The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes subprocess and docker compose to orchestrate deployment and health checks. All system calls are constructed using argument lists, avoiding shell execution and associated injection risks. Potentially destructive operations are gated by explicit user confirmation requirements.
[EXTERNAL_DOWNLOADS]: Assets are retrieved from the vendor's official GitHub organization and well-known Docker image registries. The skill follows safe patterns for downloading and verifying model data without executing unverified remote scripts.
[INDIRECT_PROMPT_INJECTION]: User-provided inputs, such as RTSP stream URLs and camera identifiers, undergo validation before use. The skill employs regex and schema-based checks to ensure that untrusted data does not introduce malicious configurations into the system.
[SAFE]: Secret management is handled locally, with credentials and certificates generated at runtime and stored using restricted filesystem permissions (mode 0600). This approach prevents credential exposure while maintaining service functionality.

scenescape-setup