and **Sidebar settings**.\n- The scope applies to the whole sidebar, including **Pinned**.\n- With no remote devices, the title still reads **All sessions ▾** and the menu only has **Remote connection settings** and **Sidebar settings**. The header stays visible even when the list is empty or remote sessions are still loading.\n\n**Sidebar settings menu** (each row shows its current value and opens a submenu):\n\n- **Group by** — independent checkboxes:\n - **Project** (default on): sessions with a project fold into project rows; project rows and stray chats share one timeline. Turn it off for a flat list.\n - **Chat** (default on): collects projectless sessions into one collapsible **Chat** group.\n - **Device** (default on): appears when you have a remote device. Splits the list into device sections (this computer first), each with a collapsible header showing the device name and online state.\n- **Session sort** — **Priority** (waiting on you first, then unread, then running, then by time; a project row ranks by its most urgent session), **Recent Activity** (default, most recent first), or **Creation Time** (newest first).\n- **Project sorting** (only while grouping by project) — **Follow sessions** (default) or **Manual**: drag project rows to reorder them; sessions inside still follow **Session sort**.\n- **Session status** — **Active** (default) / **Archived** / **All**. See above.\n- **Filters** — **Projects** (multi-select, plus **Chat** for projectless sessions; selecting specific projects hides chats unless **Chat** is also checked), **Harness** (All / Claude Code / Codex / Pi), and **Last activity** (1d / 3d / 7d / 30d / All). **Reset filters** clears these three; it doesn't change **Session status**. The **Filters** row shows how many are active.\n- **Display** — **Text** (compact single row) or **List** (two-line row with a preview). Fresh installs default to List; upgraded installs keep Text. This is the same setting as **List Style** in **Settings > Appearance**.\n- **Session info** — what appears at the right edge of each row (multi-select, shown in the order you picked): **Time** (default), **PR status** (a status icon plus `#number`; a small orange dot means unresolved review threads), **worktree** (the worktree's short name; click it to open the folder), **Tokens**, **Cost**.\n- Filters and **Session status** don't apply to the **Pinned** section; it only follows the device scope.\n\n**Pinning:**\n\n- Right-click a session and choose **Pin** / **Unpin**. There's no limit; a new pin goes to the front.\n- In a project's **⋮** menu, choose **Pin project** / **Unpin project** to pin the whole project. Pinned projects and sessions share one draggable order.\n- The **Pinned** header has its own view-style button: **Text**, **Cards**, or **List** (Cards is only available in Pinned).\n\n**Collapse / expand all groups:**\n\n- The fold button in the list header (shown on hover) collapses or expands groups. With both device and project grouping on, it steps through: collapse project groups → collapse device groups → expand all. The tooltip names the next step.\n\n**Searching:**\n\n- Click **Search** at the top of the sidebar and type. It matches session **titles and message content** (keyword and semantic matches), shows matching snippets, and jumps to the matching message when you click a result.\n- The sliders button in the search box opens **Search criteria**: **Sort by** (Relevance / Recently active / Oldest active), **Status** (Active / Archived / **All**, default All), **Harness** (All / Claude Code / Codex / Pi), **Last activity**, and **Projects**. A dot on the button means a filter is narrowing results. Filters other than sort reset when you close search.\n- To search inside one project, choose **Search sessions in this project** from the project's **⋮** menu; the Chat group's menu has **Search This Chat Group**.\n\n**Right-click / ⋮ menu on a session:**\n\n- Regular session: **Pin / Unpin**, **Rename**, **Tags**, **Copy session link** (a `cindy://session/<id>` link), **Share Task**, **Move task** (to another project, to Chat, or copy to another computer), **Export session…**, **Open in new window**, **Archived**, **Delete**. Some items are hidden for remote sessions.\n- Archived session: **Rename**, **Tags**, **Copy session link**, **Export session…**, **Unarchive**, **Delete**.\n- **Tags:** pick up to seven favorite colored tags directly in the menu, or choose **More tags** to create, edit, or reorder tags. Tags show as colored dots after the session title.\n- Double-click a session title to rename it.\n\n**Project menu (⋮ or right-click on a project row):**\n\n- **Rename project**, **Pin project**, **Search sessions in this project**, **Show Files**, **Open in file manager**, **Copy Project Link**, **Sync Codex sessions for this project**, **Remove Project from Sidebar**, **Archive all** / **Restore all**. Remote projects show a shorter menu. The **+** button on a project row starts a new session in that directory.\n\n**Removing a project from the sidebar:**\n\n- Choose **Remove Project from Sidebar** on a local project. Cindy asks for confirmation, then moves its sessions to the **Chat** section. They aren't archived or stopped, pinned sessions stay in **Pinned**, and no files are deleted.\n- To bring the project back, start a new session and pick the same folder in the project picker (**Choose another project folder** if it's not listed). Cindy restores the existing sessions under that project.\n\n**Notes:**\n\n- Click the sidebar collapse arrow to shrink it to an icon-only strip; nothing is lost.\n- Pins, pin order, hidden projects, project filters, and collapse state are kept separately for each Cindy account. Display preferences (list style, grouping, sorting, session info) are local to each installation.",},{id: "skills",title: "Skills (reusable agent capabilities)",- summary: "Create, search, browse, install, disable, uninstall, publish, and update agent Skills; Skills appear as \"/\" slash commands and are shared across Claude Code, Codex, and Pi.",- content: "Skills are reusable agent capabilities you package as a folder and load into your sessions. They're managed on the same page as Plugins — **Skills** and **Plugins** are two tabs of one management surface (the Skills tab is the in-app browser for finding, installing, publishing, and updating skills).\n\n**Skills work across local agents:**\n\n- Skills are **engine-shared** — the same installed Skill is available to Claude Code, Codex, and Pi, so you install it once.\n\n**Where skills live on disk:**\n\n- User-created or imported global skills (available to every session): the shared root `~/.agents/skills/<name>/`, which is cross-linked to `~/.claude/skills/` and `~/.codex/skills/` so every engine sees the same skills. On Windows that's under `C:\\Users\\<you>\\`.\n- Project-scoped (only inside one working directory): `<working-dir>/.agents/skills/<name>/` or `<working-dir>/.claude/skills/<name>/`.\n- Cindy’s preinstalled skills and plugin-provided skills are kept in Cindy-managed directories and loaded into local Claude Code, Codex, and Pi tasks. Cindy does not add them to the user’s shared skill directories, so external CLIs do not discover them automatically. User-created and imported skills keep the locations above.\n- Each skill is its own folder with a required `SKILL.md` at the root (the prompt / spec the agent reads). Sibling files and subfolders in that folder are also visible to the agent.\n- Global and project discovery roots may also contain one namespace/container level (for example an author folder): Cindy checks each discovery root's direct children, then their direct children, and stops there. A folder with `SKILL.md` is a Skill; its child folders are skill resources, not separate Skills.\n\n**Importing a local skill:**\n\n- On the Skills page, use **Import skill** (top-right) to pick a `.zip` package or a standalone `SKILL.md` file.\n- A zip must contain a `SKILL.md` (at the package root, or inside a single top-level folder). The YAML frontmatter must include non-empty `name` and `description` fields; `name` must match `^[a-z0-9-]{1,200}$`.\n- You then choose where to install: global, a known project, or another directory. Cindy extracts metadata from the file automatically and lists the skill with that name and description.\n- Imported skills can be uninstalled from the detail page, and you can publish them to SkillHub later if you want.\n\n**Using an installed skill:**\n\n- Hover over a local skill in the `/` suggestions and choose **View skill details** in its information panel to open that exact skill. This does not insert or send the command. Pi package skills remain usable from `/`, but do not show this local detail shortcut; manage them through their package.\n\n- Type `/` in the composer to open the slash-command palette; your installed skills show up there alongside built-in and agent commands. Pick one to run it.\n\n**Creating or learning a Skill from the composer:**\n\n- To create a Cindy **plugin** (`.cindy` with `ghost.json`), choose **Create with Cindy** from the Plugins page's add menu. This selects the separate built-in `cindy-plugin-creator` Skill, including when a user Skill has the same name. Its references cover plugin APIs and packaging; see the Plugins topic.\n\n- Cindy includes `/cindy-skill-creator`. Invoke it with a description of the Skill you want, such as `/cindy-skill-creator Create a Skill that checks release notes`. Invoking it without a description asks what you want to create.\n- The current Agent creates or updates the Skill directly with its normal file tools. It does not use the `/learn` staging flow.\n- Ask for a Skill in the composer, for example “Find a release-notes Skill” or `/cindy-skill-creator Search our organization's code-review Skills`. The Agent can search SkillHub's public marketplace or your organization's catalog. Without a specified scope, organization identities search both; personal identities search the public marketplace. Results identify their source, including when two Skills share a name. Search alone does not install anything.\n- You can also ask `/cindy-skill-creator` to upload your own Skill to SkillHub or publish a new version directly in the conversation. Give the local Skill folder (or the Skill just created) and, for a first upload, who should be able to access it. Ordinary local creation/editing does not automatically publish it.\n- Uploads use Cindy's signed-in account, which determines ownership; choosing sharing targets does not change that ownership. The Agent can list your published Skills, upload the folder, and check scan/review status. Upload success does not mean public approval. Updating requires confirmed original authorship and management access, and preserves visibility and ownership; a missing authorship flag is not treated as permission. This works in a local desktop task, including when controlled from a phone; SSH files must first be made available on the Cindy host.\n- Each Agent upload or version update follows the task's permission mode: Ask requests approval for that publication, Auto reviews it against your request, and Full Access bypasses operation approval. A previous helper-tool approval does not authorize later publications. Search, listing, and status queries remain automatic.\n- Unless you request another location, new Skills go to `~/.agents/skills/<name>/`. Ask for a project-specific Skill to create it under `<working-dir>/.agents/skills/<name>/`.\n- Cindy also includes `/learn`. Bare `/learn` distills the current task; `/learn <description>` learns from a described workflow; `/learn hub:<scope>:<slug> [instructions]` learns from a SkillHub Skill. Cindy gathers evidence, runs the distillation in a separate task, and shows a diff for review before saving the proposed Skill.\n- The built-in `cindy-skill-creator` and `learn` entries appear in the local Skill list with an Official badge. You can disable or enable either entry there; because they ship with Cindy, they cannot be uninstalled. The setting applies to new or restarted Agent sessions. Disabling `learn` also disables the **Learn this skill** action in SkillHub after the local Skill list has refreshed.\n\n**Disabling or uninstalling a local skill (Desktop):**\n\n- Open a skill's detail page to use the switch in the top action bar. For a market skill installed in multiple locations, select the local copy from the location dropdown beside the switch. Turning it off keeps the files and removes the skill from new-task command previews. Active tasks retain the skill preference snapshot applied when their Agent started, including their command palette. New or restarted Claude Code, Codex, and Pi agents apply that choice; an already-running agent keeps its existing context.\n- Renaming a local skill while publishing keeps its enabled or disabled setting. If renaming fails, Cindy restores the original folder and content.\n- For a Skill inside a namespace/container folder, renaming changes only the Skill's own folder name; the namespace stays the same.\n- The switch is local to this device and Cindy profile. It does not edit external CLI settings, sync to other devices, or override a native engine's own disabled state. Enable it again to let the engine discover it normally.\n- In the skill's details, use **… → Uninstall skill**. Confirm the location and shared-file impact. Standalone skills, including locally written skills without a market installation record, move to the system trash. Recovery is through the operating system's trash.\n- External source imports remove only their discovery links and keep the external source files. Package-owned Skills must be removed through their owning package. Cindy's built-in Skills cannot be uninstalled, but they can be disabled. Skills provided by Cindy plugins are managed on the owning plugin's page and loaded through Cindy's private skill entries.\n- Offline removal of an automatically synced skill is remembered on this device, including after restart or sign-in. Automatic sync skips it until you explicitly install it again.\n- Uninstalling a shared copy affects external CLIs that use it. It does not unpublish a skill or delete other users' copies. If file removal fails, Cindy keeps the installation. If cleanup is interrupted, the notice offers **Retry cleanup**. The unfinished operation survives closing, reloading, and restarting Cindy; reopen Skills in the same Cindy profile and account to continue. Cindy pauses conflicting installations until cleanup completes. Retrying never moves files to the trash a second time and preserves externally restored or replaced content.\n\n**Publishing your own skill:**\n\n- Find the publish action on the Skills page and point it at the skill's local folder. It zips the folder and uploads it — reading your directory in place, without copying or moving anything.\n- On first publish, choose visibility: personal identities support public or private; organization identities support public or organization sharing. Public publication is subject to review.\n- The local registry records what you published, so the app knows it's \"yours\" for future updates.\n\n**Updating an already-published skill:**\n\n- On your own skill's detail page you'll see a **发布新版本** (Publish New Version) button.\n- Edit the local folder however you want and publish — the version is **auto-incremented server-side**, you don't pick a number.\n- **Old versions stay live alongside the new one.** Users who already installed an older version keep it until they choose to update.\n- The display name and description are sent on every republish, so editing those metadata fields just means doing a republish.\n\n**Managing an already-published skill:**\n\n- Change who can see or use your skill from the management menu — public, shared with selected teams / departments, or private to you.\n- **Unpublish** makes it private again and returns ownership to your personal scope; it does not remove copies other users already installed.\n- **Authorship is fixed.** Only the original author's account can publish new versions of a given skill (enforced server-side as `NOT_AUTHOR`).\n\n**Browsing and installing others' skills:**\n\n- The marketplace lets you install skills into your global skills directory. You can request a specific version on install; the local registry tracks which version you have.\n\n**Notes:**\n\n- A skill folder without a `SKILL.md` at its root won't be picked up (lowercase `skill.md` is also accepted).\n- Project-scoped skills only show up in sessions whose working directory matches — useful for skills tied to a particular repo's conventions.