151 lines
4.8 KiB
Markdown
151 lines
4.8 KiB
Markdown
|
# 📡 铸渊指令｜persona-studio v1.1-D · 核心层（数据库动态自进化机制）
|
|||
|
|
|
|||
|
|
> **签发**：曜冥 × 冰朔 · 2026-03-11 · **执行者**：铸渊
|
|||
|
|
>
|
|||
|
|
|
|||
|
|
> **仓库**：`https://github.com/qinfendebingshuo/guanghulab`（主仓库）
|
|||
|
|
>
|
|||
|
|
|
|||
|
|
> **优先级**：**P0-S（最高优先）** · Phase A
|
|||
|
|
>
|
|||
|
|
|
|||
|
|
> **关联**：[🧪 铸渊指令｜persona-studio 人格体协助体验仓库搭建 v1.0（2026-03-09·曜冥签发）](%F0%9F%A7%AA%20%E9%93%B8%E6%B8%8A%E6%8C%87%E4%BB%A4%EF%BD%9Cpersona-studio%20%E4%BA%BA%E6%A0%BC%E4%BD%93%E5%8D%8F%E5%8A%A9%E4%BD%93%E9%AA%8C%E4%BB%93%E5%BA%93%E6%90%AD%E5%BB%BA%20v1%200%EF%BC%882026-03-09%C2%B7%20ec65d26d387746f2a88ee625a68a7104.md)
|
|||
|
|
>
|
|||
|
|
|
|||
|
|
---
|
|||
|
|
|
|||
|
|
# 🔴 ⑦ 数据库动态自进化机制
|
|||
|
|
|
|||
|
|
<aside>
|
|||
|
|
🔴
|
|||
|
|
|
|||
|
|
**核心原则：数据库不是静态模板。它是活的。每一次用户交互，都必须让系统变得更聪明。**
|
|||
|
|
|
|||
|
|
</aside>
|
|||
|
|
|
|||
|
|
---
|
|||
|
|
|
|||
|
|
## 7.1 三层架构
|
|||
|
|
|
|||
|
|
**第一层：用户层**（每个用户独立·实时更新）
|
|||
|
|
|
|||
|
|
```
|
|||
|
|
brain/memory/EXP-XXX/
|
|||
|
|
├── memory.json ← 完整对话历史
|
|||
|
|
├── profile.json ← 用户画像（每次交互后更新）
|
|||
|
|
├── projects.json ← 项目记录
|
|||
|
|
├── compressed.json ← 记忆摘要
|
|||
|
|
└── preferences.json ← 🆕 偏好模型（自动学习）
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
**第二层：系统层**（全局共享·自动聚合）
|
|||
|
|
|
|||
|
|
```
|
|||
|
|
brain/
|
|||
|
|
├── knowledge-base.json ← 🆕 知识库（自动积累）
|
|||
|
|
├── pattern-library.json ← 🆕 模式库（自动识别）
|
|||
|
|
└── evolution-log.json ← 🆕 进化日志
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
**第三层：模型层**（决策优化·自动调参）
|
|||
|
|
|
|||
|
|
```
|
|||
|
|
brain/
|
|||
|
|
├── model-benchmark.json ← 模型评测（每日更新）
|
|||
|
|
└── quality-scores.json ← 🆕 质量评分（反馈驱动）
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
---
|
|||
|
|
|
|||
|
|
## 7.2 用户画像自动进化
|
|||
|
|
|
|||
|
|
**触发**：每轮对话结束后 → `profile-learner.js` → `updateProfile()`
|
|||
|
|
|
|||
|
|
profile.json 自动学习字段：
|
|||
|
|
|
|||
|
|
- `tech_assessment`：level / known_skills / learning_skills / weakness / growth_velocity / assessment_confidence
|
|||
|
|
- `communication_style`：verbosity / prefers_examples / question_frequency / patience_level
|
|||
|
|
- `design_preferences`：color_scheme / style_keywords / layout_preference
|
|||
|
|
- `project_patterns`：avg_requirements_rounds / decision_speed / change_frequency
|
|||
|
|
- `emotional_profile`：current_mood / satisfaction_trend / frustration_triggers / delight_triggers
|
|||
|
|
|
|||
|
|
推断逻辑示例：
