The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/setup_mcp.py configures the Claude Code environment to run a remote package using npx -y @ycse/nanobanana-mcp. This allows for the download and execution of arbitrary code from the npm registry every time the MCP server is initialized.
[EXTERNAL_DOWNLOADS]: The skill's setup process fetches external code from the npm registry and directs users to install external binaries such as ImageMagick and FFmpeg for post-processing tasks.
[PROMPT_INJECTION]: The file references/prompt-engineering.md contains a section titled 'Safety Filter Rephrase Strategies' which provides the agent with specific patterns (abstraction, metaphor, artistic framing) to circumvent safety filters that block restricted content.
[COMMAND_EXECUTION]: The skill uses shell commands via npx for MCP execution and instructs the agent to use subprocesses for image manipulation tools like magick, convert, and ffmpeg.
[CREDENTIALS_UNSAFE]: The fallback scripts scripts/generate.py and scripts/edit.py transmit the Google AI API key as a query parameter in the URL string (?key={api_key}), which is a discouraged practice as it can lead to credential leakage in server logs and browser histories.

banana