summoner

Production-Ready Skills for Claude Code

![Status]() ![Version]()  

ClaudeShack is a curated collection of powerful Claude Code skills designed to enhance productivity, ensure code quality, and enable intelligent multi-agent workflows - all while maintaining privacy, minimizing context bloat, and preventing hallucinations.

🎯 Core Principles

• •Privacy First: No PII collection, opt-in telemetry, transparent data handling
• •Minimal Context: Focused context passing to subagents, no full conversation dumps
• •Facts Over Fiction: Documentation verified against code, no hallucinations
• •Read-Only Subagents: Analysis and suggestions only, modifications require user approval
• •Community Driven: GitHub-native feedback, telemetry-informed improvements

📦 Available Skills

🧠 Oracle (Project Memory & Learning)

Project memory system that tracks interactions, learns from corrections, and maintains knowledge across sessions.

Core Features:

• •Session recording and timeline tracking
• •Learn from mistakes and corrections
• •Strategic context injection (KISS - load only what's needed)
• •Pattern detection for automation opportunities
• •Token-efficient knowledge storage

Use When: Remembering project patterns, avoiding repeated mistakes, maintaining institutional knowledge

Location: skills/oracle/

Integrates With: All skills (provides memory and learning foundation)

🛡️ Guardian (Quality Gates & Session Health)

Automatic quality monitoring that spawns focused Haiku agents for code review when degradation detected.

Core Features:

• •Triggers on code volume (>50 lines), repeated errors (3+), file churn (5+ edits)
• •Read-only subagents (analysis only, no modifications)
• •Oracle validation (cross-checks suggestions against known patterns)
• •Learning from feedback (adjusts sensitivity based on acceptance rates)
• •Session health tracking

Use When: Automatic code review, detecting session degradation, task planning for complex work

Location: skills/guardian/

Integrates With: Oracle (+31% suggestion acceptance when active)

🧙 Summoner (Multi-Agent Orchestration)

Coordinates multiple specialized agents for complex, multi-component tasks.

Core Features:

• •Task decomposition into atomic subtasks
• •Mission Control Documents (MCD) as single source of truth
• •Parallel agent execution where dependencies allow
• •Quality gates (DRY, CLEAN, SOLID enforcement)
• •Minimal context passing to each agent

Use When: Tasks with 3+ distinct components, multi-phase execution, complex research coordination

Location: skills/summoner/

Integrates With: Oracle (loads patterns), Guardian (validates quality), Wizard (coordinates research)

📝 Wizard (Documentation Maintenance)

Intelligent documentation that stays accurate through research, fact-checking, and validation.

Core Features:

• •Research-first approach (Oracle + code + conversation history)
• •No-hallucination policy (facts only with references)
• •Spawns read-only research agents via Summoner
• •Guardian validates accuracy
• •Auto-detects outdated documentation

Use When: Updating docs, generating skill documentation, validating accuracy, cross-referencing with code

Location: skills/wizard/

Integrates With: Oracle (knowledge source), Summoner (research coordination), Guardian (doc validation)

🎨 Style Master (CSS & Frontend Styling)

Expert in CSS, design systems, and frontend styling.

Core Features:

• •Codebase style analysis and design token extraction
• •Living style guide generation
• •Modern CSS techniques (container queries, custom properties, logical properties)
• •Accessibility compliance (WCAG AA/AAA)
• •Framework expertise (Tailwind, CSS-in-JS, Sass)

Use When: Design systems, visual consistency, CSS guidance, modernizing legacy styles

Location: skills/style-master/

Integrates With: Oracle (remembers style preferences), Summoner (complex refactors)

📊 Evaluator (Privacy-First Telemetry)

Anonymous, opt-in telemetry and feedback collection for continuous improvement.

Core Features:

• •Opt-in only (disabled by default)
• •No PII collection (anonymous session IDs, daily rotation)
• •Local-first storage (you control what's sent)
• •GitHub-native feedback (issues and projects)
• •Aggregate metrics only (no individual events shared)

Use When: Enabling telemetry to help improve skills, submitting feedback, viewing usage analytics

Location: skills/evaluator/

Integrates With: All skills (tracks usage and acceptance rates)

🚀 Getting Started

Marketplace Installation

Add the ClaudeShack marketplace and install the plugin:

# Add the marketplace
/plugin marketplace add Overlord-Z/ClaudeShack

# Then install the claudeshack plugin from the marketplace
# (Use the /plugin menu to browse and install)

All skills included:

• •oracle - Project memory and learning
• •guardian - Quality gates and session health
• •summoner - Multi-agent orchestration
• •wizard - Documentation maintenance (no hallucinations)
• •style-master - CSS and frontend styling
• •evaluator - Privacy-first telemetry

Manual Installation

Clone this repository to use skills locally:

git clone https://github.com/Overlord-Z/ClaudeShack.git
cd ClaudeShack

Skills are automatically available when Claude Code runs in this directory.

