add-behavioral-tests

To add behavioral tests, generate a pytest suite and an EvalHub fixture within your agent repository. This establishes end-to-end behavioral coverage, including runner compatibility verification and golden query design, without modifying the agent code.

Score tool calls by extracting expectations directly from MLflow traces rather than relying on content heuristics. This approach ensures accurate tracing compatibility and reliable tool-call verification during end-to-end behavioral evaluation.

Yes, you can use EvalHub fixtures with strict phase-based guardrails that prevent modifying the agent under test. The workflow allows adding test-only artifacts and updating testing documentation while keeping the core agent code unchanged.

The best way is implementing a complete behavioral testing setup using pytest suites and EvalHub fixtures. This workflow guides the sequence from scope intake through validation and on-cluster e2e execution, ensuring MLflow tracing is present.

Behavioral testing requires MLflow tracing verification because tool-call scoring depends on traces rather than unreliable content heuristics. Verifying MLflow enrichment ensures the structural inspection of traces accurately captures tool calls for e2e evaluation.