yolo-permissions

To classify tool-call risk before execution, provide the sample invocation to the classifier to obtain a risk assessment. It combines static policy guidance with dynamic scoring to return a clear risk level and a recommended action.

AI-assisted risk classification for tool calls is a safety mechanism that evaluates invocations before execution. It assesses contexts like destructive file operations and credential exposure to return a risk level and suggested action.

Dynamic scoring evaluates bash commands by integrating static policy rules with AI-based risk analysis. It checks for destructive file operations, credential exposure, and external communication to assign a risk level and action.

Yes, you can use static policy rules combined with AI risk analysis to block destructive file operations. The classifier returns a HIGH risk level for commands like rm -rf / and suggests blocking or requiring confirmation.

You need AI safety classification for local and remote environments when executing tool invocations that pose potential risks. It guards against unsafe operations by returning a clear risk level and a suggested action before execution.

Limitations of AI-based tool-call risk classification include its reliance on both static policy guidance and dynamic scoring. While it covers common operations, paths, and contexts, highly novel or complex attack vectors might not be fully captured.