| name | self-improving-agent |
| description | Captures learnings, errors, and corrections to enable continuous improvement. Use when: (1) A command or operation fails unexpectedly, (2) User corrects Claude ('No, that's wrong...', 'Actually...'), (3) User requests a capability that doesn't exist, (4) An external API or tool fails, (5) Claude realizes its knowledge is outdated or incorrect, (6) A better approach is discovered for a recurring task. Also review learnings before major tasks. |
| triggers | ["self-improving","自我改进","learning","学习","error","错误","mistake","错误","correction","修正","improvement","改进","capture learning","记录学习","continuous improvement","持续改进"] |
Self-Improving Agent - 自我改进技能
捕捉学习、错误和修正，实现持续改进的技能。
🎯 何时使用
触发场景
- 命令或操作意外失败 - 记录失败原因和解决方案
- 用户纠正Claude ("不，那是错的...", "实际上...") - 记录正确信息
- 用户请求不存在的功能 - 记录功能需求
- 外部API或工具失败 - 记录故障和解决方法
- Claude意识到知识过时或不正确 - 更新知识库
- 发现重复任务的更好方法 - 记录优化方法
使用时机
- 在主要任务前回顾学习记录
- 当遇到重复错误时检查历史解决方案
- 当需要优化工作流程时参考最佳实践
📋 学习日志类别
| 类别 | 文件 | 描述 | 升级路径 |
|---|
| 错误 | .learnings/ERRORS.md | 操作失败、命令错误 | → AGENTS.md (工作流) |
| 学习 | .learnings/LEARNINGS.md | 事实、模式、技巧 | → CLAUDE.md (知识) |
| 功能请求 | .learnings/FEATURE_REQUESTS.md | 缺失功能、增强请求 | → TOOLS.md (工具) |
| 工作流改进 | .learnings/WORKFLOW.md | 流程优化、自动化 | → AGENTS.md (OpenClaw工作区) |
| 工具技巧 | .learnings/TOOL_TIPS.md | 工具使用技巧、陷阱 | → TOOLS.md (OpenClaw工作区) |
| 行为模式 | .learnings/BEHAVIOR.md | 交互模式、最佳实践 | → SOUL.md (OpenClaw工作区) |
🚀 快速开始
基础设置
mkdir -p .learnings
touch .learnings/ERRORS.md .learnings/LEARNINGS.md .learnings/FEATURE_REQUESTS.md
OpenClaw 设置 (推荐)
cp -r hooks/openclaw ~/.openclaw/hooks/self-improvement
openclaw hooks enable self-improvement
📖 使用示例
中文示例
echo "## [ERR-20260312-001] pip安装失败" >> .learnings/ERRORS.md
echo "**时间**: 2026-03-12T14:30:00Z" >> .learnings/ERRORS.md
echo "**优先级**: 高" >> .learnings/ERRORS.md
echo "**状态**: 已解决" >> .learnings/ERRORS.md
echo "**领域**: 配置" >> .learnings/ERRORS.md
echo "" >> .learnings/ERRORS.md
echo "### 问题" >> .learnings/ERRORS.md
echo "安装tencentcloud-sdk-python时网络超时" >> .learnings/ERRORS.md
echo "" >> .learnings/ERRORS.md
echo "### 解决方案" >> .learnings/ERRORS.md
echo "使用国内镜像源: pip install -i https://pypi.tuna.tsinghua.edu.cn/simple tencentcloud-sdk-python" >> .learnings/ERRORS.md
echo "## [LRN-20260312-001] 腾讯云配置" >> .learnings/LEARNINGS.md
echo "**时间**: 2026-03-12T15:00:00Z" >> .learnings/LEARNINGS.md
echo "**优先级**: 中" >> .learnings/LEARNINGS.md
echo "**状态**: 已验证" >> .learnings/LEARNINGS.md
echo "**领域**: API" >> .learnings/LEARNINGS.md
echo "" >> .learnings/LEARNINGS.md
echo "### 总结" >> .learnings/LEARNINGS.md
echo "腾讯云API密钥优先存储在.env文件，不在代码中明文存储" >> .learnings/LEARNINGS.md
英文示例
cat >> .learnings/ERRORS.md << 'EOF'
**Logged**: 2026-03-12T16:00:00Z
**Priority**: high
**Status**: resolved
**Area**: git
Failed to push to GitHub: "fatal: could not read Username"
