seedance-2-0
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data via the --prompt, --prompt-file, and image frame parameters in scripts/generate_video.sh. Evidence chain: (1) Ingestion points: --prompt and image path arguments in scripts/generate_video.sh. (2) Boundary markers: Absent; inputs are interpolated directly into the API payload. (3) Capability inventory: Network requests and file downloads via curl, plus dynamic script execution via python3. (4) Sanitization: Parameters are passed through environment variables and escaped using Python's json.dump utility during payload generation.
- [DYNAMIC_EXECUTION]: scripts/generate_video.sh executes embedded Python code blocks using the 'python3 -' pattern to perform utility tasks such as Base64 encoding local images and assembling the JSON request body.
- [COMMAND_EXECUTION]: The skill relies on executing system binaries including curl for API communication and python3 for data transformation tasks within the main generation script.
- [EXTERNAL_DOWNLOADS]: The skill automatically downloads generated MP4 files using curl from URLs provided in the API response, which may include external storage providers.
Audit Metadata