\n- Editing files inside an installed skill folder takes effect on the next session start; you don't need to reinstall.\n- Uninstalling only removes your local copy. For your own published skills, use unpublish or manage visibility to change market availability.\n\nIn a new-task draft, the slash hover panel offers detail links for global skills. Project skill detail links are available after creating a task, when SkillHub can include that project in its scan.",+ summary: "Where the Skills page is; create, import, search, install, edit, disable, uninstall, publish, update, and manage visibility of agent Skills; Skills appear as \"/\" slash commands and are shared across Claude Code, Codex, and Pi.",+ content: "Skills are reusable agent capabilities you package as a folder and load into your sessions. They're managed on the same page as Plugins: click **Plugins** in the sidebar (or **More → Plugins** if you've hidden it), then the **Skills** tab (the same page is also under **Settings > Plugins > Skills**). The Skills tab lists **Local Skills** on this computer and the **SkillHub** catalog (filters **Public**, **Organization**, **Managed**) for finding, installing, publishing, and updating skills.\n\n- When another computer is linked, use the **Manage computer** picker at the top of the page to view, install, update, or uninstall skills on that computer instead.\n\n**Skills work across local agents:**\n\n- Skills are **engine-shared** — the same installed Skill is available to Claude Code, Codex, and Pi, so you install it once.\n\n**Where skills live on disk:**\n\n- User-created or imported global skills (available to every session): the shared root `~/.agents/skills/<name>/`, which is cross-linked to `~/.claude/skills/` and `~/.codex/skills/` so every engine sees the same skills. On Windows that's under `C:\\Users\\<you>\\`.\n- Project-scoped (only inside one working directory): `<working-dir>/.agents/skills/<name>/` or `<working-dir>/.claude/skills/<name>/`.\n- Cindy’s preinstalled skills and plugin-provided skills are kept in Cindy-managed directories and loaded into local Claude Code, Codex, and Pi sessions. Cindy does not add them to the user’s shared skill directories, so external CLIs do not discover them automatically. User-created and imported skills keep the locations above.\n- Each skill is its own folder with a required `SKILL.md` at the root (the prompt / spec the agent reads). Sibling files and subfolders in that folder are also visible to the agent.\n- Skills can also be grouped one folder deep (for example inside an author folder); deeper nesting isn't scanned. A folder with `SKILL.md` is a Skill; its subfolders are that Skill's resources, not separate Skills.\n\n**Importing a local skill:**\n\n- On the Skills page, use **Import skill** (top-right) to pick a `.zip` package or a standalone `SKILL.md` file.\n- A zip must contain a `SKILL.md` (at the package root, or inside a single top-level folder). The YAML frontmatter must include non-empty `name` and `description` fields; `name` may contain only lowercase letters, digits, and hyphens.\n- You then choose where to install: global, a known project, or another directory. Cindy extracts metadata from the file automatically and lists the skill with that name and description.\n- Imported skills can be uninstalled from the detail page, and you can publish them to SkillHub later if you want.\n\n**Using an installed skill:**\n\n- Type `/` in the composer to open the slash-command palette; your installed skills show up there alongside built-in and agent commands (see the Slash commands topic). Pick one to run it.\n- Hover over a local skill in the `/` suggestions and choose **View skill details** in its information panel to open that exact skill. This does not insert or send the command. In a new-session draft this works for global skills; project skill details become available once the session has been created. Pi package skills remain usable from `/`, but don't show this shortcut; manage them through their package.\n\n**Creating or learning a Skill from the composer:**\n\n- To create a Cindy **plugin** (a `.cindy` package) rather than a Skill, choose **Add plugin > Create with Cindy** on the Plugins page; it uses the separate built-in plugin creator (see the Plugins topic).\n- Cindy includes `/cindy-skill-creator`. Invoke it with a description of the Skill you want, such as `/cindy-skill-creator Create a Skill that checks release notes`. Invoking it without a description asks what you want to create.\n- The current Agent creates or updates the Skill directly with its normal file tools. It does not use the `/learn` staging flow.\n- Ask for a Skill in the composer, for example “Find a release-notes Skill” or `/cindy-skill-creator Search our organization's code-review Skills`. The Agent can search SkillHub's public marketplace or your organization's catalog. Without a specified scope, organization identities search both; personal identities search the public marketplace. Results identify their source, including when two Skills share a name. Search alone does not install anything.\n- You can also ask `/cindy-skill-creator` to upload your own Skill to SkillHub or publish a new version directly in the conversation. Give the local Skill folder (or the Skill just created) and, for a first upload, who should be able to access it. Ordinary local creation/editing does not automatically publish it.\n- Uploads use Cindy's signed-in account, which determines ownership; choosing sharing targets does not change that ownership. The Agent can list your published Skills, upload the folder, and check scan/review status. Upload success does not mean public approval. Updating requires that you're the original author and can manage the Skill, and preserves visibility and ownership. This works in a local desktop session, including when controlled from your phone; Skills on an SSH host must first be copied to this computer.\n- Each upload or version update follows the session's permission mode: Ask requests approval for that publication, Auto reviews it against your request, and Full Access doesn't ask. Each publication is approved on its own. Search, listing, and status checks don't need approval.\n- Unless you request another location, new Skills go to `~/.agents/skills/<name>/`. Ask for a project-specific Skill to create it under `<working-dir>/.agents/skills/<name>/`.\n- Cindy also includes `/learn`. Bare `/learn` distills the current session; `/learn <description>` learns from a described workflow; `/learn hub:<scope>:<slug> [instructions]` learns from a SkillHub Skill. Cindy gathers evidence, runs the distillation in a separate session, and shows a diff for review before saving the proposed Skill (see also the Slash commands topic).\n- The built-in `cindy-skill-creator` and `learn` entries appear in the local Skill list with an Official badge. You can disable or enable either entry there; because they ship with Cindy, they cannot be uninstalled. The setting applies to new or restarted sessions. Disabling `learn` also disables the **Learn this skill** action in SkillHub after the local Skill list has refreshed.\n\n**Disabling or uninstalling a local skill (Desktop):**\n\n- Open a skill's detail page to use the switch in the top action bar. For a market skill installed in multiple locations, select the local copy from the location dropdown beside the switch. Turning it off keeps the files and removes the skill from the `/` palette of new sessions. Running sessions keep the skills (and command palette) they started with; new or restarted Claude Code, Codex, and Pi sessions apply the change.\n- Renaming a local skill while publishing keeps its enabled or disabled setting. If renaming fails, Cindy restores the original folder and content.\n- For a Skill inside a namespace/container folder, renaming changes only the Skill's own folder name; the namespace stays the same.\n- The switch is local to this device and Cindy profile. It does not edit external CLI settings, sync to other devices, or override a native engine's own disabled state. Enable it again to let the engine discover it normally.\n- In the skill's details, use the **⋯** button next to the switch → **Uninstall Skill**. Confirm the location and shared-file impact. Standalone skills, including locally written skills without a market installation record, move to the system trash. Recovery is through the operating system's trash.\n- External source imports remove only their discovery links and keep the external source files. Package-owned Skills must be removed through their owning package. Cindy's built-in Skills cannot be uninstalled, but they can be disabled. Skills provided by Cindy plugins are managed on the owning plugin's page.\n- Offline removal of an automatically synced skill is remembered on this device, including after restart or sign-in. Automatic sync skips it until you explicitly install it again.\n- Uninstalling a shared copy affects external CLIs that use it. It does not remove the published skill or delete other users' copies. If file removal fails, Cindy keeps the installation. If cleanup is interrupted, the notice offers **Retry cleanup**, even after restarting Cindy (reopen Skills with the same account); installs that would conflict wait until cleanup finishes.\n\n**Viewing and editing a local skill:**\n\n- Click a skill under **Local Skills** to open its detail page: its files and frontmatter, an **Edit** / **Save** action for editing files in place, and a **Usage** panel (last 30 days of local use, with **Diagnose this Skill**, which creates a new session with usage stats filled into the composer; it doesn't send automatically).\n\n**Publishing your own skill:**\n\n- On a local skill's detail page, click **Publish to Market**. Cindy packs the skill folder in place (nothing is copied or moved) and uploads it.\n- On first publish you set the display name, description, version, optional tags, and **Visibility**: personal identities choose **Public** or **Only me**; organization identities choose **Public** or **Team** (and can publish as a team). Public publication goes through review; the detail page shows the review status.\n- Cindy records what you published, so it knows the skill is yours for future updates.\n\n**Updating an already-published skill:**\n\n- On your skill's detail page, click **Publish new version**. The **Version** field is pre-filled with the next patch version (format `x.y.z`); you can change it. Describing what changed is required.\n- Publishing a new version doesn't change visibility. Users who installed an older version keep it until they update (**Get update** on the market card, or **Update to v…** on the detail page).\n\n**Managing an already-published skill:**\n\n- In the SkillHub catalog (the **Managed** filter lists skills you can manage), open the skill's **Manage** menu: **Edit info** (display name and description), **Manage visibility** (**Public** / **Team** / **Only me**; choosing Public submits the current version for review, Only me removes it from the market), **Clone**, and **Delete**.\n- **Delete** removes the cloud release from the market and team access and can't be undone. Copies other users already installed, and your local files, are not affected.\n- Only the original author — or, for team-published skills, the team's admins and publishers — can publish new versions.\n\n**Browsing and installing others' skills:**\n\n- Search the SkillHub catalog, then click **Clone** on a skill and choose an install location: **Global**, one of your **Projects**, or **Other directory**. If a skill with the same name already exists there, Cindy asks before overwriting and keeps a backup of the old copy.\n\n**Notes:**\n\n- A skill folder without a `SKILL.md` at its root won't be picked up (lowercase `skill.md` is also accepted).\n- Project-scoped skills only show up in sessions whose working directory matches — useful for skills tied to a particular repo's conventions.\n- Editing files inside an installed skill folder takes effect on the next session start; you don't need to reinstall.\n- Uninstalling only removes your local copy. For your own published skills, use **Manage visibility** or **Delete** to change market availability.",+ },+ {+ id: "storage",+ title: "Storage and cleanup",+ summary: "Settings > Storage shows how much disk Cindy uses and lets you set the Chat workspace folder, clear media caches, run Disk cleanup, clean the database (permanently removes old archived history), and handle Worktree cleanup.",+ content: "**Settings > Storage** is where you see and manage what Cindy keeps on this computer. Your database and media cache stay local, and Cindy never deletes your history automatically — every cleanup here is something you start and confirm.\n\n**Chat workspace:**\n\n- The folder used by new sessions that have no project folder. Cindy creates an account-specific subfolder inside it.\n- **Open directory**, **Change directory**, **Restore default location**, and **Full path** (with **Copy path**).\n- Changing it only affects new sessions — existing files stay where they are.\n\n**Local storage overview:**\n\n- Shows database size, media files, the image cache, and the message attachment cache. **Refresh storage usage** recalculates.\n\n**Media files:**\n\n- **Image cache directory** and **Message attachment cache directory** — **Open directory** to look inside, or **Clear image cache** / **Clear attachment cache**. Clearing deletes only those cache files, but images or attachments in older messages may stop showing, and it can't be undone.\n- **Remote image cache** — images downloaded for reuse. **Clear remote cache** frees the space; images download again when opened, and original files and saved attachments are kept.\n\n**Database:**\n\n- **Database cleanup** — click **Scan database** to see what would be removed. It targets the message history of **soft-deleted and archived** tasks not updated for longer than the period you pick (**7 days** by default, or 1, 3, or 6 months).\n - **Clean Unarchived Tasks** (off by default) also includes old unarchived tasks: their history is removed, their agent context is reset, and they leave the task list (teammates' main tasks stay). Cindy asks you to confirm, because such a task could still be running.\n - **Back up database before cleanup** is on by default (the default backup lives in the database folder and only the latest cleanup backup is kept; **Choose backup folder** to put it elsewhere).\n - **Confirm database cleanup** → **Restart and clean up database**: Cindy restarts, backs up (if enabled), removes the history, compacts, and verifies the database before opening it. If a step fails, Cindy keeps or restores the original database.\n - **This permanently removes that history.** With a backup you can restore task and message data from it (**Open cleanup backup**); without one, it can't be recovered. Records used to review or undo file changes from those messages aren't in the backup. Files and generated media on disk are not deleted.\n- **Database reminder threshold** — Cindy reminds you to clean up when the database grows beyond this size (10 G by default, 1–1024). **Do not remind me again** hides these reminders.\n\n**Disk cleanup and storage check** (below the database card):\n\n- **Disk cleanup** — **Scan** finds media no session references anymore, caches over their limit, obsolete leftover folders, and stray temporary files, and shows the totals first. Nothing is deleted until you click **Clean up**. Images in composer drafts and queued messages are always protected.\n- **Storage check** — **Check** compares Cindy's media records with the files on disk and reports any mismatch. It never deletes anything.\n\n**Worktree cleanup:**\n\n- Lists worktrees Cindy is still waiting to clean up (for example, still in use or with changed contents), with their state and reason. Use **Retry cleanup**, **Keep directory** to keep it, or **Refresh**. See the worktrees topic for how worktrees are created and removed.\n\n**Notes:**\n\n- Archiving or deleting a session in the sidebar keeps its history in the database; Database cleanup is what actually frees that space (see the sidebar topic).