The Agent Skills Directory

[SAFE]: The execution script scripts/edit.ts uses Bun to perform image editing tasks through the Gemini API. It implements logical fallbacks to locate its shared utilities within the local environment without executing remote code.
[SAFE]: File system operations are limited to reading input images, reading local configuration files for installation path discovery, and writing output images to the specified directory.
[SAFE]: The skill follows context discipline by providing file paths to the user instead of reading generated image data back into the LLM context, preventing potential context window exhaustion.
[SAFE]: Although the skill processes untrusted user prompts and image data (Indirect Prompt Injection surface), this is central to its functionality and handled by the underlying model's safety guardrails.
Ingestion points: Input image and prompt arguments in scripts/edit.ts.
Boundary markers: None; the prompt is passed directly to the model.
Capability inventory: Image generation via the Gemini API.
Sanitization: Standard prompt interpolation without specific sanitization.

edit-image