Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/cavinhuang/lume/computer-usenpx skills add CavinHuang/lume --skill computer-usegit clone --depth 1 https://github.com/CavinHuang/lumeWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/cavinhuang/lume/computer-use)<a href="https://agentmods.dev/skills/cavinhuang/lume/computer-use"><img src="https://agentmods.dev/badge/skills/cavinhuang/lume/computer-use.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00011 | $0.00776 |
| Opus 5 | $0.00005 | $0.00388 |
| Sonnet 5 | $0.00002 | $0.00155 |
| Haiku 4.5 | $0.00001 | $0.00078 |
Grade A, and why
computer-use scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
What it actually says
Computer Use
Use this skill for native desktop applications. Browser pages should use the Browser or Chrome capability when it is available.
Treat connection setup as internal. Do not mention Node REPL, JavaScript sessions, module imports, or sky in user-facing updates unless the user asks about the implementation.
Before the first desktop action in a task, bootstrap the trusted client through mcp__node_repl__js using the absolute path of this plugin's scripts/computer-use-client.mjs:
if (!globalThis.sky) {
const { setupComputerUseRuntime } = await import("${PLUGIN_DIR}/scripts/computer-use-client.mjs");
await setupComputerUseRuntime({ globals: globalThis });
}
Then read the runtime guidance, exact API, and confirmation policy once:
globalThis.computerUseGuidance = await sky.documentation("guidance")
globalThis.computerUseApi = await sky.documentation("api")
globalThis.computerUseConfirmations = await sky.documentation("confirmations")
nodeRepl.write(`${computerUseGuidance}\n\n${computerUseApi}\n\n${computerUseConfirmations}`)
The Node tool only returns string values passed to nodeRepl.write. Bare final expressions are invisible; use nodeRepl.write(JSON.stringify(value)) for structured output.
Use the exact snake_case Window2 methods from the API. Start discovery with await sky.list_apps(); do not translate method names to listApplications, focusApplication, or observe.
list_apps() returns application descriptors, not Windows. Select a Window from an application's windows array before calling get_window. Window is plain data and has no methods; every action is a sky.* call. WindowState uses accessibility and plural screenshots, never text or singular screenshot:
globalThis.apps = await sky.list_apps()
// Replace this exact matcher with the application requested by the user.
globalThis.targetApp = apps.find(app => app.displayName === "微信" && app.windows.length === 1)
if (!targetApp || targetApp.windows.length !== 1) {
nodeRepl.write(JSON.stringify({
error: "Select one unique application window",
candidates: apps.filter(app => /微信|wechat|weixin/i.test(`${app.displayName ?? ""} ${app.id}`))
}))
} else {
globalThis.targetWindow = targetApp.windows[0]
targetWindow = await sky.get_window({ id: targetWindow.id, app: targetWindow.app })
globalThis.state = await sky.get_window_state({ window: targetWindow })
targetWindow = state.window
globalThis.screenshot = state.screenshots[0]
}
For a coordinate action, call the client directly and pass the current Window and screenshot ID:
await sky.click({ window: targetWindow, x, y, screenshotId: state.screenshots[0].id })
await sky.type_text({ window: targetWindow, text })
state = await sky.get_window_state({ window: targetWindow })
targetWindow = state.window
Keep reusable top-level bindings such as apps, targetWindow, and state. After every observation, replace the old target with targetWindow = state.window. Never invent or reconstruct a window ID.
If the trusted client or desktop host is unavailable after one retry, stop and report the failure. Do not fall back to PowerShell SendKeys, WScript, pyautogui, ctypes, robotjs, terminal UI automation, or system-key shortcuts.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 3d ago First seen · 68 lines · 11 tokens per session scan A 06b968920193
computer-use is a skill published in the GitHub repository CavinHuang/lume (2 stars, last pushed 3d ago), licensed MIT. It adds 11 tokens to every session and 776 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
in-app-browser
Proma 内嵌受管浏览器使用指南。当用户要求打开、展示、访问、浏览或操作网页,或提到小红书、X/Twitter、LinkedIn、BOSS 直聘、登录后站内搜索、动态页面、截图或本地 HTML/React 预览时使用。对邮件、消息、文档、项目管理等已有匹配专用 MCP/API/CLI 的服务,必须优先使用专用工具;仅在没有匹配工具、工具无法完成当前能力、网络搜索工具不可用或无法取得足够好的结果、或用户明确要求网页时改用 Browser。浏览器工具出现在当前工具列表时,必须先阅读本 Skill 再进行网页操作;不要因为工具直接可见就跳过。.
ha-browser
Hope Agent browser automation — the standard status → tabs → snapshot → act loop, stale-ref recovery rules, and what to do when login / 2FA / captcha / camera-prompt / dialog blocks progress. Load this skill whenever you reach for the browser tool. Trigger on: user asks the agent to open / control / click / scrape /…
guizang-ppt-skill
生成横向翻页网页 PPT(单 HTML 文件),含 WebGL 背景、章节幕封、数据大字报、图片网格等模板。提供两种风格:① "电子杂志 × 电子墨水"(衬线 + 流体背景 + 暖色) ② "瑞士国际主义"(无衬线 + 网格点阵 + IKB/柠檬黄/柠檬绿/安全橙高亮)。当用户需要制作分享 / 演讲 / 发布会风格的网页 PPT,或提到"杂志风 PPT"、"瑞士风 PPT"、"Swiss Style"、"horizontal swipe deck"时使用。.
proma-coach
Proma 使用顾问,主动把用户在 Proma/Agent/Skill/Chat 工具/项目里的摩擦、疑惑、重复解释和低效流程,转成更顺手的使用方式或合适的知识维护动作。触发要积极:用户表达不满、困惑、重复提醒、"为什么没用/不会自动/又要我说"、"算了,我自己来"、"你上次不是说..."、"你又忘了"、"以后都这样/能不能记住/少让我选/下次自动"、询问 Proma 怎么用更好、某事能不能固化、该用 Agent 还是 Chat 工具、有没有现成 Skill、Skill 为什么没触发、想优化已有 Skill description、想减少步骤/降低认知负担/让 Proma…
skill-creator
Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.
automation
Proma 内嵌自动任务与定时任务 Skill,属于 Proma 自带能力而不是用户临时安装的外部 Skill。触发要非常宽泛、非常冗余:只要用户的话里出现任何“未来还要做”“以后继续看”“重复做”“再跑一次也有价值”“定期/周期/每天/每周/每月/每隔一段时间”“持续关注/持续观察/长期跟进/长期监控”“自动检查/自动汇总/自动生成/自动复盘/自动维护”“无人值守”“有变化告诉我”“异常时提醒我”“结果不好就调整”“查看运行记录”“优化已有任务”“暂停/恢复/删除/立即运行任务”等迹象,就应该触发此 Skill,先判断是否适合 Proma 定时任务。也要覆盖一次性与有限次的延时执行信号:“X…