seedance-prompt-library
Pass
Audited by Gen Agent Trust Hub on Oct 2, 2026
Risk Level: SAFEPERSISTENCECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PERSISTENCE]: The skill includes an installation script (
bin/install.mjs) that creates and modifies files within the user's persistent agent directories. - Evidence: The script writes to
~/.claude/skills/seedance-prompt-libraryand~/.codex/skills/seedance-prompt-libraryusingmkdirSyncandcpSyncto ensure the skill remains available across sessions. - [COMMAND_EXECUTION]: The package defines a binary entry point that executes a Node.js script to perform file system operations.
- Evidence:
package.jsonmapsseedance-prompt-librarytobin/install.mjs, which is designed to be executed vianpxor local installation. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted user input and interpolate it into structured prompt templates, which is a common vector for indirect injection.
- Ingestion points: The
SKILL.mdworkflow instructs the agent to identify user intent and "Fill the structure" using specifics from the user's request. - Boundary markers: Absent. The instructions do not specify the use of delimiters (like XML tags or triple quotes) or provide warnings to ignore embedded instructions within user-provided scene descriptions.
- Capability inventory: The skill reads local reference files (
references/style-library.md) and generates complex instructions for external video generation models. - Sanitization: Absent. The workflow does not include a step for sanitizing or escaping user-provided content before it is placed into the prompt structure.
Audit Metadata