The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill configures an MCP server by fetching the @ycse/nanobanana-mcp package from the NPM registry using npx during the setup process. NPM is a well-known service for package distribution.
[DATA_EXFILTRATION]: Fallback scripts (generate.py, edit.py) communicate directly with Google's official generative language API (generativelanguage.googleapis.com). This domain belongs to a well-known service provider and is used for the skill's primary function.
[COMMAND_EXECUTION]: The skill utilizes standard system-level command-line tools such as ImageMagick (magick/convert), FFmpeg, and Potrace for image post-processing tasks like resizing, background removal, and format conversion.
[COMMAND_EXECUTION]: The setup script (setup_mcp.py) modifies the local agent configuration (~/.claude/settings.json) to integrate the MCP server, which is a standard procedure for enabling third-party tool extensions.
[COMMAND_EXECUTION]: Local scripts manage data within the user's home directory, specifically for cost logging (~/.banana/costs.json), brand presets (~/.banana/presets/), and generated image storage (~/Documents/nanobanana_generated/). These operations are consistent with the skill's documented functionality.
[PROMPT_INJECTION]: The skill implements a 'Reasoning Brief' system that explicitly instructs the agent to analyze and wrap user input rather than passing it directly to the API, which serves as a mitigation against potential prompt injection attacks.

banana