forge-tool-use

What problem does it solve?

Unchecked LLM tool/function loops waste tokens, can spin forever, and often fail by throwing exceptions or returning unstructured results, leading to brittle agent behavior.

Core Features & Use Cases

• Hygienic tool-call loop design: enforces a hard iteration cap, structured loop events, and predictable termination.
• Validated structured inputs/outputs: validates tool arguments and tool outputs against schemas to enable recovery from “model called tool wrong” vs “tool execution failed.”
• Robust error-as-data handling: returns tool failures as structured data instead of letting exceptions crash the loop.

Quick Start

Instruct your AI agent to follow the forge-tool-use discipline by using a tool-use loop with a fixed max iterations, validating tool inputs/outputs, and returning tool results as structured content (not stringified prose) even when failures occur.