seedance-interview

Pass

Audited by Gen Agent Trust Hub on Sep 21, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied scene descriptions and feelings to construct production briefs for downstream tasks.\n
  • Ingestion points: The skill ingests untrusted user input through the "Priority Question Pool" and the "Starting Points" menu in SKILL.md.\n
  • Boundary markers: There are no explicit delimiters or system instructions to ignore potential commands embedded within user descriptions.\n
  • Capability inventory: The skill is capable of routing data to other tools, including seedance-prompt, seedance-prompt-short, and seedance-pipeline.\n
  • Sanitization: The skill lacks sanitization, explicitly instructing the agent to "capture their direction verbatim" and preserve specific vocabulary.\n- [METADATA_POISONING]: The skill exhibits inconsistent naming in its metadata. The author is listed as Iamemily2050, whereas the repository URL and vendor context reference Emily2040. This numeric substitution is an identifier discrepancy that could be misleading.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 21, 2026, 07:30 AM
Security Audit — agent-trust-hub — seedance-interview