The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs users to install and run external code packages (nanobanana-mcp and nano-banana-mcp) from the npm registry using npx. While these are central to the skill's function, they originate from unverified third-party sources and one is attributed to an entity ('ConechoAI') different from the skill author.
[COMMAND_EXECUTION]: The setup guide involves executing shell commands to add MCP servers to the agent's configuration, which involves downloading and running arbitrary code on the host machine via npx.
[INDIRECT_PROMPT_INJECTION]: The 'Translator Workflow' and the provided tools take natural language briefs and instructions from users to generate image prompts and structured JSON specifications. This creates a surface where malicious instructions hidden in the user-provided data could influence the agent's behavior.
Ingestion points: Natural language briefs, subject descriptions, and editing instructions processed by the agent in the Translator Workflow and gemini_edit_image tool.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present in the prompting guidelines.
Capability inventory: The skill utilizes image generation and editing tools (gemini_generate_image, gemini_edit_image) which have access to the local file system (via imagePath) and the Gemini API.
Sanitization: The instructions do not specify any validation or sanitization steps for the user input before it is used to construct final prompts or JSON schemas.

nano-banana