adaptive-learning-agents（自适应学习 Agent）

Adaptive Learning Agent

Learn from errors and corrections in real-time. Continuously improve by capturing failures, user feedback, and successful patterns.

Free and open-source (MIT License) • Zero dependencies • Works locally

🚀 Why This Skill?

Problem Statement

Working with Claude or any AI agent means encountering:

• Mistakes that need correction
• Unexpected API behaviors
• Better approaches discovered through experimentation
• Knowledge gaps that get revealed during use

But there's no systematic way to learn from these moments and apply the knowledge next time.

The Solution

Adaptive Learning Agent captures every error, correction, and successful pattern automatically. Then retrieves relevant learnings before tackling similar problems again.

Real Use Cases

• Bug discovery: Record an error once, never struggle with it again
• Prompt optimization: Keep track of what prompt variations work best
• API integration: Remember quirky behaviors and workarounds
• Workflow improvement: Document shortcuts and best practices
• Team knowledge: Export and share learnings across projects

✨ What You Get

Four Core Functions

1. Record Learnings

agent.record_learning(
    content="Use claude-sonnet for 90% of tasks—faster and cheaper",
    category="technique",
    context="Model selection"
)

Capture successful patterns, insights, and best practices.

2. Record Errors

agent.record_error(
    error_description="JSON parsing failed on null values",
    context="Processing API response",
    solution="Add null check before parsing"
)

Document failures and solutions automatically.

3. Search & Retrieve Learnings

results = agent.search_learnings("JSON parsing")
recent = agent.get_recent_learnings(limit=5)
by_category = agent.get_learnings_by_category("bug-fix")

Find relevant knowledge instantly when you need it.

4. View Summaries

summary = agent.get_learning_summary()
print(agent.format_learning_summary())

Understand what you've learned at a glance.

Key Features

✅ Zero dependencies - Pure Python, works everywhere ✅ Local-only storage - All data on your machine, no uploads ✅ MIT Licensed - Free to use, modify, fork, redistribute ✅ Automatic categorization - Errors become learnings ✅ Search and filter - Find knowledge by keyword or category ✅ Export capability - Share learnings as JSON ✅ No API keys - Works without any external credentials

📊 Real-World Example

from adaptive_learning_agent import AdaptiveLearningAgent

# Initialize agent
agent = AdaptiveLearningAgent()

# Day 1: Discover a bug
agent.record_error(
    error_description="Anthropic API rejects prompts with excessive newlines",
    context="Testing prompt with formatted lists",
    solution="Use \\
.strip() to clean whitespace before sending"
)

# Day 2: Same bug, but now you have the solution
similar_errors = agent.search_learnings("newlines")
# Result: [Previous learning with solution] ✅

# Week 1: Document successful pattern
agent.record_learning(
    content="Always use temperature=0 for deterministic output in tests",
    category="best-practice",
    context="Prompt engineering"
)

# Get weekly summary
summary = agent.get_learning_summary()
print(f"You've recorded {summary['total_learnings']} learnings this week!")
print(f"Resolved {summary['error_statistics']['resolved']} errors")

🔧 Installation

No installation needed! The skill is pure Python with zero dependencies.

# Copy the adaptive_learning_agent.py file to your project
# Or import it directly:

from adaptive_learning_agent import AdaptiveLearningAgent

💡 Use Cases

Software Development

Record bugs you find and their fixes. Next time you hit a similar error, you have the solution ready.

agent.record_error(
    error_description="Port 8000 already in use",
    context="Running local dev server",
    solution="Use `lsof -i :8000` to find process, then kill it"
)

Prompt Engineering

Keep track of prompting techniques that work for your specific use cases.

agent.record_learning(
    content="Chain-of-thought works better for math problems, direct answers for facts",
    category="technique"
)

API Integration

Remember quirky behaviors and workarounds for each provider.

agent.record_learning(
    content="OpenAI API requires explicit 'assistant' role messages",
    category="api-endpoint",
    context="Chat completion endpoint"
)

Team Knowledge

Export learnings and share with your team or future projects.

agent.export_learnings("team_learnings.json")
# Share this file with teammates

Continuous Improvement

Before major tasks, review what you've learned to avoid repeating mistakes.

summary = agent.get_learning_summary()
unresolved = summary['error_statistics']['unresolved']
if unresolved > 0:
    print(f"⚠️ {unresolved} unresolved errors—review before proceeding")

📚 Categories

When recording learnings, choose from these categories:

🎯 Sources

When recording learnings, specify the source:

• user-correction - User told you something was wrong
• error-discovery - You found the solution to an error
• successful-pattern - You discovered something that works well
• user-feedback - User suggested an improvement

📖 API Reference

Core Methods

record_learning(content, category, source, context)

Record a successful pattern or insight.

Parameters:

• content (str, required): What was learned
• category (str): One of the category types above
• source (str): One of the source types above
• context (str): Optional context about where this applies

Returns: Learning object with ID and timestamp

record_error(error_description, context, solution, prevention_tip)

Record an error and optionally its solution.

Parameters:

• error_description (str, required): What went wrong
• context (str, required): What was being attempted
• solution (str): How to fix it
• prevention_tip (str): How to avoid it

Returns: Error object with ID

search_learnings(query)

Search learnings by keyword or category.

Parameters:

• query (str): Search term

Returns: List of matching Learning objects (sorted by relevance)

get_recent_learnings(limit)

Get the most recent learnings.

Parameters:

• limit (int): Number to return (default: 10)

Returns: List of Learning objects, newest first

get_learning_summary()

Get comprehensive summary of learnings and errors.

Returns: Dictionary with statistics and recent items

export_learnings(output_file)

Export all learnings and errors to JSON file.

Parameters:

• output_file (str): Path to save JSON (default: "learnings_export.json")

🔒 Privacy & Security

• ✅ Zero telemetry - No data sent anywhere
• ✅ Local-only storage - Everything stored in .adaptive_learning/ on your machine
• ✅ No API calls - Works completely offline
• ✅ No authentication - No accounts, keys, or logins needed
• ✅ Full transparency - Source code included and open-source

🤝 Contributing

This is MIT Licensed and community-maintained. You're encouraged to:

• Fork the repository
• Submit improvements and features
• Integrate it into your projects
• Share learnings with others

📝 Changelog

[1.0.0] - 2026-02-14

✨ Initial Release

• Core learning system - Record and retrieve learnings
• Error tracking - Capture errors with solutions
• Search functionality - Find learnings by keyword or category
• Local storage - All data stays on your machine
• Export capability - Share learnings as JSON files
• Zero dependencies - Pure Python, no external packages
• MIT Licensed - Free to use, modify, redistribute
• Comprehensive API - Simple, Pythonic interface

📞 Support

• GitHub: https://github.com/clawhub-skills/adaptive-learning-agent
• Issues & Contributions: Open an issue or PR on GitHub
• Community: Share your learnings and improvements!

📄 License

MIT License - Free and open-source

Use, modify, fork, and redistribute freely. See LICENSE.md for full details.

Copyright © 2026 UnisAI Community

Last Updated: February 14, 2026 Current Version: 1.0.0 Status: Active & Community-Maintained

Free to use, modify, and fork. No restrictions.