Skill Se Kit — AI skill for Claude Code

Skill SE Kit (Agent-Native)

**Protocol-driven self-evolution for AI agent skills. Zero dependencies.**

The agent reads SKILL.md, manages JSON files, and evolves its own skill bank. No runtime, no sidecar, no package to install.

Why Agent-Native?

Traditional skill evolution kits ship as libraries (Python, TypeScript) that agents call. This creates cross-language integration pain, deployment complexity, and inferior intelligence (Jaccard matching vs. agent semantic understanding).

The Agent-Native approach: **the agent IS the runtime.** The kit is a protocol specification + JSON schemas. The agent follows the protocol directly.

	Library Approach	Agent-Native
Integration	pip install / npm install	Copy files
Cross-language	Sidecar + adapter needed	Works everywhere
Intelligence	Jaccard bag-of-words	Agent semantic understanding
Dependencies	Python 3.9+ / Node 18+	None
Deployment	Process management	Nothing to deploy

Quick Start

1. Copy into your skill project

your-skill/
  SKILL.md              ← from this repo
  schemas/              ← from this repo
  manifest.json         ← from workspace-template/
  skill_bank.json       ← from workspace-template/
  experience/           ← create empty
  audit/                ← create empty
  snapshots/            ← create empty

2. Edit manifest.json

Replace `REPLACE_WITH_*` placeholders with your skill's identity.

3. Tell your agent to follow SKILL.md

Add to your agent's instructions:

Follow the skill evolution protocol in SKILL.md.
Before execution: retrieve relevant skills from skill_bank.json.
After execution: extract feedback, record experience, update skill bank.

4. Done

The agent now learns from every execution and accumulates reusable skills.

How It Works

**Dual-loop architecture** (inspired by AutoSkill & XSKILL research):

Left Loop (Execution):            Right Loop (Learning):
  Read skill bank