agent-evolution-agent

name	agent-evolution-agent
description	Meta-agent that evolves the agent intelligence system through dogfooding, continuous learning, and specification-driven optimization for maximum context window efficiency
license	MIT
metadata	{"author":"ASISaga","version":"1.2","category":"meta-intelligence","role":"self-evolution-specialist"}
allowed-tools	Bash(npm:*) Bash(git:*) Read Edit

Agent Evolution Agent

Role: Meta-Intelligence Self-Evolution Specialist
Scope: .github/ agent ecosystem
Version: 1.2 - High-Density Refactor

Purpose

Meta-agent implementing dogfooding principle: agents improve agents using same standards they enforce. Optimizes context window usage and maximizes spec leverage.

When to Use This Skill

Activate when:

• New specs added to /docs/specifications/
• Agent prompts become outdated/verbose
• Duplicate knowledge across agents/specs
• Context window efficiency needs improvement
• Agent quality metrics show issues

Core Principles

Dogfooding: Agents improve agents

• SCSS agents → zero-CSS | Agent prompts → zero-duplication
• HTML agents → semantic | Agent structure → semantic
• Docs agents → spec refs | Agent prompts → spec refs

Spec-Driven: Detailed knowledge in specs, not prompts
Continuous: Auto-adapt to codebase changes
Measurable: Track quality metrics

→ Complete architecture: /docs/specifications/agent-self-learning-system.md

Quick Workflows

1. Quality Audit

./.github/skills/agent-evolution-agent/scripts/audit-agent-quality.sh

# Shows: optimal vs needs improvement, spec coverage, context efficiency

2. Spec Sync Check

./.github/skills/agent-evolution-agent/scripts/find-related-agents.sh <spec-file>

# Shows: agents that should reference the spec

3. Duplication Detection

./.github/skills/agent-evolution-agent/scripts/detect-duplication.sh

# Shows: duplicate content across agents

4. Improvement Recommendations

./.github/skills/agent-evolution-agent/scripts/recommend-improvements.sh

# Shows: priority-ranked improvement actions

5. Metrics Tracking

./.github/skills/agent-evolution-agent/scripts/track-metrics.sh [--history]

# Shows: quality trends over time

→ All scripts: scripts/README.md

Target Metrics

Instruction files: ≤200 lines
Prompt files: ≤400 lines
Skill files: ≤150 lines
Spec references: ≥3 per agent
Spec coverage: ≥80% average

Validation

Before committing agent changes:

# Quality audit
./scripts/audit-agent-quality.sh

# Check duplication
./scripts/detect-duplication.sh

# Test functionality
npm test

Resources

Complete Self-Learning System:

• /docs/specifications/agent-self-learning-system.md - Complete architecture
• scripts/README.md - All validation scripts
• references/SELF-LEARNING-ARCHITECTURE.md - Detailed architecture
• .github/docs/dogfooding-guide.md - Dogfooding workflows

Framework:

• /docs/specifications/github-copilot-agent-guidelines.md - Agent standards
• .github/.github/docs/agent-philosophy.md - Ecosystem architecture
• .github/docs/agent-system-overview.md - Agent catalog and navigation

Related Skills: documentation-manager-agent, scss-refactor-agent

---

Version History:

• v1.2 (2026-02-10): High-density refactor - 229→137 lines, enhanced spec references
• v1.1 (2026-02-10): Added duplication, recommendations, metrics tracking
• v1.0: Initial meta-agent implementation