智能体工具调用评测

1. 1. Prepare redacted UTF-8 JSON by following examples/input.json.
2. Run python scripts/evaluate_tool_calls.py --input examples/input.json --output examples/output.json.
3. Review warnings, thresholds, and human-review items before using the result in another system.

This skill reads local test data and writes a reviewable local result. It does not call models, execute tools, change production systems, or transmit traces.