The Agent Skills Directory

[DATA_EXFILTRATION]: The skill transmits the contents of local files provided via the -f flag and images via the -i flag to the external endpoint https://zenmux.ai/api/v1. This behavior is the primary intended function of the tool to provide design context to the remote Grok model.
[COMMAND_EXECUTION]: The grok-designer CLI script executes local system utilities (sips on macOS and cwebp) to retrieve image dimensions and perform optimization. These subprocess calls are executed safely using argument lists to prevent shell command injection.
[PROMPT_INJECTION]: The skill provides a surface for indirect prompt injection by ingesting untrusted local project data which is then interpolated into prompts sent to the external design advisor.
Ingestion points: Local source files and screenshots processed by the grok-designer script in scripts/grok-designer.
Boundary markers: The script uses dedicated system prompts and structured templates to define the role and constraints of the advisor.
Capability inventory: Performs network requests via urllib and writes advisor output to the local filesystem.
Sanitization: Implements structural integrity checks on generated HTML and SVG outputs to prevent the inclusion of malicious file:// URLs, markdown fences, or placeholder text.

grok-designer