seedance-prompt-en
Fail
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: CRITICALINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONMETADATA_POISONING
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates processing of multiple external media inputs (up to 9 images, 3 videos, and 3 audio files) through a structured referencing system, creating an attack surface for instructions embedded in data.
- Ingestion points: Defined in SKILL.md and zh/SKILL.md via natural language instructions for referencing external assets using the @ syntax.
- Boundary markers: The skill lacks specific instructions for the agent to differentiate between descriptive content and malicious directives that could be contained within the referenced assets' metadata or descriptions.
- Capability inventory: The skill is primarily used to generate prompts for multimodal video generation, which are then processed by external AI systems.
- Sanitization: No input validation or content sanitization is described or required by the instructions.
- [COMMAND_EXECUTION]: The installation guide in README.md includes shell commands (mkdir, cp) and a Node.js CLI execution (npx skills add dexhunter/seedance2-skill). These instructions involve local file system manipulation and remote package retrieval from the author's repository.
- [METADATA_POISONING]: An automated scanner flagged SKILL.md with a FileRepMalware reputation alert. Upon manual inspection, the file contains natural language text describing dramatic and violent scenes for video templates. While the alert is present, the content appears functional and aligned with the skill's purpose of creative video generation.
Recommendations
- CRITICAL: 1 file(s) identified as malware by FileRep - DO NOT USE
Audit Metadata