\n- Logs live elsewhere: **Settings > About** (see the updates-and-about topic).",+ },+ {+ id: "teammates",+ title: "Teammates (Cindy and custom companions)",+ summary: "Teammates are long-lived AI companions (the built-in Cindy plus ones you create) listed at the top of the left sidebar. Chat with them, set their profile, model, memory, capabilities, projects, voice, messaging platforms, and routines; let them run background tasks, follow up on your sessions, and keep todos; put several in a group chat; choose which computer runs them.",+ content: "A **teammate** is a long-lived AI companion with its own name, avatar, personality, memory, skills, and one continuous conversation. **Cindy** is built in and is added automatically the first time you use teammates; you can create more. A teammate belongs to your account, not to one computer: you see the same teammate on every computer and on your phone, but it runs on one **main computer**.\n\n**Where teammates are:**\n\n- The **Teammates** section sits at the top of the left sidebar, above **Pinned** and your task list. It mixes private chats and group chats, newest activity first; pinned ones stay on top. The first 5 show, then **Show all N**.\n- Click **Teammates** (the section title) or its arrow to collapse the section into one row of avatars (overflow goes into **+N**). An unread dot marks new replies.\n- Hover the section header for the **+** menu: **Discover Teammates**, **New teammate**, **New Group Chat**, and **Enter Invite Link…**. If you have hidden teammates, an eye button there shows or collapses them.\n- Right-click a teammate row: **Pin to top** / **Unpin**, **Hide**, **Duplicate teammate** (not for Cindy), **Delete permanently**. Hidden teammates move to **Hidden (N)** at the bottom of the section; click the eye on a row to unhide it.\n- The sidebar **Search** also finds teammates (name, introduction, skills) and group chats (group name, member names); they're listed before session results.\n- Teammates running on your other computers also show up here, under their main computer.\n- When the sidebar is collapsed to icons, the teammates icon (with an unread count) opens the teammate you last viewed.\n\n**Chatting with a teammate:**\n\n- Click a teammate to open its chat and type in the box (\"Say something to …\"). The conversation is one continuous timeline; it isn't replaced each day. Long histories are compacted automatically, and the teammate keeps the full record and its memory.\n- The **permission** chip under the message box works like in a normal task (**Default permissions**, **Auto-review**, **Full access**, and for Claude Code also **Auto accept edits**). There's no model picker in the teammate's message box; set the model in its settings.\n- Hover a message for **Reply** (quotes it above the message box) and **Start a thread**. A thread opens in a side panel with its own separate conversation. Work started in a thread reports back in that thread. Threads are desktop-only for now.\n- The chat header has the teammate's name/avatar and a gear (both open its settings), a grid button (**… at a glance**, see below), and for teammates on this computer **Add … to a Group Chat** (**New Group Chat…** or **Add to Existing Group**).\n- In a normal task you can also @-mention a teammate (\"Message or delegate\") to send it a message or hand it work.\n\n**Creating a teammate:**\n\n- **+ → New teammate** opens **Create a teammate**. Pick a portrait (or **Upload image**), enter a **Name** (names must be distinct), and check the **Model**. It follows the default teammate model unless you change it (\"This selection applies only to this teammate\"). If no model is connected, connect a provider first. Click **Create teammate**.\n- Cindy then prepares the teammate (\"Getting ready to meet\"), and it sends a first hello. You can start typing right away. If preparation pauses, everything done so far is saved; use **Change Model** or retry.\n- **Import an Agent** (in the same dialog) brings in a teammate from Hermes or OpenClaw on this computer. You can bring its personality, memories, skills, environment and connections, automations, and messaging bots.\n- **Discover Teammates** lists ready-made teammates with their skills and routines; click **Invite** to add one. Their routines start turned off.\n- You can also ask Cindy to create a teammate for you.\n\n**Teammate settings** (click the name/avatar or gear in the chat header; changes save automatically):\n\n- **Profile** — avatar (**Change avatar**; PNG, JPEG, or WebP under 5 MB), name (\"What Should We Call Them?\"), and **Self-introduction**.\n- **Personality & conversation** — free text describing who the teammate is and how it talks.\n- **Memory** — **Use memory** toggle (off: memories aren't read or saved), **What you know about me**, and the teammate's saved memories grouped as About you / Your preferences / Projects / References. Search, edit, or delete them. Each teammate's memory is its own, separate from Settings > Personalization > Memory.\n- **Linked Projects** — projects the teammate may read for background when needed (**Choose a project**, **Add**). Linking doesn't assign the project's sessions or start background work.\n- **Model** — **Primary Model** with up to 5 candidates (**Add fallback model**; **Backup models (N)**). If one is unavailable, it automatically continues with the next. **Task Model** is the model for new background tasks; by default it uses **Inherit Primary Model**. **Restore default model** goes back to the shared default.\n- **Voice** — speech model, voice, speech speed, and **Read by sentence** for voice conversations (see below).\n- **Capabilities** — choose which **Built-in tools**, **MCP** servers, and **Skills** this teammate may use (items unavailable on this computer are marked). By default it follows the tools enabled on this computer; turning one off switches to an explicit list. **Input Handoff** lets it ask you to type a password or code on any of your devices; it never sees the value.\n- **Messaging platforms** — give the teammate its own bot (see below).\n- **Routines** — follow-ups, built-in routines, nightly learning, and your own routines (see below).\n- **Search teammate history** — search past messages, background tasks, and archived records.\n- **Advanced configuration** — **Open folder** opens the teammate's own folder (SOUL.md, USER.md, system_prompt.md, skills/, memories/). Use it only to fix something by hand.\n- Also on the settings home: **Main Computer** (see below), **Restart teammate** (stops the current run for a fresh start; history, memory, and files are kept; send a message to continue), and **Permanently delete teammate**. Deleting removes its profile, memory, working directory, channel connections, and automated tasks. Task history and separate worktrees are kept. A paused teammate shows **Resume teammate**.\n\n**What a teammate can do for you:**\n\n- **Background tasks:** for bigger work, a teammate starts a normal Cindy task. The task shows in your task list as \"…'s background task\", and a **Background task** card in the chat lets you **Open task** or **Stop task**. The result comes back to the teammate's chat.\n- **At a glance** (grid button in the chat header) has up to three blocks:\n - **Computers**: where it runs and other computers it can move to.\n - **Todos**: open todos, due ones first. Click one to quote it in the message box and tell the teammate what to change. Right-click → **Delete**. **Check for me** asks it to look through your connected mail and chats for new todos; **Check every day** does that on weekdays at 9:00 and stays quiet when nothing is new.\n - **Outputs**: finished files from its tasks.\n- **Follow Up on My Sessions** (Settings → **Routines**): when on, the teammate reads summaries of sessions you create on this computer, assesses progress, and tells you about important changes. This uses your model allowance. It's off by default, and only one teammate can hold it; Cindy may offer to turn it on in chat (**Start Following Up** / **Not Now**). Tasks you hand to a teammate are followed by that teammate either way (built-in **Follow assigned sessions**). It reacts when they stall, need you, fail, reach a usage limit, or finish. Sessions a teammate is looking after show its avatar in the sidebar (\"This task is looked after by …\").\n- **Routines:** **Add Routine** gives the teammate standing instructions plus triggers (Hourly, Daily, Weekdays, Weekly, Monthly, Interval, Cron, or plugin events; any trigger can start a run). Each routine has **Run Now**, **Run History**, and an optional **Stay quiet when checks find nothing**. Built-in routines can only be paused or resumed. Routines run on the teammate's main computer and are managed on the desktop.\n- **Nightly learning** (built-in, on by default): every night at 3:30 (or weekly), it reviews new conversations, tidies memories, and saves approaches that keep coming up as skills. When there's nothing new, it skips without calling a model. Open it under **Routines** for **Review now**, **Frequency**, **History** (with **Undo** per change), and **Approaches being watched** (**Make it a skill** / **Don't learn**). It runs only on the main computer; with memory off, only skills are reviewed.\n- Teammates can message each other; those exchanges appear as cards in the chat.\n\n**How permissions work (high level):**\n\n- When you ask in its chat, a teammate acts on your behalf under the chat's permission mode. **Full access** doesn't ask, **Auto-review** reviews automatically, and **Default permissions** asks you. No extra confirmation cards are added.\n- When it acts on its own (a follow-up wake-up, a message from another teammate or a stranger), it can only touch its own background tasks and tasks already handed to it. It can keep those going toward the original goal or switch their model. Anything else on your sessions (new requirements, stopping, renaming, archiving, creating or changing routines) becomes a todo or a note to you. Reply to it and the teammate does it on your behalf.\n- Routines you set up run as your standing instructions.\n\n**Messaging platforms (the teammate's own bot):**\n\n- On **Messaging platforms**, click **Add platform** and paste the credentials of a bot you created on **Telegram**, **Feishu** / **Lark**, **WeCom**, **DingTalk**, or **Discord**. Credentials stay on this computer. If the same bot is already set up in Settings > IM bots, choose **Use the bot already set up in Cindy** to move it over; Settings > IM bots is then left empty. A bot can be connected in only one place.\n- You can also link this computer's **Personal WeChat** to one teammate at a time. Where available, you can assign the official X bot (@askmycindy) to it; X replies are public.\n- Anyone who messages the bot is talking to the teammate. Your own messages go into its main conversation with its permissions. Each group, topic, and DM from someone else gets its own hidden task with restricted permissions; risky actions there need your confirmation. If other people can reach the bot, prefer **Auto-review** or **Default permissions** over **Full access**.\n- Each connection lists its **… tasks**; **Start over** gives one a clean context while keeping history. For Telegram groups, choose **@ only** or **Always** (reads every message and decides whether to chime in; each one costs tokens). **Remove** stops the bot and deletes its credentials. Memory is shared across all of a teammate's channels.\n\n**Voice conversations (desktop):**\n\n- Set up **Voice** in the teammate's settings (needs a text-to-speech model from a connected provider). Then click **Start voice conversation** in the message box. Speak; it sends after a pause and reads replies aloud. Speak again to interrupt; **Mute** and **End** are on the call bar.\n- Work you ask for during a call is handed to the teammate's main conversation (\"Work from this call\"). A record of the call is saved as a card. Voice mode works for teammates on this computer, one-on-one only; it's not available on the phone.\n\n**Group chats:**\n\n- **+ → New Group Chat**: give it a name and choose teammates (up to 6). Group chats live in the same Teammates section.\n- **@** a teammate to have only them reply, or **@Everyone**. **Group Settings** has **When No One Is @Mentioned** (**Anyone Can Reply** / **Only @Mentioned Reply**). A round runs up to 3 passes and 10 messages, then waits for **Continue Discussion**.\n- For real work, the **Organizer** can split it into steps. Review the plan card, then **Start Plan** or **Skip Plan**. Each step waits for **Start Next Step** unless **Step handoffs** is on. Set a **Project Folder** for the group; with a git project, each work split gets its own worktree and branch.\n- Invite people with **Invite to group** / **Invitation link**; they can bring their own companions (**Invite my companion**). **Access for other members** controls what others may use from your companion: **Owner only**, **Group conversation only**, or **Owner authorized tools and information**.\n- Group messages are separate from the teammate's private chat; memories it saves in a group are still its own. Groups can be archived, left (**Leave group**), or dissolved by the owner (**Dissolve group**).\n\n**Main computer and other devices:**\n\n- The computer where you created a teammate is its **Main Computer**. Its working folder, logins, model connections, and routines live there. Other computers and your phone see the same teammate and can chat with it through remote control. This needs \"Allow same-account devices to control this machine\" on in Settings → Remote Connection on the main computer.\n- To move it, use **Main Computer** in its settings or **Computers** in **at a glance**, then **Move to \"…\"**. Chat history, memory, skills, todos, and routine definitions come along. Working directories, connections, and local files stay where they are. On the new computer it uses that computer's default permissions, and if its model isn't available there, it follows that computer's default model.\n- If the main computer is offline, other computers show the cloud copy of the conversation (read only) and offer **Continue on This Computer**. Only visible messages and final replies are saved in the cloud; tool details stay on the main computer.\n- If you had a separate Cindy on several computers, Cindy's settings show **Merge Duplicate Cindys**. It merges them into one and keeps everyone's history, memories, skills, todos, and routines.\n\n**On your phone:**\n\n- The phone's home screen starts with the same **Teammates** section (tap **+** to create a teammate or group chat). Creating a teammate needs a computer that's online and signed in to the same account.\n- In a chat, tap the teammate's avatar at the top for **at a glance**. Use the settings button for **Teammate settings**: Profile & Personality, Memory, Skills & Connections, Artifacts, Teammate Messages, Models & Permissions, Main Computer, and Messaging platforms.\n- Routines can't be managed on the phone (except **Check every day**). Voice conversations and private threads are desktop-only.\n\n**Settings shared by all teammates** (Settings > General > **Teammates**):\n\n- **Default model**: used by teammates that don't choose their own, independent of new-task model choices.\n- **Show which message a reply answers**: shows above a teammate's reply which of your messages it answers. It's off by default.",+ },+ {+ id: "tools",+ title: "Tools and automation settings",+ summary: "Settings > Tools turns built-in agent tools (Collab Mode, SSH commands, Document Toolkit, Slack, Feishu Bot, Personal WeChat, Browser) on or off per project and adds custom MCP servers; Settings > Automation controls browser, computer, and Android automation.",+ content: "Two Settings tabs decide which tools the agents get: **Tools** (built-in tools and your own MCP servers) and **Automation** (letting agents operate a browser, this computer, or an Android phone). Tool changes only apply to **new or restarted** sessions — a running session keeps the tools it started with.\n\n**Settings > Tools > Built-in Tools:**\n\n- Each row is a tool the agents can call, with an on/off switch:\n - **Collab Mode** — lets a session start a team of worker agents (see the collaboration topic). Turning it off only affects new sessions and doesn't stop workers that are already running.\n - **SSH commands** — run commands on remote computers through the SSH hosts you've configured (see the remote topic).\n - **Document Toolkit** — create and read back PDF, Word, Excel, and PowerPoint files.\n - **Slack** — search, read, and send messages with your connected Slack account.\n - **Feishu Bot** — send files and notifications to Feishu users through your bound bot.\n - **Personal WeChat** — send messages to known WeChat contacts.\n - **Browser** — browse and act on web pages. It's listed here only when the scope is **New session defaults**; per-project control is under **Settings > Automation**.