dev-swarm-use-computer
Pass
Audited by Gen Agent Trust Hub on Mar 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes a local Python script
use_computer.pyusing theuvtool. This script processes JSON commands to perform system-level actions like moving the mouse and typing text. - [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8).
- Ingestion points: Visual data from the user's screen captured via the
screenshotaction (SKILL.md). - Boundary markers: Absent. The skill lacks delimiters or instructions to prevent the agent from obeying text found within screenshots.
- Capability inventory: Full GUI interaction capabilities including
mouse_move,click,type, andhotkey(SKILL.md). - Sanitization: Absent. There is no mention of filtering or validating screen content before the agent processes it.
Audit Metadata