The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's environment validation script (scripts/validate-cli.sh) provides instructions to download and execute the Alibaba Cloud CLI installation script from https://aliyuncli.alicdn.com/setup.sh via a pipe to bash. This resource is provided by the official vendor and is intended for necessary tooling setup.
[COMMAND_EXECUTION]: The skill extensively uses the aliyun CLI tool across multiple scripts (fw-switch.sh, auto-protect.sh, validate-cli.sh) to perform cloud management operations. It also invokes python3 via command substitution in scripts/auto-protect.sh to perform JSON parsing and configuration merging for the firewall's auto-protection settings.
[INDIRECT_PROMPT_INJECTION]: The skill defines a multi-step workflow in SKILL.md where the agent is instructed to query assets and then filter them based on user-provided conditions, such as matching instance names (e.g., 'instances whose name contains Dify'). Because instance names and other resource metadata are externally controlled, they could potentially contain malicious instructions intended to influence the agent's behavior during the filtering or preview phases.
Ingestion points: Resource metadata fields including Name and BindInstanceName returned by the DescribeAssetList API via scripts/fw-switch.sh query (referenced in SKILL.md).
Boundary markers: None explicitly defined in the instructions for the filtering step.
Capability inventory: Execution of CLI management tools with write permissions, file system access, and network operations through the Cloud Firewall API.
Sanitization: While the shell scripts include validation for specific fields like IP addresses and regions, the agent-side natural language filtering of descriptive metadata fields lacks explicit sanitization or instruction boundary protection.

alibabacloud-cfw-internet-firewall-protect