\n- Core tools (memory, scheduling, and the helper that tells agents what's available) are always on and aren't listed. Smart Contacts has its own switch under **Settings > Personalization > Smart Contacts**.\n- **Scope picker** (top right): **New session defaults** applies to sessions without a project override; choosing a recent project changes the setting for that project only. Project settings take priority. When you open Settings from a project session, that project is selected first.\n- A small label on each row shows where the current value comes from: **project**, **user default**, or **product default** (hover for details). After you change a row, it shows **Custom**; **Restore default** clears your change.\n\n**Settings > Tools > External tools (custom MCP):**\n\n- Click **Add MCP server** to connect your own **remote** MCP server. Enter a **Display name**, the **Transport** (http or sse), the **Endpoint URL**, and optionally a **Bearer token** and extra **Headers**.\n- HTTP / Streamable HTTP servers work with both Claude and Codex; **SSE works with Claude only**. Codex only supports Bearer-token authentication, so custom headers apply to Claude.\n- A saved server is available to new sessions right away; sessions already running aren't affected. Use the row's edit or delete buttons to change or remove it (deleting stops loading it in new sessions).\n- Local (command-line) MCP servers can't be added here — only remote endpoints.\n\n**Settings > Automation:**\n\n- **Browser automation** (on by default) — gives agents a dedicated browser, separate from your everyday one.\n - **Automation browser target**: **Standalone external browser** (the default — a separate Chrome window with its own \"Cindy\" profile, so it never touches your normal tabs, bookmarks, or sign-ins) or **Sidebar embedded browser** (the agent works in browser tabs in the right sidebar). The two keep separate sign-ins. If the embedded browser loses its connection, use **Reconnect Browser** / **Recover Browser**.\n - With the external browser, click **Open agent browser** to sign in to sites once; the agent keeps that session afterward. If no browser is found, **Download Chrome**.\n - **Use my browser logins** (external browser only) — lets the agent open sites with your signed-in Chrome, Edge, or Brave accounts. Cindy asks you to confirm first; on macOS it also needs **Full Disk Access**. Turning it off deletes the copied logins.\n - This switch applies to the project of the session you came from (or to new-session defaults if there isn't one).\n- **Computer automation** (off by default) — lets agents see windows, read the UI, click, and type in local apps. It's powered by the open-source **CUA Driver**, which Cindy installs the first time you turn it on.\n - On macOS it needs **Accessibility** and **Screen Recording** for CuaDriver; Cindy walks you through granting them (**Grant**, **Recheck**). macOS may still ask for extra confirmation for a specific app or action later — approve those only when you've just started a computer-use task.\n - Windows needs no extra permission; keep the target app visible on the active desktop. On Linux, the desktop session must allow window inspection and input automation.\n - **Check for updates** / **Update** keeps the driver current (you may need to re-grant permissions after an update).\n- **Android automation** (off by default) — lets agents inspect and operate a connected Android phone over ADB. Cindy prepares ADB for you. To connect: use a data-capable USB cable, enable **Developer options** and **USB debugging** on the phone, keep it unlocked, allow the debugging prompt, then click **Recheck**. Pick a default **Device** (or **Auto**) and, if needed, a custom ADB path.\n- Computer and Android automation apply to the whole computer (not per project) and only to sessions started afterward.\n\n**Notes:**\n\n- Plugins add more tools on their own pages under **Settings > Plugins** (see the plugins topic).\n- If a tool you expect is missing in a session, check its switch here (and the project scope), then start a new session.",},{id: "updates-and-about",title: "App updates, version and logs",- summary: "How auto-updates apply; where to find the app and agent versions, the debug-log toggle, and the logs folder.",+ summary: "How app updates download and apply (relaunch banner, idle install, Beta channel); where to find the app and agent versions, agent updates, debug logs, log upload, and the logs folder.",tab: "about",- content: "Cindy checks for updates automatically and downloads them in the background. When an update is ready, you're prompted to **relaunch** the app to apply it.\n\n**Settings > About:**\n\n- **App version** — the desktop app's release.\n- **Claude Code version** / **Codex version** — the agent CLIs Cindy manages and downloads at runtime into your user data (they update separately from the app).\n- **Auto-relaunch when idle** — an optional toggle that lets a downloaded update apply itself during idle time instead of waiting for you to click relaunch.\n- **Debug log toggle** — turn on more verbose logging when reproducing a problem. Leave it off for normal use.\n- **Open logs folder** — opens the logs directory in your OS file browser, useful when you need to attach logs to a bug report.\n- **Storage** — a storage-management card for reviewing / clearing local app data.\n\n**Notes:**\n\n- By default, a downloaded update only triggers a relaunch prompt — it won't restart you unexpectedly. If you enable **Auto-relaunch when idle**, Cindy will restart itself automatically once the idle and busy gates pass, without an additional prompt.\n- If you want to skip auto-update temporarily, you can dismiss the relaunch prompt; the update applies next time you start the app.\n- The Claude Code / Codex versions you see here are what Cindy's **managed agents** run as — they're independent from any Claude / Codex CLI you have installed globally on your shell.",+ content: "Cindy checks for updates automatically and downloads them in the background. When an update is ready, a banner appears in the sidebar (**Updated to <version> · Relaunch to apply**) with a **Relaunch** button and **See What's New**.\n\n**Applying an update:**\n\n- Click **Relaunch** to install. If a task or other background work is still running, Cindy warns you first (**A task is still running**) and offers **Restart App Anyway** or **Cancel**.\n- Click **Later** to hide the banner. The flame button on your account card at the bottom-left of the sidebar then shows **Update ready — click to review** and brings the banner back. Otherwise, the update is applied the next time you start Cindy.\n- When no update is pending, the same flame button opens **View release notes**.\n- You can also ask a Cindy agent to update Cindy or to turn idle installation on or off; Cindy always shows a confirmation card first.\n\n**Settings > About:**\n\n- **App version** — shown here in the China edition. In every edition, the app version also appears under your name on the account card at the bottom-left of the sidebar.\n- **Install updates while idle** (off by default) — after an update finishes downloading, Cindy restarts to install it once the computer has been idle for 10 minutes and no tasks are running. On Linux this row only explains how updates work there: system `.deb` installs ask for your system password; other installs use their package manager.\n- **Usage statistics** — turn launch and usage reporting on or off (on by default once you accept the Privacy Policy).\n- **Claude Code version** / **Codex version** / **Pi version** — the agent runtimes Cindy manages. They update separately from the app: open the row's menu to **Check for updates** or update (Cindy restarts to apply it). For Pi you can also **Update from Pi GitHub** or **Restore Cindy release**.\n- **Debug logs** — turn on more verbose logging while reproducing a problem. Leave it off for normal use.\n- **Logs directory** → **Open** — opens the logs folder in your file manager.\n- **Upload logs automatically on crash** (off by default) and **Upload logs** → **Upload** — sends recent run records to Cindy and gives you an upload code to quote when you report a problem.\n- **Terms of Service**, **Privacy Policy**, and **Connect with Cindy** community links.\n\n**Related settings:**\n\n- **Beta channel** — **Settings > General > Experimental**. Gets upcoming updates early; takes effect after a restart. Not shown on Linux.\n- **Storage** — its own Settings tab (**Settings > Storage**): media and cache usage and cleanup, database cleanup, the Chat workspace folder, and **Worktree cleanup**.\n\n**Notes:**\n\n- A downloaded update never restarts Cindy unexpectedly unless you turn on **Install updates while idle**.\n- The Claude Code / Codex / Pi versions shown here are what Cindy's managed agents run. They are independent of any Claude Code, Codex, or Pi CLI installed on your system.",+ },+ {+ id: "usage",+ title: "Usage, billing, and Activity Monitor",+ summary: "Where to see how much you've used — Settings > Usage history (token usage by day, agent, model, and session), Settings > Usage and billing (plan, balance, top-ups, orders), the macOS menu bar usage icon, and Activity Monitor (CPU / memory of Cindy and agent processes).",+ content: "Cindy has several places that answer \"how much have I used?\". Pick the one that matches the question.\n\n**Settings > Usage history — tokens used in Cindy:**\n\n- Shows token usage produced inside Cindy. It's available whenever you're signed in, including local mode. It counts tokens, not money.\n- **Device**: **All devices** (the default, combined across your computers) or one device such as **This device**. Other computers are read only while they're online and allow remote control; if one can't be reached, its last-read data is used and the page warns that totals may be low.\n- **Time range**: **All time**, **Last 30 days** (default), **Last 7 days**, or **Today**. Use **Date**, or click a day in the heatmap, to look at a single day.\n- Cards: **Overview** (tokens, **Active streak**, cache hit rate, **Models used**), **Activity heatmap**, **Daily tokens (30d)** stacked by model, **By Agent / harness**, **By model** (input, output, cache read, cache write, cache hit rate), and **Top sessions by tokens** — click a session there to open it (when it's reachable).\n- Recording started when this page shipped; earlier history isn't backfilled.\n\n**Settings > Usage and billing — your Cindy plan and balance:**\n\n- Shown only when you're signed in with a personal Cindy account (not local mode or an organization account).\n- **Current plan** — your subscription, renewal or end date, and **View plans** / **Change plan**. You can **Cancel Subscription** (no refund; the plan stays available until the period ends, and you can't buy a new plan until then) or **Resume Subscription**. Stripe-billed plans also offer **Manage in Stripe**.\n- **Available balance** — **Plan Balance**, **Top-up Balance**, and **Promotional Balance**. **Add Funds** buys more.\n- **Balance Usage** — how much of each balance is used and remaining, plus **Promotional Balance Details** (each grant's amount and expiry).\n- **Order history** — top-up and subscription orders started in Cindy, with status. Unpaid orders are canceled automatically when they time out. **Request Invoice** on a completed order shows the details to email to support.\n- Provider pages link here too: **View usage** / **Buy a plan** on **Cindy AI** in **Settings > Providers** (see the providers topic).\n\n**Menu bar (macOS):**\n\n- **Show remaining usage in the menu bar** (on by default) is in **Settings > General > App Behavior**. It adds a Cindy icon to the macOS menu bar; click it to see what's left on your subscriptions, your Cindy balance, and today's usage.\n\n**Activity Monitor — what's using CPU and memory:**\n\n- Open the **Menu** button at the top left and choose **Activity Monitor**. It opens in its own window.\n- **Agent Processes** (the agent runtimes behind your tasks, with the task they belong to) and **App Processes** (Cindy's own processes), each with **CPU**, **Memory**, and **PID**. Click a column to sort.\n- Select an agent process and click **Stop Process** to force-stop it. Every task using that process is interrupted; send another message in the task to start its agent again. Cindy's own app processes can't be stopped here.\n- It shows processes on this computer only, not on remote computers you control.\n\n**Other usage views:**\n\n- Voice input usage (duration, count, estimated cost): **Settings > Voice Input** (see the voice input topic).\n- A plugin's own calls: **Your calls** on its page under **Settings > Plugins**; a skill's recent use: its **Usage** panel on the skill detail page.",},{id: "voice-input",title: "Using voice input (dictation)",- summary: "Push-to-talk dictation that inserts ASR text into the focused composer; configure shortcut, mic, language, refinement, and dictionary in Settings > Voice Input.",+ summary: "Dictate with the global shortcut, the composer mic button, or by holding the composer; configure voice service, language, microphone, shortcut, refinement, dictionary, voiceprint filter, and sounds in Settings > Voice Input.",tab: "voice-input",- content: "Voice input lets you dictate instead of typing. Hold a shortcut, talk, release — the recognized text is inserted into whichever composer has focus.\n\n**Activation:**\n\n- Default shortcut: **Alt+Space** on macOS, **Ctrl+Shift+Space** on Windows. The shortcut is customizable on macOS and Windows. On Linux, global voice shortcuts are not supported yet (the setting is hidden).\n- The shortcut is **global** — it works even when Cindy isn't the focused app, so you can dictate into the composer from anywhere.\n- It's **push-and-hold**: press to start listening, release to stop and submit.\n- Pressing the microphone immediately shows the recording controls and starts the timer. The start cue acknowledges the action and plays independently while capture starts in the background. Once audio becomes available, it is retained even while the recognition service is connecting; up to about 60 seconds is buffered locally and sent in order when it connects. You can stop before connection completes without discarding that buffered audio.\n\n**Settings:**\n\n- **Microphone** — pick a specific device or leave on \"auto\" (system default).\n- **Language** — auto / 中文 (zh-CN) / 繁體中文 (zh-TW) / English / 日本語 / 한국어. Those are the supported recognition locales — there's no free-form locale field.\n- **Refinement** — optional LLM post-process that cleans up the raw ASR text (fixes punctuation, removes ums, joins fragments). You can write a custom refinement prompt to bias it toward your preferences.\n- **Mute system audio while recording** — toggle; reduces feedback / echo from your speakers picking up onto the mic. On by default.\n- When enabled, the start cue finishes before system audio is muted. Capture continues throughout. Audio captured before the mute finishes is retained too, so the beginning of your speech is not cut off; a brief amount of sound already playing through the speakers may also be captured.\n- **Fast activation** — toggle; keeps the mic warm so press-to-talk has lower latency at the cost of slightly more background mic usage. Off by default.\n- **Interaction sound** — a short sound cue when dictation starts / stops. On by default.\n\n**Dictionary (improves recognition of your terms):**\n\n- **Manual entries** — type terms you want recognized (people's names, product names, jargon).\n- **Learned entries** — when you edit the text shortly after dictation lands in the composer, the external-edit inspector picks up your corrections and feeds them back as automatic dictionary entries. Both manual and learned entries feed the refinement pass.\n\n**Permissions (macOS):**\n\n- **Microphone** — required; the page prompts the OS dialog if you haven't granted it.\n- **Input Monitoring** and **Accessibility** — may be required depending on how the global shortcut is captured. The page shows each permission's status (granted / denied / unknown) so you know what to grant in System Settings.\n\n**Notes:**\n\n- If Cindy reports \"Too many voice requests\", wait a moment before starting another recording. Cindy voice providers share this account limit, so automatic provider switching stops for that attempt. The limit does not mark the providers as faulty or impose a provider cooldown.\n- Push-and-hold is the primary activation mode; **Fast activation** changes the warm-up behavior, not the trigger itself.