lov-nano-banana-pro
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-suppliedRequirements and reference image descriptions to generate output, creating a potential surface for indirect prompt injection.\n
- Ingestion points: Natural language input triggers defined in SKILL.md.\n
- Boundary markers: The skill lacks explicit boundary markers or instructions to the agent to disregard embedded instructions within user-provided requirements.\n
- Capability inventory: The skill possesses the capability to access the host file system, execute CLI commands, and perform browser operations as specified in the SKILL.md execution boundary.\n
- Sanitization: The profile utility in scripts/profile_store.py sanitizes keys to prevent secret storage, but no explicit sanitization is applied to the input content before interpolation into prompts.\n- [SAFE]: The profile management script scripts/profile_store.py includes proactive security controls that explicitly block the persistence of sensitive fields such as tokens, secrets, and passwords within the user profile.\n- [SAFE]: The skill follows security best practices for resource management by using safe configuration parsing (yaml.safe_load) and enforcing strict file permissions (0o600) for local data storage.
Audit Metadata