Back to skill

Security audit

AWS智能体编排专业版

Security checks for vulnerabilities and agentic risk

Overview

This AWS integration guide needs review because its ordering example lacks safeguards for creating real orders.

Test with a mock gateway and restricted AWS permissions before connecting production systems. Repair the examples and require confirmation of the environment, item, and quantity before creating real orders. Decide what conversation data may enter cloud memory and configure access, retention, and deletion controls. Missing references and broken examples limit confidence in production readiness.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)

The executable template sends user-message content to AWS-managed memory without a privacy warning.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

The quick-start template records user_msg as an event payload in AgentCore Memory. Although the document describes cross-session persistence, it does not warn that message content may leave the local environment and be retained in AWS-managed storage; the generic HTTPS and output-secret precautions do not address this transfer.

Content

No source excerpt is available for this finding.

The shopping workflow creates an order without warning about the resulting external state change.

Medium
Category
Output Handling
Confidence
92% confidence
Finding

The executable shopping example calls create_order with a fixed SKU and quantity, rather than merely listing an available tool. It provides no advance warning or confirmation that running the example against a connected production gateway can create a real order; reporting the order number afterward does not warn before the change.

Content

No source excerpt is available for this finding.