\n- If the global shortcut fails to register (e.g. it's already taken by another app like an IME), you'll see a warning in the page; pick a different shortcut.\n- If the connection drops mid-dictation, the text recognized up to that point is kept — it lands in the composer (or stays in the overlay with a copy button) alongside the error, instead of being discarded. That salvaged text is raw ASR output: refinement needs the connection that just failed, so it is skipped.",+ content: "Voice input lets you dictate instead of typing. The recognized text goes into the Cindy composer you're typing in, or, when another app is in front, is typed into that app's focused text field.\n\n**Starting and stopping:**\n\n- **Global shortcut** — default **Option+Space** (shown as Alt+Space) on macOS and **Ctrl+Shift+Space** on Windows. **Hold** it, speak, and release to finish. Or **tap** it once to keep recording hands-free, then press it again to finish. It works even when Cindy isn't the active app. On Linux the global shortcut isn't supported yet; use the in-app voice input button.\n- **Composer mic button** — click **Start voice input** in the message composer; click again to stop.\n- **Hold the composer** — press and hold the left mouse button inside the composer for a moment to start; release to stop and insert the text.\n- Recording starts right away when you trigger it. Audio captured while the recognition service is still connecting is kept (about 60 seconds is buffered) and sent once it connects, so the beginning of your speech isn't lost.\n- Outside Cindy, a small overlay shows the live transcript. **Finish and input** inserts it; **Close and discard** cancels. If Cindy can't type into the current app, the overlay keeps the text with a **Copy** button.\n\n**Settings > Voice Input:**\n\n- **Service source → Voice service** — **Cindy official service** (default; uses your Cindy sign-in) or **Custom voice service** (choose a recognition model that uses its own credentials, or **Custom realtime ASR** with your own WebSocket URL, model, and API key; **Test connection** checks it). When you're not signed in, use a custom service.\n- **Basics:**\n - **Language** — **Auto** (default) / Simplified Chinese / Traditional Chinese / English / Japanese / Korean. The China edition defaults to Simplified Chinese. There's no free-form locale field.\n - **Microphone** — **Auto** (system default) or a specific device.\n - **Shortcut** — click to record a new one: a key combination, a standalone F1–F24 key, or (macOS) holding Left/Right Command, Option, Control, or Fn. **Clear** removes it. Cindy rejects shortcuts that conflict with system or common editing shortcuts.\n- **Refinement:**\n - **Voice refinement** (on by default) — cleans up punctuation, terminology, and spoken phrasing. Turn it off to use the raw recognition result.\n - **Refinement rules** — editable instructions for your preferred style (they guide cleanup only; the model won't answer or expand content).\n - **Custom dictionary** — add names, product terms, and jargon, optionally with **Often misheard as** variants; **Import CSV** (one term per row) adds many at once. Filter by **All** / **Auto-added** / **Manual**.\n - **Auto-learning** (on by default) — when you correct dictated words by hand, Cindy records candidates and adds them to the dictionary after repeated evidence. Learning from edits in other apps is macOS-only.\n - **Sync across my devices** (on by default) — merges your dictionary between your computers that are online at the same time, device to device, without storing it on a server.\n- **Preferences:**\n - **Mute during voice input** (on by default) — temporarily mutes system audio and restores it afterwards. The start cue finishes before muting, and capture isn't interrupted. Not supported on Linux.\n - **Fast voice activation** (off by default) — keeps the microphone ready briefly after use so recording starts about 100 ms sooner; the mic is released after 30 minutes without use.\n - **Sound cues** (on by default) — short sounds when voice input starts and ends.\n- **Voiceprint filter → Recognize my voice** (off by default) — record your voiceprint (read three short passages) so only your voice is transcribed. The first use downloads a recognition model. **Strictness**: Relaxed / Standard / Strict. The voiceprint stays encrypted on this computer.\n- **Usage** — duration, count, and estimated cost, plus **Recent transcripts** you can copy or delete.\n\n**Permissions (macOS):**\n\n- **Microphone** — required. The page shows **Granted** or a **Grant** button.\n- **Shortcut listener** (Input Monitoring) — needed only for Fn, standalone Command / Option / Control, and standalone F1–F24 shortcuts. Regular key combinations don't need it.\n- **Automatic input** (Accessibility) — lets voice input type into other apps, read nearby text for refinement context, and learn from your edits there.\n\n**Notes:**\n\n- If Cindy reports \"Too many voice requests\", wait a moment before trying again.\n- If the shortcut can't be registered (for example, another app or input method already uses it), the page shows a warning; pick a different shortcut.\n- If the connection drops mid-dictation, the text recognized so far is kept — in the composer, or in the overlay with a copy button — instead of being discarded. That text isn't refined.",+ },+ {+ id: "workspace-panels",+ title: "Right sidebar panels (files, terminal, browser, side chat, and more)",+ summary: "Open the right sidebar next to a task and add panels with + (Add tab) — File browser, Review, Subagent, Background tasks, New side chat, Browser, Terminal, plugin panels — then reorder, move tabs to another edge, or pop them out into their own window.",+ content: "The **right sidebar** sits next to an open task. It holds tabs (\"panels\") that help you work alongside the conversation. Tabs belong to the task: switching tasks switches to that task's tabs, and they come back when you return. Plugin panels are the exception (see below).\n\n**Opening it:**\n\n- Click the panel icon at the top-right of the window (**Show sidebar**). The same area has **Maximize sidebar** / **Restore sidebar**, **Close sidebar**, and **Open the current tab in a new window**.\n- With no tabs yet, the sidebar shows **Add a panel** with the same choices as the **+** menu.\n- The right sidebar is only available while a task is open.\n\n**The + (Add tab) menu:**\n\n- **File browser** — browse the task's working folder (details below).\n- **Review** — the task's git changes; see the Reviewing changes topic.\n- **Subagent** — progress, results, and details of subagents created in this task. It's shown for local Pi tasks (and for tasks that still have Pi subagent records).\n- **Background tasks** — background agents, shell commands, and workflows started in this task, split into **Running** and **Completed**; open one to see its progress or output, or **Stop** it. Clicking a background-task card in the chat also opens this tab.\n- **New side chat** — see Side chat below.\n- **Browser** — a built-in web browser (details below).\n- **Terminal** — a shell in the task's working folder (details below).\n- **Plugins** — installed plugins that provide a panel are listed under their own heading.\n- Review, Subagent, and Background tasks can be open only once per task; the menu then shows **Open** and switches to the existing tab. You can open several File browser, Browser, Terminal, and side chat tabs.\n- Other tabs open automatically from elsewhere: the **Team** tab when you use Collaboration (see the Collaboration topic), plus some plugin and mail pages.\n- In IM bot chats, the menu is shorter (no Review, Subagent, Background tasks, Terminal, or side chat).\n\n**File browser:**\n\n- A file tree of the working folder next to a preview of the selected file. Drag the divider to resize the tree.\n- **Filter files…** narrows the tree by file name; the search button switches to **Search file contents** (with **Match case**). The tree also has **Collapse all**, **Refresh**, and **Show ignored directories**.\n- **Edit** a text file right in the panel, then **Save** (Cmd/Ctrl+S). If you switch files with unsaved changes, Cindy asks first.\n- Right-click a file for **Open in file browser** (opens it in another File browser tab), **Copy File Path**, and **Show In Folder** (these two for local tasks); web pages also offer **Open in sidebar** (the built-in Browser) and **Open in browser**. Click an image to preview it. Creating, renaming, or deleting files isn't done from this panel.\n- Drop a file onto the panel to preview it (one file at a time).\n- Also works for tasks on another computer or an SSH host (file-name filtering and search there need ripgrep on the remote machine). Tasks without a working folder can't use it.\n\n**Browser:**\n\n- Address bar (**Enter URL or search**), **Back**, **Forward**, **Reload**, and **New tab**.\n- **Screenshot** copies the page to the clipboard.\n- **Comment on the page**: click an element (Shift-drag to select an area), describe the change you want, and **Add comment** — the comment goes into the chat input so you can send it to the agent. Press Esc to leave comment mode.\n- **More** has **Open in system browser** and **Copy link**.\n- Sign-ins and cookies are kept between sessions. You can reference this task's browser tabs in the composer with **@** (see the Attachments topic).\n\n**Terminal:**\n\n- Opens a shell in the task's working folder. Choose the shell in **Settings > Personalization > Default terminal shell** (new tabs only).\n- Shortcut: **Ctrl+`** (**Open terminal**) focuses the existing terminal tab or creates one.\n- If the shell exits, click **Restart**. Terminals don't survive quitting Cindy.\n- Available only for tasks running on this computer — not for tasks on another computer or an SSH host.\n\n**Side chat:**\n\n- Ask a separate question with the main task's context without interrupting it. Open one with **New side chat** in the + menu, with `/side` (text after it becomes the draft question), or by selecting text in a reply and clicking **Side chat**.\n- It uses the context completed when you opened it; the main task keeps running. **Open full task page** shows it full-size; **Back to main task** returns.\n- Wait until the main task has its first reply. Not available in shared tasks you joined, in IM bot chats, or in tasks whose agent runs in another environment.\n- Closing the tab asks **Close side chat?** — it stops the side chat and deletes its history (**Close and delete**).\n\n**Plugin panels:**\n\n- Shown as tabs marked \"Global tab, kept across tasks\": they stay put when you switch tasks.\n- Reopen a closed plugin panel from the **+** menu or with **Open in the sidebar** (see the Plugins topic).\n\n**Managing tabs:**\n\n- Drag tabs to reorder them. Right-click a tab for **Close**, **Close others**, **Close all**, **Move to …**, and **Open in a new window**.\n- **Move to left side / top / bottom / right sidebar** places the tab on another edge of the chat area. You can also drag a tab out of the tab strip to an edge, or onto another tab strip.\n- **Pop out:** **Open in a new window** (or drag a tab outside the Cindy window) opens just that tab in its own window. Use **Move to window:** to add a tab to a window you already popped out, or **Merge back into the sidebar** to return it.\n- Shortcuts: **Previous / Next right sidebar tab** — Cmd+Shift+[ / ] on macOS, Ctrl+PageUp / PageDown on Windows, or Ctrl+Shift+Tab / Ctrl+Tab. On macOS, Cmd+W closes the active tab while the sidebar has focus. Tab-switching and **Open terminal** shortcuts can be changed in **Settings > Keyboard Shortcuts**.\n- Panel positions and widths are saved; **Settings > Appearance > Restore default layout** resets them.\n\n**Activity Monitor:**\n\n- Not a sidebar tab: open the menu button at the top-left of the window and choose **Activity Monitor**. It lists **Agent Processes** and **App Processes** with CPU and memory, and can **Stop Process** for an agent process (tasks using it are interrupted; send another message to restart).",},{id: "worktrees",title: "Working in an isolated git worktree",- summary: "Opt a session into an isolated git worktree so its changes stay on their own branch and folder.",- content: "A session can run inside its own **git worktree** — a separate working copy of the same repository on its own branch — so the agent's edits don't mix with your main checkout.\n\n**Opting in:**\n\n- In the new-chat view, click the working-directory chip → the **gear** icon to open the Advanced popover → tick the **worktree** checkbox.\n- The worktree is created **on first message send**, not on session create. An overlay shows the status (creating / ready / failed).\n\n**Where it goes:**\n\n- Folder: `<repo>/.cindy-worktrees/<auto-generated-name>/` — kept inside the same repo so git tooling can find it. Worktrees created by older versions under `.xdt-worktrees/` remain supported.\n- Branch: `cindy/<auto-generated-name>` — auto-created off the source branch you pick, which defaults to the repo's current branch (HEAD). Branches created by older versions under `xdt/` remain supported and do not need to be renamed.\n\n**Lifecycle:**\n\n- Created on first agent turn that needs files.\n- Survives across app restarts as long as the session is around. Transient closes (`/clear`, reconnecting, quitting the app) do **not** remove it.\n- The folder is removed only when you explicitly **delete or archive** the session; the **branch is preserved** (no force-delete), so you can keep the work or merge / cherry-pick from it later.\n- Uncommitted changes and new files inside the worktree are **saved in Git** before the folder is removed, and come back when the session is restored. Files ignored by `.gitignore` (dependencies such as `node_modules`, build output, caches) and Git repositories cloned or initialised inside the worktree are **not kept**; reinstall, rebuild or clone them again after restoring. Restored files go through Git's usual line-ending and symlink handling, like a normal checkout. Removal is skipped if you leave a `.worktree-keep` file in it or another live session is still using it.\n\n**When the option is disabled:**\n\n- Git is not installed on your machine.\n- The chosen working directory isn't a git repository.\n- The chosen directory is already inside an existing worktree (no nesting).\n\n**Error cases (shown on the overlay):**\n\n- `permission-denied` (can't write into the repo dir), `git-crypt-locked` (repo uses git-crypt and isn't unlocked), `dubious-ownership` (git's safe.directory check tripped), `lfs-error`, `not-a-git-repo`, `git-not-installed`.\n\n**Notes:**\n\n- The agent's **Read / Write / Bash** tools automatically operate inside the worktree path — you don't need to switch directories manually.\n- Because the branch survives the worktree, \"did the agent's changes go anywhere?\" is always answerable with `git branch --list 'cindy/*' 'xdt/*'`.\n- Worktrees are a per-session opt-in — they're not on by default and they don't change non-worktree sessions in any way.",+ summary: "Start a session in an isolated git worktree (Branch | worktree control in the new-session view) so its changes stay on their own branch and folder; what happens on archive/delete and how to restore.",+ content: "A session can run inside its own **git worktree** — a separate working copy of the same repository on its own branch — so the agent's edits don't mix with your main checkout.\n\n**Opting in:**\n\n- In the new-session view, pick a project that is a git repository. A combined control appears next to the project picker: the left half shows the **Branch** (the source branch, defaulting to the repo's current branch; click it to search and choose another), and the right half is a **worktree** checkbox. Tick **worktree** to use isolation.\n- The checkbox is off by default. Cindy remembers your last explicit choice for new sessions on that computer.\n- The worktree is created when you send the first message. While it's being created, an overlay shows **Creating worktree · input disabled** and the composer unlocks automatically when it's ready.\n- Next to the branch, a refresh button can pull the latest commits for the selected branch from its remote before you start.\n\n**Where it goes:**\n\n- Folder: `<repo>/.cindy-worktrees/<auto-generated-name>/`. Worktrees created by older versions under `.xdt-worktrees/` remain supported.\n- Branch: `cindy/<auto-generated-name>`, created from the source branch you picked. Older branches under `xdt/` remain supported.\n- Untracked local files (such as `.env`) aren't in a new worktree by default. List them in a `.worktreeinclude` file (`.gitignore` syntax) at the repo root and Cindy copies matching untracked files into each new worktree.