定义现状 (Context)：The current implementation in @module/service.py has a performance bottleneck under concurrent loads.
定义优化目标 (Goal)：Refactor this module to use an asynchronous pattern with a connection pool to handle up to 500 concurrent requests.
提供示例 (Example)：比如定义输入和输出。For instance, a call to process_data(items) should now return a Future object.
施加约束 (Constraints)：Only use libraries present in requirements.txt. The solution must be compatible with Python 3.9+. Maintain the existing public API signature to avoid breaking changes.
TDD (Methodology)：`First, write a failing benchmark test that reproduces the bottleneck. Then, implement the async changes. Finally, ensure all existing unit tests and the new benchmark pass.
一些技巧：
加入 think, think hard, think harder, ultrathink 等关键词，可以强制 AI 分配更多资源进行规划
Shift + Tab 切换到计划模式（Plan Mode），强制 AI 在动手前提交一份详细的行动计划。
Before fulfill a task or refactor some submodules, you should firstly use N parallel subagents or different aspects/perspectives to brainstorm possible plans. Compare their trade-offs in terms of performance, scalability, readability and maintainability, security etc. Do not implement any code.
质量与合规：利用 Hooks 系统，在项目中强制执行代码格式化、安全扫描等质量门禁。
Hooks 是在 Claude Code 生命周期特定事件点（如 PreToolUse, PostToolUse, UserPromptSubmit 等）执行的、用户定义的 Shell 命令。 它们提供了对 AI 行为的确定性控制，能将团队的非功能性需求（如安全、合规、质量）强制注入到 AI 对代码的每一次操作中，实现从“建议”到“规则”的转变。
