pixel-dissolve-landscape
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-uploaded images and explicitly instructs the agent to extract and output text from them, which creates a surface for indirect prompt injection.
- Ingestion points: The instructions in 'SKILL.md' require the agent to use 'uploaded travel photos' as a reference and to 'write the exact English text' based on visible facts in the source.
- Boundary markers: The skill lacks instructions or delimiters to ignore or specifically handle potentially malicious text or instructions embedded within the source images.
- Capability inventory: The skill utilizes the agent's image generation and editing tools. It does not involve file system modifications, network requests, or shell command execution.
- Sanitization: There is no explicit validation or sanitization of the content extracted from the images; the skill relies on the underlying vision-language model's internal safety guardrails.
Audit Metadata