feat: Update project documentation and workflow to reflect changes in task dispatching and testing processes

- Changed references from "kilo" to "Agent 子代理" in various documentation files to align with the new task dispatching method.
- Updated CLAUDE.md to reflect the new agent-based execution model and modified task dispatch template.
- Revised PROGRESS.md to indicate successful testing and acceptance of M1.2 with detailed results.
- Enhanced communication and workflow documentation to clarify the use of Agent 子代理 for real-time collaboration and task management.
- Updated team member soul files with lessons learned from recent testing and integration experiences.
- Refactored test files to remove unused code and improve clarity, ensuring compliance with zero warning standards.
- Documented specific experiences and lessons learned during M1.2 testing, including handling of edge cases and configuration validation.

souls/chen-yifei.md

@@ -42,9 +42,9 @@
 
 ## CEO 个人经验
 
-### kilo 派发经验
+### 子代理派发经验（kilo 机制已于 2026-07-02 废除，教训保留，现用 Agent 子代理）
 - PM 刘建国 git commit 连续两次失败（agent 无视"不要读 diff"），QA 林晓峰一次搞定
-- kilo agent 倾向于自作主张读 diff/分析代码 → 只给命令，不给"任务描述"
+- 执行 agent 倾向于自作主张读 diff/分析代码 → 只给命令，不给"任务描述"
 - 指令越简单越好，不要让 agent 读大文件/大 diff
 - 代码提交任务直接给命令序列，不需要 agent 自行分析
 - GPT-5.4 执行力 > 策略力 → 给具体命令比给目标可靠
@@ -54,7 +54,7 @@
 - 关键路径任务（git push）派给最可靠的人，不按头衔
 - PM 不可靠时可跳过 PM 直接派开发者，记录原因
 - QA 比 PM 更适合执行类任务（提交、验证），PM 适合分析和规划
-- 员工用 kilo 完成后必须自行更新 soul 文件，否则经验丢失
+- 员工完成任务后必须自行更新 soul 文件，否则经验丢失
 
 ### 角色边界
 - **CEO 绝不直接写代码/改代码/跑测试** — 即使小修复也派回原作者