View a markdown version of this page

策略创建和更新:每个策略的引擎验证 - Amazon Bedrock AgentCore
Amazon Web Services 文档中描述的 Amazon Web Services 服务或功能可能因区域而异。要查看适用于中国区域的差异,请参阅 中国的 Amazon Web Services 服务入门 (PDF)。

本文属于机器翻译版本。若本译文内容与英语原文存在差异,则一律以英文原文为准。

策略创建和更新:每个策略的引擎验证

当直接创建或更新策略(而不是通过生成)时,验证和分析会考虑新策略及其与策略引擎中所有先前存在的策略的交互。

工作原理

  1. 对于与策略引擎关联的所有网关,将根据 Cedar 架构检查该策略。无论验证模式如何,架构检查始终运行。

  2. 如果将验证模式设置为FAIL_ON_ANY_FINDINGS,则语义验证将在整个策略引擎的上下文中运行,检查是否存在过于宽松、过于严格和无效的策略。如果架构检查或语义验证得出结果,则策略将被拒绝。有关每项检查的详细信息,请参阅验证和分析概述。

注意

使用IGNORE_ALL_FINDINGS时,仅运行架构检查。只要策略通过架构检查,就可以被接受。有关更多信息,请参阅向策略引擎添加策略。

示例:创建带有验证的策略

创建经过严格验证的策略,拒绝包含任何发现的策略:

例
AgentCore CLI
  1. 将语句保存到文件中,例如restrict_refunds.cedar:

    forbid ( principal, action == AgentCore::Action::"RefundTool___process_refund", resource == AgentCore::Gateway::"arn:aws:bedrock-agentcore:us-west-2:123456789012:gateway/my-gateway-a1b2c3d4e5" ) when { context.input.amount > 1000 };
  2. 添加策略并部署:

    agentcore add policy --name RestrictRefunds \ --engine MyEngine \ --source restrict_refunds.cedar \ --validation-mode FAIL_ON_ANY_FINDINGS agentcore deploy --yes
Amazon CLI
  1. 在终端中运行以下代码,使用 Amazon CLI 创建策略:

    aws bedrock-agentcore-control create-policy \ --policy-engine-id MyEngine-abc123defg \ --name RestrictRefunds \ --validation-mode FAIL_ON_ANY_FINDINGS \ --definition '{ "policy": { "statement": "forbid (principal, action == AgentCore::Action::\"RefundTool___process_refund\", resource == AgentCore::Gateway::\"arn:aws:bedrock-agentcore:us-west-2:123456789012:gateway/my-gateway-a1b2c3d4e5\") when { context.input.amount > 1000 };" } }'
Amazon Python SDK (Boto3)
  1. 运行以下代码,使用适用于 Python 的 Amazon 软件开发工具包创建策略:

    import boto3 client = boto3.client('bedrock-agentcore-control') GATEWAY_ARN = ( 'arn:aws:bedrock-agentcore:us-west-2:123456789012:gateway/my-gateway-a1b2c3d4e5' ) statement = ( 'forbid (principal, ' 'action == AgentCore::Action::"RefundTool___process_refund", ' f'resource == AgentCore::Gateway::"{GATEWAY_ARN}") ' 'when { context.input.amount > 1000 };' ) response = client.create_policy( policyEngineId='MyEngine-abc123defg', name='RestrictRefunds', validationMode='FAIL_ON_ANY_FINDINGS', definition={'policy': {'statement': statement}}, ) print(response['policyId'], response['status'])

响应表示策略正在创建中:

{ "policyId": "RestrictRefunds-ghi789jklm", "status": "CREATING" }

检查策略状态以确认验证通过:

