co-op-game-intro-generator
Pass
Audited by Gen Agent Trust Hub on Nov 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by ingesting untrusted user data (player names, game titles, and custom visual styles) and interpolating it directly into prompts for image and video generation. The agent could be manipulated if a user provides inputs containing instructions instead of names or titles.
- Ingestion points: User-provided inputs collected in STEP 1 (visual style) and STEP 2 (player names, game title) in
SKILL.mdandSKILL.cn.md. - Boundary markers: The skill utilizes simple placeholder interpolation (e.g.,
{player1_name},{game_title},{visual_style}) within the markdown templates provided inreferences/h3-confirmation-image-template.mdandreferences/h3-video-prompt-template.md. While the templates provide structured instructions, they lack explicit 'ignore embedded instructions' warnings for the interpolated data. - Capability inventory: The skill utilizes the MiniMax H3 generation tool to create images and 15-second videos based on the constructed prompts.
- Sanitization: The instructions do not define any sanitization, validation, or escaping logic to prevent the user-provided text from containing prompt injection payloads that could override the generation parameters.
Audit Metadata