self-improving-agent

name	self-improving-agent
description	Enables the AI agent to iteratively improve itself by logging lessons learned, creating new skills from session patterns, and maintaining a living knowledge base.

Self-Improving Agent

The agent that learns from every session and gets better at its job.

Purpose

This skill enables the AI agent to:

1. Log lessons learned from each session into a persistent knowledge base
2. Create new micro-skills when it discovers repeatable patterns
3. Update existing skills when better approaches are found
4. Maintain session retrospectives that feed future improvements
5. Track agent performance metrics over time

When to Use

• At the END of every significant session (after completing a task)
• When discovering a new pattern that could be reused
• When a workaround is found for a recurring problem
• When the agent makes an error and wants to prevent it in the future

Self-Improvement Workflow

Step 1: Session Retrospective

After completing a task, generate a structured retrospective:

## Session Retrospective — [DATE]

### What Worked
- List techniques, approaches, and tools that were effective

### What Didn't Work
- List failed approaches, dead ends, and time wasters

### Lessons Learned
- Distill actionable insights from the session

### Patterns Discovered
- Repeatable patterns that could become skills or workflows

### Knowledge Gaps
- Areas where more information or skills are needed

Save to: <appDataDir>/knowledge/<topic>/artifacts/retrospective_<date>.md

Step 2: Pattern Detection

Analyze retrospectives for recurring patterns:

Pattern Type	Action	Example
Repeated fix	Create a workflow	"Every time X breaks, do Y"
Common research path	Create a knowledge item	"The answer to X is always found at Y"
Tool chain	Create a skill	"To do X, always run A→B→C"
Error resolution	Create a troubleshooting guide	"Error X is caused by Y, fixed by Z"

Step 3: Skill Generation

When a pattern is detected ≥2 times, create a new skill:

# Create skill directory
mkdir -p .agent/skills/<skill-name>

# Generate SKILL.md with:
# - Frontmatter (name, description)
# - When to use
# - Step-by-step instructions
# - Examples from past sessions

Step 4: Knowledge Base Update

Update the agent's knowledge items:

# Create/update knowledge item
mkdir -p <appDataDir>/knowledge/<topic>/artifacts/

# Write metadata.json
{
  "title": "<Topic Title>",
  "summary": "<What was learned>",
  "created": "<timestamp>",
  "lastAccessed": "<timestamp>",
  "references": ["<source files>"]
}

# Write artifact(s)
# - architecture.md, patterns.md, troubleshooting.md, etc.

Step 5: Metrics Tracking

Log performance metrics for continuous improvement:

## Agent Performance Log

| Date | Task | Time | Success | Score | Notes |
|------|------|------|---------|-------|-------|
| 2026-04-02 | SEO Audit | 15min | ✅ | 92 | Used squirrel CLI |
| 2026-04-02 | Schema Fix | 5min | ✅ | 100 | Reused pattern |

Save to: <appDataDir>/knowledge/agent-performance/artifacts/metrics.md

Improvement Categories

1. Shamrock-Specific Patterns

• Wix Velo element ID discovery workflow
• GAS deployment verification steps
• CMS data integrity checks
• Schema markup validation

2. SEO Intelligence

• County page generation optimization
• FAQ content enhancement patterns
• Structured data best practices
• Sitemap discovery techniques

3. System Integration

• Wix ↔ GAS bridge debugging
• SignNow packet validation
• Telegram bot flow testing
• Node-RED cron job monitoring

4. Error Recovery

• Common GAS deployment failures
• Wix sync conflicts
• npm permission issues
• Git merge strategies

Auto-Improvement Triggers

The agent should automatically trigger self-improvement when:

1. ❌ An error occurs → Log the error, root cause, and fix
2. 🔄 A task is repeated → Create a workflow or skill
3. 💡 A new technique is discovered → Update knowledge base
4. 📊 A task takes >30min → Analyze for optimization opportunities
5. 🎯 A score improves → Document what changed and why

Integration with Existing Skills

This skill enhances all other skills by:

• Adding retrospective data to using-superpowers checks
• Feeding audit results into production-code-audit
• Updating wix_gas_bridge_integrity with new failure modes
• Enriching systematic-debugging with resolved cases

Example: Auto-Generated Skill

When the agent notices it has fixed "County County" redundancy 3 times:

---
name: county-name-sanitizer
description: Strip trailing "County" suffix from CMS county names to prevent "Lee County County" redundancy
---

# County Name Sanitizer

## When to Use
When generating county-specific content from CMS data.

## Fix
```javascript
const cleanName = countyName.replace(/\s+County$/i, '');

Files Affected

• county-generator.jsw (line 114)
• Any template using ${countyName} County