48 lines
1.2 KiB
Markdown
48 lines
1.2 KiB
Markdown
---
|
||
title: "Self-Improving Skill"
|
||
type: concept
|
||
tags: [openclaw, memory, agent]
|
||
last_updated: 2026-04-17
|
||
---
|
||
|
||
## Definition
|
||
Self-Improving Skill（自改进技能）是 OpenClaw 中的一种结构化经验记录系统，使 AI Agent 能够在每次遇到问题、做出决策、或发现值得记住的东西时自动记录学习内容，实现持续改进。
|
||
|
||
## Structure
|
||
```markdown
|
||
## [LRN-YYYYMMDD-NNN] type
|
||
|
||
**Logged**: YYYY-MM-DDTHH:MM:SS+08:00
|
||
**Priority**: high|medium|low
|
||
**Status**: pending|resolved
|
||
**Area**: config|workflow|correction
|
||
|
||
### Summary
|
||
一句话描述学到了什么
|
||
|
||
### Details
|
||
具体发生了什么、问题出在哪
|
||
|
||
### Suggested Action
|
||
以后遇到类似情况该怎么做
|
||
|
||
### Metadata
|
||
- Pattern-Key: xxx
|
||
- Recurrence-Count: N
|
||
- See Also: LRN-YYYYMMDD-NNN
|
||
```
|
||
|
||
## Key Fields
|
||
- **Pattern-Key**：经验检索键，用于追踪同一类型问题的生命周期
|
||
- **Recurrence-Count**：重复次数，区分一次性错误和系统性重复
|
||
- **Suggested Action**：具体可执行的改进建议，而非抽象的注意事项
|
||
|
||
## Use Cases
|
||
- 记录 AI Agent 犯过的错误及修复方法
|
||
- 记录工作流优化发现
|
||
- 记录配置技巧和环境差异
|
||
|
||
## Related
|
||
- [[双层记忆架构]]
|
||
- [[每日复盘机制]]
|
||
- [[Pattern-Key]] |