The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The documentation for the application.injects feature explicitly supports loading and executing JavaScript from remote http(s):// URLs. Although recommended only for debugging purposes in the documentation, this capability allows for the execution of unverified remote code.
[COMMAND_EXECUTION]: The skill provides examples of using Go templates to interpolate user-provided deployment parameters directly into shell commands via the backend_launch_command field in the application manifest. This pattern is vulnerable to command injection if the underlying platform does not sanitize the interpolated strings before execution.
[DYNAMIC_EXECUTION]: The platform utilizes Go's text/template engine to dynamically render the lzc-manifest.yml configuration at deployment time, allowing for runtime generation of environment variables, commands, and script parameters based on external inputs.
[INDIRECT_PROMPT_INJECTION]: The skill documents an architecture where untrusted data provided by users during deployment (lzc-deploy-params.yml) is ingested and processed by the agent/platform to configure the application. 1. Ingestion points: Deployment parameters defined in lzc-deploy-params.yml and accessed via the .U template variable. 2. Boundary markers: The provided documentation and examples do not include explicit boundary markers or instructions for the agent to ignore potentially malicious content within these parameters. 3. Capability inventory: The platform supports shell command execution, file system operations, and front-end script injection. 4. Sanitization: There is a lack of documented sanitization or validation logic to ensure that user-provided parameters do not contain malicious payloads intended to exploit the manifest rendering or script injection processes.

lazycat-dynamic-deploy