The Agent Skills Directory

[PERSISTENCE]: The setup documentation instructs the user to modify persistent shell profile files (~/.zshrc, ~/.bashrc) to store the GEMINI_API_KEY environment variable.
[COMMAND_EXECUTION]: The skill relies on Bash commands to perform file system operations, encode binary images into base64 format, and execute network requests to Google's API via curl.
[INDIRECT_PROMPT_INJECTION]: The skill exposes an attack surface for indirect prompt injection as it ingests untrusted user input and image files to construct API requests.
Ingestion points: The skill accepts user-provided prompts via the EDIT_PROMPT variable and local image files via IMG_PATH as described in SKILL.md.
Boundary markers: While the skill uses heredocs (JSONEOF) to structure the JSON payload, it does not include instructions for the model to ignore potentially malicious instructions embedded within the user-provided prompt or image metadata.
Capability inventory: The skill utilizes Bash (including curl, cat, and base64), Read, and Write tools to process data and interact with external services.
Sanitization: The skill performs basic MIME type detection based on file extensions but does not sanitize the text content of the prompt before interpolating it into the JSON request body.
[DYNAMIC_EXECUTION]: The skill uses python3 -c to execute inline Python scripts for parsing JSON API responses and decoding base64 image data into output files.

image-generator