The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs users to install MCP servers using npx -y nanobanana-mcp and npx nano-banana-mcp. These commands download packages from the public NPM registry.
[REMOTE_CODE_EXECUTION]: The installation process involves executing downloaded third-party code via the npx command. While the recommended package name aligns with the skill's identity, the alternative reference to ConechoAI introduces an external dependency from an untrusted third party.
[METADATA_POISONING]: The skill documentation includes misleading information, claiming to support 'Gemini 3.1' models and referencing a launch date of 'February 2026'. This metadata deception may lead to a misjudgment of the skill's actual capabilities or safety.
[INDIRECT_PROMPT_INJECTION]: The skill accepts natural language briefs from users which are used to generate parameters for image generation and editing tools.
Ingestion points: User briefs processed by the system prompt defined in references/translator-prompt.md.
Boundary markers: The instructions do not provide delimiters or negative constraints to prevent the agent from following instructions embedded in user-supplied data.
Capability inventory: The skill calls MCP tools to generate images and write them to the local file system (~/Documents/nanobanana_generated/).
Sanitization: The skill translates free-text user input into structured JSON schemas before tool execution, acting as a normalization layer.

nano-banana