The Agent Skills Directory

[COMMAND_EXECUTION]: The script invokes external utilities sops and magick (ImageMagick) using subprocess.run. All calls use argument lists rather than shell strings, effectively preventing command injection. The tool usage is limited to necessary tasks like secret decryption and image post-processing.
[CREDENTIALS_SAFE]: The skill implements professional secret management practices. It uses SOPS (Secrets Operation) with age encryption to store the Gemini API key in secrets.enc.yaml, ensuring sensitive credentials are not stored in plain text. An onboarding wizard (init) is provided to guide the user through secure setup.
[EXTERNAL_DOWNLOADS]: The skill communicates with Google's official Gemini API (generativelanguage.googleapis.com) to fulfill its primary purpose of image generation. These network operations are targeted at a well-known service and are restricted to sending prompts and receiving image data.
[INDIRECT_PROMPT_INJECTION]: While the skill ingests untrusted user input (text prompts and image files), the attack surface is minimal. The resulting output is confined to image files and JSON metadata sidecars, which do not contain instructions that could be executed by the agent in a feedback loop.

nano-banana