The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references an external installation source in references/setup.md (https://github.com/gemini-cli-extensions/nanobanana). It instructs the agent to run gemini extensions install, which downloads and executes code from a repository that is not part of a known or verified organization.
[COMMAND_EXECUTION]: The skill workflow in SKILL.md and references/setup.md explicitly instructs the use of the --yolo flag. This flag is intended to enable non-interactive execution by suppressing safety confirmations, which reduces user oversight for shell commands.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it interpolates user-controlled text directly into shell commands (e.g., gemini --yolo "/generate 'prompt'").
Ingestion points: User requests for image generation, edits, and styles are processed in the initial workflow step of SKILL.md.
Boundary markers: The skill does not implement boundary markers or instructions to ignore embedded commands within the prompt text.
Capability inventory: The agent is granted the capability to execute shell commands using the gemini CLI tool, as seen in SKILL.md and references/setup.md.
Sanitization: There is no evidence of input validation or escaping for shell metacharacters (such as single quotes or backticks) that could be used to break out of the command string.

nano-banana