The Agent Skills Directory

[SAFE]: The skill is developed by 'aliyun' and all external resources, including the Alibaba Cloud CLI and official GitHub repositories, are vendor-owned or well-known service endpoints. These are treated as safe under the trust scope rules for vendor resources.
[COMMAND_EXECUTION]: The skill executes Alibaba Cloud CLI commands through a sandboxed MCP tool (CallCLI) to provision cloud resources such as ECS instances, RDS databases, and VPC networks. These actions are the primary intended functionality of the skill and are guarded by multiple confirmation gates.
[EXTERNAL_DOWNLOADS]: The skill automatically installs the Alibaba Cloud CLI from aliyuncli.alicdn.com. This is an official vendor source, and the installation process includes SHA256 integrity checks or TOFU (Trust On First Use) verification to ensure secure delivery.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it processes user input to drive powerful cloud management capabilities.
Ingestion points: User chat messages are ingested to determine SKU selection and provide payment confirmation.
Boundary markers: The skill employs hard gates, such as the PAYMENT GATE and SKU GATE, which require explicit, affirmative user responses before proceeding with resource creation or charging.
Capability inventory: The skill can create, manage, and delete a wide range of cloud resources (VPC, ECS, RDS, OSS, ALB, ESS) via CLI commands.
Sanitization: User-supplied SKU names are validated against a strict list of seven supported tokens. CLI commands are generated using fixed YAML templates, limiting the impact of malicious parameter injection.
[CREDENTIALS_SAFE]: The skill strictly follows security best practices for credential management. It uses RamRoleArn for temporary, least-privilege access, never reads or echoes plaintext secrets (AK/SK), and enforces chmod 600 permissions on local configuration and private key files.

alibabacloud-opc-deploy