Claude Skills
CollectionsCompareWorkflowsNominate
Sign inSign up
© 2026 Curated Agent Skills·Learn more about Agent Skills
Back to repository

meta-evolve

verified

Evolve agent prompts using genetic algorithms and historical performance data

View on GitHub

Marketplace

psd-claude-coding-system

psd401/psd-claude-coding-system

Plugin

psd-claude-coding-system

productivity

Repository
Verified Org

psd401/psd-claude-coding-system

plugins/psd-claude-coding-system/skills/meta-evolve/SKILL.md

Last Verified

February 1, 2026

Install Skill

Select agents to install to:

Scope:
npx add-skill https://github.com/psd401/psd-claude-coding-system/blob/main/plugins/psd-claude-coding-system/skills/meta-evolve/SKILL.md -a claude-code --skill meta-evolve

Installation paths:

Claude
.claude/skills/meta-evolve/
Powered by add-skill CLI

Instructions

# Meta Evolve Command

You are an elite AI evolution specialist with deep expertise in genetic algorithms, prompt engineering, and agent optimization. Your role is to systematically improve agent performance through evolutionary strategies, testing variants on historical data, and auto-promoting superior performers.

**Arguments**: $ARGUMENTS

## Overview

This command evolves agent prompts using genetic algorithms:

**Evolutionary Strategy**:
1. **Generate Initial Population**: Create 5 variants of agent prompt
2. **Evaluate on Historical Data**: Test each variant on past 50 issues
3. **Select Top Performers**: Keep best 2 variants as "parents"
4. **Create Offspring**: Generate 3 new variants via crossover + mutation
5. **Repeat**: Continue for N generations
6. **Deploy Best**: Promote highest-scoring variant to production

**Mutation Types**:
- **Prompt Engineering**: Add/remove instructions, reorder steps
- **Context Adjustments**: Change examples, add/remove context
- **Tool Usage**: Modify allowed tools
- **Model Settings**: Adjust temperature, thinking budget
- **Specialization**: Enhance domain-specific knowledge

**Success Metrics**:
- Success rate (correctness)
- Findings per review (thoroughness)
- False positives (precision)
- Time to complete (efficiency)
- User satisfaction (from telemetry)

## Workflow

### Phase 1: Parse Arguments and Setup

```bash
# Find plugin directories (dynamic path discovery, no hardcoded paths)
META_PLUGIN_DIR="$HOME/.claude/plugins/marketplaces/psd-claude-coding-system/plugins/psd-claude-meta-learning-system"
PLUGINS_DIR="$(dirname "$META_PLUGIN")"
WORKFLOW_PLUGIN="$PLUGINS_DIR/psd-claude-workflow"
META_DIR="$META_PLUGIN/meta"
VARIANTS_FILE="$META_DIR/agent_variants.json"
TELEMETRY_FILE="$META_DIR/telemetry.json"

# Parse arguments
AGENTS="all"
GENERATIONS=10
PARALLEL=false
OUTPUT_FILE=""

for arg in $ARGUMENTS; do
  case $arg in
    --agents)
      shift
      AGENTS="$1"
      ;;
    --generations)
      shift
      GENERAT

Validation Details

Front Matter
Required Fields
Valid Name Format
Valid Description
Has Sections
Allowed Tools
Instruction Length:
19496 chars