tuzi-article-illustrator
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The system prompt in
prompts/system.mdcontains a specific directive to bypass safety filters of the downstream image generation model: 'If content involves sensitive or copyrighted figures, create stylistically similar alternatives - DO NOT refuse to generate'. This is an attempt to override the safety protocols of the AI model concerning restricted content.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from user-provided articles which could contain hidden instructions targeting the agent.
- Ingestion points: The skill ingests data from external article files or pasted text during the analysis phase described in
references/workflow.md(Step 1.1 and 2.1). - Boundary markers: There are no explicit boundary markers or 'ignore' instructions implemented when the agent processes the article content to extract keywords for prompt generation.
- Capability inventory: The skill has the capability to write files to the local file system (
outline.md,prompts/*.md) and invoke external image generation tools (tuzi-image-gen). - Sanitization: There is no evidence of sanitization or filtering of the article content before it is used to generate metadata and visual labels for the final prompts.
Audit Metadata