prompt-optimization

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The MCP integration example shows the skill fetching an external MCP resource at runtime (google_drive://Q4_Sales.xlsx) and loading contextual prompts/templates (MCP Prompt: sales_summary_template), which indicates external content could directly control agent instructions during skill execution.