Feat/issue 91 eval optimize loop

[zzp1221]

概要

本 PR 在 examples/optimization/eval_optimize_loop/ 下新增一个可复现的“评测 -> 失败归因 -> prompt 优化 -> 验证 -> 接
受/拒绝 gate -> 审计报告”闭环示例。

示例覆盖：

• 使用 AgentEvaluator 执行 baseline 评测
• 按最终回复不匹配、工具调用错误、工具参数错误、rubric 不达标、知识召回不足、格式不符合要求进行失败归因
• 默认使用离线 deterministic fake backend 进行 prompt 优化，避免依赖 API key
• 可选支持 --backend agent_optimizer 调用真实 AgentOptimizer
• 使用独立 validation split 验证候选 prompt
• 通过可配置 gate 判断 accept/reject
• 输出 optimization_report.json、optimization_report.md 和 trace evalset 审计产物

交付物

包含以下文件：

• run_pipeline.py
• train.evalset.json
• val.evalset.json
• prompts/system.md
• prompts/skill.md
• optimizer.json
• README.md
• design.md
• optimization_report.json
• optimization_report.md
• trace_evalsets/ 示例输出

样例 evalset 共 6 条 case：3 条训练、3 条验证，覆盖可优化成功、优化无效、优化后退化三类情况。

验证

本地已运行：

py -3.14 examples\optimization\eval_optimize_loop\run_pipeline.py

输出：

Optimization gate decision: reject
Report written to: optimization_report.json

这里的 reject 是预期结果：候选 prompt 修复了训练集和一个验证集 case，但引入了验证集退化，包括关键 case 退化，因此 gate
正确拒绝。

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (main@099b571). Learn more about missing BASE report.

Additional details and impacted files

@@            Coverage Diff             @@
##             main        #132   +/-   ##
==========================================
  Coverage        ?   87.86295%           
==========================================
  Files           ?         467           
  Lines           ?       44014           
  Branches        ?           0           
==========================================
  Hits            ?       38672           
  Misses          ?        5342           
  Partials        ?           0           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[github-actions]

CLA Assistant Lite bot All contributors have signed the CLA ✍️ ✅

[zzp1221]

I have read the CLA Document and I hereby sign the CLA