More screenshots
Works with the AI Agents you already use
Limit What Your Desktop Agent Can Open
A permission list for agents that click and type on a real desktop — anything not on the list can't be opened.
$12
Limit What Your Desktop Agent Can Open
Example session with this skill installed
Use the desktop-agent-safety skill to review my desktop agent setup before I let it loose: check the allowlist, test the denials, and tell me what's still unsafe.
- Read your context and instructions
- Compiled the limit what your
Safety review: 2 issues found, 1 blocked by design
Allowlist check: 3 apps allowlisted (MyApp, Notes, Calculator). Default deny is on — anything not named is refused.
Denial tests (all passed):
- Password manager launch: REFUSED outright
- Credential store access: REFUSED outright
- System settings: REFUSED outright
- Browser with a signed-in profile: REFUSED — only throwaway profiles allowed
Re-check at every input step: An allowed app was asked to open a second, non-allowlisted app as a trampoline. The harness re-checked the allowlist at that input step and refused it.
Issues to fix before running:
- One allowlisted app had broad file access — narrow its scope.
- The operator checklist was not signed off — run it and record the answers to the five review questions.
Verdict: Do not run the agent unattended until both issues are fixed. The threat model in this skill lists exactly what the allowlist does and does not protect against.
Connects securely to your tools. The creator never sees your data.
About this skill
An agent that can click and type across your whole desktop can open your mail, your password manager, and a browser where you're already signed in. "The agent promised to be careful" is not a control. This skill is a permission model enforced by the software around the agent, not by the prompt: a list of named apps the agent may open, written by you, with everything else denied by default.
Inside the package: the skill file with the full model, a plain-language threat model that says honestly what this does and does not protect against, an operator checklist, a security-model reference, and a working example config from deskcheck-ai.
A run goes like this: you list only the apps the work truly needs. Permission is checked when the agent launches an app and again at every click and keystroke, so an allowed app can't be used as a doorway into a denied one. Password managers, credential stores, and system-settings apps are refused even if someone puts them on the list. Browsers may only run with a fresh, throwaway profile — never the one carrying your real logins. A denied attempt fails loudly and gets logged, so the agent can't quietly route around it.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Security.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 6 days ago
- Software
- deskcheck-ai (free, MIT)