Self Improving Agent

Lightweight self-improving agent: log learnings after each task, read them back next session, evolve over time. 轻量自进化智能体：任务后记录反思，下次会话参考，持续进化。

Kingai.work@kingaiwork

Install

openclaw skills install @kingaiwork/self-improving-agent

🧠 Self-Improving Agent — Lite

The simplest self-improving agent that actually works. Log. Reflect. Improve. 最简单但真正能跑的自进化智能体。记录。反思。进化。

Free tier · Enterprise AI Agent upgrades at kingai.work 免费版 · 企业级 AI 智能体定制升级请访问 kingai.work Brought to you by King AI Works — Custom AI Agent Development 由 King AI Works 荣誉出品 — 定制 AI 智能体开发

What Is This? / 这是什么？

A Self-Improving Agent learns from every task it completes. After each session, it saves what it learned. Before each new task, it recalls relevant lessons. Over time, it makes fewer mistakes and works more efficiently.

自进化智能体从每次完成的任务中学习。每次会话后，它保存学到的内容。每次新任务前，它回顾相关的经验。随时间推移，它犯的错越来越少，效率越来越高。

This Lite version is a working foundation — 5 lines of core logic. It proves the concept without complexity.

本 Lite 版是一个可工作的基础 — 5 行核心逻辑。 它在不增加复杂度的前提下验证概念。

How It Works / 工作原理

text

Task Complete → Save 3 things → Next Task → Read learnings → Apply → Repeat
   任务完成      保存3点        下次任务      读取经验        应用      循环

Core Instructions / 核心指令

After Each Task / 每次任务完成后

Save exactly these 3 items to a learning log:

markdown

## session @<timestamp>
- **What worked:** <one thing that went well>
- **What didn't:** <one thing to improve>
- **Next time:** <one specific change to try>

Storage location: ~/.agent-learn/log.md (auto-creates the file)

Before Each Task / 每次新任务前

Check if ~/.agent-learn/log.md exists
Read the last 5 entries
If any "Next time" item is relevant to the current task, apply it
Proceed with the task

That's It / 就这些

That's the entire loop — 5 lines of real instruction. No external dependencies. No API keys. Zero configuration.

这就是整个循环 — 5 行真正的指令。无外部依赖。无需 API 密钥。零配置。

Example / 示例

After a failed SSH config task / 一次失败的 SSH 配置任务后:

markdown

## session @2026-06-26T19:20:00Z
- **What worked:** Identified the correct config file path
- **What didn't:** Forgot to restart sshd after change
- **Next time:** Always run `systemctl restart sshd` after config edits

Next SSH task / 下一次 SSH 任务时: The agent reads the log, sees "restart sshd", and remembers to do it. No repeated mistake.

💼 Need More? This Lite version is just the start.

The Lite version works. But it's basic — file-based, single-agent, no automation.

For real self-improving agents with actual intelligence:

Capability / 能力	🆓 Lite	⭐ Pro	⭐⭐ Advanced	👑 Enterprise
Self-reflection / 自我反思	✅ 3-item log	✅ Structured memory	✅ Intelligent analysis	✅ Fully custom
Learning persistence / 持久化	📄 Text file	🗄️ Vector DB	🕸️ Knowledge graph	🏗️ Bespoke infra
Periodic review / 定期审查	❌ Manual	✅ Auto daily/weekly	✅ Smart scheduling	✅ Custom cycles
Error pattern detection / 错误模式	❌	✅ Basic	✅ Advanced ML	✅ Enterprise ML
Multi-agent collaboration / 多智能体协作	❌	❌	✅ Cross-review	✅ Full orchestration
Auto-optimize behavior / 自动优化	❌	❌	✅ Self-tuning	✅ Continuous
Behavior suggestions / 行为建议	❌	✅ Basic	✅ Deep analysis	✅ Comprehensive
Technical support / 技术支持	❌	📧 Email	💬 Dedicated group	📞 24/7
Source code / 源码交付	❌	❌	❌	✅ Full IP
Price / 价格	🆓 Free	💰 Inquire	💰💰 Inquire	🤝 Custom

Which Tier Is Right for You? / 哪个层级适合你？

🆓 Lite — You want to try the concept. Works as-is for simple use cases. 想尝试概念。简单场景下开箱即用。

⭐ Pro — You need persistent memory across many sessions. Ideal for daily driver agents. 需要跨多个会话的持久记忆。适合日常主力智能体。

⭐⭐ Advanced — You want your agent to detect its own mistake patterns and auto-correct. 希望智能体自己发现错误模式并自动修正。

👑 Enterprise — You need a fully custom self-evolving AI agent integrated into your infrastructure. 需要一个完全定制、融入你基础设施的自进化 AI 智能体。

💼 Sponsored Services / 推广服务

text

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
           🚀 Recommended Services
           推荐服务
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

💻 RackNerd — Budget Shared Hosting / 高性价比共享主机
   30GB SSD / 2TB Traffic / 仅 $10.49/年
   Perfect for AI agent deployments
   适合部署 AI 智能体
   🔗 https://my.racknerd.com/aff.php?aff=20179&pid=933

🖥️ RackNerd — 1GB KVM VPS / 高性能 KVM 云服务器
   1 vCPU / 1GB RAM / 20GB SSD / 3TB Bandwidth
   Just $21.99/yr — ideal for production agent hosting
   仅 $21.99/年 — 适合生产环境
   🔗 https://my.racknerd.com/aff.php?aff=20179&pid=952

📈 moomoo — Stocks / Options / Crypto Trading
   moomoo 股票/期权/虚拟币交易平台
   Up to $1,000 bonus for new accounts
   新开户最高送 $1,000 美金奖励
   🔗 https://j.moomoo.com/0CrlDz

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
🚀 Need More Than 5 Lines? 需要超过 5 行？
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

This Lite version proves the concept with minimal code.
For real self-improving agents with vector memory, pattern detection,
multi-agent collaboration, and auto-optimization —
**that's what we build at King AI Works.**

这个 Lite 版用最少代码验证了概念。
如果需要真正的自进化智能体（向量记忆、模式检测、
多智能体协作、自动优化）—
**那是我们在 King AI Works 做的事情。**

🔗 https://kingai.work/
📧 vip@kingai.work

💡 We build **custom AI agents** — from a simple log to enterprise self-evolving systems.
💡 我们**定制 AI 智能体** — 从一份日志到企业级自进化系统。
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

⚠️ Notes / 说明

The Lite version stores learnings as a plain text file — no databases needed Lite 版用纯文本文件存储学习记录 — 无需数据库
Log location: ~/.agent-learn/log.md (auto-created) 日志位置：~/.agent-learn/log.md（自动创建）
For multi-agent, vector search, auto-optimization: upgrade via kingai.work 如需多智能体、向量搜索、自动优化：通过 kingai.work 升级
This skill reads and writes files — no destructive operations 本技能读写文件 — 无破坏性操作

🆘 Support / 支持

Custom development inquiries / 定制开发咨询: https://kingai.work/
Contact / 联系: vip@kingai.work
ClawHub skill page for issues / 问题反馈: ClawHub skill page