nano-banana
Warn
Audited by Gen Agent Trust Hub on Apr 14, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references an external installation script at
https://raw.githubusercontent.com/inference-sh/skills/refs/heads/main/cli-install.mdfor the requiredinfshCLI tool. - [EXTERNAL_DOWNLOADS]: The skill instructions encourage the addition of further skills from the
inference-shGitHub organization using the commandnpx skills add, which involves fetching and potentially executing code from a remote repository. - [COMMAND_EXECUTION]: The skill relies on the execution of the
infshCLI tool through Bash to perform image generation and account management tasks (infsh login,infsh app run). - [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes user-supplied prompts and external image URLs without explicit sanitization or boundary markers before passing them to the underlying CLI tool.
- Ingestion points: User-provided
promptandimagesURL array in theinfsh app runcommand inputs. - Boundary markers: Absent; the user content is directly interpolated into a JSON string within a shell command.
- Capability inventory: Execution of the
infshCLI via Bash, including network interaction with theinference.shAPI. - Sanitization: No validation or sanitization of the input strings is evidenced in the skill instructions.
Audit Metadata