The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides configuration instructions for AI Gateway features, including ai-proxy, ai-agent, and ai-rag. These components ingest and process untrusted external user input to interface with Large Language Models, creating a surface for indirect prompt injection.
Ingestion points: User prompts and requests processed through the WasmPlugin (e.g., in SKILL.md).
Boundary markers: Not explicitly defined in the configuration templates, though the gateway model inherently separates control and data planes.
Capability inventory: The configured plugins can perform complex operations including network proxying, data transformation, and agentic reasoning.
Sanitization: The skill documents specific mitigation tools such as ai-security-guard and ai-data-masking to filter and sanitize content.
[EXTERNAL_DOWNLOADS]: The documentation includes references to Wasm plugins hosted on an external OCI registry at higress-registry.cn-hangzhou.cr.aliyuncs.com. These plugins are downloaded and loaded into the gateway's memory at runtime.
[DYNAMIC_EXECUTION]: The skill describes the lifecycle of WasmPlugin resources which involve fetching and executing binary code from remote OCI images. This is a core feature of the Higress gateway architecture for extending Envoy functionality.

higress-operator