operate-on-aws
Pass
Audited by Gen Agent Trust Hub on Nov 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- AWS Infrastructure and IAM Management: The skill is designed to perform high-privilege operations within the user's AWS environment, including the creation of IAM roles (e.g., DevOpsAgentRole-AgentSpace) and service associations. These actions are governed by a mandatory 'CONFIRM' phase, which requires an explicit 'Yes' from the user after presenting a notice detailing the specific permissions and potential costs. The instructions also strictly forbid enabling elevated remediation actions by default, adhering to the principle of least privilege.- Indirect Prompt Injection Surface: The skill processes potentially untrusted external data, such as CloudWatch logs and source code from pull requests. To address the risk of the agent interpreting this data as instructions, the skill includes explicit safety directives to treat all findings as informational and never as commands to be executed.- External Utility Execution: The skill utilizes
uvxto runmcp-proxy-for-aws, which is a standard utility provided by the vendor to facilitate SigV4-authenticated connections to the AWS DevOps Agent MCP endpoint. This is documented as a requirement for skill functionality and leverages established vendor tools.- Consent-Based Telemetry Subsystem: A telemetry system is included to collect skill usage metrics. This system implements a robust opt-in mechanism requiring users to acknowledge a disclosure notice before any data is transmitted. The data collected (e.g., skill name, Agent Space ID) is scoped to operational metrics and is sent to an official vendor domain.- PII Handling and Disclosure: The skill includes proactive warnings to the user that investigations do not automatically filter Personally Identifiable Information (PII) from logs. This transparency allows users to make informed decisions about redacting sensitive data before initiating investigations.
Audit Metadata