AGENT LAB: SKILLS

drawio

Fail

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTION
Full Analysis
  • [Unverifiable Dependencies & Remote Code Execution] (HIGH): The skill configures an MCP server to run via npx --yes @next-ai-drawio/mcp-server@latest in .mcp.json. This downloads and executes arbitrary code from the npm registry without source verification or version pinning. Neither the package nor its author are on the trusted list.\n- [Data Exposure & Exfiltration] (SAFE): Analysis of the provided source files shows no evidence of hardcoded credentials or unauthorized data exfiltration logic.\n- [Indirect Prompt Injection] (LOW): The skill has an attack surface for indirect prompt injection as it processes user-provided natural language into structured diagram XML.\n
  • Ingestion points: Natural language descriptions and image metadata processed via the /drawio commands.\n
  • Boundary markers: Not present in the conversion logic; the system relies on regex parsing.\n
  • Capability inventory: The MCP server can open local ports, control browser instances, and export files to the local filesystem.\n
  • Sanitization: Node labels and formulas are escaped using XML entities in src/math/index.js.\n- [Prompt Injection] (SAFE): No malicious instruction overrides or bypass attempts were found in the workflow documentation or templates.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 17, 2026, 04:43 PM