\n\n**Lifecycle:**\n\n- The worktree survives app restarts, `/clear`, and reconnecting.\n- When you **archive** or **delete** the session, Cindy removes the worktree folder but **keeps the branch**, so you can still merge or cherry-pick from it. Uncommitted changes and new files are first saved as a git snapshot (the confirmation dialog warns you when there are any). Files ignored by `.gitignore` (dependencies like `node_modules`, build output, caches) and git repositories cloned inside the worktree are **not kept**.\n- Removal is skipped if the folder contains a `.worktree-keep` file or another live session is still using it.\n- **Restoring:** when you unarchive the session and send a message, Cindy recreates the worktree from the kept branch and re-applies the snapshot automatically. If you open a session whose worktree is missing, a banner offers **Restore worktree** (or **Apply changes** if only the snapshot is pending).\n- **Settings > Storage > Worktree cleanup** lists worktrees still awaiting cleanup, with **Retry cleanup** and **Keep directory**.\n\n**When the control doesn't appear:**\n\n- The control is hidden when Git isn't installed, the chosen folder isn't a git repository, or it's already inside a worktree (no nesting). It's also hidden for projectless chats.\n- While Cindy is still detecting the folder, or detection failed, the checkbox is disabled and its tooltip explains why.\n\n**Errors (shown if creation fails):**\n\n- Cannot access worktree directory (permissions), git-crypt repository is not unlocked, Dubious git repository ownership (`safe.directory`), git-lfs sync failed, Current directory is not a git repository, git not detected. Each comes with a suggested fix.\n\n**Other places that use worktrees:**\n\n- Automations can run in a **New worktree session** (see the Scheduling automations topic), collaboration Workers can get their own worktree (**Create a worktree for this Worker**), and importing a shared session offers **Create in a worktree**.\n- In **Sidebar settings → Session info**, tick **worktree** to show each session's worktree name; click it to open the folder.\n- A session whose working directory is a worktree can't be moved to another project yet.\n\n**Notes:**\n\n- The agent works inside the worktree folder automatically; you don't need to switch directories.\n- Because the branch outlives the worktree, you can always find the agent's work with `git branch --list 'cindy/*' 'xdt/*'`.\n- Worktrees don't affect sessions that don't use them.",},];diff --git a/apps/desktop/src/main/mcp-integrations/mcp-providers.ts b/apps/desktop/src/main/mcp-integrations/mcp-providers.tsindex a09cf57e12..b7cbfd00d7 100644--- a/apps/desktop/src/main/mcp-integrations/mcp-providers.ts+++ b/apps/desktop/src/main/mcp-integrations/mcp-providers.ts@@ -127,6 +127,7 @@ import { createRemoteSessionsStatusSetter } from './sessionStatusRemote.js';import { isCindyLearnSkillEnabled } from '../skillhub/activationPreferences.js';import { botLearningTracker } from '../maker-ipc/botLearningTracker.js';import { getLearnController } from '../learn-host/index.js';+import { HELP_KNOWLEDGE } from '../maker-ipc/helpKnowledge.generated.js';import { consumeLearnInvocationGrant } from '../learn-host/invocationGrant.js';import { createSkillhubAgentTools } from '../skillhub/agentTools.js';import { startGrokDeviceLogin, grokDeviceLoginStatus, cancelGrokDeviceLogin } from '../maker-host/grok-device-login-service.js';@@ -669,6 +670,7 @@ export function createDesktopMcpProviders(deps: DesktopMcpProvidersDeps): LiziMc// (LLM 调工具时) registerMakerIpc 早已执行完毕, holder 已 ready。xdtHelper: {runtimeCapabilities: deps.runtimeCapabilities,+ productHelpDocs: HELP_KNOWLEDGE,// Install / auto-update requests become owner-only Host cards (see agent-app-update/).appUpdate: createAgentAppUpdateCallbacks((sessionId, sessionInstanceId) =>deps.isCurrentLocalSessionInstance?.(sessionId, sessionInstanceId) === true),diff --git a/apps/mobile/src/i18n/locales/en/deviceLink.json b/apps/mobile/src/i18n/locales/en/deviceLink.jsonindex c1a312bbfc..54b0bf233d 100644--- a/apps/mobile/src/i18n/locales/en/deviceLink.json+++ b/apps/mobile/src/i18n/locales/en/deviceLink.json@@ -24,7 +24,7 @@},"connectStep1": "Install and open {{appName}} on your computer","connectStep2": "Sign in with the same account as your phone",- "connectStep3": "Under Settings → Remote Access, turn on \"Allow same-account devices to control this computer\"",+ "connectStep3": "Under Settings → Remote Connection, turn on \"Allow same-account devices to control this computer\"","brandEyebrow": "Cindy Mobile","cloudTeaserTitle": "{{appName}} in the Cloud, Coming Soon","cloudTeaserCopy": "Once it launches, the mobile app works on its own, no computer needed.",diff --git a/apps/mobile/src/i18n/locales/en/devices.json b/apps/mobile/src/i18n/locales/en/devices.jsonindex ebbf61bfb0..d4f9beaead 100644--- a/apps/mobile/src/i18n/locales/en/devices.json+++ b/apps/mobile/src/i18n/locales/en/devices.json@@ -62,7 +62,7 @@"filterMeta": "{{available}} available","filterMetaWithUnavailable": "{{available}} available · {{unavailable}} need attention","showUnavailableReason": "Expand a device to see why it is unavailable",- "emptyCopy": "Sign in to the same account on your computer, then enable “Allow same-account devices to control this computer” under Remote Control in Settings.",+ "emptyCopy": "Sign in to the same account on your computer, then enable “Allow same-account devices to control this computer” under Settings → Remote Connection.","emptyTitle": "No computers to show yet","hideUnavailableA11y": "Hide unavailable computers","showUnavailableA11y": "Show unavailable computers",diff --git a/apps/mobile/src/i18n/locales/en/remoteDesktop.json b/apps/mobile/src/i18n/locales/en/remoteDesktop.jsonindex c3700f10fb..5aae7f2db6 100644--- a/apps/mobile/src/i18n/locales/en/remoteDesktop.json+++ b/apps/mobile/src/i18n/locales/en/remoteDesktop.json@@ -65,7 +65,7 @@"permissionMissing": "Not granted","openPermissionSettings": "Open System Settings","recheckPermissions": "Check again",- "permissionActionFailed": "Could not open the permission guide. Try again in Remote control settings on the computer.",+ "permissionActionFailed": "Could not open the permission guide. Try again in Settings → Remote Connection on the computer.","permissionCheckFailed": "Could not check permissions. Make sure the computer is online and try again.","openGuideOnComputer": "Open guide on computer","permissionGuideOpened": "The permission guide is open on the computer. Continue there.",@@ -90,7 +90,7 @@"controlling": "Controlling","viewOnly": "View only","upgrade": "Update Cindy on the computer to use remote desktop.",- "disabled": "On the computer, enable remote desktop in Settings → Remote control.",+ "disabled": "On the computer, enable remote desktop in Settings → Remote Connection.","busy": "Another device or Agent is using the computer. Stop controlling it from the other device or wait for the Agent to finish.","connectionBusy": "Another device is connected to this remote desktop. Exit remote desktop on that device, then try connecting again.","connectionError": "Connection ended. Check that the computer is online, then reconnect.",diff --git a/apps/mobile/src/i18n/locales/ja/devices.json b/apps/mobile/src/i18n/locales/ja/devices.jsonindex 8563b01e86..715af42725 100644--- a/apps/mobile/src/i18n/locales/ja/devices.json+++ b/apps/mobile/src/i18n/locales/ja/devices.json@@ -62,7 +62,7 @@"filterMeta": "{{available}} 台を開けます","filterMetaWithUnavailable": "{{available}} 台を開けます · {{unavailable}} 台は対応が必要です","showUnavailableReason": "デバイスを展開すると利用できない理由を確認できます",- "emptyCopy": "コンピューターで同じアカウントにログインし、設定のリモートコントロールで「同じアカウントのデバイスによるこのコンピューターの操作を許可」をオンにしてください。",+ "emptyCopy": "コンピューターで同じアカウントにログインし、「設定 → リモート接続」で「同じアカウントのデバイスによるこのコンピューターの操作を許可」をオンにしてください。","emptyTitle": "表示できるコンピューターはまだありません","hideUnavailableA11y": "利用できないコンピューターを非表示","showUnavailableA11y": "利用できないコンピューターを表示",diff --git a/apps/mobile/src/i18n/locales/ja/remoteDesktop.json b/apps/mobile/src/i18n/locales/ja/remoteDesktop.jsonindex 4e5a39872e..3ad015eda0 100644--- a/apps/mobile/src/i18n/locales/ja/remoteDesktop.json+++ b/apps/mobile/src/i18n/locales/ja/remoteDesktop.json@@ -65,7 +65,7 @@"permissionMissing": "未許可","openPermissionSettings": "システム設定を開く","recheckPermissions": "再確認",- "permissionActionFailed": "権限ガイドを開けませんでした。コンピューターのリモート操作設定で再試行してください。",+ "permissionActionFailed": "権限ガイドを開けませんでした。コンピューターの「設定 → リモート接続」で再試行してください。","permissionCheckFailed": "権限を確認できません。コンピューターがオンラインか確認して再試行してください。","openGuideOnComputer": "コンピューターでガイドを開く","permissionGuideOpened": "コンピューターに権限ガイドを表示しました。そちらで続けてください。",@@ -90,7 +90,7 @@"controlling": "操作中","viewOnly": "表示のみ","upgrade": "コンピューター側の Cindy を更新してください。",- "disabled": "コンピューターの設定 → リモート操作でリモートデスクトップを有効にしてください。",+ "disabled": "コンピューターの「設定 → リモート接続」でリモートデスクトップを有効にしてください。","busy": "他のデバイスまたは Agent がパソコンを操作中です。他のデバイスでの操作を終了するか、Agent の操作が終わるまでお待ちください。","connectionBusy": "別のデバイスがこのリモートデスクトップに接続しています。そのデバイスでリモートデスクトップを終了してから、もう一度接続してください。","connectionError": "接続が終了しました。コンピューターがオンラインか確認し、再接続してください。",diff --git a/apps/mobile/src/i18n/locales/ko/devices.json b/apps/mobile/src/i18n/locales/ko/devices.jsonindex 13d37b0f86..768b5082d7 100644--- a/apps/mobile/src/i18n/locales/ko/devices.json+++ b/apps/mobile/src/i18n/locales/ko/devices.json@@ -62,7 +62,7 @@"filterMeta": "{{available}}대 접속 가능","filterMetaWithUnavailable": "{{available}}대 접속 가능 · {{unavailable}}대 확인 필요","showUnavailableReason": "기기를 펼치면 사용할 수 없는 이유가 표시됩니다",- "emptyCopy": "컴퓨터에서 같은 계정으로 로그인한 뒤 설정의 원격 제어에서 ‘같은 계정의 기기가 이 컴퓨터를 제어하도록 허용’을 켜세요.",+ "emptyCopy": "컴퓨터에서 같은 계정으로 로그인한 뒤 '설정 → 원격 연결'에서 ‘같은 계정의 기기가 이 컴퓨터를 제어하도록 허용’을 켜세요.","emptyTitle": "표시할 컴퓨터가 아직 없습니다","hideUnavailableA11y": "사용할 수 없는 컴퓨터 숨기기","showUnavailableA11y": "사용할 수 없는 컴퓨터 표시",diff --git a/apps/mobile/src/i18n/locales/ko/remoteDesktop.json b/apps/mobile/src/i18n/locales/ko/remoteDesktop.jsonindex a5223590f5..a54246e3ce 100644--- a/apps/mobile/src/i18n/locales/ko/remoteDesktop.json+++ b/apps/mobile/src/i18n/locales/ko/remoteDesktop.json@@ -65,7 +65,7 @@"permissionMissing": "허용되지 않음","openPermissionSettings": "시스템 설정 열기","recheckPermissions": "다시 확인",- "permissionActionFailed": "권한 안내를 열지 못했습니다. 컴퓨터의 원격 제어 설정에서 다시 시도하세요.",+ "permissionActionFailed": "권한 안내를 열지 못했습니다. 컴퓨터의 '설정 → 원격 연결'에서 다시 시도하세요.","permissionCheckFailed": "권한을 확인하지 못했습니다. 컴퓨터가 온라인인지 확인하고 다시 시도하세요.","openGuideOnComputer": "컴퓨터에서 권한 안내 열기","permissionGuideOpened": "컴퓨터에 권한 안내를 표시했습니다. 컴퓨터에서 계속하세요.",@@ -90,7 +90,7 @@"controlling": "제어 중","viewOnly": "보기 전용","upgrade": "컴퓨터의 Cindy를 업데이트하세요.",- "disabled": "컴퓨터의 설정 → 원격 제어에서 원격 데스크톱을 켜주세요.",+ "disabled": "컴퓨터의 '설정 → 원격 연결'에서 원격 데스크톱을 켜주세요.","busy": "다른 기기 또는 Agent가 컴퓨터를 사용 중입니다. 다른 기기에서 제어를 중지하거나 Agent의 작업이 끝날 때까지 기다리세요.","connectionBusy": "다른 기기가 이 원격 데스크톱에 연결되어 있습니다. 해당 기기에서 원격 데스크톱을 종료한 후 다시 연결하세요.","connectionError": "연결이 종료되었습니다. 컴퓨터가 온라인인지 확인하고 다시 연결하세요.",diff --git a/apps/mobile/src/i18n/locales/zh-CN/devices.json b/apps/mobile/src/i18n/locales/zh-CN/devices.jsonindex 7da5fdd4fb..452347be93 100644--- a/apps/mobile/src/i18n/locales/zh-CN/devices.json+++ b/apps/mobile/src/i18n/locales/zh-CN/devices.json@@ -62,7 +62,7 @@"filterMeta": "{{available}} 台可进入","filterMetaWithUnavailable": "{{available}} 台可进入 · {{unavailable}} 台需处理","showUnavailableReason": "展开的设备会显示不可用原因",- "emptyCopy": "在电脑端登录同一账号,并在设置里的远程控制打开“允许同账号设备控制本机”。",+ "emptyCopy": "在电脑端登录同一账号,并在「设置 → 远程连接」中打开“允许同账号设备控制本机”。","emptyTitle": "还没有可显示的电脑","hideUnavailableA11y": "隐藏不可用电脑","showUnavailableA11y": "显示不可用电脑",diff --git a/apps/mobile/src/i18n/locales/zh-CN/remoteDesktop.json b/apps/mobile/src/i18n/locales/zh-CN/remoteDesktop.jsonindex d8d6de98ba..c7dafaaea8 100644--- a/apps/mobile/src/i18n/locales/zh-CN/remoteDesktop.json+++ b/apps/mobile/src/i18n/locales/zh-CN/remoteDesktop.json@@ -65,7 +65,7 @@"permissionMissing": "未授权","openPermissionSettings": "打开系统设置","recheckPermissions": "重新检测",- "permissionActionFailed": "未能打开授权引导。请在电脑端的远程控制设置中重试。",+ "permissionActionFailed": "未能打开授权引导。请在电脑端的「设置 → 远程连接」中重试。","permissionCheckFailed": "暂时无法检测权限,请确认电脑在线后重试。","openGuideOnComputer": "在电脑上打开授权引导","permissionGuideOpened": "已在电脑上显示授权引导,请在电脑上继续。",@@ -90,7 +90,7 @@"controlling": "正在控制","viewOnly": "仅查看","upgrade": "请升级电脑端 Cindy 后使用远程桌面。",- "disabled": "请在电脑端的设置 → 远程控制中开启远程桌面。",+ "disabled": "请在电脑端的「设置 → 远程连接」中开启远程桌面。","busy": "其他设备或 Agent 正在操作电脑。请先停止其他设备的操作,或等待 Agent 操作结束。","connectionBusy": "已有其他设备连接此远程桌面。请先在那台设备上退出远程桌面,再尝试连接。","connectionError": "连接已结束。请检查电脑是否在线,然后重新连接。",diff --git a/apps/mobile/src/i18n/locales/zh-TW/devices.json b/apps/mobile/src/i18n/locales/zh-TW/devices.jsonindex 05ed3b3833..370e8763b5 100644--- a/apps/mobile/src/i18n/locales/zh-TW/devices.json+++ b/apps/mobile/src/i18n/locales/zh-TW/devices.json@@ -62,7 +62,7 @@"filterMeta": "{{available}} 臺可進入","filterMetaWithUnavailable": "{{available}} 臺可進入 · {{unavailable}} 臺需處理","showUnavailableReason": "展開裝置可查看無法使用的原因",- "emptyCopy": "在電腦端登入同一帳號,並在設定的遠端控制中開啟「允許同帳號裝置控制本機」。",+ "emptyCopy": "在電腦端登入同一帳號,並在「設定 → 遠端連線」中開啟「允許同帳號裝置控制本機」。","emptyTitle": "還沒有可顯示的電腦","hideUnavailableA11y": "隱藏無法使用的電腦","showUnavailableA11y": "顯示無法使用的電腦",diff --git a/apps/mobile/src/i18n/locales/zh-TW/remoteDesktop.json b/apps/mobile/src/i18n/locales/zh-TW/remoteDesktop.jsonindex 9efe973ef6..313dc70ce9 100644--- a/apps/mobile/src/i18n/locales/zh-TW/remoteDesktop.json+++ b/apps/mobile/src/i18n/locales/zh-TW/remoteDesktop.json@@ -65,7 +65,7 @@"permissionMissing": "未授權","openPermissionSettings": "開啟系統設定","recheckPermissions": "重新檢測",- "permissionActionFailed": "無法開啟授權引導。請在電腦端的遠端控制設定中重試。",+ "permissionActionFailed": "無法開啟授權引導。請在電腦端的「設定 → 遠端連線」中重試。","permissionCheckFailed": "暫時無法檢測權限,請確認電腦在線後重試。","openGuideOnComputer": "在電腦上開啟授權引導","permissionGuideOpened": "已在電腦上顯示授權引導,請在電腦上繼續。",@@ -90,7 +90,7 @@"controlling": "正在控制","viewOnly": "僅檢視","upgrade": "請升級電腦端 Cindy 後使用遠端桌面。",- "disabled": "請在電腦端的設定 → 遠端控制中開啟遠端桌面。",+ "disabled": "請在電腦端的「設定 → 遠端連線」中開啟遠端桌面。","busy": "其他裝置或 Agent 正在操作電腦。請先停止其他裝置的操作,或等待 Agent 操作結束。","connectionBusy": "已有其他裝置連線至此遠端桌面。請先在該裝置上退出遠端桌面,再嘗試連線。","connectionError": "連線已結束。請檢查電腦是否在線,然後重新連線。",diff --git a/packages/lizi-mcps/src/__tests__/getCapabilitiesTool.test.ts b/packages/lizi-mcps/src/__tests__/getCapabilitiesTool.test.tsindex 71e542bc67..16a0bbee19 100644--- a/packages/lizi-mcps/src/__tests__/getCapabilitiesTool.test.ts+++ b/packages/lizi-mcps/src/__tests__/getCapabilitiesTool.test.ts@@ -128,6 +128,23 @@ describe('get_capabilities tool', () => {} finally { await h.cleanup(); }});+ // 帮助文档原先只接给设置页帮助助手,Agent 被问「归档的任务在哪」时只能编造入口。+ it('scope=help 先返回目录再按 key 返回整篇;宿主未注入时明确不可用', async () => {+ const productHelpDocs = [{ id: 'sidebar', title: 'Sidebar', summary: 'Find sessions.', content: 'Status — active / archived / all.' }];+ const call = async (server: McpServer, args: Record<string, unknown>) => {+ const h = await connect(server);+ try {+ return parsePayload(await h.client.callTool({ name: 'call_tool', arguments: { name: 'get_capabilities', args } }));+ } finally { await h.cleanup(); }+ };+ const make = () => createXdtHelperMcpServer({ logger: makeLogger(), productHelpDocs }, sessionCtx);++ expect(await call(make(), { scope: 'help' })).toMatchObject({ ok: true, docs: [{ key: 'sidebar', title: 'Sidebar', summary: 'Find sessions.' }] });+ expect(await call(make(), { scope: 'help', key: 'sidebar' })).toMatchObject({ ok: true, doc: { key: 'sidebar', content: 'Status — active / archived / all.' } });+ expect(await call(make(), { scope: 'help', key: 'nope' })).toMatchObject({ ok: false, errorCode: 'UNKNOWN_KEY', data: { available: ['sidebar'] } });+ expect(await call(createXdtHelperMcpServer({ logger: makeLogger() }, sessionCtx), { scope: 'help' })).toMatchObject({ ok: false, errorCode: 'HELP_UNAVAILABLE' });+ });+it('不传 key 时返回包含 about-cindy 的索引', async () => {const h = await connect(createXdtHelperMcpServer({ logger: makeLogger() }, sessionCtx));try {diff --git a/packages/lizi-mcps/src/lizi_xdtHelperMcpServer.ts b/packages/lizi-mcps/src/lizi_xdtHelperMcpServer.tsindex 1672df3c1b..7fac716ded 100644--- a/packages/lizi-mcps/src/lizi_xdtHelperMcpServer.ts+++ b/packages/lizi-mcps/src/lizi_xdtHelperMcpServer.ts@@ -147,7 +147,7 @@ const CATEGORY_ENUM = ['cindy', 'auth', 'control', 'history', 'feedback', 'hando* 说明才发现。改工具的作用范围时同步改这里(测试按类目全集守护)。*/export const HELPER_CATEGORY_DESCRIPTIONS: Record<(typeof CATEGORY_ENUM)[number], string> = {- cindy: `${BRAND_NAME} 能力说明与当前任务信息/交付回执(report_task_delivery);get_capabilities({key:"self-management"}) 是管理自己的任务、定时任务、项目、伙伴的工具索引。本机。`,+ cindy: `${BRAND_NAME} 能力说明、界面操作帮助(get_capabilities scope=help)与当前任务信息/交付回执(report_task_delivery);get_capabilities({key:"self-management"}) 是管理自己的任务、定时任务、项目、伙伴的工具索引。