The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the infsh CLI tool to perform image generation and management tasks.\n
Evidence: Use of infsh login, infsh app run, and infsh app list throughout the SKILL.md documentation.\n
Context: Command execution is strictly limited to the infsh prefix via the allowed-tools configuration.\n- [EXTERNAL_DOWNLOADS]: The documentation suggests installing additional tools and related skills via npx.\n
Evidence: Command npx skills add inference-sh/skills@agent-tools found in SKILL.md.\n
Context: These downloads target the official and verified vendor namespace ('inference-sh-9') and represent standard installation procedures.\n- [PROMPT_INJECTION]: The skill handles natural language prompts as input for AI models, creating a potential surface for indirect prompt injection.\n
Ingestion points: The prompt and images fields within the JSON input string for the infsh command (SKILL.md).\n
Boundary markers: Inputs are encapsulated within a structured JSON object, which helps separate user data from command-line instructions.\n
Capability inventory: The skill's capabilities are limited to executing the infsh binary via Bash.\n
Sanitization: Input validation and safety filtering are managed by the underlying inference.sh platform and the Google Gemini models.

nano-banana