The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input via the prompt and images parameters within the belt app run command. While this establishes an injection surface, it is limited to the functionality of the image generation app and does not grant unauthorized access to the system or sensitive data. (Ingestion points: prompt and images fields in JSON input; Boundary markers: Absent; Capability inventory: Bash(belt *); Sanitization: Not explicitly implemented in prompt instructions).
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill references external resources including an installation script and related skills from inference.sh on GitHub and npx. These are documented as trusted references to well-known AI services and do not involve suspicious remote script piping or untrusted sources.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute belt CLI commands. This is the intended primary purpose of the skill and is appropriately scoped via the allowed-tools frontmatter.

nano-banana