google-maps-platform

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 1.00). The file contains an explicit "⚠️ OVERRIDE" that instructs agents to ignore the MCP server's "call me first" instruction and forbids calling the retrieve-instructions/getInstructions tool, which is a direct attempt to override/contradict external tool/system directives (a classic prompt-injection).

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

• Potentially malicious external URL detected (high risk: 0.90). This skill mandates fetching remote skill/index content at runtime (primary URL https://www.gstatic.com/googlemapsplatform-agent-skills/index.json) and shows a runtime REST call to the MCP ('https://mapscodeassist.googleapis.com/v1:retrieveContexts') whose returned content would directly control agent prompts/instructions, making these runtime dependencies high-risk.