The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill is designed to configure and execute arbitrary Lua code strings at runtime within the APISIX environment. The documentation describes how these strings are compiled and cached (LRU cache) upon route creation, which is a form of script generation and execution.
[INDIRECT_PROMPT_INJECTION]: The skill provides examples of processing untrusted external data (headers, URIs, and authentication tokens) directly within executable Lua functions, creating a vulnerability surface.
Ingestion points: HTTP request data accessed via ngx.var, ngx.req.get_headers(), and ngx.req.get_body_data() in SKILL.md.
Boundary markers: The examples lack explicit delimiters or validation logic to separate untrusted data from the Lua execution logic.
Capability inventory: The Lua environment has access to core ngx APIs, request modification capabilities, and external libraries like resty.http and cjson.
Sanitization: The provided examples do not demonstrate sanitization or escaping of external inputs before they are used in logic (e.g., string concatenation for redirects or header injection).
[COMMAND_EXECUTION]: The skill documentation encourages the execution of the a6 CLI tool to perform sensitive administrative operations, such as creating and updating routes with embedded code.
[CREDENTIALS_UNSAFE]: Example 8 ("Custom Authentication Guard") contains a hardcoded placeholder secret (secret123) within the Lua code logic. While intended as a placeholder, it demonstrates an unsafe pattern for managing authentication credentials.

a6-plugin-serverless