seedance-food-asmr
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill accepts user-provided information (such as food subjects and settings) and incorporates this data into structured templates based on cases in
references/cases.md. The absence of explicit boundary markers or sanitization for this user input creates a surface where a user could potentially inject instructions meant to influence the downstream video generation model. - Ingestion points: Workflow step 1 in
SKILL.md(collecting subject, product, and setting). - Boundary markers: No specific delimiters (e.g., XML tags or triple backticks) are mandated for the user-supplied content within the template structure.
- Capability inventory: The skill is limited to text generation; it does not have tools for file writing, network requests, or shell execution.
- Sanitization: The instructions do not define methods for escaping or filtering user-provided text before interpolation.
- [EXTERNAL_DOWNLOADS]: The skill's documentation refers to the author's public GitHub repositories (
github.com/LearnPrompt/awesome-seedanceandgithub.com/LearnPrompt/goodcase-lite) as the source for daily updates to the reference cases. These links are for informational purposes and data synchronization within the vendor's ecosystem. - [SAFE]: The skill follows best practices by providing a fixed workflow and grounding its output in human-verified examples, which reduces the likelihood of hallucinations or unpredictable agent behavior during the prompt drafting process.
Audit Metadata