Agent skill
continuous-learning-v2
Instinct-based learning system with confidence scoring. Automatically learns patterns from sessions and evolves them into skills. Supports import/export for team sharing.
Install this agent skill to your Project
npx add-skill https://github.com/WenJunDuan/Rlues/tree/main/vibeCoding/old version/codex/config-agent_v7.9.1-codex/.codex/skills/continuous-learning-v2
SKILL.md
Continuous Learning v2 - Instinct System
Overview
Instincts are micro-patterns learned from coding sessions. Unlike full skills, instincts are:
- Lightweight - Single pattern, minimal context
- Confidence-scored - Track success rate
- Evolvable - Cluster into skills when mature
Instinct Structure
{
"id": "inst_abc123",
"pattern": "When encountering CORS errors in Next.js API routes...",
"solution": "Add headers object with Access-Control-Allow-* fields",
"confidence": 0.85,
"uses": 12,
"successes": 10,
"created": "2026-01-15T10:30:00Z",
"lastUsed": "2026-01-28T14:20:00Z",
"tags": ["nextjs", "api", "cors", "debugging"],
"context": {
"framework": "next.js",
"version": "15.x",
"category": "debugging"
}
}
Commands
/instinct-status
View learned instincts with confidence scores:
๐ Instinct Status
Total: 47 instincts | Avg Confidence: 0.78
High Confidence (>0.8):
โ
CORS handling in Next.js API [0.92] - 15 uses
โ
Prisma transaction patterns [0.88] - 8 uses
โ
React useEffect cleanup [0.85] - 23 uses
Medium Confidence (0.5-0.8):
โก Supabase RLS policies [0.72] - 5 uses
โก Tailwind responsive patterns [0.68] - 7 uses
Low Confidence (<0.5):
โ Edge function cold starts [0.45] - 2 uses
/instinct-export
Export instincts for sharing:
# Export all instincts
/instinct-export
# Export by tag
/instinct-export --tags=nextjs,react
# Export high confidence only
/instinct-export --min-confidence=0.8
Output: .ai_state/instincts/export-2026-01-28.json
/instinct-import <file>
Import instincts from team:
/instinct-import shared-instincts.json
Imported instincts start with 0.5 confidence and adjust based on local use.
/evolve
Cluster related instincts into a skill:
# Interactive evolution
/evolve
# Target specific tags
/evolve --tags=authentication
Learning Triggers
Instincts are captured when:
- Successful Debugging - Problem โ Solution pattern
- Pattern Recognition - Repeated similar solutions
- User Confirmation - Explicit "remember this" signals
- Code Review Feedback - Accepted improvements
Confidence Scoring
confidence = successes / uses * decay_factor
where:
decay_factor = 0.95^(days_since_last_use / 30)
Confidence increases with successful uses, decreases with failures or time.
Storage
.ai_state/instincts/
โโโ instincts.json # Main instinct database
โโโ index.md # Human-readable index
โโโ exports/ # Export history
โ โโโ export-*.json
โโโ evolved/ # Evolved skills
โโโ skill-*.md
Evolution Process
When instincts cluster around a topic:
1. Identify Related Instincts
- Same tags (>3 instincts)
- Similar patterns
- High combined confidence
2. Generate Skill Draft
- Merge patterns
- Synthesize solutions
- Create SKILL.md
3. User Review
- Present draft
- Cunzhi confirmation
- Install or iterate
4. Deprecate Instincts
- Mark as "evolved"
- Link to new skill
Integration
With Experience Skill
Instincts โ lightweight, auto-captured
Experience โ heavyweight, manually curated
Workflow:
1. Instinct captured automatically
2. High-confidence instincts โ candidate for experience
3. User confirms โ promote to experience
With /learn Command
/learn # Capture to instincts (default)
/learn --experience # Capture to experience (manual)
/learn --dry-run # Preview without saving
Quality Filters
Instincts must meet criteria:
- Non-trivial - Not basic language features
- Reusable - Applies to multiple contexts
- Specific - Clear trigger and solution
- Tested - At least one successful use
Team Sharing Best Practices
# Export team-relevant instincts
/instinct-export --min-confidence=0.7 --tags=our-stack
# Import with namespace
/instinct-import team-patterns.json --namespace=team
# Review imported before trusting
/instinct-status --namespace=team
Recommended Agent Skills
Expand your agent's capabilities with these related and highly-rated skills.
agent-teams
Path D ๅนถ่ก็ผๆ โ ๅญไปฃ็ๅๅทฅ + Agent Teams ๅไฝใ้่ฆ CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1
tdd
Path ๅ็บง TDD ็ญ็ฅ
context7
E ้ถๆฎตๆ้ๆๅๅบๆๆกฃ โ ็จ ctx7 ่ทๅๅ็กฎ API ไฟกๆฏ
knowledge
่ทจไผ่ฏ็ป้ช็ฎก็ โ ๅจ .ai_state ๅ ๆฒๆทๅๅค็จ้กน็ฎ็ป้ช
code-quality
Rev ้ถๆฎต Plugin ็ผๆ้กบๅบ
verification
V ้ถๆฎต Path ๅ็บง้ช่ฏๆธ ๅ
Didn't find tool you were looking for?