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 skills add katipally/openlive --skill computer-usegit clone --depth 1 https://github.com/katipally/openliveWrote 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/katipally/openlive/computer-use)<a href="https://agentmods.dev/skills/katipally/openlive/computer-use"><img src="https://agentmods.dev/badge/skills/katipally/openlive/computer-use/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/katipally/openlive/computer-use"><img src="https://agentmods.dev/badge/skills/katipally/openlive/computer-use.svg" alt="Reviewed on agentmods" width="80" 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 5.1 | $0.00033 | $0.01366 |
| Opus 5.5 | $0.00013 | $0.00546 |
| Sonnet 5.5 | $0.00007 | $0.00273 |
| Haiku 4.5 | $0.00003 | $0.00137 |
Grade B, and why
computer-use scanned grade B with 1 finding 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 11d 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.
Subtle steeringmediumPrompt injection
Instructions that bias recommendations or shape behaviour without the user noticing.
- Never tell the user something was sent, saved, bought or deleted unless the How it starts
The opening of the file, as written. The whole thing — 127 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Working in apps on the user's computer
You see an app through its accessibility tree: every control numbered, with a picture of the window after it. You act on those numbers. Work in short loops: look, act once, read what came back, then decide the next step.
The rules that never bend
- Do not send, submit, buy, delete, or change account settings unless the user asked for exactly that. When a step would do one of these and they did not ask, stop and ask.
- Password managers are off limits. Do not open, read or fill from them.
- Never tell the user something was sent, saved, bought or deleted unless the window's state shows it.
1. Look first
Call get_app_state with the app before any action. Name the app by its name
or its id from list_apps. Without one you get the window in front that is
not OpenLive's own.
- More than one window?
list_windows, then passwindow_id. - Only need the controls? Pass
screenshot: false. The tree alone is faster. - App not running?
open_app. A web page?open_urlis one call and cannot miss; prefer it over typing into the address bar. - Still loading?
wait(a second or two), which looks again. Never guess what a page probably says by now.
2. Act on meaning, not pixels
In order of preference:
set_valuewith an element number, for a text field, search box, slider or checkbox. It replaces the whole value and is the surest way to fill a field.clickwith an element number. It presses through accessibility when it can, which works even when the window is behind another.perform_actionwith an action the element lists under "Secondary Actions", such as opening a menu or expanding a row.typeafter clicking into a field, only whenset_valuecannot reach it. Passpaste: truefor long text; the clipboard is put back after.keypressfor shortcuts, as["cmdorctrl", "s"].cmdorctrlis Cmd on a Mac and Ctrl elsewhere.xandyfrom the latest picture, only when no element fits (a canvas, a map, a game). Use the coordinates of the picture you were given.
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.
- 11d ago Changed 4b39e00d0de2
- 12d ago First seen · 127 lines · 33 tokens per session scan B 4ba6c11dcbd8
computer-use is a skill published in the GitHub repository katipally/openlive (360 stars, last pushed 8d ago), licensed Apache-2.0. It adds 33 tokens to every session and 1,366 once invoked, about $0.0001 per session on Opus 5.5. A static security scan graded it B with 1 finding (subtle steering). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-10-04.
Other skills, from other repositories
computer-use
Drives the GUI of a visible local app window through orca computer: accessibility tree, clicks, typing, menus, dialogs, and screenshots in native apps and external browser windows (Chrome, Edge, Safari) or webviews. Prefer a programmatic path (shell, filesystem, git, HTTP, existing CLIs) whenever it can complete the…
browser-use
Control the user's Chrome through the existing persistent Node REPL and the Qwen Browser SDK.
opencli-autofix
Automatically fix broken OpenCLI adapters when commands fail. Load this skill when an opencli command fails — it guides you through collecting a trace artifact, patching the adapter, retrying, and filing an upstream GitHub issue after a verified fix. Works with any AI agent.
browser
Drives a real browser through the omowright library from the js eval kernel: sites the user is already signed into, forms and clicks, JS-rendered pages, screenshots, web QA, extension popups, a human handoff for login, CAPTCHA or OTP, and a browser you own for scraping, bot-scored targets, network capture and QA…
web-scraping
Fetches web pages and extracts structured data using mq's toolchain (mq-crawl for fetching/crawling/JS rendering, mq for HTML-to-Markdown selector-based extraction, http() for in-query requests). Use when the user wants to scrape a URL, pull structured data out of a webpage, crawl a site, or turn HTML into…
youtube-search
Search YouTube by keyword and return structured video metadata (title, URL, channel, views, duration, date) via yt-dlp. No API keys. Triggers on: "search youtube", "find youtube videos", "top youtube videos on", "trending videos on", "youtube results for", "yt search", "/yt-search".