ltx2
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill accepts user-provided text prompts and images to generate video content, which is a standard vector for indirect prompt injection.
- Ingestion points: The
--promptand--inputflags in thetools/ltx2.pyscript, as described in SKILL.md. - Boundary markers: Not present; the documentation does not show delimiters or instructions to the model to ignore potential commands within the user prompt.
- Capability inventory: The tool writes video files to the disk and executes shell commands via
uv run(SKILL.md). - Sanitization: No evidence of input sanitization or filtering is provided in the documentation.
- [EXTERNAL_DOWNLOADS]: The skill utilizes external repositories and cloud platforms for model weights and execution.
- Evidence: Mentions of Hugging Face for LoRAs and Google Gemma models, and the use of Modal for remote GPU execution. These are established well-known services and trusted organizations.
Audit Metadata