lov-nano-banana-pro
Pass
Audited by Gen Agent Trust Hub on Oct 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes bundled Python scripts (
scripts/profile_store.pyandscripts/validate_skill.py) to manage user profile data and validate skill integrity. These scripts are executed locally via subprocess calls to handle persistent state and development checks. - [INDIRECT_PROMPT_INJECTION]: The skill's primary function involves processing untrusted image descriptions and reference material requirements to generate structured prompts. This creates an attack surface where instructions embedded in user-provided data could potentially influence the agent's output.
- Ingestion points: Image requirements and reference materials (documented in SKILL.md).
- Boundary markers: Not explicitly defined to delimit or ignore instructions within input data.
- Capability inventory: Includes local script execution and atomic file read/write operations for profile management.
- Sanitization: Input data is processed directly for prompt extraction without explicit sanitization mechanisms.
- [SAFE]: The profile management script implements a security blacklist (
SENSITIVE_PARTS) to prevent the inadvertent storage of sensitive information, such as tokens, secrets, or passwords, in the persistent user profile store.
Audit Metadata