Add new work pattern: iterative optimization work pattern

[caiy0233]

When submitting a PR, please confirm the following points and put [x] in the boxes one by one. | 在提出pr时，请确认了以下几点，并逐一使用[x]符号确认勾选。

关联issue:#257

Checklist | 检查项

• I have read and understood the contributor guidelines. | 我已阅读并理解贡献者指南 。
• I have checked for any duplicate features related to this request and communicated with the project maintainers. | 我已检查没有与此请求重复的功能并与项目维护者进行了沟通。
• I accept the suggestion of the maintainers to make changes to or close this PR. | 我接受此PR配合维护人员的建议进行修改或关闭。
• I have submitted the test files and can provide screenshots of the test results (required for feature or bug fixes) | 我已经提交了测试文件并可提供测试结果截图(功能修改、BUG修复类PR必须提供，其他按需)
• I have added or modified the documentation related to this PR | 我已经添加或修改了本次pr对应的文档说明(非必要，根据实际PR内容按需添加)
• I have added examples and notes if needed | 我已经添加了使用案例代码与文档说明(非必要，根据实际PR内容按需添加)

Please fill in the specific details of this PR: | 请详细填写本次PR的内容:

• Iterative Optimization Work Pattern: Iterative Prompt Optimization with Batches, Scoring, and Memory

• 自适应优化：支持批量样本评估、质量评分与持久化记忆的迭代式提示优化
  Summary / 摘要

• Adds optimization_work_pattern to enable iterative prompt optimization over batched samples, guided by scoring and feedback loops, with persistent memory of past successes and failures.

• 新增 optimization_work_pattern ，实现基于批量样本的迭代式提示优化，通过评分与反馈闭环驱动，并持久化过去的成功与失败样例，从其中的关键case吸收经验。

Motivation / 动机

• Single-sample evaluation often produces unstable outcomes; batched evaluation provides more robust signals.
• Persisting historical insights accelerates convergence and avoids repeated mistakes.
• 单样例评价不稳定，批量样本能提供稳健信号；历史记忆的持久化可加速收敛并避免重复错误。

Highlights / 特色功能

• Batched evaluation for stability
  • samples are processed in batches across iterations, producing averaged scores and pass rates that smooth variance.
  • 在多轮迭代中按批评估 samples ，生成平均分与通过率，降低方差提升稳定性。
• Persistent memory of insights
  • Persists iteration records with before/after prompts to ~/.agentuniverse/optimization_memory.json , enabling reuse of successful patterns and awareness of failure modes.
  • 迭代记录（优化前后提示）持久化到 ~/.agentuniverse/optimization_memory.json ，可复用成功模式并感知常见失败原因。
• Intelligent history selection and summarization
  • Selects important records by score change, extremes, diversity, and recency; summarizes unselected records to fit context limits.
  • 按分数变化、极端表现、多样性与时序选择关键历史；对未选记录生成摘要以适配上下文。
• Flexible stop criteria
  • Supports avg_score_threshold and pass_rate_threshold for early stopping when quality meets targets.
  • 支持 avg_score_threshold 与 pass_rate_threshold 作为停止条件，达到目标即提前停止。

Main Functions / 主要功能

• Iterative optimization over batched samples with configurable iteration counts.
• Continuous quality scoring and early stopping by thresholds.
• Feedback-driven prompt improvement with contextual history and prior insights.
• 基于批量样本的迭代优化，可配置迭代次数；连续质量评分并支持阈值提前停止；基于反馈与历史上下文优化智能体提示词。

Parameters / 参数解释

• samples :
  • List of strings or dicts. The evaluation/optimization corpus. Required.
  • 字符串或字典列表，用于优化的测试评估样例。必需。
• initial_prompt :
  • Starting prompt text; if omitted, executing agent’s current instruction applies.
  • 初始提示文本；未提供则使用执行agent的当前prompt。与agent_name_for_optimization提供任一即可。
• agent_name_for_optimization :
• Optionally reference another agent to source its instruction as initial_prompt .
• 可引用其他agent的提示词作为 initial_prompt 来源。与initial_prompt提供任一即可。注意：提示词中如包括多个动态变量，必须与samples中的key一一对应。
• batch_size :
  • Number of samples per iteration batch. Default 2–3 depending on template/profile.
  • 每轮使用的样例数量。
• max_iterations :
  • Maximum optimization rounds. Default 5.
  • 最大迭代轮次。默认 5。
• scoring_standard :
  • Natural language rubric for the scoring agent to evaluate answer quality.
  • 面向评分agent的自然语言评分标准。
• avg_score_threshold :
  • Stop when average score ≥ threshold.
  • 当平均分 ≥ 阈值时停止。
• pass_rate_threshold :
  • Stop when fraction of items with score ≥ pass_score ≥ threshold.Optional.
  • 当 score ≥ pass_score 的占比 ≥ 阈值时停止。非必需。
• pass_score :
  • Minimum score considered “pass”; Optional.
  • 认为“通过”的最低分；非必需。
• max_history_records :
  • Upper bound of retained detailed history records used in feedback.Optional.
  • 反馈中保留的详细历史记录上限。非必需。
• max_feedback_chars :
  • Soft limit for feedback payload size; controls summarization to avoid exceeding model context.Optional.
  • 反馈长度软限制；用于摘要以避免超出模型上下文。非必需。

Usage / 使用示例

• Minimal unittest pattern:
  • examples/sample_apps/optimization_agent_app/intelligence/test/test_optimization_agent.py:1
• Sample run script with multiple input presets:
  • examples/sample_apps/optimization_agent_app/intelligence/test/optimization_agent.py:16

Validation / 验证

• Unit test added mirroring peer agent test structure.
  • examples/sample_apps/optimization_agent_app/intelligence/test/test_optimization_agent.py:1
• Manual run via sample script demonstrates multi-dataset capability and threshold stopping.

Checklist / 自检清单

• Tests included: yes ( test_optimization_agent.py )
• Sample app runnable: yes ( examples/sample_apps/optimization_agent_app )

Please provide the path of test files and submit screenshots or files of the test results(fill in as needed): | 请填写测试文件路径并提供测试结果截图或文件(按需填写):
examples/sample_apps/optimization_agent_app/intelligence/test/test_optimization_agent.py

[caiy0233]

#257

[caiy0233]

#257

[caiy0233]

#257