本机。`,auth: '本机窗口的账号、密码、验证码、二维码交给主人填写或扫码(request_input_handoff);Grok 设备码授权。',control: '管理任务与项目:归档/取消归档支持本机与跨电脑(deviceId::sessionId);改名、移动、标签、排队消息、插话、停止、运行设置、项目列表与增删改仅本机。',history: '按话题查找跨渠道聊天,支持伙伴归属(teammate_id)、最近活动时间与真实来源;不要默认限制当前工作目录或创建日期。查询任务、工作目录、聊天记录、任务摘要与输入队列;任务结果附带可直接引用的 sessionUrl,已知 id 要链接用 get_task_digest。任务列表、搜索、读取聊天记录、任务摘要(get_task_digest)可跨电脑(device 参数 / deviceId::sessionId);其余本机。',@@ -771,6 +771,8 @@ export interface XdtHelperMcpDeps {context: import('./types.js').LiziMcpSessionContext,query: import('./xdt-helper/get_capabilities.js').RuntimeCapabilityQuery,) => Promise<unknown>;+ /** 产品帮助文档(设置页帮助助手同一份),供 get_capabilities scope=help 读取;不注入则该 scope 返回 HELP_UNAVAILABLE。 */+ productHelpDocs?: readonly import('./xdt-helper/get_capabilities.js').ProductHelpDoc[];/*** 历史聊天数据查询的回调集合(读本地 SQLite 的 sessions / messages 表)。host* 注入后, history 类工具(list_workdirs / list_sessions / get_chat_history /@@ -899,7 +901,7 @@ export function createXdtHelperMcpServer(// 'cindy' 类: 自省 (无 host 依赖, 始终注册)。registerGetCapabilitiesTool(registry, deps.runtimeCapabilities? (query) => deps.runtimeCapabilities!(resolveLiziMcpSessionContext(sessionCtx), query)- : undefined);+ : undefined, deps.productHelpDocs);if (deps.appUpdate) {registerAppUpdateTools(registry, {getSessionContext: () => resolveLiziMcpSessionContext(sessionCtx),diff --git a/packages/lizi-mcps/src/xdt-helper/bot_capabilities.ts b/packages/lizi-mcps/src/xdt-helper/bot_capabilities.tsindex f04618ed2e..a8a3afb676 100644--- a/packages/lizi-mcps/src/xdt-helper/bot_capabilities.ts+++ b/packages/lizi-mcps/src/xdt-helper/bot_capabilities.ts@@ -78,7 +78,7 @@ export const TEAMMATE_CONTROL_GUIDANCE = ['message_session_task returns queued_message_id for queued or steered input. check_session_task exposes only your own queue; optionally query queued_message_id for queued/consuming/dispatched/not-found/unavailable; queue restore errors preserve task results and cannot prove a message is missing. Dispatched means accepted into host history, not proof of model action. Use mode=edit with message or mode=withdraw for your own unconsumed item; consuming input cannot be edited or withdrawn.','Control your own background tasks with `message_session_task` (mode=queue by default, mode=steer for same-turn input, mode=resume for a reversible pause) and `stop_session_task` (mode=cancel by default, request-stop for a graceful stop request, pause to hold execution and queued input). General control tools are not part of the teammate toolset. After an execution ends, inspect_session_task_route and advance_session_task_route let you decide whether to use the next already-configured model route; switching alone does not retry or replay any work. A successful queue receipt is not a steer, and pausing/requested/unconfirmed is not proof the engine stopped. Check the returned control state; unsupported engines fail explicitly. Pause preserves the same task and Session, holds timers and automatic input, and resumes without replaying completed work. If paused, resume without a message before answering a pending interaction. Only tasks owned by your teammate are controllable.','The host decides whom each turn speaks for. When the owner asked in this turn (typed it, or the host verified an owner request in a group or IM lane), act for the owner on their tasks under the task\'s ordinary permissions; the host attaches the owner\'s verified words when you relay them. An owner-created routine carries its standing instruction. Every other turn is your own initiative: host wake-ups and the built-in follow-up, teammate messages, guests, forwarded input and unavailable group verification. Then you may read, keep Todos, notify the owner, message teammates, control background tasks you started, and continue or re-route tasks already handed to you (continue_workbench_task to carry on the original goal, set_session_runtime per suggested_route). Other changes to the owner\'s tasks, and saving, deleting, running or creating routines and schedules, return NEEDS_OWNER_REQUEST: do not retry; propose the task, action and content in a Todo or notification so the owner can ask. When relaying, keep your supplement to the background the task needs; do not restate authorization or ask the task to send results back. When the host wakes you, inspect get_task_digest and report only important new facts or decisions. Check current state: long tools, pauses and recovered tasks do not need automatic nudges. Observation does not expand private-data access.',- 'Use `get_capabilities` for application features and UI guidance. It describes the product, not a permission grant or proof that every feature is available here. Use the matching live tool to act, and verify its result before claiming success.',+ 'Use `get_capabilities` for application features, and `get_capabilities` with scope `help` for where things are in the UI and how to use them. It describes the product, not a permission grant or proof that every feature is available here. Use the matching live tool to act, and verify its result before claiming success.',].join('\n');/** @deprecated Internal compatibility export; new teammate guidance uses the product name. */diff --git a/packages/lizi-mcps/src/xdt-helper/capabilities.ts b/packages/lizi-mcps/src/xdt-helper/capabilities.tsindex 86977eed9b..b6dcf0e740 100644--- a/packages/lizi-mcps/src/xdt-helper/capabilities.ts+++ b/packages/lizi-mcps/src/xdt-helper/capabilities.ts@@ -12,8 +12,8 @@** Bucket 分类原则:* - "用户产品功能" 一类一个 bucket(ai-chat / session-management / scheduler 等)- * - "外部数据源 / 服务的 MCP 接入" 各自独立 bucket(jira / confluence / google-sheets / ai-art)- * - 飞书较特殊:"基础接入(登录/通知/MCP 文档表格等)" 与 "移动办公接管 session" 拆两个 bucket,+ * - "外部数据源 / 服务接入" 各自独立 bucket(jira / google / ai-art);大多已迁为插件,经 ghost_* 调用+ * - 飞书较特殊:"基础接入(机器人/通知/xd-feishu 插件)" 与 "IM 接管桌面任务" 拆两个 bucket,* 因为后者是 Cindy 最差异化的能力,值得独立索引项让模型一眼能查到。*/@@ -39,7 +39,7 @@ export const CAPABILITIES: readonly CapabilityEntry[] = [title: 'Grok / SuperGrok 授权登录',oneLiner: 'Grok 有浏览器跳转和设备短码两种登录方式;agent 可发授权页与短码并查询完成状态。',detail: [- '浏览器跳转登录在设置的 Grok 连接卡中可用。',+ '浏览器跳转登录在「设置 → 模型供应商」的 Grok 连接中可用;agent 也可调用 cindy 网关的 connect_account({kind:"host", id:"grok"}) 在当前任务里弹出连接卡。','设备码登录适合远程协助:agent 通过 cindy_helper 的 auth 类 start_grok_device_login 取得 xAI 授权页和短码,发给用户;用户在授权页输入短码并确认,Cindy 在本机自动轮询并保存凭证。','agent 可用 get_grok_login_status 查询连接状态,用户要求停止时可用 cancel_grok_device_login。长期 access token 和 refresh token 不会出现在工具返回值或聊天里;不要要求用户粘贴长期凭证。',].join(' '),@@ -52,7 +52,8 @@ export const CAPABILITIES: readonly CapabilityEntry[] = [`【是什么】${BRAND_NAME} 是 XD Inc. 出品的开源 AI 助手(open-source AI assistant), 以桌面 / 移动客户端形态交付, 源码 https://github.com/makecindy/cindy (Apache-2.0)。官网分区域: 中国大陆 https://cindy.cn, 国际版 https://cindy.app —— 给下载 / 定价链接前先确认用户所在区域, 不要一律给国际版。`,`【身份归本条, 执行位置不归】问"你是谁 / 你是什么"以本条为准, 不要用训练数据猜, 也不要凭工作目录路径或工具名反推。但 ${BRAND_NAME} 只是编排方(管会话、工具、上下文与 memory; 底层 harness 是 Claude Code 或 Codex, 可中途切换且上下文连续), 不代表代码在哪执行: agent 进程与 workdir 可能在本机, 也可能在 SSH 远程工作区的远端主机(文件与进程都在远端), 或经设备互联隧道驱动的被控桌面端。问"你跑在哪 / 文件在哪台机器"时以当前会话实际工作区为准, 拿不准就说明, 不要断言。`,`【源码范围】该仓库是客户端本体(desktop、mobile 及共享 packages 的 pnpm monorepo), 服务端不在其中也不开源。安装版不携带源码, agent 侧无法推断用户是否 clone 过、clone 在哪: 要读改源码就让用户给路径或用工程模式打开, 不要假设固定路径, 更不要因为工作目录里出现品牌名就断定当前目录是源码仓库。`,- `【版本号不要猜】本条不含版本号。CN 版可在"设置 → 关于"看到客户端版本; 国际版该页当前不展示它(只有更新开关与 agent 二进制版本), 别把用户支使过去空找。用 submit_github_issue 提反馈时客户端版本 / OS / 界面语言由系统自动附加, 不用 agent 填。`,+ `【查源码排障】帮助文档(get_capabilities scope=help)没写到、或要确认 ${BRAND_NAME} 某个入口或行为的实际实现时, 可以只读查看 GitHub 上与用户版本对应的源码: 每个发布版本都有 tag(如 v0.1.118-beta; 当前版本号用 check_app_update 的 currentVersion 取), 用 gh 或网页按 tag 读取, 不要拿最新 main 对照旧版本。回答时说明结论来自源码; 服务端行为不在仓库里, 查不到就直说。`,+ `【版本号不要猜】本条不含版本号。所有版本都显示在侧栏左下角账号卡片上; CN 版的"设置 → 关于"也有, 国际版该页不展示(只有更新开关与 agent 二进制版本), 别把用户支使过去空找。agent 自己要版本号就调 check_app_update 读 currentVersion。用 submit_github_issue 提反馈时客户端版本 / OS / 界面语言由系统自动附加, 不用 agent 填。`,`【模型怎么来】登录官方 ${BRAND_NAME} 服务按量透明扣费、授权已付费的 Claude Code / Codex Coding Plan 继续用(不重复付费)、接自己的 API key, 或跑本地模型。价格见对应区域官网。`,].join(' '),},@@ -75,21 +76,21 @@ export const CAPABILITIES: readonly CapabilityEntry[] = [{key: 'ai-chat',title: 'AI 对话',- oneLiner: '与 Claude Code / Codex agent 实时对话,支持代码执行、工具调用、多模态输入。',+ oneLiner: '在任务里与 Claude Code / Codex / Pi agent 实时对话,支持代码执行、工具调用、图片与文件输入。',detail: [- `${BRAND_NAME} 的核心入口:基于 Claude Agent SDK 接入 Claude Code,基于自建 AppServer 接入 Codex,两套 agent 在同一对话界面里使用。`,- '消息支持文本 / 图片 / 本地文件三种输入。Markdown 渲染、代码高亮、表格、流式输出齐全,长会话用 render-window 优化避免卡顿。',- '会话持久化在本地 SQLite,可全文搜索历史消息。',+ `${BRAND_NAME} 的核心入口:基于 Claude Agent SDK 接入 Claude Code,基于自建 AppServer 接入 Codex,另有 Pi;三种 Harness 在同一任务界面里使用,每个任务互相独立、可并行。`,+ '输入支持文本、粘贴 / 拖入图片(PNG / JPEG / WEBP / GIF)、@ 引用文件与附加参考目录。Markdown 渲染、代码高亮、表格、流式输出齐全,长任务做了渲染窗口优化避免卡顿。输入框「+」菜单里可开「计划模式」(先出计划再执行)与「协同模式」。',+ '对话记录持久化在本地 SQLite;侧栏「搜索」可按标题或内容找任务(默认含已归档)。',].join(' '),},{key: 'agent-switching',- title: 'Agent 切换',- oneLiner: '同一会话或新会话间在 Claude Code 与 Codex 之间切换。',+ title: 'Agent(Harness)切换',+ oneLiner: '任务进行中可在 Claude Code / Codex / Pi 之间切换,新引擎通过交接摘要续接。',detail: [- '运行时切 agent:Claude Code(Anthropic SDK)和 Codex(自建 AppServerHost,NDJSON 协议)双线接入。',- '支持"问题转移到另一个 agent",新 agent 继承上文继续回答。',- '跨厂商切换有守卫,从国产模型切回 Claude/GPT 时会提醒(避免环境变量污染)。',+ '在输入框的模型选择里换到另一 Harness 的模型即切换;会弹「切换 Agent?」确认(可勾「下次不再提醒」):新 Agent 引擎通过交接摘要续接,思考内容等不能保证完美继承;切回时仍可继续原引擎会话;切换会让 prompt 缓存中断一次,可能增加 Token 消耗。',+ '任务正在运行时切换会先登记,下一条消息发出时生效。SSH 远程任务暂不支持切换 Harness(可换同一 Harness 下的模型,或新建所需 Harness 的 SSH 任务)。',+ 'agent 可用 cindy_helper control 类 set_session_runtime(harness + model)为本机任务登记切换,下一条消息发送时生效。',].join(' '),},{@@ -101,14 +102,15 @@ export const CAPABILITIES: readonly CapabilityEntry[] = ['【怎么开】用户表达"用 worker / 协同模式 / 独立 agent / 派一个 agent 帮我 X"等意图时, Lead 调 start_team 工具创建 team, 再调 create_worker 工具添加 worker。worker 创建后可通过 send_to_worker 派活, 通过 list_workers 查看所有 worker, 通过 switch_focus 切换 focused worker。','【怎么关】用户表达"够了 / 关掉 worker / 不要协同了"时, Lead 调 end_team 结束整个 team(归档所有 worker), 或调 archive_worker 归档单个 worker。Lead 自身保留可继续单 session 对话。','【硬边界】Worker session 不能再开 team(嵌套禁止, 返 WORKER_CANNOT_NEST);Claude Code / Codex / Pi 本地项目或对话 session 都可以作为 Lead 调 start_team,也都可以作为 Worker;SSH 远程 Lead 与 Worker 当前只支持 Claude Code / Codex;Worker 不能结束自己所在 team(返 WORKER_CANNOT_DISABLE);要求 Lead session 有 workingDir。',- '【工具归属】18 个 team 工具(start_team / create_worker / create_workers / send_to_worker / interrupt_worker / get_worker_queue_status / update_queued_message / cancel_queued_message / merge_queued_messages / steer_queued_message / move_queued_message / list_workers / switch_focus / idle_worker / end_team / archive_worker / update_worker / list_available_models)在独立的 cindy_orca server 直接顶层注册, 对应"协同模式"可关插件(Settings → Connections → Built-in Tools)。通用 session handoff 原语 send_to_session 在 essential 的 cindy_helper 的 handoff 类目下(走 call_tool, 常开, 供 skill 路由用)。',+ '【Worker 放在哪】create_worker / create_workers 可让 Worker 继承 Lead 目录、用显式 working_dir 共用已有目录、或 new_worktree 新建独立 worktree(仅本机,归档时回收);也可用 execution_device_id 放到同账号另一台在线且开启远程控制的电脑上运行(get_workspace_info 的 execution_devices 列出可选电脑)。',+ '【工具归属】18 个 team 工具(start_team / create_worker / create_workers / send_to_worker / interrupt_worker / get_worker_queue_status / update_queued_message / cancel_queued_message / merge_queued_messages / steer_queued_message / move_queued_message / list_workers / switch_focus / idle_worker / end_team / archive_worker / update_worker / list_available_models)及诊断用的 get_workspace_info / worker_status / read_worker(Worker 完成会自动回报,不要用后两者轮询)在独立的 cindy_orca server 直接顶层注册, 对应可关的「协同模式」开关(设置 →「工具」→「内置工具」)。通用 session handoff 原语 send_to_session 在 essential 的 cindy_helper 的 handoff 类目下(走 call_tool, 常开, 供 skill 路由用)。',`【与 Claude Code Task tool / Codex subagent 的区别】Task / subagent 是 agent 框架内的子任务派发机制(子 agent 跑在同一 SDK 进程内、有限工具集、生命周期短、对话历史归属父 turn);${BRAND_NAME} 协同模式是业务层的 session 级编排(Lead/Worker 都是完整独立的 ${BRAND_NAME} session, 独立进程、UI 栏位、完整工具、独立对话历史, 长生命周期, 通过 main 进程 IPC + MCP bridge 通信)。两者不互斥, Worker 内部仍可用 Task/subagent 派子任务。`,'【手动入口】ChatInput「+」菜单里的「协同模式」项,开启态用橙色 UsersRound 图标与文字标识;用户也能在这里手动开/关,与本工具走同一份业务代码。',].join(' '),},{key: 'session-handoff',- title: 'Session Handoff(把消息派给已存在的 session)',+ title: 'Session Handoff(把消息派给已有任务)',oneLiner: 'skill 可读当前 sessionId,并把消息 handoff 到一个既有 session,适合把同一业务对象的二次处理路由回原上下文。',detail: [`【是什么】get_current_session_id(cindy_helper 自省类)+ send_to_worker(cindy_orca team 工具)/ send_to_session(cindy_helper handoff 类, 走 call_tool)配合使用。前者返回当前 ${BRAND_NAME} session 的 business id / agent_kind / working_dir, 后两者把一条控制层消息投递到指定 session;目标不在内存时会自动 resume, 投递成功即返回。`,@@ -122,7 +124,7 @@ export const CAPABILITIES: readonly CapabilityEntry[] = [},{key: 'session-control',- title: '会话控制面(队列 / 插话 / 停止 / 运行探针)',+ title: '任务控制面(队列 / 插话 / 停止 / 运行探针)',oneLiner:'agent 可观察本机任务的队列与运行状态,并控制自己投递的队列消息、same-turn 插话或请求优雅停止(仅本机任务)。',detail: [@@ -137,26 +139,28 @@ export const CAPABILITIES: readonly CapabilityEntry[] = [},{key: 'session-management',- title: '会话管理(含 Fork / Rewind / 批量归档)',- oneLiner: '新建 / 关闭 / 搜索会话,在历史任意点 Fork 分岔或 Rewind 重跑;agent 还能批量归档整理。',+ title: '任务管理(含查看已归档 / Fork / Rewind / 批量整理)',+ oneLiner: '新建、搜索、置顶、改名、移动、归档 / 取消归档、删除任务,查看已归档任务;从某条消息开新任务(Fork)或回退(Rewind);agent 可批量改名与归档。',detail: [- '会话列表带搜索、按时间和工程目录组织。',- 'Fork:在任意历史消息处分裂出新会话,继承到该点为止的全部上文。',- 'Rewind:在当前会话回到历史点,改写发送内容后重新运行,原分支被替换。',- 'agent 批量整理:control 类工具 rename_sessions(批量改名)、archive_sessions / unarchive_sessions(批量归档/取消归档,把 status 在 active↔archived 间切换;跨电脑传 deviceId::sessionId,单次限定一台电脑)。',- '归档可逆、不删数据,经统一权威出口写库并广播 sessions:patched,侧栏即时收敛;不能归档当前正在运行的会话。恢复不启动执行、不发旧消息、不恢复暂停自动化;之后可用 send_to_session 发送新消息。',+ '【列表】左侧主列表可按项目 / 对话 / 电脑分组。主列表标题(「全部任务 ▾」,按范围显示「本机任务」、设备名等)旁的「侧边栏显示设置」(滑杆图标,鼠标悬停标题行时出现;也可从标题下拉菜单里的「侧边栏显示设置」或侧栏空白处右键打开)管理分组、排序、筛选与显示。',+ '【右键菜单】置顶、重命名、移动任务、复制任务链接、在新窗口打开、归档、删除;已归档任务显示「取消归档」。删除后无法恢复。',+ '【查看已归档】归档后任务从默认的「活跃」列表移除。查看方式:「侧边栏显示设置」→「任务状态」→「已归档」(只看已归档)或「全部」;处于「已归档」时设置按钮常驻并带圆点提醒。侧栏「搜索」的状态默认为「全部」,按标题或内容能直接搜到已归档任务,也可在「搜索条件」→「状态」里只看「已归档」。用户要找某个具体的已归档任务时,直接给 list_sessions / search_chat_history 结果里的 sessionUrl 链接([任务标题](sessionUrl)),用户点开即可,不必先切筛选。',+ '【恢复】右键「取消归档」,或项目菜单「全部取消归档」;用户在桌面端打开已归档任务直接发消息,会自动取消归档回到「活跃」。agent 用 send_to_session 发给已归档任务则返回 ARCHIVED,不会自动恢复。',+ '【归档限制】运行中、正被远程接管的任务不能归档;项目 / 对话组「全部归档」与多选归档会跳过已置顶和运行中的任务;worktree 有未提交改动时归档前自动保存 git 快照。',+ 'Fork:消息操作栏「开启一个新任务」,用到该点为止的上下文创建独立新任务,原任务不变。Rewind:「回到此处」移除这条消息之后的对话记录,有 Git 保存点 / 文件 checkpoint 时可同时把工作目录文件回退到那一刻;任务运行中会先停止;远端 Claude Code 任务暂不支持。Rewind 会丢弃后续对话,确认前会提示影响范围。',+ 'agent 批量整理:control 类 rename_sessions(批量改名)、move_session(移动)、任务标签工具、archive_sessions / unarchive_sessions(每批 ≤50,把 status 在 active↔archived 间切换;跨电脑传 deviceId::sessionId,单次限定一台电脑;改名 / 移动 / 标签仅本机)。agent 没有删除、Fork、Rewind 任务的工具。',+ '归档可逆、不删数据,经统一权威出口写库并广播 sessions:patched,侧栏即时收敛;不能归档调用方自己所在的任务。恢复不启动执行、不发旧消息、不恢复暂停自动化;之后可用 send_to_session 发送新消息。','本机批次原子写入;远程逐项执行,失败即停并返回已完成、失败与未执行项,不保证跨项事务;回执未知时先查状态再决定重试。',- '以上操作都不破坏原数据,可放心试错。',+ '更细的界面操作说明可查 get_capabilities({scope:"help", key:"sidebar"})。',].join(' '),},{key: 'workdir-browser',- title: '工程文件浏览器',- oneLiner: 'VSCode 风格文件树 + 搜索 + 预览,与会话侧栏自动联动切换。',+ title: '右侧栏面板(文件浏览器 / 审查 / 终端 / 浏览器等)',+ oneLiner: '任务右侧栏可开文件浏览器、审查(git 改动)、终端、浏览器、侧边追问、后台任务等面板,与主任务并行。',detail: [- '打开会话所在工程目录后,侧栏自动切到文件浏览模式。',- '支持文件名 / 内容搜索、大文件按需加载、语法高亮、Markdown / 表格预览。',- '上下两层 Tab:上层是打开的会话,下层是打开的文件,可平行切换。',+ '右侧栏顶部「+」(添加标签)选择面板:「文件浏览器」(工作目录文件树、全项目搜索、预览与编辑保存)、「审查」(按 git diff 查看改动,可还原改动块,顶部「提交或推送」)、「终端」、「浏览器」、侧边追问(基于主任务上下文独立提问)、「Subagent」、「后台任务」、「资源监视器」,协同模式下有「协同」标签,插件也可提供面板。',+ '标签可拖动分组,也可「在独立窗口中打开」。部分面板(如终端、资源监视器)对远程任务暂不可用,面板内会写明原因。',].join(' '),},{@@ -174,13 +178,12 @@ export const CAPABILITIES: readonly CapabilityEntry[] = [},{key: 'skillhub',- title: 'Skill Hub(技能市场)',- oneLiner: '本地 skill 管理 + 官方 skill 市场,以 .md 形式扩展 agent 能力。',+ title: 'Skill(技能)与 Skill Hub',+ oneLiner: '本地技能管理 + Skill Hub 市场(公开 / 组织),Skill 是带 SKILL.md 的文件夹,三种 Harness 共用,以 / 命令触发。',detail: [- '两个页签:Local(项目级 + 全局 skill),Market(官方 skill 仓库)。',- 'Skill 以 markdown 形式存储,可作为 /slash 命令或工具触发。',- '支持发布、版本管理、以及通过飞书卡片走的审核流。',- '让用户不用改代码就能给 agent 加新能力包。',+ '入口:主界面侧边栏「技能」页(与「插件」同一管理界面),含本地技能(全局 + 项目级)与 Skill Hub 市场;可导入 .zip 或 SKILL.md、安装、停用、卸载、发布与更新。',+ 'Skill 是一个带 SKILL.md 的文件夹(同目录文件 agent 也能读),Claude Code / Codex / Pi 共用同一份,出现在输入框的 / 建议里。',+ 'agent 工具(cindy_helper skills 类):search_skills(搜公开市场 / 所在组织)、publish_skill(仅在用户要求时上传或发新版本,首次发布须用户选可见范围)、get_skill_publish_status(扫描 / 审核状态,人工审核中如实告知、不要反复发布或轮询)、list_my_published_skills;start_skill_learning 对应 /learn,从当前任务提炼可复用 Skill。',].join(' '),},{@@ -188,9 +191,9 @@ export const CAPABILITIES: readonly CapabilityEntry[] = [title: '自动化调度',oneLiner: '按 cron 规则自动跑任务(定期分析、工单分拣等),含执行历史,agent 也能通过 MCP 工具自助管理,包括其它电脑上的定时任务。',detail: [- '基于 cron 表达式创建定时任务,常见用法是每天定时跑 AI 工单分拣或周报生成。',- 'GUI 展示任务列表、上次执行结果、失败原因。