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/sshh12/claude-plugins/windows-computer-usenpx skills add sshh12/claude-plugins --skill windows-computer-usegit clone --depth 1 https://github.com/sshh12/claude-pluginsWrote 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/sshh12/claude-plugins/windows-computer-use)<a href="https://agentmods.dev/skills/sshh12/claude-plugins/windows-computer-use"><img src="https://agentmods.dev/badge/skills/sshh12/claude-plugins/windows-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 | $0.00060 | $0.00743 |
| Opus 5 | $0.00030 | $0.00371 |
| Sonnet 5 | $0.00012 | $0.00149 |
| Haiku 4.5 | $0.00006 | $0.00074 |
Grade A, and why
windows-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.
How it starts
The opening of the file, as written. The whole thing — 44 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Windows computer use
Drive the local Windows desktop through the windows-computer-use MCP tools. The server runs on
this machine and controls the real displays, mouse, and keyboard.
Tools
- screenshot — see the screen:
desktop,display:N/display:primary|left|right, a window (window:<title>, even if it is behind others),foreground, or aregion. Returns a downscaled image + acapture_id. - act — input, batched:
left_click,type,key,scroll,drag,hold_key,paste,click_element,mouse_move_relative, … Coordinates are in the last screenshot's image space. - record — record N seconds → one timestamped frame montage + an mp4 (judge motion/animation).
- play — run a timed input script at a cadence while recording (games):
hold w 2,look 200 0,tap space,probe/untilfor telemetry-driven early stop. - window — find / focus / close /
get_text(read via UI Automation, no screenshot) /click_element(click a control by accessible name) — all token-cheap, no image. Usecloseto close UWP/Store apps (their window PID isn't the killable process). - process —
launch(incl.shell:truefor URLs /ms-settings:/ Store apps),kill,wait,shell,wait_for_file,wait_for_window(readiness — use instead of guessing a sleep). - system —
displays(monitor layout, DPI, scale),cursor, clipboard get/set.
Rules that prevent the common mistakes
- Coordinates are in the last
screenshot's image space; the server maps them to physical pixels. A newer capture invalidates old coordinates — each screenshot returns acapture_idandacterrors if you click against a stale frame. Pass thatcapture_idtoactwhen precision matters. - Keyboard input goes to the FOREGROUND window. To type into an app, pass
act focus="<title>"(focuses atomically before the batch) or callwindow focusfirst.acttells you which window actually received the input. - To read text, use
window get_text(UIA + OCR fallback) instead of a screenshot — far cheaper, and it works without focusing the window. To click a labeled control, useclick_element— it invokes via UI Automation and works even when the window isn't foreground. - Multi-monitor: call
system displaysfirst; target a monitor withdisplay:0/display:primary|left|right. Displays are 0-indexed. - Readiness: after
process launch, usewait_for_windowrather than sleeping. - Screenshots are inline + downscaled; video and full-resolution captures are written to a file and the path/link is returned.
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 · 44 lines · 60 tokens per session scan A eafd0f652dad
windows-computer-use is a skill published in the GitHub repository sshh12/claude-plugins (22 stars, last pushed 2mo ago), licensed MIT. It adds 60 tokens to every session and 743 once invoked, about $0.0003 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 2026-08-30.
Other skills, from other repositories
pipefy-ai-agents
Use this skill when the user wants to create, read, update, delete, or troubleshoot AI agents (conversational agents with behaviors). Covers 7 MCP tools including pre-flight validation, plus pipe-scoped knowledge bases (list, plain text/document/data lookup CRUD, access probe) attached via dataSourceIds. For…
pipefy-automations
Use this skill when the user wants to create, read, update, or delete traditional automations (if/then rules) or AI automations (prompt-driven). Covers 16 MCP tools. For AI agents (conversational), see skills/ai-agents/.
pipefy-portal-setup
Use this skill when the user wants to list, create, or configure Pipefy portals (main hub, pages, page elements, sub-portals, publish/unpublish). Covers 20 MCP tools on Interfaces + internalapi. Not for pipes/cards.
pipefy-ipaas
Use when the user wants to build, test, publish, or manage iPaaS (Advanced Automations) flows: multi-step integrations with external apps (Slack, Gmail, Google Sheets), incoming webhooks, schedules, routers, code steps, or iPaaS data tables. MCP-only, driven through 4 meta-tools over a per-pipe catalog discovered at…
pipefy-api-fallback
Use this skill when an MCP tool fails AND the introspection skill could not resolve the problem. This is the last-resort fallback (Tier 3): call the Pipefy GraphQL API directly using curl or httpx, authenticating with the Service Account (OAuth2) or a Personal Access Token (PAT) available as env var. Follow the 3-tier…
pipefy-attachments
Use this skill when the user wants to upload a file to a Pipefy card or table-record attachment field. Covers both input modes — a local filepath and a downloaded fileurl — plus a presigned-URL handshake for uploading a file the server can't read, through the MCP tools and the CLI.