The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses npx -p @vm0/cli to download and execute the vendor's command-line interface from the NPM registry. This is the primary mechanism for the skill's functionality.
[COMMAND_EXECUTION]: The skill performs multiple shell commands through the CLI to manage desktop state, including listing running applications, opening new app instances, and simulating user input (clicks, typing, key presses).
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it processes untrusted data from external application accessibility trees and screenshots.
Ingestion points: Reads application state files (appState) and screenshots from potentially untrusted desktop applications like browsers or Slack.
Boundary markers: None present; the agent parses raw text and JSON filtered by rg or node -e.
Capability inventory: Full GUI control including click, type-text, set-value, and open-app.
Sanitization: Uses filtering tools like rg and node -e to locate specific elements, but does not implement validation to prevent the execution of instructions embedded in the UI data.

computer-use