qianwenai-operate

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill provides instructions and commands to download the official Alibaba Cloud Command Line Interface (aliyun-cli) from a well-known service provider domain (aliyuncli.alicdn.com).
  • [COMMAND_EXECUTION]: The skill extensively uses the aliyun CLI to interact with cloud infrastructure, including ECS instances, RDS databases, and VPC network settings. This is the primary function of the skill and is governed by strict user-confirmation requirements for any action that modifies resource states.
  • [DYNAMIC_EXECUTION]: The skill dynamically generates shell scripts (e.g., for service restarts, Nginx reloads, and disk cleanup) which are then executed on remote ECS instances via the Alibaba Cloud Assistant (RunCommand API). These scripts are constructed from predefined templates and encoded in Base64 for API compatibility.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted external data by reading application logs and command output during the diagnostic phase.
  • Ingestion points: System logs retrieved via journalctl and cloud assistant output retrieved via aliyun ecs DescribeInvocations in the references/diagnose.md and references/workflow.md files.
  • Boundary markers: No explicit delimiters are used when interpolating external data into the prompt context.
  • Capability inventory: The skill possesses the capability to modify cloud resource states (reboot, start, resize disks) and write local audit logs (operate_audit.jsonl).
  • Sanitization: The skill includes explicit developer instructions to desensitize and mask all evidence, ensuring no passwords, tokens, or full connection strings are displayed or logged.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 07:44 AM
Security Audit — agent-trust-hub — qianwenai-operate