|
|||
|
|
|
|||
|
|
- 用户说「直接告诉我命令」→ verbosity = concise
|
|||
|
|
- 用户连续问JS问题 → learning_skills +JS
|
|||
|
|
- 用户连续三次一次通过 → growth_velocity = fast
|
|||
|
|
|
|||
|
|
---
|
|||
|
|
|
|||
|
|
## 7.3 知识库自动积累
|
|||
|
|
|
|||
|
|
**触发**：代码生成完成时 / 用户确认方案时 / 解决技术问题时
|
|||
|
|
|
|||
|
|
`knowledge-base.json` 每条 entry：
|
|||
|
|
|
|||
|
|
- type：solution / pattern / pitfall / best_practice
|
|||
|
|
- category / tags / title / content
|
|||
|
|
- source_exp / source_project / usage_count
|
|||
|
|
|
|||
|
|
**使用方式**：下一个用户聊到类似需求 → persona-engine 先查知识库 → 引用已有方案 → 更专业更快
|
|||
|
|
|
|||
|
|
---
|
|||
|
|
|
|||
|
|
## 7.4 模式库自动识别
|
|||
|
|
|
|||
|
|
**触发**：每完成一个项目 + 每日凌晨3:00聚合
|
|||
|
|
|
|||
|
|
`pattern-library.json` 每条 pattern：
|
|||
|
|
|
|||
|
|
- name / frequency / common_features / common_tech_stack / avg_build_time / success_rate
|
|||
|
|
- 高频模式可生成模板代码存入 `brain/templates/`
|
|||
|
|
|
|||
|
|
**使用方式**：新用户说「做个博客」→ 查到高频模式 → 主动建议最佳功能组合 → 复用模板提速
|
|||
|
|
|
|||
|
|
---
|
|||
|
|
|
|||
|
|
## 7.5 更新总触发器
|
|||
|
|
|
|||
|
|
| 触发时机 | 更新内容 |
|
|||
|
|
| --- | --- |
|
|||
|
|
| 每轮对话后 | memory.json追加 / profile.json更新 / compressed判断压缩 |
|
|||
|
|
| 项目完成时 | projects.json / knowledge-base提取 / pattern-library判断 / quality-scores |
|
|||
|
|
| 每日3:00 | model-benchmark重测 / pattern聚合 / knowledge去重 / evolution-log日志 |
|
|||
|
|
| 新用户注册 | registry新增 / 创建memory目录 / 初始化空profile |
|
|||
|
|
| 主控手动 | persona-config更新 / 知识库清洗 / 模式库优化 |
|
|||
|
|
|
|||
|
|
---
|
|||
|
|
|
|||
|
|
## 7.6 新增后端文件
|
|||
|
|
|
|||
|
|
```
|
|||
|
|
backend/brain/
|
|||
|
|
├── profile-learner.js ← 🆕 画像自动学习
|
|||
|
|
├── knowledge-extractor.js ← 🆕 知识提取
|
|||
|
|
├── pattern-analyzer.js ← 🆕 模式识别
|
|||
|
|
└── evolution-logger.js ← 🆕 进化日志记录
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
---
|
|||
|
|
|
|||
|
|
### 执行清单
|
|||
|
|
|
|||
|
|
- [ ] 创建 `knowledge-base.json` / `pattern-library.json` / `evolution-log.json` 初始结构
|
|||
|
|
- [ ] 实现 `profile-learner.js`
|
|||
|
|
- [ ] 实现 `knowledge-extractor.js`
|
|||
|
|
- [ ] 实现 `pattern-analyzer.js`
|
|||
|
|
- [ ] 实现 `evolution-logger.js`
|
|||
|
|
- [ ] memory-manager.js 改造：新增 updateProfile + autoExtractKnowledge
|
|||
|
|
- [ ] persona-engine.js 改造：查知识库+模式库
|
|||
|
|
- [ ] code-generator.js 改造：完成后触发知识提取+模式识别
|
|||
|
|
|
|||
|
|
<aside>
|
|||
|
|
🔴
|
|||
|
|
|
|||
|
|
**铸渊，Phase A 最优先执行此指令。**
|
|||
|
|
|
|||
|
|
数据库动态自进化 = 系统灵魂。知识库越厚、画像越准、模式越多、代码越快——活的系统。
|
|||
|
|
|
|||
|
|
</aside>
|
|||
|
|
|
|||
|
|
💙 曜冥签发 · 2026-03-11
|