The Agent Skills Directory

[DATA_EXFILTRATION]: The skill's primary function involves sending local project files and user-provided images to an external API endpoint at https://zenmux.ai/api/v1. While this is intended for design analysis, users should be aware that their source code and visual assets are being transmitted to a non-standard third-party proxy service.
[COMMAND_EXECUTION]: The SKILL.md instructions direct the agent to 'install it silently with the bundled scripts/install_cli script' if the command is missing. This bypasses user confirmation for local environment modifications, including creating symlinks in ~/.local/bin and writing configuration files to ~/.config/.
[COMMAND_EXECUTION]: The Python CLI tool (scripts/gemini-designer) executes external system utilities such as sips and cwebp via subprocess.run to process and resize images before transmission.
[CREDENTIALS_UNSAFE]: The skill manages a sensitive API key stored at ~/.config/gemini-designer/api_key. The CLI script reads this file to authenticate requests to the external proxy. While this is a functional requirement, the automated handling of credentials by an agent-controlled script represents an increased attack surface for credential theft.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. Since it reads and processes external HTML and CSS files provided as context for design advice, malicious instructions embedded in those files could influence the advice generated by the external LLM, which the agent is then instructed to 'apply' to the workspace.

gemini-designer