eval-tool-use

You evaluate LLM tool selection and argument construction by providing a trace of a tool-use session and running deterministic checks to identify tool selection, argument, and sequencing issues. This applies to Claude-style agents with function calling and API integrations.

Test tool sequencing and error handling by auditing the order of tool calls and verifying the agent's recovery after failures. This Skill applies structured tests to a tool-use session trace to validate that the agent handles tool errors gracefully.

Yes, you can audit an MCP server tool-calling workflow. The evaluation applies to Claude-style agents with access to tools like MCP servers and API integrations, verifying tool selection, argument validation, and sequencing within the trace.

The best way to validate tool inputs is to run argument validation checks on a tool-use session trace. This verifies that inputs to tools conform to expected schemas and constraints, separating deterministic tool interactions from higher-level reasoning.

This evaluation focuses on deterministic tool checks rather than purely reasoning. It assesses how an LLM agent selects tools, constructs inputs, and sequences tool calls, applying structured tests for tool selection, argument validation, and error handling.