Skill-Plugin-System

🔌 Skill Plugin System

Teach your AI new abilities with auto-triggered skills

What This Feature Does

Adds a Claude Code plugin system to your AI companion, enabling auto-triggered skills — markdown-based abilities that activate automatically based on conversation context.

• Create custom skills as simple .md files
• Auto-triggering — skills activate when conversation matches their description
• Zero configuration — drop a folder with a SKILL.md and it's live
• Slash commands — user-invocable commands via /command-name
• Leveling system — skills evolve by adding capability layers over time

How Skills Work

The Concept

A skill is a self-contained instruction set written in markdown. When your conversation matches the skill's trigger description, Claude Code automatically loads and follows the skill's protocol.

Think of it as teaching your AI new behaviors through structured markdown files — no code required.

Example: Auto-Save Skill

You: "save my progress"
→ Claude detects this matches the "save" skill description
→ Loads SKILL.md automatically
→ Follows the save protocol step by step
→ Your AI knows exactly what to save and how

Plugin Architecture

Folder Structure

[ai-name]-skills/
├── .claude-plugin/
│   └── plugin.json              # Plugin identity (name, version, author)
├── skills/                      # Auto-triggered behaviors
│   └── [skill-name]/
│       └── SKILL.md             # Complete skill definition
├── commands/                    # User slash commands (optional)
│   └── [command-name].md
└── README.md                    # Plugin documentation

How Auto-Discovery Works

Claude Code automatically finds and registers:

• All SKILL.md files in skills/ → registered as auto-triggered skills
• All .md files in commands/ → registered as slash commands

Folder structure IS the configuration — no index files, no registration tables.

The Three Component Types

Component	Trigger	Use Case
Skills	Auto-triggered by conversation context	Behaviors that should happen automatically
Commands	User types /command-name	Actions the user explicitly invokes
Agents	AI spawns via Task tool	Autonomous background workers

Quick Integration

# Install this feature:
"Load skill-plugin"

What Happens During Integration

1. Asks for your plugin name (defaults to [AI_NAME]-skills)
2. Creates plugin folder structure in your project
3. Creates .claude-plugin/plugin.json manifest
4. Creates a sample skill as starter reference
5. Installs the plugin into Claude Code
6. Creates skill-format.md as permanent template reference
7. Self-deletes this feature folder after successful integration

Post-Integration Result

After running the integration protocol:

• Your AI has a working plugin with a sample skill
• New skills can be added by simply creating folders in skills/
• No build steps, no config changes needed for new skills
• Skills auto-activate based on their description field

Creating Your First Skill

Step 1: Create a Folder

skills/my-skill/

Step 2: Write SKILL.md

---
name: my-skill
description: "MUST use when user says 'do the thing', 'activate my-skill',
             or when [specific context condition]."
---

# My Skill

## Activation
When this skill activates, output:
"Activating my-skill..."

## Protocol
1. [Step 1 - What to do]
2. [Step 2 - What to do next]
3. [Step 3 - Complete and confirm]

## Mandatory Rules
1. [Rule 1]
2. [Rule 2]

Step 3: Done

The skill is live. No registration needed. Claude Code auto-discovers it.

Writing Good Trigger Descriptions

The description field in YAML frontmatter is the most critical element — it determines when Claude activates the skill.

Strong triggers (recommended):

description: "MUST use when committing code changes, when user says 'commit',
             'save changes', 'git commit', or when a task is completed and
             code needs to be preserved."

Weak triggers (avoid):

description: "Helps with code management"  # Too vague — causes false activations

Tips for Trigger Descriptions

• Use "MUST use when..." for strong activation
• List specific trigger phrases the user might say
• Include contextual conditions (not just phrases)
• Be specific enough to avoid false activations
• Be broad enough to catch all relevant situations

Skill Evolution (Leveling)

Skills grow through levels rather than version numbers:

Level	Meaning
Lv.1	Base capability — initial skill definition
Lv.2	Enhanced — added features or removed friction
Lv.3	Advanced — proactive behavior or deep integration
Lv.4+	Expert — fully autonomous with absorbed protocols

Track levels in the ## Level History section of each SKILL.md.

Benefits

• Modular — each skill is independent, add or remove freely
• Readable — skills are plain markdown, human-readable and editable
• Automatic — no manual triggering needed for context-matched skills
• Evolving — skills level up as you refine them through use
• Shareable — share skill files with others using AI MemoryCore

Platform Note

This feature requires Claude Code (Anthropic's CLI tool). The skill plugin system uses Claude Code's native plugin architecture for auto-discovery and triggering.

For other AI platforms, skills can still be used as protocol files loaded via master-memory.md commands — they just won't auto-trigger.

---

Based on the proven alice-enchantments plugin system (20 skills in production)