amazon-bedrock

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

• Potentially malicious external URL detected (high risk: 0.90). The AgentCore Harness fetches skills from external Git or S3 locations at runtime (e.g., https://github.com/example/skills-repo and s3://my-bucket/skills/my-skill/), and those fetched skills are injected as trusted input and may include scripts that the agent will execute or that control prompts, so this is a runtime external dependency that can execute remote code or directly control agent instructions.

MEDIUM W009: Direct money access capability detected (payment gateways, crypto, banking).

• Direct money access detected (high risk: 1.00). The skill explicitly includes an AgentCore "Payments" service that enables agents to pay via specific payment gateways: "Coinbase CDP, Stripe Privy" (microtransactions). This is a concrete, named payments integration (payment gateways) which grants agents the ability to execute financial transactions.