Go: Phase 11 — Evaluation Framework

[juliomenendez]

Summary

• EvalItem — test case with input messages, expected output, expected tool calls
• Evaluator function type — pluggable evaluation logic
• Built-in checks: KeywordCheck (case-insensitive), TextContainsCheck, ToolCalledCheck
• CombineEvaluators — composes multiple evaluators (all must pass, fail-fast)
• EvaluateAgent — runs agent against eval items, returns EvalResults
• EvalResults.RaiseForStatus() — returns error if any items failed

Test plan

• go test ./... -count=1 — all tests pass across 10 packages
• go vet ./... — no issues
• go mod tidy — module is clean
• EvaluateAgent tests (all pass, keyword fail, mixed results)
• TextContainsCheck tests (found, not found)
• ToolCalledCheck tests (called, not called)
• CombineEvaluators tests (all pass, partial fail)

🤖 Generated with Claude Code