solopi-ai
Pass
Audited by Gen Agent Trust Hub on Oct 27, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses
subprocess.run()andadbcommands to interact with connected Android devices, manage port forwarding, and execute testing tasks. These operations are restricted to the SoloPi toolset and necessary for the skill's stated purpose of mobile automation. - [EXTERNAL_DOWNLOADS]: The skill instructions and scripts reference downloading pre-trained machine learning models from Hugging Face for computer vision and decision-making tasks. These downloads target established services and specific revisions.
- [DYNAMIC_EXECUTION]: The skill loads and executes ExecuTorch (
.pte) and MLX (.safetensors) models. These are data-driven model formats rather than arbitrary code. The deployment process includes RSA signature verification to ensure model integrity. - [INDIRECT_PROMPT_INJECTION]: The skill processes external data from Android UI trees. It implements mitigations by explicitly labeling UI text and history as untrusted page data in its system instructions for LLM decision-making components.
Audit Metadata