screenshot

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Screenshotting

屏幕截图功能

Take screenshots to see what's on screen. Captures persist as files (unlike browsermcp snapshots which only exist in context).

截取屏幕截图以查看屏幕内容。截图会保存为文件（与仅存在于上下文的browsermcp快照不同）。

Quick Reference

快速参考

bash

undefined

bash

undefined

Capture specific app window

捕获特定应用窗口

~/.claude/skills/screenshot/scripts/look.py --app Ghostty

Capture window by title match

通过标题匹配捕获窗口

~/.claude/skills/screenshot/scripts/look.py --app Chrome --title "LinkedIn"

Capture full screen

捕获全屏

~/.claude/skills/screenshot/scripts/look.py --screen

List available windows

列出可用窗口

~/.claude/skills/screenshot/scripts/look.py --list

List windows grouped by category

按类别分组列出窗口

~/.claude/skills/screenshot/scripts/look.py --categories

List only browser windows

仅列出浏览器窗口

~/.claude/skills/screenshot/scripts/look.py --category browsers

Native resolution (skip resize)

原始分辨率（跳过缩放）

~/.claude/skills/screenshot/scripts/look.py --app Safari --native


**Categories:** browsers, terminals, editors, communication, documents, media, other

~/.claude/skills/screenshot/scripts/look.py --app Safari --native


**类别：** browsers, terminals, editors, communication, documents, media, other

When to Use

适用场景

Reactive (user asks):

"Have a look at this"
"Can you see what's on screen?"
"What does it look like?"
"Check the browser"

Proactive (verify state):

After uncertain CLI operations (did it background?)
When tool prompt state is unclear
After browsermcp actions when snapshot isn't enough
To verify visual changes actually happened

Documentation:

Capture steps in a workflow
Before/after comparisons
Bug evidence with screenshots

响应式（用户主动请求）：

"Have a look at this"
"Can you see what's on screen?"
"What does it look like?"
"Check the browser"

主动式（验证状态）：

执行不确定的CLI操作后（是否成功后台运行？）
当工具提示状态不明确时
执行browsermcp操作后，快照不足以满足需求时
验证视觉变化是否实际发生

文档记录：

捕获工作流中的步骤
前后对比截图
用截图作为Bug证据

When NOT to Use

不适用场景

Browser-only tasks where browsermcp snapshot suffices
When you just need to describe what's visible
High-frequency captures that would clutter the directory

仅浏览器任务，browsermcp快照已足够时
仅需描述可见内容时
高频率捕获会导致目录杂乱时

Anti-Patterns

反模式

Pattern	Problem	Fix
Capture without purpose	Clutters context	Only screenshot when you need the visual
Skip --list first	Wrong window captured	List windows to find exact app/title
Native on large screens	Huge files, slow upload	Use default 1568px resize

模式	问题	解决方法
无目的捕获	占用上下文空间	仅在需要视觉信息时才截图
未先执行--list	捕获到错误窗口	先列出窗口以找到准确的应用/标题
大屏幕使用原生分辨率	文件过大，上传缓慢	使用默认的1568px缩放

Resolution Strategy

分辨率策略

Default: 1568px max dimension (~1,600 tokens, optimal for API)

Option	Tokens	Use case
Default (1568px)	~1,600	Full detail, no resize penalty
`--max-size 735`	~500	Quick look, text readable
`--native`	varies	When original resolution needed

Why 1568px: Images larger than this get resized server-side anyway. Pre-resizing avoids upload latency while getting the same visual fidelity.

默认：最大尺寸1568px（约1600个token，API最优尺寸）

选项	Token数	使用场景
默认（1568px）	~1600	完整细节，无缩放损耗
`--max-size 735`	~500	快速查看，文本可读
`--native`	可变	需要原始分辨率时

为什么选择1568px： 大于此尺寸的图片会在服务器端被缩放。预先缩放可避免上传延迟，同时保持相同的视觉保真度。

Output

输出

Ephemeral (no path): Screenshots go to

/tmp/claude-screenshots/

— auto-cleaned by OS, won't clutter project directories:

/tmp/claude-screenshots/2025-12-15-143022-chrome.png

Persistent (explicit path): For documentation workflows, specify where screenshots should live:

bash

look.py --app Chrome ./docs/step3.png

Design rationale: Quick looks are ephemeral by default. Documentation requires intentional placement. If a subagent is documenting, it should think about where artifacts belong.

临时文件（无指定路径）： 截图保存到

/tmp/claude-screenshots/

— 由系统自动清理，不会杂乱项目目录：

/tmp/claude-screenshots/2025-12-15-143022-chrome.png

持久文件（指定路径）： 用于文档工作流时，可指定截图保存位置：

bash

look.py --app Chrome ./docs/step3.png

设计理念： 快速查看的截图默认是临时的。文档记录需要有意指定保存位置。如果子代理负责文档记录，应考虑工件的存放位置。

How It Works

工作原理

Window enumeration: Uses macOS CGWindowList API (pure Quartz, no AppleScript)

Capture: Uses

screencapture -l<windowid>

for windows,

screencapture -x

for screen

Resize: Uses
```
sips --resampleHeightWidthMax
```
for efficient scaling

Key capability: Can capture windows even when covered or minimized.

窗口枚举： 使用macOS CGWindowList API（纯Quartz，无AppleScript）

捕获： 对窗口使用

screencapture -l<windowid>

，对屏幕使用

screencapture -x

缩放： 使用
```
sips --resampleHeightWidthMax
```
进行高效缩放

核心能力： 即使窗口被覆盖或最小化，也能捕获。

Limitations

局限性

Scrollback: Only captures visible viewport. If content scrolled off screen, it won't be in the screenshot. Workaround: increase window size or pipe output to file.

Multiple monitors: Untested.

--screen

with

-m

flag captures main monitor only.

Window selection: Takes first match when multiple windows match filters. No "frontmost" heuristic yet.

滚动内容： 仅捕获可见视口。如果内容滚动出屏幕，不会出现在截图中。解决方法：增大窗口尺寸或将输出管道到文件。

多显示器： 未测试。带

-m

标志的

--screen

仅捕获主显示器。

窗口选择： 当多个窗口匹配筛选条件时，捕获第一个匹配项。目前没有“最前端”启发式逻辑。

Integration with browsermcp

与browsermcp的集成

browsermcp's

browser_screenshot

injects images directly into context but doesn't persist them as files. Use this skill when you need:

Screenshots that persist beyond the conversation
Captures of non-browser apps
Captures of windows behind the browser
Files to upload to Drive or include in docs

browsermcp的

browser_screenshot

直接将图片注入上下文，但不会保存为文件。在以下场景使用本技能：

需要在对话之外持久化的截图
捕获非浏览器应用
捕获浏览器后方的窗口
需要上传到Drive或包含在文档中的文件

Permissions

权限

Requires Screen Recording permission in System Preferences > Privacy & Security.

If capture fails with "check Screen Recording permissions", the user needs to grant permission to the terminal app (Ghostty, Terminal, iTerm, etc.).

需要在系统设置 > 隐私与安全性中开启“屏幕录制”权限。

如果捕获失败并提示“check Screen Recording permissions”，用户需要为终端应用（Ghostty、Terminal、iTerm等）授予权限。