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/metahub-tech/agent-fleet/using-winnpx skills add metahub-tech/agent-fleet --skill using-wingit clone --depth 1 https://github.com/metahub-tech/agent-fleetWhat 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 | $0.00048 | $0.01689 |
| Opus 5 | $0.00024 | $0.00844 |
| Sonnet 5 | $0.00010 | $0.00338 |
| Haiku 4.5 | $0.00005 | $0.00169 |
Grade A, and why
using-win 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 2d 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 — 120 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Using win-device
Drive a remote Windows test machine via the win-device MCP server (FastMCP, streamable-http on Tailscale). Multi-client native; advisory single-holder coordination; 42 tools spanning state / screen / window / mouse / keyboard / process / file / search / shell.
Critical patterns
Screenshot click coordinates are SCREEN pixels, NOT image-display pixels
take_screenshot returns a PNG of the actual screen (e.g. 1920x1080). The viewer rendering it (Read tool, image preview, etc.) MAY downscale to a smaller display size like 960x540. Always pass click coordinates in actual screen pixels.
get_screen_size # returns {"width": 1920, "height": 1080}
take_screenshot # PNG of full screen; rendered display may be smaller
tap(x=1096, y=600) # x,y MUST be in 1920x1080 space
If taps miss: the displayed image is a thumbnail, not the source of truth. Call get_screen_size first; reason about coordinates in that coordinate system.
UI introspection and app lifecycle
dump_ui() # UI-Automation tree of the foreground window
find_elements(query="Save") # locate elements by UIA attributes -> ranked candidates + center coords
tap_element(query="Save") # find + click an element's live center
inspect_window(...) # richer per-window introspection (platform extension)
terminate_app(target="notepad.exe") # terminate by process name / path substring
kill_process(pid=...) # kill by PID (platform extension)
dump_ui, find_elements / tap_element, and terminate_app are the canonical cross-platform tools. inspect_window (richer window introspection) and kill_process (PID-based kill) remain as Windows-specific extensions for finer control.
Element-first, screenshot as fallback
Prefer find_elements / tap_element(query) over hardcoded tap(x, y). Element location queries the live UIA tree by accessibility attributes (name / automation-id / control-type), so it survives layout / scroll / resolution drift that breaks fixed coordinates.
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.
- 2d ago First seen · 120 lines · 48 tokens per session scan A de4d18a02c7b
using-win is a skill published in the GitHub repository metahub-tech/agent-fleet (2 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 48 tokens to every session and 1,689 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
langbot-testing
Test LangBot WebUI and core product flows with an automated browser and backend logs. Use when validating the configured LangBot frontend, pipeline Debug Chat, model provider setup and test buttons, bot and knowledge-base UI flows, or troubleshooting failed LangBot end-to-end tests.
penguin-harness-manual-test
Use when standing PenguinHarness up to try a change by hand — launching the Web App, the desktop shell, the landing page or the docs site to click through it, screenshot it, or reproduce a report. Covers the four dev entry points and their ports, which data root each writes to, and the four ways a healthy setup looks…
browser-testing-with-screenshots
Use when testing web applications with visual verification - automates Chrome browser interactions, element selection, and screenshot capture for confirming UI functionality.
e2e-verify
面向 Bifrost 管理端的端到端 UI 与 API 验证工具。 适用于浏览器测试、场景回归、管理端接口验证与页面快照排查。 Use when: 端到端验证、功能验证、E2E 测试、UI 测试、浏览器测试、API 测试、接口验证.
browser-qa
Drive a real headless browser against a locally-served web UI and capture screenshot evidence, reporting step-level PASS/FAIL/BLOCKED with extracted DOM text and console errors. Use when a phase on the docker QA image must exercise a rendered UI and attach screenshots — it complements the text-evidence verify/qa-test…
tlamatini-daily-chat-test
Run the daily automated Tlamatini chat regression — drive a visible Chrome via Playwright, log into agentpage.html, ask up to 1000 curated safe questions one-by-one (Multi-Turn ON, ACPX/Ask-Execs/Exec-Report/Internet OFF), wait for and qualify each answer (heuristic + LLM judge on failures), then write a dated report…