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 Panchangam18/background-computer-use --skill computer-usegit clone --depth 1 https://github.com/Panchangam18/background-computer-useWrote 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/panchangam18/background-computer-use/computer-use)<a href="https://agentmods.dev/skills/panchangam18/background-computer-use/computer-use"><img src="https://agentmods.dev/badge/skills/panchangam18/background-computer-use/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/panchangam18/background-computer-use/computer-use"><img src="https://agentmods.dev/badge/skills/panchangam18/background-computer-use/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 | $0.00106 | $0.01987 |
| Opus 5.5 | $0.00042 | $0.00795 |
| Sonnet 5.5 | $0.00021 | $0.00397 |
| Haiku 4.5 | $0.00011 | $0.00199 |
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 1mo 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 — 174 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Background computer use (macOS)
You have access to a Codex-style computer-use tool surface that drives macOS
applications in the background — without activating them, bringing them to
the front, or moving the user's cursor. The tools are provided by the
background-computer-use MCP server and are named:
list_appsget_app_stateclickdragperform_secondary_actionpress_keyscrollset_valuetype_text
The only interaction pattern that works
Use this loop every time:
- Discover (optional). If you don't know which app to drive, call
list_apps. Theappargument on every other tool accepts either the display name ("Finder") or the bundle identifier ("com.apple.finder"). - Observe. Call
get_app_state(app)once at the start of every turn before acting on that app. Its response contains:- an indexed accessibility tree of the key window
- a window screenshot
- (sometimes) bundled app-specific hints under
<app_hints>
- Decide. Pick element indexes from the tree wherever possible. Those
indexes are valid only until your next
get_app_statecall — treat them as snapshot-local, never durable. - Act. Invoke
click/type_text/press_key/scroll/set_value/drag/perform_secondary_actionusing the index you just picked (or raw coordinates as a fallback). - Re-observe after any action that likely changed the UI (menu opened, view navigated, sheet presented, async result arrived). Stale indexes are the #1 cause of silent failures.
If you skip step 2, tools that take element_index will fail with
"No captured app state for this app."
Accessibility vs. coordinates
Prefer accessibility-based targeting. It is more robust to window moves, scaling, and theme changes.
- Use
click(app, element_index=N)when the target appears in the tree. For left single-clicks on elements that supportAXPress, the server callsAXPressdirectly, which works even when the app is occluded. - Use
click(app, x=..., y=...)only when the tree does not surface the target (e.g. custom canvases, WebViews without AX, games). Coordinates are screen points in the top-left origin. - Use
perform_secondary_action(app, element_index, action)when the element advertises an action like"open","Increment","Decrement","Pick","ScrollToVisible","Confirm". Most secondary actions are background-safe.- Refused by default:
AXShowMenu,AXShowAlternateUI, andAXRaisealmost always bring the target app to the foreground because they need to present a visible menu or modal. The tool raisesToolErroron these unless you passallow_foreground_activation=True. Only use that flag when you've confirmed with the user that activating the app is acceptable for this turn.
- Refused by default:
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.
- 1mo ago First seen · 174 lines · 106 tokens per session scan A 5d91441b9036
computer-use is a skill published in the GitHub repository Panchangam18/background-computer-use (5 stars, last pushed 6mo ago), licensed MIT. It adds 106 tokens to every session and 1,987 once invoked, about $0.0004 per session on Opus 5.5. 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
axstream
Fast deterministic macOS UI automation via replayable macros. Use BEFORE driving any macOS app's UI by hand (clicking/typing via computer-use or screenshots) — if a macro exists, replay finishes the task in seconds with no per-step reasoning; after doing a UI task manually, save it as a macro so the next run is fast.…
ios-test
Build, launch, and visually test iOS/SwiftUI apps in the Simulator using computer use. Automated screen navigation, crash log analysis, state testing (empty/error/loading), and memory leak detection. Use when you need to test an iOS app, run it in the Simulator, check for crashes, or verify UI flows.
add-accessibility
Scan SwiftUI views and add missing accessibility identifiers using a consistent {screen}-{type}-{name} naming convention. Also flags Dynamic Type compatibility issues. Use when you need to add accessibility support, make views testable, add VoiceOver labels, or prepare for UI testing.
swiftui-autotest-skill
AI-powered visual testing and accessibility setup for iOS/SwiftUI apps. Build, launch in Simulator, test with computer use, detect crashes, analyze memory leaks, and add accessibility identifiers — all from the terminal.
scroll-craft
Build premium scroll-driven landing pages for service, product, food, and drink brands. Plan the visitor journey, page grammar, emotional peak, and bespoke signature move. Create dimensional heroes with independent visual planes, restrained motion, and separate mobile composition. Use supplied photos and footage or…
macos-harness
Read and operate the apps on this Mac from the shell. Use it when a task needs an app's UI and no API, CLI or purpose-built tool can do it.