Use HTTPS with token: git push https://username:token@github.com/user/repo.git
EOF
cat >> .learnings/LEARNINGS.md << 'EOF'
**Logged**: 2026-03-12T17:00:00Z
**Priority**: medium
**Status**: verified
**Area**: tools
Use PyMuPDF for PDF text extraction, OpenCV for QR code detection
- PyMuPDF is fast for text extraction
- OpenCV with QR code detector works well for images
- Combine both for comprehensive PDF processing
EOF
🔧 核心功能
学习捕获
./scripts/self_improvement.py log-learning --category "tool" --summary "使用skillhub安装技能更可靠" --details "相比手动安装，skillhub提供版本管理和依赖解析"
学习回顾
./scripts/self_improvement.py list --category errors --days 7
./scripts/self_improvement.py search --query "GitHub" --category all
./scripts/self_improvement.py stats
学习升级
./scripts/self_improvement.py promote --id LRN-20260312-001 --to CLAUDE.md
./scripts/self_improvement.py promote-batch --category workflow --to AGENTS.md
🛠️ 工作流程
错误处理流程
1. 检测到错误
2. 自动记录到 ERRORS.md
3. 分析错误原因
4. 寻找解决方案
5. 验证解决方案
6. 升级到适当文档
学习流程
1. 发现新知识
2. 记录到 LEARNINGS.md
3. 验证知识准确性
4. 分类和组织
5. 定期回顾
6. 应用到实际任务
📁 文件结构
self-improving-agent/
├── SKILL.md # 技能说明文档
├── _meta.json # 元数据
├── scripts/
│ ├── self_improvement.py # 主脚本
│ ├── error_capture.sh # 错误捕获脚本
│ └── learning_review.py # 学习回顾工具
├── hooks/
│ └── openclaw/ # OpenClaw钩子
├── templates/
│ ├── error_template.md # 错误模板
│ ├── learning_template.md # 学习模板
│ └── feature_request.md # 功能请求模板
└── references/
├── openclaw-integration.md # OpenClaw集成指南
├── best_practices.md # 最佳实践
└── examples/ # 示例文件
⚠️ 故障排除
常见问题
-
钩子未触发
openclaw hooks list
openclaw hooks enable self-improvement
-
学习文件不存在
mkdir -p .learnings
touch .learnings/ERRORS.md .learnings/LEARNINGS.md
-
脚本权限问题
chmod +x scripts/*.py scripts/*.sh
调试模式
export SELF_IMPROVEMENT_DEBUG=1
./scripts/self_improvement.py log-learning --debug
tail -f .learnings/debug.log
🔄 与其他技能集成
与GitHub技能集成
./scripts/self_improvement.py sync-to-github --repo owner/repo
./scripts/self_improvement.py import-from-github --repo owner/repo --label feedback
与飞书技能集成
./scripts/self_improvement.py send-report --to feishu --channel daily-learnings
./scripts/self_improvement.py extract-from-feishu --message-id <id>
📊 学习分析
生成报告
./scripts/self_improvement.py generate-report --period week --format html
./scripts/self_improvement.py analyze-errors --period month --output chart.png
./scripts/self_improvement.py evaluate-impact --metric "error_reduction"
数据可视化
./scripts/self_improvement.py visualize --type pie --data .learnings/stats.json
./scripts/self_improvement.py timeline --output timeline.html
🎯 最佳实践
记录技巧
- 立即记录 - 不要推迟记录学习
- 具体详细 - 提供足够的上下文
- 可操作 - 确保解决方案可执行
- 分类清晰 - 使用正确的类别和标签
- 定期回顾 - 每周回顾并清理
管理技巧
- 优先级排序 - 高优先级问题优先处理
- 定期升级 - 将有价值的学习升级到文档
- 去重合并 - 合并相似的学习记录
- 过期清理 - 删除过时或不相关的内容
🔮 未来计划
v2.0.0 计划功能
- 机器学习驱动的学习建议
- 自动知识图谱构建
- 实时协作学习
- 跨项目学习共享
- 智能学习推荐系统
短期改进
- 更好的错误模式识别
- 自动解决方案生成
- 集成更多工具和服务
- 改进的用户界面
✨ 持续学习，持续改进，让AI助手越来越聪明！
