Self Improving Agent 中文使用指南 - Skill ZIP 下载与风险提示

输入：任务目标、上下文材料、文件路径、约束条件或需要处理的内容。

输出：按 Skill 说明生成的文档、代码、检查结果、计划、建议或操作步骤。

示例任务

安装方式

风险边界

SkillHub 提供了源站安全报告入口，但本站不替代人工审查。使用前仍需检查权限、外部依赖和敏感数据边界。

SKILL.md 文档介绍

Self-Improving Agent

A continuous learning agent system for OpenClaw that learns from:

✅ Interactions - Learn from every conversation
✅ Errors - Learn from mistakes to avoid them
✅ Recoveries - Learn what works to recover successfully

---

✨ Features

🧠 Continuous Learning - Learns from every interaction
❌ Error Learning - Learns from mistakes to avoid repetition
✅ Recovery Learning - Learns successful recovery patterns
🔄 Auto-Improvement - Automatically applies improvements
📚 Memory System - Stores and retrieves learnings
🔌 Hook System - Extensible hook system for custom improvements
📊 Progress Tracking - Track improvement over time
🔧 Python CLI - Easy to use command-line interface
🧩 OpenClaw Skill - Seamless integration with OpenClaw

---

🚀 Quick Start

Prerequisites

Python 3.10+
OpenClaw installed
pip or uv package manager

Installation

As OpenClaw Skill

# Clone to OpenClaw skills directory
cd ~/.openclaw/workspace/skills
git clone https://github.com/leohuang8688/self-improving-agent.git

Auto-learning is enabled by default! After each OpenClaw session, the agent will automatically learn and improve.

As Python Package

# Clone the repository
git clone https://github.com/leohuang8688/self-improving-agent.git
cd self-improving-agent

# Install with pip
pip install -e .

# Or install with uv
uv pip install -e .

Basic Usage

Automatic Mode (Recommended)

Just use OpenClaw normally! Learning happens automatically after each session, error, or recovery.

# Start OpenClaw
openclaw

# Use it normally...

# Exit OpenClaw
# → Auto-learning triggers automatically!

Manual Mode

# Run the self-improving agent
python -m self_improving_agent run

# Learn from last session
python -m self_improving_agent learn

# Learn from errors
python -m self_improving_agent learn-errors

# Learn from recoveries
python -m self_improving_agent learn-recoveries

# Review all learnings
python -m self_improving_agent review

# Export learnings to file
python -m self_improving_agent export

---

📖 Commands

`run` - Run the Agent

Executes the self-improving agent with all applied improvements.

python -m self_improving_agent run --workspace /path/to/workspace

`learn` - Learn from Session

Analyzes the last session and extracts learnings.

python -m self_improving_agent learn --verbose

`learn-errors` - Learn from Errors

Analyzes error logs and extracts learnings.

python -m self_improving_agent learn-errors --verbose

`learn-recoveries` - Learn from Recoveries

Analyzes recovery logs and extracts successful patterns.

python -m self_improving_agent learn-recoveries --verbose

`review` - Review Learnings

Reviews all stored learnings.

python -m self_improving_agent review --verbose

`export` - Export Learnings

Exports all learnings to a markdown file.

python -m self_improving_agent export

---

🪝 Hook System

Available Hooks

`onSessionEnd(session)`

Triggered: When session ends
Purpose: Post-session learning, cleanup, save state
Parameter: session - Session object

`onError(error)`

Triggered: When an error occurs
Purpose: Error logging, learning from mistakes
Parameter: error - Error object

`onRecovery()`

Triggered: When recovering from error
Purpose: Learn successful recovery patterns
Parameter: None

---

📝 Learning Types

1. Session Learning

Triggered by: onSessionEnd
Learns from: Conversation patterns, user preferences
Stored in: learnings/sessions.json

2. Error Learning

Triggered by: onError
Learns from: Mistakes, errors, failures
Stored in: learnings/errors.json
Purpose: Avoid repeating mistakes

3. Recovery Learning

Triggered by: onRecovery
Learns from: Successful recoveries
Stored in: learnings/recoveries.json
Purpose: Repeat successful recovery patterns

---

📁 Project Structure

self-improving-agent/
├── src/
│   ├── hooks.py              # Hook manager
│   └── ...
├── hooks/
│   └── error_learning.py     # Error learning hook
├── learnings/
│   ├── sessions.json         # Session learnings
│   ├── errors.json           # Error learnings
│   └── recoveries.json       # Recovery learnings
├── main.py
├── README.md
└── SKILL.md

---

🔧 Configuration

In OpenClaw configuration file:

{
  "hooks": {
    "internal": {
      "enabled": true,
      "entries": {
        "self-improve": {
          "enabled": true
        },
        "error-learning": {
          "enabled": true
        }
      }
    }
  }
}

---

📊 Learning Files

`learnings/errors.json`

[
  {
    "timestamp": "2026-03-15T12:30:00",
    "error_type": "ValueError",
    "error_message": "Invalid parameter",
    "context": {
      "traceback": "..."
    }
  }
]

`learnings/recoveries.json`

[
  {
    "timestamp": "2026-03-15T12:35:00",
    "recovery_method": "automatic_recovery"
  }
]

---

🎯 Best Practices

1. Review learnings regularly - Review and apply learnings weekly

2. Export learnings - Export learnings for backup

3. Apply learnings - Apply learnings to improve future performance

4. Clean old learnings - Remove outdated learnings periodically

---

📝 License

MIT License

👨‍💻 Author

PocketAI for Leo - OpenClaw Community

Self Improving Agent

快速判断

适合任务

输入与输出