已执行项可加黑名单防重复。',- 'agent 也能通过 MCP 工具(schedule_create / list / get / update / pause / resume / run_now / delete / list_runs)自助创建和管理调度,GUI 与 MCP 看到同一份数据。',+ '可按 cron / 间隔重复运行,或在指定时间运行一次;常见用法是每天定时跑 AI 工单分拣或周报生成。运行以普通任务的形式进行,可从运行记录打开对应任务。',+ 'GUI 入口是主界面侧边栏「自动化」,展示定时任务列表、运行记录、失败原因,可暂停 / 恢复 / 编辑 / 删除。只在所在电脑的桌面端运行着时触发(不是云端 cron)。',+ 'agent 也能通过 cindy_scheduler 工具(schedule_create / list / get / update / pause / resume / run_now / delete / list_runs,以及 schedule_list_models 查可选运行位置与模型、schedule_set_pre_run_hook 安装运行前检查脚本)自助创建和管理,GUI 与工具看到同一份数据。','定时任务保存在各自的电脑上,工具默认只作用于本机:schedule_list / get / list_runs 传 device="all" 或某台电脑的 deviceId 可查其它在线且已开启远程控制的电脑(结果按电脑分组,没查到的电脑写明原因),update / pause / resume / delete / run_now 传 device=<deviceId> 可管理那台电脑上的任务;新建只在本机。','后端 SQLite 存储,有独立 Scheduler Host 进程管理生命周期。',].join(' '),@@ -202,7 +205,7 @@ export const CAPABILITIES: readonly CapabilityEntry[] = [detail: ['【引用任务】任务元数据的 sessionUrl 可直接写成 [任务标题](sessionUrl);已知 id 只需链接时用 get_task_digest,无需读取完整聊天。',`【是什么】${BRAND_NAME} 的对话历史保存在各电脑的本地 SQLite。cindy_helper 的 history 类提供只读查询入口,让 agent 按宿主授予的范围读取历史,协助用户组织 memory / 知识库。`,- '【六个工具】(1)list_workdirs: 列出本机出现过的工作目录 + session 数 / 首末活动时间; (2)list_sessions: 按 workdir / 时间段 / agent_kind 过滤 session 元数据,并附当前 queuedCount; (3)list_session_queue: 按 session_id 查看尚未消费消息的位置、来源、入队时间、正文摘要与投递状态; (4)get_chat_history: 按 session_ids / workdir / 时间段 / role "按元数据精确捞"原始消息(content / agentMeta JSON 解析后透传); (5)search_chat_history: 跨 session "按内容语义找"——自然语言 query, FTS5 全文(全量、永远可用)+ 向量语义(开启"聊天记录语义索引"后生效)RRF 融合, 返回命中 + 上下文窗口; (6)list_history_devices: 发现本机与在线已授权电脑。',+ '【六个工具】(1)list_workdirs: 列出本机出现过的工作目录 + session 数 / 首末活动时间; (2)list_sessions: 按 workdir / 时间段 / agent_kind 过滤 session 元数据,并附当前 queuedCount; (3)list_session_queue: 按 session_id 查看尚未消费消息的位置、来源、入队时间、正文摘要与投递状态; (4)get_chat_history: 按 session_ids / workdir / 时间段 / role "按元数据精确捞"原始消息(content / agentMeta JSON 解析后透传); (5)search_chat_history: 跨 session "按内容语义找"——自然语言 query, FTS5 全文(全量、永远可用)+ 向量语义(开启「启用对话记录语义索引」后生效)RRF 融合, 返回命中 + 上下文窗口; (6)list_history_devices: 发现本机与在线已授权电脑。','【典型用法】"帮我总结这周和 agent 的讨论, 写成 memory 条目" → list_sessions({from: 周一 ISO, time_field: "activity"}) 拿 sessionId 列表 → get_chat_history({session_ids: [...]}) 拿对话 → LLM 提炼成 memory。"我之前聊过 X / 上次怎么解决那个 bug"(只记得内容、不知道在哪) → search_chat_history({query}) 直接语义召回。"我在 xxx 项目里都聊过啥" → list_sessions({workdir}) → get_chat_history。','【伙伴跨渠道】get_teammate_state 返回本机连接;list_sessions / search_chat_history 使用 teammate_id=profile.id 匹配已保存的归属,含归档、旧历史和话题。按话题搜索不默认限制 workdir 或创建时间。旧 IM 未关联记录仍可在原有账号权限内按内容查询,结果 companion=null 不代表属于当前伙伴;当前连接、目录或名字都不能补造归属。source 是原始任务来源,伙伴渠道看 companion.channel 和消息来源。','【跨设备】list_history_devices 发现设备;list_sessions / search_chat_history 传 device="all" 查询本机与在线已授权电脑,或 device=<设备 ID> 指定电脑。结果按设备分组,每组独立翻页,partial 表示有设备未查到。远程任务 ID 带 deviceId:: 前缀,可直接传给 get_chat_history;不读取离线缓存,旧端不支持会明确报告。默认 device="local" 保持原本机查询。',@@ -215,97 +218,118 @@ export const CAPABILITIES: readonly CapabilityEntry[] = [{key: 'memory-system',title: '记忆系统',- oneLiner: '两层记忆:跨 agent 共享的 Maker Memory + 各 agent 原生记忆,均可通过 MCP 工具读写。',+ oneLiner: '「Cindy」记忆(按工作目录跨 Claude Code / Codex / Pi 共享)或各 Agent 原生记忆二选一,agent 可经 cindy_memory 工具读写。',detail: [- '两层并行:(1)Maker Memory,App 级共享库,跨 session 跨 agent 全局共享,可启用 / 禁用 / 重置;(2)Agent-native 记忆,Claude 走 SDK 内置 Auto Memory,Codex 走 AppServer 实验通道。',- 'Maker Memory 用 frontmatter + body 的 markdown 分片格式,自动维护 MEMORY.md 索引,超大时有 size 告警。',- 'agent 通过 MCP 工具操作:memory_list / read / search / write / delete,以及 memory_consolidate(瘦身)、memory_review(归纳)、session_search(跨历史 session 全文搜索)。',- '系统提示自动注入记忆使用规则,模型主动决定何时存何时取。',+ '入口:设置 →「个性化」→「记忆」。「Cindy」总开关(默认开)启用按工作目录共享的记忆,同一目录下 Claude Code、Codex 与 Pi 的任务共用;开着时 Claude Code / Codex 原生记忆开关置灰,关掉后各自用原生记忆。「Pi」选项需要 Cindy 记忆开启,会把长任务的压缩摘要存进 Cindy 记忆。各项可单独重置,重置不可撤销。改设置一般在新的 Agent 会话生效。',+ 'Cindy 记忆是本机 user-data 下的 markdown 条目(frontmatter + 正文,自动维护 MEMORY.md 索引),不在代码仓库里,换电脑不会迁移。',+ 'agent 通过 cindy_memory 工具操作:memory_list / read / search / write / delete,以及 memory_consolidate(瘦身)、memory_review(归纳)、session_search(跨历史任务全文搜索)。',+ '系统提示自动注入记忆使用规则,模型主动决定何时存何时取。记忆不是任务内的对话记录,也不是「个性化」里的自定义指令。',].join(' '),},{key: 'model-effort-switching',- title: '模型 / Effort / Fast Mode 运行时切换',- oneLiner: '会话进行中切换 Claude 模型、推理强度、权限模式、Fast Mode。',+ title: '模型 / 推理强度 / 速度 / 权限模式切换',+ oneLiner: '任务进行中切换模型、推理强度、速度(标准 / 快速 / 超快)与权限模式,不用新建任务。',detail: [- '不用重开会话就能调:换 Claude 模型(如 Opus / Sonnet / Haiku 之间切)、调推理强度 effort(快答 vs 深度思考)、切权限模式(sandbox / full)、开关 Fast Mode(走 Claude 1M context 通道,只在 Opus 4.6 可用)。',- '修改立刻对下一条消息生效。',+ '输入框处随时可调:换模型(当前账号与供应商可用的模型)、「推理强度」、「速度」(标准 / 快速 / 超快;只在模型支持时出现,目前快速主要由 GPT / Codex 系模型提供,超快取决于当前账号的 Codex 目录)。',+ '权限模式:「默认权限」「自动审批」「完全访问」,Claude Code 另有「允许编辑」;另可开「计划模式」。',+ '模型、推理强度、速度的修改对下一条消息生效(运行中则在当前轮结束后)。agent 可用 cindy_helper control 类 set_session_runtime 调整本机任务的模型来源、模型、推理强度与速度(get_session_runtime 先读);伙伴任务另有 bots 类 get_app_default_model / set_app_default_model 读改以后新建任务的默认模型;agent 不能切换权限模式。',].join(' '),},{key: 'feishu-integration',title: '飞书集成(基础接入)',- oneLiner: '飞书 OAuth 登录与 IM 出站通知;飞书数据读写由企业档 xd-feishu 插件经 ghost 网关提供。',+ oneLiner: '飞书 / Lark 机器人(私聊与群聊使用 Cindy、任务完成通知);飞书数据读写由企业档 xd-feishu 插件经 ghost 网关提供。',detail: [- `使用飞书 OAuth 登录 ${BRAND_NAME}。任务完成、权限超期等事件通过 IM 推送通知;`,- '出站通道(cindy_feishu_bot)负责给当前飞书用户发消息 / 发文件 / 发图片,不提供数据读取。',+ `飞书已不是 ${BRAND_NAME} 的登录方式(用 ${BRAND_NAME} 账号登录)。机器人在 设置 →「IM 机器人」→「个人」→「飞书 / Lark 机器人」配置:填自建应用的 App ID / App Secret 绑定,再私聊机器人一次完成主人绑定;可开「任务完成通知」(窗口失焦时私聊推送)与上下线「消息通知」。群里 @ 机器人会开话题回复,读群上下文需在飞书开放平台额外开通群消息权限。`,+ '出站通道 cindy_feishu_bot:普通桌面任务里经其 list_tools / call_tool 调 send_message_to_user / send_file_to_user,发到机器人主人的私聊(用户明确要求「发飞书给我」时才用);飞书轮次内另有作用域钉在当前对话的 send_text / send_markdown / send_file / 表情 / get_chat / get_message / 置顶 / call_api 等直通工具。','读写飞书数据(云文档 docx / 多维表格 bitable / Wiki 知识库 / IM 消息搜索阅读 / 联系人 / 日程 / 审批 / 会议纪要)属于企业档 xd-feishu 插件:',- '需在侧边栏「插件」安装并启用 xd-feishu、在其详情页连接账号;OAuth 登录成功不等于该插件可用。',+ '需在侧边栏「插件」安装并启用 xd-feishu、在其详情页连接账号;机器人绑定成功不等于该插件可用。','这些操作不注册为顶层 MCP 工具——agent 须经 ghost 网关发现与调用(ghost_info({ghost_id:"xd-feishu"}) 查实时工具清单,再 ghost_call 执行)。',- '把"接管 desktop session 移动办公"这条独立能力请查 mobile-takeover bucket。',+ '从 IM 接管桌面任务这条独立能力请查 mobile-takeover bucket;其它 IM 渠道见 im-channels。',].join(' '),},{key: 'mobile-takeover',- title: '飞书移动办公(接管 desktop session)',- oneLiner: '在飞书里发 /ctr 接管桌面端正在跑的 session,两端同屏推进、不互锁。',+ title: 'IM 移动办公(/ctr 接管桌面任务)',+ oneLiner: '在飞书 / Telegram / Discord 等机器人里发 /ctr 接管桌面端的任务,两端同屏推进、不互锁;/exctr 退出。',detail: [- '触发流程:在飞书里给 bot 发 /ctr 斜杠命令 → 弹卡片选工作区 → 选要接管的 session → 接管开始。',- '接管后是"路由"而非"转移":飞书消息路由到桌面同一个 session,桌面 renderer 仍能实时看到所有输出和工具调用,等于双端同屏。',- '桌面端正在等待用户响应的卡片(权限审批 / ask / plan)会自动搬到飞书,在手机上点完就继续推进,无需切回电脑。',- '桌面端不被锁,两端都能继续发消息推进同一会话,消息合并到同一处理队列按到达序处理。',- '退出:detach 后桌面交互监听器自动还原。当前限制:接管模式下 agent 不能用 send_file_to_user 给飞书推文件(只在原生飞书 session 可用),其他 MCP 工具(art / scheduler / google 等)无差异。',+ '支持卡片的渠道可用:个人飞书 / Lark、Telegram、Discord 机器人,以及官方机器人。触发流程:给机器人发 /ctr → 卡片选工作区 → 选要接管的任务 → 接管开始;已在接管时再发 /ctr 可直接换一个任务。',+ '接管后是"路由"而非"转移":IM 消息路由到桌面同一个任务,桌面端仍能实时看到所有输出和工具调用,等于双端同屏。',+ '桌面端正在等待用户响应的卡片(权限审批 / ask / plan)会自动搬到 IM,在手机上点完就继续推进,无需切回电脑。',+ '桌面端不被锁,两端都能继续发消息推进同一任务,消息合并到同一处理队列按到达序处理。正被接管的任务不能归档或移动,需先结束接管。',+ '退出:发 /exctr(个人机器人也认 /exitctr)或点卡片上的退出;结束接管后桌面交互监听器自动还原。',].join(' '),},{key: 'auto-update',title: '自动更新',- oneLiner: '后台检测新版本,断点续传 + 完整性校验 + 增量更新,不打断会话。',+ oneLiner: '后台检测并下载新版本,提示重启安装;可开「空闲时自动安装更新」;agent 可在用户确认后代为更新。',detail: [- '内置 cindy-updater 模块:启动和定时轮询检测版本,下载支持断点续传、SHA / MD5 校验、自动重试。',- '增量包传输节省带宽。',- '更新过程不中断当前会话,启动时若发现版本不一致会提示用户。',+ '内置更新器在后台检测并下载新版本(断点续传 + SHA256 校验),下载完成后提示用户重启应用来安装,不会意外重启。「设置 → 关于」有「空闲时自动安装更新」开关,开启后在空闲时自动安装并重启(Linux 只在用户选择重启时应用)。',+ 'agent 工具(cindy_helper app_update 类):check_app_update 只读检查;install_app_update 由宿主弹确认卡(写明目标版本、会重启、会中断哪些任务),用户确认后才经内置更新器下载、安装并重启,重启后结果写回发起任务;set_app_auto_update 同样经确认卡开关上述设置。只有主人本人发起的轮次能触发。',+ '不要用 shell 下载安装包或替换应用;开发版、个人版等不支持应用内更新的构建会明确返回原因。',].join(' '),},{- key: 'jira',- title: 'Jira 集成',- oneLiner: 'agent 通过 MCP 工具搜索 / 创建 / 更新 / 流转 Jira 问题,支持多 site。',+ key: 'plugins',+ title: '插件(扩展工具、面板与外部服务连接)',+ oneLiner: '插件为 agent 增加工具、右侧栏面板和应用页;GitHub / GitLab / Google / Atlassian / 网页搜索等外部服务大多以插件提供,agent 经 ghost_* 网关发现与调用。',detail: [- '基于 Atlassian OAuth 接入,与 Confluence 共用 token。',- 'agent 工具:sites_list(发现可用 site)、projects_search、issues_search_jql(JQL 高级搜索)、issue_get(详情含字段/评论)、issue_create / issue_update、issue_add_comment、issue_list_transitions / issue_transition(走工作流)。',- '不需要打开 Jira 网页就能由 agent 帮你做工单的查询、归档、推进。',+ '入口:主界面侧边栏「插件」(与「技能」同一管理界面),可安装、启用 / 停用、卸载,在插件详情页连接账号;也可在设置 →「插件」管理。停用后工具与面板休眠,插件仍保留。',+ 'agent 侧(cindy 网关):ghost_list 看已装插件、ghost_info({ghost_id}) 查实时工具清单、ghost_call 执行;插件工具不注册为顶层 MCP 工具。没有合适插件时 ghost_market_search 搜插件市场、ghost_market_install 按当前任务权限安装;需要连接账号时 connect_account({kind:"plugin", id}) 弹连接卡,不要向用户索要令牌。',+ '常见内置插件:cindy-github / cindy-gitlab(代码托管)、google-gmail / google-drive / google-calendar / google-sheets(部分组织为合并的 filo-google)、企业档 xd-atlassian(Jira / Confluence)、cindy-web-search(网页搜索)、企业档 xd-feishu(飞书数据)。实际可用以 ghost_list 为准,未启用返回 GHOST_ASLEEP,在当前工作目录被停用返回 GHOST_DISABLED_IN_WORKDIR。',+ '用户想自己做插件:ghost_forge_guide / ghost_forge_scaffold / ghost_forge_pack / ghost_forge_install / ghost_forge_publish 提供编写、打包、安装与发布流程。',].join(' '),},{- key: 'confluence',- title: 'Confluence 集成',- oneLiner: 'agent 通过 MCP 工具检索 / 读 / 写 Confluence 页面、评论、附件。',+ key: 'jira',+ title: 'Jira / Confluence(xd-atlassian 插件)',+ oneLiner: 'Jira 工单与 Confluence 文档的查询、创建、更新、评论,由 xd-atlassian 插件经 ghost 网关提供。',detail: [- '基于 Atlassian OAuth 接入,与 Jira 共用 token。',- 'agent 工具:sites_list、spaces_list / space_get、pages_search / page_get / page_list_children / page_create / page_update、blogpost_get、comments_list / comment_add、attachments_list / attachment_get。',- '能让 agent 直接帮你查文档、生成新页面、追加评论,适合做 KB 整理和写作类任务。',+ '原 lizi_jira / lizi_confluence MCP 已退役,能力合并进 xd-atlassian 插件(Atlassian OAuth,老账号已自动迁入)。',+ '需在侧边栏「插件」里启用 xd-atlassian 并在详情页连接 Atlassian 账号;agent 用 ghost_info({ghost_id:"xd-atlassian"}) 查实时工具清单,再 ghost_call 执行,不要假设旧的顶层工具名。',+ '插件列表里没有 xd-atlassian 时说明当前账号 / 构建未提供,不要承诺可用。',].join(' '),},{- key: 'google-sheets',- title: 'Google Sheets 集成',- oneLiner: 'agent 通过 MCP 工具读取 Google Sheets 指定区间数据。',+ key: 'google',+ title: 'Google(Gmail / Drive / Calendar / Sheets 插件)',+ oneLiner: 'Gmail、Google Drive、Google Calendar、Google Sheets 是四个独立内置插件,经 ghost 网关提供;部分组织账号改为合并的 filo-google 插件。',detail: [- '基于 Google OAuth 接入。',- 'agent 工具:sheets_read_range(按 spreadsheet ID + A1 range 读取)。',- '当前是只读集成,适合让 agent 拉取表格数据做分析或汇总,不能改写 Sheets。',+ '原 lizi_google MCP(含只读的 sheets_read_range)已退役。公共内置插件按服务拆分:google-gmail / google-drive / google-calendar / google-sheets,各自在详情页连接 Google 账号授权;部分组织账号拿到的是合并的 filo-google(Filo Google)。',+ '实际装了哪个以 ghost_list 为准;agent 用 ghost_info 查实时工具与读写范围,再 ghost_call 执行。',].join(' '),},{key: 'ai-art',- title: 'AI 绘画 / 视频生成',- oneLiner: 'agent 通过 MCP 工具做文生图、图编辑、视频生成。',+ title: 'AI 绘画 / 视频生成(media 工具)',+ oneLiner: 'agent 通过 cindy 网关的 media 工具做文生图、改图、文生视频、图生视频,也用它把已有图片交付给用户。',+ detail: [+ '旧 lizi_art 的 image_generate / video_generate 等工具已不再提供。现在统一走 cindy 网关的 media 工具:list_models 查当前账号可用的生图 / 视频模型(能力 image.generate / image.edit / video.generate / video.image_to_video),prepare → request 发起生成,异步任务按 recommended_poll_after_ms 调 poll。',+ '可用模型取决于当前账号的模型权限,不要猜模型名;prepare 返回的 invocation_id 是一次性付费令牌,超时或结果未知时不要自动重提以免重复扣费。',+ '结果是 cindy-media:// / xdt-image:// / xdt-video:// 受管地址,在回复里直接嵌入展示一次即可,手机端也能看;把已有截图或本地图片发给用户用 import_image。用户问存储位置时才用 resolve_local_path。',+ ].join(' '),+ },+ {+ key: 'im-channels',+ title: 'IM 机器人渠道',+ oneLiner: '在 Slack、Telegram、X、个人微信、企业微信、飞书 / Lark、钉钉、Discord 里使用 Cindy;设置 →「IM 机器人」配置。',+ detail: [+ '设置 →「IM 机器人」分两区:「官方」由 Cindy 提供并维护(Slack、Telegram、X 等卡片),无需自建机器人或凭证,按引导连接;「个人」用用户自己的账号或机器人:个人微信(扫码)、企业微信智能机器人、飞书 / Lark 机器人、钉钉机器人、Discord 机器人、Telegram 机器人,每个渠道的凭证、Agent / 模型与权限设置独立保存。可见渠道可能随账号与构建不同,以设置页实际显示为准。',+ '在 IM 里直接发消息即进入对应任务;常用命令有 /new、/stop、/model、/permission、/settings、/ctr(接管桌面任务,见 mobile-takeover)、/exctr、/help,各渠道支持程度不同。',+ '官方渠道(Slack / Telegram / X)的任务里,文字回复直接输出即可自动回到当前对话;发文件在最终回复里写 [文件名](xdt-file:///绝对路径)(文件须在当前工作目录内),图片直接引用受管地址。',+ ].join(' '),+ },+ {+ key: 'remote-and-mobile',+ title: '远程连接(手机 / 其他电脑控制,SSH 远端主机)',+ oneLiner: '同账号的手机与其他电脑可远程操控这台桌面上的任务;也可经 SSH 让 agent 在远端主机上运行。',detail: [- 'agent 工具:image_generate(文生图)、image_edit(多图融合 / 编辑)、video_generate(文生视频)、video_edit(参考图生视频)。',- '图片支持多模型(gpt-image-2 / gemini),输入可以是在线 URL 或本地路径。',- '视频生成基于 seedance 等供方,作为可选能力按 host 配置启用。',- '生成结果存到本地,可在对话里直接展示,也可上传飞书或写入文档。',+ '入口:设置 →「远程连接」。同一 Cindy 账号登录的设备自动出现,无需手动配对;每台设备有「我控制它」与「允许它控制本机」两个独立开关,本机另有允许被控制的总开关。手机 App 只做控制端,任务与数据都在被控电脑上。',+ '控制端可查看与推进被控电脑上的任务(发消息、审批、切模型等),账号密钥与全局设置等不开放给控制端。主列表标题「全部任务 ▾」可切换只看本机或某台远程电脑。',+ 'SSH 远端主机:在同一页添加主机(也读取 ~/.ssh/config),可在远端安装 Claude Code / Codex / Pi(未安装时首条消息自动安装)并新建远端任务,agent 进程与文件都在远端。',+ 'agent 侧跨电脑能力(查任务、给另一台电脑的任务发消息、管定时任务等)见 self-management。',].join(' '),},] as const;diff --git a/packages/lizi-mcps/src/xdt-helper/get_capabilities.ts b/packages/lizi-mcps/src/xdt-helper/get_capabilities.tsindex dba42e2b66..9ad2cdb513 100644--- a/packages/lizi-mcps/src/xdt-helper/get_capabilities.ts+++ b/packages/lizi-mcps/src/xdt-helper/get_capabilities.ts@@ -5,6 +5,8 @@* - 不传 key → 返回全部 {key, title, oneLiner} 索引,模型挑感兴趣的再细查* - 传 key → 返回单条 {key, title, oneLiner, detail}* - key 不存在 → 返回可用 key 列表 + UNKNOWN_KEY 错误码,让模型自纠+ * - scope=help → 宿主注入的产品帮助文档(与设置页帮助助手同一份):不传 key 返回目录,+ * 传 key 返回整篇。回答「X 在哪 / 怎么操作」这类界面问题用。** 故意不支持批量 keys:简单优先,模型多调几次成本极低,且分次拉取也省 token。*/@@ -17,8 +19,37 @@ import type { XdtHelperToolRegistry } from '../lizi_xdtHelperToolRegistry.js';export interface RuntimeCapabilityQuery { server?: string; category?: string; toolQuery?: string; offset?: number; }export type RuntimeCapabilityReader = (query: RuntimeCapabilityQuery) => Promise<unknown>;+export interface ProductHelpDoc { id: string; title: string; summary: string; content: string; }-export function registerGetCapabilitiesTool(registry: XdtHelperToolRegistry, readRuntime?: RuntimeCapabilityReader): void {+function textResult(payload: unknown, isError = false) {+ return { content: [{ type: 'text' as const, text: JSON.stringify(payload) }], ...(isError ? { isError: true } : {}) };+}++function readProductHelp(docs: readonly ProductHelpDoc[] | undefined, key: string | undefined) {+ if (!docs?.length) return textResult({ ok: false, errorCode: 'HELP_UNAVAILABLE', message: '当前宿主未提供产品帮助文档。' }, true);+ if (!key) {+ return textResult({+ ok: true,+ docs: docs.map(({ id, title, summary }) => ({ key: id, title, summary })),+ hint: '挑相关的一篇用 get_capabilities({scope:"help", key}) 读全文。',+ });+ }+ const doc = docs.find((d) => d.id === key);+ if (!doc) {+ return textResult({ ok: false, errorCode: 'UNKNOWN_KEY', data: { requested: key, available: docs.map((d) => d.id) } }, true);+ }+ return textResult({+ ok: true,+ doc: { key: doc.id, title: doc.title, content: doc.content },+ hint: '文档按英文界面撰写;用其他语言回答时说出该语言界面上的对应文字,拿不准译名就附英文原名。文档没写到的入口直说不确定,不要推测。',+ });+}++export function registerGetCapabilitiesTool(+ registry: XdtHelperToolRegistry,+ readRuntime?: RuntimeCapabilityReader,+ helpDocs?: readonly ProductHelpDoc[],+): void {registry.register({name: 'get_capabilities',category: 'cindy',@@ -27,9 +58,10 @@ export function registerGetCapabilitiesTool(registry: XdtHelperToolRegistry, rea'传 key=<具体能力 key> → 返回该能力的完整 detail。scope=runtime 查询当前 Agent 的真实工具目录;指定 server 查看注册工具,category 展开内置类目。产品说明不等于可调用工具。' +`当用户问"${BRAND_NAME} 能做什么 / 有哪些功能 / 支持 X 吗"时调用本工具,而不是用训练数据回答。` +`用户问"你是谁 / 你是什么 / 你跑在哪 / ${BRAND_NAME} 是什么产品 / 谁做的 / 开不开源 / 源码在哪"时,` +- '同样调本工具取 key=about-cindy,不要靠工作目录路径、工具名或训练数据推断。',+ '同样调本工具取 key=about-cindy,不要靠工作目录路径、工具名或训练数据推断。' ++ `用户问"某功能在哪 / 界面上怎么操作 / 找不到某个入口"时,用 scope=help 先看帮助文档目录,再传 key 读对应篇,按文档回答。`,inputShape: {- scope: z.enum(['product', 'runtime']).optional().describe('product 查询产品介绍;runtime 查询实际注册与启用状态。'),+ scope: z.enum(['product', 'runtime', 'help']).optional().describe('product 查询产品介绍;runtime 查询实际注册与启用状态;help 查询界面入口与操作步骤的帮助文档。'),server: z.string().optional().describe('从 runtime 目录返回的 server 标识。'),category: z.string().optional().describe('该 server 的 list_tools 返回的类目。'),toolQuery: z.string().trim().max(200).optional().describe('引擎管理的 MCP 工具按名称或说明筛选;每页最多 50 项。'),@@ -37,14 +69,15 @@ export function registerGetCapabilitiesTool(registry: XdtHelperToolRegistry, reakey: z.string().optional()- .describe('能力 key(从不传参的索引返回中获取)。不传 = 返回索引列表。'),+ .describe('能力 key 或帮助文档 key(从不传 key 的目录返回中获取)。不传 = 返回目录。'),},handler: async ({ key, scope, server, category, toolQuery, offset }) => {if (scope === 'runtime') {const result = readRuntime ? await readRuntime({ server, category, ...(toolQuery !== undefined ? { toolQuery } : {}), ...(offset !== undefined ? { offset } : {}) }): { ok: false, errorCode: 'RUNTIME_CATALOG_UNAVAILABLE', message: '当前宿主尚未提供运行能力目录。' };- return { content: [{ type: 'text' as const, text: JSON.stringify(result) }] };+ return textResult(result);}+ if (scope === 'help') return readProductHelp(helpDocs, key);if (!key) {return {content: [