vercel-ai-sdk

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

• Potentially malicious external URL detected (high risk: 0.90). The skill spawns an external MCP server at runtime via "npx @picahq/mcp", which executes remote package code and exposes dynamic tool definitions that the agent ingests (directly controlling prompts/tool behavior), so this is a risky runtime dependency.