The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references the google-genai Python library for image generation tasks.
Evidence: Found in references/api-integration.md. This is a well-known official library from Google used for interacting with the Gemini API.
[PROMPT_INJECTION]: The skill processes user-provided descriptions and constraints to generate code, presenting an indirect prompt injection surface.
Ingestion points: SKILL.md indicates the skill parses description, style, and constraints from _AGENT_CONTEXT to generate Python deliverables.
Boundary markers: The skill employs a structured prompt architecture ([Subject] + [Style] + [Composition] + [Technical]) as defined in references/prompt-patterns.md to organize user inputs.
Capability inventory: The skill generates Python scripts using the google-genai SDK which performs network requests to Google's API.
Sanitization: Implements mandatory Japanese-to-English translation and policy-safe rewriting rules to sanitize user-provided descriptions before they are embedded in the generated code.
[COMMAND_EXECUTION]: The skill's primary function is to generate code for the user to execute locally.
Evidence: SKILL.md explicitly states it delivers code and does not execute the API calls directly, reducing the risk of unintended runtime behavior within the agent environment.
[CREDENTIALS_UNSAFE]: The skill documentation provides guidance on secure credential handling.
Evidence: Instructions in SKILL.md and references/api-integration.md require reading API keys from environment variables (os.environ["GEMINI_API_KEY"]) and advise using .gitignore to prevent .env files from being committed.

Sketch