Skill Evolution is a powerful tool designed to enhance the skills of AI agents through autonomous self-improvement. This framework analyzes past sessions to identify performance gaps and generates structured proposals for improvement. By automating this feedback loop, it allows agents to evolve their capabilities seamlessly, enhancing their overall efficiency and effectiveness.

Key Features

• Self-Improvement Feedback Loop: Skill Evolution reads previous sessions and assesses them in relation to the loaded skills, generating actionable insights for improvement.
• Auto-Apply Options: High-confidence improvements can be auto-applied, streamlining the enhancement process without manual intervention.
• Host-Agnostic Design: It is compatible with multiple hosts such as Hermes and Claude Code, thanks to its extensible HostAdapter interface which facilitates integration with any compatible host.
• Simplicity and Efficiency: The framework operates without daemons, external services, or the need for GPU resources. It relies solely on the host agent, scheduled jobs, and the Python standard library.

Quick Start

To get started with Skill Evolution, follow these simplified steps for both Hermes and Claude Code hosts:

For Hermes:

# Install the skill
hermes skills install https://raw.githubusercontent.com/Carlo1911/skill-evolution/main/SKILL.md

# Load the skill
hermes -s skill-evolution

# Run analysis
echo "Run skill evolution analysis on my recent sessions"

For Claude Code:

1. Clone the repository and load the skill as a directory, ensuring to set the host variable appropriately.
2. The agent will then process unprocessed sessions, analyze skills for gaps, and produce proposal files for review.

Proposal Review and Application

By default, proposals generated are for review only. To enable automatic application of these proposals:

export SKILL_EVOLUTION_AUTO_APPLY=true
export SKILL_EVOLUTION_MIN_CONFIDENCE=0.85

Evaluation Framework

The evaluation framework encompasses multiple independent evaluators that assess proposals against various criteria, including skill correctness, proposal quality, and tool-use efficiency. The automatic proposal application requires passing evaluations under specified conditions to ensure quality improvements.

Project Structure

The structure of the project promotes clarity and maintainability, featuring the main skill file, README documentation, and scripts for various operations such as session fetching, skill indexing, analysis, and quality tracking.

Dependencies

• Python 3.10+ is required.
• Supports either Hermes or Claude Code or any host implementing the HostAdapter interface.
• No additional pip packages are necessary for core functionality, although optional extras like gepa (for optimization) and fastembed (for embedding similarity evaluation) can enhance capabilities.

Skill Evolution empowers AI agents to continually learn and adapt through structured enhancements, ensuring they remain at the forefront of their capabilities.