The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents a system where user-supplied parameters (defined in lzc-deploy-params.yml) are interpolated into a Go template (lzc-manifest.yml) using the .U prefix.
Ingestion points: Parameters are collected via a UI defined in lzc-deploy-params.yml (e.g., target_ip).
Boundary markers: No boundary markers or specific instructions for escaping user-supplied strings are provided in the template examples.
Capability inventory: The resulting manifest controls backend_launch_command, environment variables, and filesystem binds.
Sanitization: There is no evidence of sanitization or validation of the user-provided parameters before they are used in shell-equivalent commands.
[COMMAND_EXECUTION]: The skill provides examples of using user-defined parameters directly within the backend_launch_command field of lzc-manifest.yml (e.g., backend_launch_command: /lzcapp/pkg/content/netmap -target={{ .U.target }}). This demonstrates a platform capability that can lead to command injection if the underlying system does not properly sanitize the interpolated template values.
[DYNAMIC_EXECUTION]: The application.injects feature allows the injection of JavaScript into arbitrary web pages matching specific URL patterns. While the skill suggests using built-in scripts for tasks like auto-filling passwords, it also documents support for loading external scripts via http(s):// and local files via file:///, which is a vector for code execution within the user's browser context.

lazycat-dynamic-deploy