Skills
skills/agentspace-so/runcomfy-agent-skills/nano-banana-edit/Gen Agent Trust Hub

nano-banana-edit

Pass

Audited by Gen Agent Trust Hub on Apr 30, 2026

Risk Level: SAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external user-supplied strings and image URLs which are eventually passed to a generative model.
  • Ingestion points: User-provided prompt and image_urls array defined in the schema in SKILL.md.
  • Boundary markers: The prompt is structured within a JSON payload to prevent local shell expansion during command execution.
  • Capability inventory: The skill executes the runcomfy CLI and interacts with the vendor's model API.
  • Sanitization: The skill documentation explicitly warns about image-based prompt injection risks and confirms the CLI transmits the JSON body directly to the API without shell interpolation.
  • [COMMAND_EXECUTION]: The skill utilizes a CLI tool to communicate with a remote image processing service.
  • Evidence: Orchestrates local execution of runcomfy run google/nano-banana-2/edit to perform image edits.
  • [EXTERNAL_DOWNLOADS]: The skill depends on a specific vendor-provided command-line utility for its functionality.
  • Evidence: Directs the user to install @runcomfy/cli via npm as a prerequisite.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 30, 2026, 06:38 AM