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/wimi321/windows-computer-use-skill/computer-use-windowsnpx skills add wimi321/windows-computer-use-skill --skill computer-use-windowsgit clone --depth 1 https://github.com/wimi321/windows-computer-use-skillWrote 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/wimi321/windows-computer-use-skill/computer-use-windows)<a href="https://agentmods.dev/skills/wimi321/windows-computer-use-skill/computer-use-windows"><img src="https://agentmods.dev/badge/skills/wimi321/windows-computer-use-skill/computer-use-windows.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 | $0.00036 | $0.00469 |
| Opus 5 | $0.00018 | $0.00234 |
| Sonnet 5 | $0.00007 | $0.00094 |
| Haiku 4.5 | $0.00004 | $0.00047 |
Grade A, and why
computer-use-windows 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
Windows Computer-Use Skill
Use this skill when the task needs a portable Windows computer-use skill bundled with its own standalone runtime and MCP server.
What this skill does
- uses the bundled
windows-computer-use-skillproject under the installed skill directory - builds the standalone MCP server
- lets the server auto-bootstrap its Python runtime on first launch
- avoids any dependency on local Claude binaries,
.nodemodules, or extracted app assets - stays explicitly Windows-only because the underlying desktop-control backend is Windows-specific
Default bundled project path
After installation, assume the standalone project lives at:
~/.codex/skills/computer-use-windows/project
If the user installed the skill under a custom CODEX_HOME, use that equivalent path instead.
Build
Always build from the bundled project:
cd ~/.codex/skills/computer-use-windows/project
npm install
npm run build
Run
cd ~/.codex/skills/computer-use-windows/project
node dist/cli.js
The first real run will automatically create .runtime/venv and install the public Python dependencies.
Validation notes
- Version
0.1.1fixes the shared system-key blocklist logic so Windows runtime builds use the correct Windows shortcut denylist instead of a broken cross-platform branch. - This project has been statically validated from macOS for TypeScript build health, bundled project integrity, and Python helper syntax, but still needs end-to-end runtime validation on a real Windows machine.
Guardrails
- Treat this host as trusted-local only.
- Do not tell the user to search their local Claude install for binaries or hidden assets.
- Be explicit that this runtime is standalone and uses public dependencies only.
- Mention that the current runtime reports
screenshotFiltering: none, so action gating is handled at the MCP layer.
What ships with it
47 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- agents/openai.yaml 370 B
- project/.gitignore 59 B
- project/assets/hero.svg 5.0 KB
- project/examples/env.sh.example 233 B
- project/examples/mcp-config.json 292 B
- project/LICENSE 1.0 KB
- project/package-lock.json 55 KB
- project/package.json 1.1 KB
- project/README.ja.md 7.1 KB
- project/README.md 7.0 KB
- project/README.zh-CN.md 6.2 KB
- project/runtime/requirements.txt 72 B
- project/runtime/windows_helper.py 23 KB runs code
- project/src/cli.ts 258 B runs code
- project/src/computer-use/appNames.ts 3.0 KB runs code
- project/src/computer-use/common.ts 250 B runs code
- project/src/computer-use/computerUseLock.ts 3.2 KB runs code
- project/src/computer-use/drainRunLoop.ts 1.1 KB runs code
- project/src/computer-use/escHotkey.ts 907 B runs code
- project/src/computer-use/executor.ts 6.5 KB runs code
- project/src/computer-use/gates.ts 975 B runs code
- project/src/computer-use/hostAdapter.ts 1.8 KB runs code
- project/src/computer-use/inputLoader.ts 397 B runs code
- project/src/computer-use/pythonBridge.ts 4.2 KB runs code
- project/src/computer-use/swiftLoader.ts 397 B runs code
- project/src/lib/errors.ts 403 B runs code
- project/src/lib/execFileNoThrow.ts 883 B runs code
- project/src/lib/log.ts 428 B runs code
- project/src/lib/sleep.ts 110 B runs code
- project/src/lib/withResolvers.ts 260 B runs code
- project/src/server.ts 1.5 KB runs code
- project/src/session.ts 3.4 KB runs code
- project/src/vendor/computer-use-mcp/deniedApps.ts 18 KB runs code
- project/src/vendor/computer-use-mcp/executor.ts 3.0 KB runs code
- project/src/vendor/computer-use-mcp/imageResize.ts 3.4 KB runs code
- project/src/vendor/computer-use-mcp/index.ts 1.7 KB runs code
- project/src/vendor/computer-use-mcp/keyBlocklist.ts 5.0 KB runs code
- project/src/vendor/computer-use-mcp/mcpServer.ts 12 KB runs code
- project/src/vendor/computer-use-mcp/pixelCompare.ts 5.9 KB runs code
- project/src/vendor/computer-use-mcp/sentinelApps.ts 1.5 KB runs code
- project/src/vendor/computer-use-mcp/subGates.ts 469 B runs code
- project/src/vendor/computer-use-mcp/toolCalls.ts 134 KB runs code
- project/src/vendor/computer-use-mcp/tools.ts 26 KB runs code
- project/src/vendor/computer-use-mcp/types.ts 27 KB runs code
- project/tsconfig.json 422 B
- scripts/install.ps1 408 B runs code
- scripts/install.sh 260 B runs code
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 · 65 lines · 36 tokens per session scan A 897a71c419fb
computer-use-windows is a skill published in the GitHub repository wimi321/windows-computer-use-skill (1 stars, last pushed 5mo ago), licensed MIT. It adds 36 tokens to every session and 469 once invoked, about $0.0002 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
computer-use-macos
Top-level macOS computer-use skill with a bundled standalone runtime that bootstraps itself without any local Claude installation, private native modules, or extracted app assets.
exo-control
Use when any AI needs eyes and hands on Windows — desktop UIA, browser CDP, files, registry, OS ops via Exo Control (MCP, CLI, or Python). Harness-agnostic.
computer-use-linux
Top-level Linux computer-use skill with a bundled standalone runtime that bootstraps itself without any local Claude installation, private native modules, or extracted app assets.
desktop-gui-inspect
Full-desktop automation for targets that have no browser and no API at all — a legacy Java-based NMS client, a vendor's Windows-only configuration utility, a terminal emulator with no scriptable interface. Drives OpenClaw's ClawHub computer-use skill (Xvfb+XFCE virtual desktop, xdotool input automation) to read…
host-computer-use
Beta desktop control through the connected A0 CLI host. Use for the user's host/local computer screenshots, screen inspection, menus, native app UI, OS-level clicking, scrolling, typing, or checking computeruseremote status. Use instead of linux-desktop for host/local machine control. Do not use for ordinary browser…
workflow-design
接收任务 ├─ [查] uimapssearch 检索同类经验 → 有 screen/experience 直接复用,跳过重复探索 ├─ [探] 逐屏探索 → uimapssavescreen 固化布局(每屏经用户确认) ├─ [固] 生成 params.json + workflow.json + guide.md(参数即契约,全部有界面证据) ├─ [验] workflowvalidate 编译校验 → workflowrun 执行 → 连续 3 次一致 + 至少一个异常路径 └─ [馈] 跑通后 uimapssaveexperience 提炼经验;新异常回写 params.json exceptions.