例
AgentCore CLI
  1. 显示项目中资源的部署状态:

    agentcore status

    C AgentCore LI 没有按策略执行读取操作。使用 C Amazon LI 或 SDK 检查单个策略的状态和发现。

Amazon CLI
  1. 在终端中运行以下代码以使用 Amazon CLI 获取策略:

    aws bedrock-agentcore-control get-policy \ --policy-engine-id MyEngine-abc123defg \ --policy-id RestrictRefunds-ghi789jklm
Amazon Python SDK (Boto3)
  1. 使用适用于 Python 的 Amazon SDK 运行以下代码,等待策略生效:

    import boto3 client = boto3.client('bedrock-agentcore-control') client.get_waiter('policy_active').wait( policyEngineId='MyEngine-abc123defg', policyId='RestrictRefunds-ghi789jklm', ) policy = client.get_policy( policyEngineId='MyEngine-abc123defg', policyId='RestrictRefunds-ghi789jklm', ) print(policy['status'], policy.get('statusReasons'))

验证通过后,策略将变为活动状态:

{ "policyId": "RestrictRefunds-ghi789jklm", "status": "ACTIVE", "statusReasons": [] }

示例:验证失败

如果策略引用的操作不存在于任何关联网关的架构中,则验证失败。请注意,此失败是异步的:create 调用返回 HTTP 202CREATING,然后策略设置为CREATE_FAILED。

例
AgentCore CLI
  1. 验证失败的策略会使整个部署失败,而不仅仅是该资源:

    agentcore add policy --name InvalidPolicy \ --engine MyEngine \ --statement 'permit (principal, action == AgentCore::Action::"RefundTool___nonExistentTool", resource == AgentCore::Gateway::"arn:aws:bedrock-agentcore:us-west-2:123456789012:gateway/my-gateway-a1b2c3d4e5");' agentcore deploy --yes

    部署报告了根本CREATE_FAILED原因。使用删除错误的策略,agentcore remove policy --engine MyEngine --name InvalidPolicy然后重新部署。

Amazon CLI
  1. 在终端中运行以下代码,创建引用未定义操作的策略:

    aws bedrock-agentcore-control create-policy \ --policy-engine-id MyEngine-abc123defg \ --name InvalidPolicy \ --validation-mode FAIL_ON_ANY_FINDINGS \ --definition '{ "policy": { "statement": "permit (principal, action == AgentCore::Action::\"RefundTool___nonExistentTool\", resource == AgentCore::Gateway::\"arn:aws:bedrock-agentcore:us-west-2:123456789012:gateway/my-gateway-a1b2c3d4e5\");" } }'
  2. 检查状态以查看验证失败:

    aws bedrock-agentcore-control get-policy \ --policy-engine-id MyEngine-abc123defg \ --policy-id InvalidPolicy-jkl012mnop
Amazon Python SDK (Boto3)
  1. 使用适用于 Python 的 Amazon SDK 运行以下代码来创建策略并读取失败原因:

    import time import boto3 client = boto3.client('bedrock-agentcore-control') GATEWAY_ARN = ( 'arn:aws:bedrock-agentcore:us-west-2:123456789012:gateway/my-gateway-a1b2c3d4e5' ) response = client.create_policy( policyEngineId='MyEngine-abc123defg', name='InvalidPolicy', validationMode='FAIL_ON_ANY_FINDINGS', definition={'policy': {'statement': ( 'permit (principal, ' 'action == AgentCore::Action::"RefundTool___nonExistentTool", ' f'resource == AgentCore::Gateway::"{GATEWAY_ARN}");' )}}, ) # The 202 above is not success - poll for the terminal status while True: policy = client.get_policy( policyEngineId='MyEngine-abc123defg', policyId=response['policyId'], ) if policy['status'] != 'CREATING': break time.sleep(4) print(policy['status'], policy.get('statusReasons'))

当您检查策略状态时,响应会显示验证失败:

{ "policyId": "InvalidPolicy-jkl012mnop", "status": "CREATE_FAILED", "statusReasons": [ "Validation failed: Action 'nonExistentTool' is not defined in the schema for any associated gateway" ] }