drawio

Pass

Audited by Gen Agent Trust Hub on Oct 30, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute the drawio command-line utility for diagram conversion and export tasks. It also invokes system commands such as node, open, xdg-open, and cmd.exe to manage files and launch web browsers across different operating systems including macOS, Linux, Windows, and WSL2.
  • [DATA_EXFILTRATION]: The skill includes functionality to generate a web URL that transmits the diagram's XML content (compressed and encoded) to the official app.diagrams.net web editor. This allows users to view and edit diagrams online but involves sending local data to an external service.
  • [DYNAMIC_EXECUTION]: The skill employs Node.js one-line scripts to handle data processing tasks, such as deflating and Base64-encoding diagram XML for URL generation, and creating Windows .url shortcut files in temporary directories.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied diagram descriptions to create files and run system commands, creating a surface for indirect prompt injection.
  • Ingestion points: Diagram definitions in Mermaid or XML format are authored by the agent based on user prompts in SKILL.md.
  • Boundary markers: There are no specific delimiters recommended to separate untrusted user input from agent instructions.
  • Capability inventory: The skill possesses capabilities for file system writes and execution of the drawio and node command-line tools.
  • Sanitization: The instructions mandate XML escaping for special characters within the generated diagram attributes to maintain well-formedness.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 30, 2026, 02:29 AM