skills/rheinmir/setup/computer-use/Gen Agent Trust Hub

computer-use

Pass

Audited by Gen Agent Trust Hub on Jun 26, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses the orca CLI to list applications, capture screenshots, and perform UI actions such as clicking, typing, and hotkey execution. This command execution is the primary intended function of the skill for desktop automation.
  • [DATA_EXFILTRATION]: The skill has access to sensitive information displayed in application windows (e.g., Slack messages, Gmail content) via UI trees and screenshots. The instructions include explicit safeguards directing the agent not to expose secrets or perform sensitive actions (like financial transactions or data deletion) without direct user consent.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted text and elements from third-party application windows.
  • Ingestion points: Accessibility trees and screenshots retrieved from running applications via the orca computer get-app-state command (SKILL.md).
  • Boundary markers: The skill does not define specific boundary markers to distinguish between system instructions and data retrieved from application UIs.
  • Capability inventory: Full UI control including clicking, typing, and hotkey execution (SKILL.md).
  • Sanitization: No sanitization or content filtering is performed on the UI text or elements retrieved by the CLI tool.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 26, 2026, 09:59 AM
Security Audit — agent-trust-hub — computer-use