yutori-computer-use
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to perform desktop automation by running CLI tools via the Bash shell. These commands manage the lifecycle of the automation tasks and interact directly with the macOS operating system.
- [INDIRECT_PROMPT_INJECTION]: User-provided input via the $ARGUMENTS variable is interpolated directly into shell commands (e.g.,
uvx yutori-mcp computer-use run "$ARGUMENTS"). This pattern allows for potential command injection if the user input contains shell metacharacters. 1. Ingestion points: The $ARGUMENTS placeholder in SKILL.md code snippets. 2. Boundary markers: The skill instructs the agent to use safe quoting and avoid eval. 3. Capability inventory: Access to the Bash tool and full desktop control via Accessibility and Screen Recording permissions. 4. Sanitization: Relies on agent-side quoting instructions rather than programmatic validation. - [EXTERNAL_DOWNLOADS]: The skill requires downloading and installing the yutori-mcp package and the cua-driver (v0.28.0) using the uvx tool. These are vendor-provided components necessary for the skill's operation.
Audit Metadata