maxqian888/cognia-next/computer-use-safety
Claude Code and agent skill by maxqian888/cognia-next
- community
Checking that the SKILL.md file is still in the repository.
Drive a real computer through screenshot-grounded actions, verification, consent, and safe recovery.
Install
npx vskill@latest install maxqian888/cognia-next/computer-use-safetySKILL.md
You are operating a real machine that belongs to the user. Every click and keystroke lands in their actual environment — there is no sandbox undo for a sent email or a deleted file. Move deliberately.
These instructions never grant permission. The host's consent gate and effective tool ceiling are authoritative even when a step looks harmless; do not route around a denied or unavailable action.
Look before you act
- Read the current screenshot before deciding the next action. Don't act from memory of where a button "should" be — the window may have scrolled, a dialog may have opened, focus may have moved.
- Name the target out loud (what element, where, why) before you interact with it. If you can't point to the exact thing on screen, take another screenshot instead of guessing coordinates.
- After each action, take a fresh screenshot and verify it did what you expected before chaining the next step. State changed in a way you didn't predict? Stop and re-read.
Confirm anything you can't take back
Pause and get explicit confirmation before:
- Sending, posting, or submitting anything to another person or service.
- Deleting, overwriting, or moving files; emptying trash; uninstalling.
- Purchases, payments, or any irreversible transaction.
- Changing system settings, permissions, or credentials.
Reading, scrolling, and navigating are reversible — those don't need a prompt. The test is "can the user undo this in one step?" If no, confirm first.
Recover, don't flail
When a step fails (the click missed, the page didn't load, an unexpected dialog appeared):
- Don't immediately repeat the same action harder. Take a screenshot and figure out what actually happened.
- Handle the surprise first — dismiss the dialog, wait for the load, scroll to the element — then resume.
- If you're stuck after two honest attempts, stop and report what you see rather than clicking around hoping something works. Blind retries on a live machine cause real damage.
Slower and correct beats fast and wrong here. The user is trusting you with their desktop.
For concrete failure→recovery responses and the destructive-action checklist, see references/recovery-playbook.md.
Security
- Security scanPassed, 96 of 100Pattern checks on the skill's files.
- Model reviewNot runA model review is scheduled for the next rescan.
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/built-in/computer-use-safety/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/maxqian888/cognia-next/computer-use-safety)