alayarenderer-generative-world
Warn
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill clones a code repository from an external source (
https://github.com/ShandaAI/AlayaRenderer.git) and includes recursive submodules from other third-party sources. - [REMOTE_CODE_EXECUTION]: The skill downloads large-scale model weights from third-party Hugging Face repositories (
Brian9999/world_inverse_rendererandBrian9999/stylerenderer). The code downloaded from the external repository is subsequently executed via Python scripts for model inference. - [COMMAND_EXECUTION]: The skill instructions involve executing several shell commands, including
git clone,conda create,huggingface-cli download, andpythonexecution for inference scripts. - [INDIRECT_PROMPT_INJECTION]: The skill accepts user-provided text prompts which are passed directly to a CLI argument (
--prompt). This creates a vulnerability surface where a malicious input could attempt to perform command injection if the underlying agent or system does not properly sanitize the input before shell execution. - Ingestion points: The text prompt input for the Game Editing inference script.
- Boundary markers: No explicit delimiters or instructions are provided to the agent to treat the prompt purely as data.
- Capability inventory: The skill possesses file system write capabilities and shell command execution via Python.
- Sanitization: The documentation does not specify any validation or sanitization for the user-supplied prompt strings.
Audit Metadata