reasoningbank-intelligence
Build Self-Learning AI Agents with ReasoningBank
๋ํ ๋ค์์์ ์ฌ์ฉํ ์ ์์ต๋๋ค: DNYoussef
AI agents waste time repeating the same suboptimal approaches. ReasoningBank enables agents to learn from experience, recognize patterns, and continuously improve their strategies over time.
์คํฌ ZIP ๋ค์ด๋ก๋
Claude์์ ์ ๋ก๋
์ค์ โ ๊ธฐ๋ฅ โ ์คํฌ โ ์คํฌ ์ ๋ก๋๋ก ์ด๋
ํ ๊ธ์ ์ผ๊ณ ์ฌ์ฉ ์์
ํ ์คํธํด ๋ณด๊ธฐ
"reasoningbank-intelligence" ์ฌ์ฉ ์ค์ ๋๋ค. Recommend strategy for code review of TypeScript React component
์์ ๊ฒฐ๊ณผ:
- Best Strategy: component-first-approach (Score: 0.89)
- - Start with component structure analysis
- - Check prop types and state management
- - Review hooks usage patterns
- - Validate JSX best practices
- Based on 127 similar reviews with 94% success rate
"reasoningbank-intelligence" ์ฌ์ฉ ์ค์ ๋๋ค. What patterns exist in my debugging approaches?
์์ ๊ฒฐ๊ณผ:
- Pattern identified: reproduce-before-fix methodology
- - Applied in 89% of successful debugging sessions
- - Confidence: 0.92
- - Best for: production incidents, complex bugs
- - Related pattern: logging-first delays resolution by 2.3x
๋ณด์ ๊ฐ์ฌ
์์ This skill is pure documentation with TypeScript code examples. Static scanner flagged patterns that are false positives: GitHub URLs in metadata, example code blocks, and standard database integration patterns. No executable code, no credential exfiltration, no command injection risks. All findings dismissed as safe usage.
์ํ ์์ธ
๐ ๋คํธ์ํฌ ์ ๊ทผ (1)
๐ ํ์ผ ์์คํ ์ก์ธ์ค (1)
โ๏ธ ์ธ๋ถ ๋ช ๋ น์ด (19)
ํ์ง ์ ์
๋ง๋ค ์ ์๋ ๊ฒ
Build Self-Improving Agents
Create agents that learn from each task and automatically optimize their approach for better results.
Optimize Multi-Step Processes
Record outcomes from complex workflows and identify the most effective sequence of steps.
Implement Meta-Cognitive Systems
Build systems that can reflect on their own learning process and improve how they learn.
์ด ํ๋กฌํํธ๋ฅผ ์ฌ์ฉํด ๋ณด์ธ์
Initialize ReasoningBank with AgentDB persistence and set up pattern learning for my code review agent with 0.1 learning rate.
Record this experience: successful API optimization that reduced response time by 40% using caching strategy on Node.js Express app.
Recommend the best strategy for debugging a memory leak in a production Node.js microservice based on past experiences.
Analyze patterns in deployment failures over the last 30 days and suggest preventive measures with confidence scores.
๋ชจ๋ฒ ์ฌ๋ก
- Record both successes and failures to build balanced learning data
- Include rich context like environment, constraints, and relevant metadata
- Set confidence thresholds to filter low-quality learnings
ํผํ๊ธฐ
- Only recording successful outcomes creates biased learning
- Storing experiences without context reduces pattern matching accuracy
- Never learning from failures misses valuable optimization opportunities
์์ฃผ ๋ฌป๋ ์ง๋ฌธ
Is this compatible with all Claude versions?
How much data is needed before recommendations improve?
Can I integrate this with my existing agent system?
Is my learning data secure?
What if recommendations are poor quality?
How does this compare to simple logging?
๊ฐ๋ฐ์ ์ธ๋ถ ์ ๋ณด
์์ฑ์
ruvnet๋ผ์ด์ ์ค
MIT
๋ฆฌํฌ์งํ ๋ฆฌ
https://github.com/ruvnet/claude-flow/tree/main/.claude/skills/reasoningbank-intelligence์ฐธ์กฐ
main
ํ์ผ ๊ตฌ์กฐ
๐ SKILL.md