Tool Composition

Composable tools let agents build sophisticated workflows by chaining simple, focused primitives rather than relying on monolithic tools that bundle complex operations into a single call, mirroring the Unix philosophy of small tools that each do one thing well. This approach makes individual tools easier to test, debug, and reuse across different workflows, and gives the model more granular control over each step so that a failure in one operation does not silently corrupt the entire sequence. The trade-off is that heavily decomposed tools require the model to plan multi-step sequences, which increases reasoning burden and token cost per task compared to calling one large tool.