Using Skills

Skills activate automatically based on context, or you can explicitly request them:

Use oracle to remember this pattern
Use guardian to review this code
Use summoner to coordinate [complex multi-component task]
Use wizard to update the documentation
Use style master to analyze our CSS
Use evaluator to view telemetry summary

Initialize in Your Project

After installing ClaudeShack skills, initialize them in your project:

Option 1: Use the slash command

/init-claudeshack

Option 2: Run the initialization script

python /path/to/ClaudeShack/scripts/init_project.py

This will:

• •Create/update claude.md with ClaudeShack context
• •Set up .oracle/ directory structure for project memory
• •Create .guardian/config.json with default thresholds
• •Update .gitignore to exclude project-specific data

What gets created:

your-project/
├── claude.md              # Updated with ClaudeShack context
├── .oracle/               # Project memory (gitignored)
│   ├── knowledge/         # Patterns, corrections, gotchas
│   ├── sessions/          # Session logs
│   └── timeline/          # Project history
├── .guardian/             # Session health (gitignored)
│   └── config.json        # Guardian thresholds
└── .gitignore             # Updated with ClaudeShack entries

See MARKETPLACE.md for detailed installation and usage instructions.

ClaudeShack CLI

Manage skills with the marketplace CLI:

# Set up CLI
export CLAUDESHACK_HOME="/path/to/ClaudeShack"
export PATH="$PATH:$CLAUDESHACK_HOME/marketplace/scripts"

# List all skills
claudeshack list

# Get skill info
claudeshack info oracle

# Verify installation
claudeshack verify

# Show version
claudeshack version

See marketplace/INSTALL.md for complete installation guide.

📁 Repository Structure

ClaudeShack/
├── skills/                          # All ClaudeShack skills
│   ├── oracle/                      # Project memory & learning
│   ├── guardian/                    # Quality gates & session health
│   ├── summoner/                    # Multi-agent orchestration
│   ├── wizard/                      # Documentation maintenance
│   ├── style-master/                # CSS & frontend styling
│   ├── evaluator/                   # Privacy-first telemetry
│   └── documentation-wizard/        # Legacy doc sync (use wizard instead)
├── .github/                         # GitHub configuration
│   ├── ISSUE_TEMPLATE/              # Bug reports, feature requests, feedback
│   └── PULL_REQUEST_TEMPLATE.md     # PR template
├── README.md                        # This file
├── CONTRIBUTING.md                  # Contribution guidelines
├── CODE_OF_CONDUCT.md               # Community standards
└── REPO_SETUP_GUIDE.md              # Public release setup guide

🛠️ Skill Development

Want to create your own skill? Check out:

• •Anthropic's Skill Creator Guide
• •Official Claude Code Skills Documentation

🎨 Creating Templates

ClaudeShack skills are extensible through templates. Add your own without modifying core skills.

Guardian Templates

Create custom review templates in skills/guardian/Templates/:

{
  "name": "API Security Review",
  "description": "Review API endpoints for security issues",
  "version": "1.0.0",
  "focus_areas": [
    "Authentication and authorization",
    "Input validation and sanitization",
    "Rate limiting",
    "SQL injection prevention",
    "API key exposure"
  ],
  "oracle_patterns": [
    "api-security",
    "authentication",
    "rate-limiting"
  ],
  "questions": [
    "Are all endpoints authenticated?",
    "Is input properly validated?",
    "Are API keys in environment variables?"
  ],
  "severity_thresholds": {
    "critical": ["sql_injection", "auth_bypass", "key_exposure"],
    "high": ["missing_auth", "weak_validation"],
    "medium": ["missing_rate_limit", "verbose_errors"]
  }
}

Available Templates:

• •security_review.json - OWASP Top 10 security review
• •performance_review.json - Performance optimization
• •feature_planning.json - Complex task breakdown
• •session_health.json - Session degradation monitoring

Summoner MCD Templates

Create Mission Control Document templates in skills/summoner/References/:

• •Define standard task structures
• •Include quality gate checklists
• •Specify agent specifications
• •Add context boundaries

Custom Slash Commands

Add your own commands in .claude/commands/:

# /mycommand.md

You are helping with [specific task].

## Context
[Provide context for this command]

## Your Task
1. [Step 1]
2. [Step 2]

## Remember
- [Key principle 1]
- [Key principle 2]

See existing commands:

• •/handoff - Session handoff with context preservation

🔗 Skill Synergies

Skills work better together:

• •Guardian + Oracle: +31% suggestion acceptance rate (Oracle provides validation patterns)
• •Wizard + Summoner: Comprehensive docs through coordinated research agents
• •Wizard + Guardian: Validated documentation (Guardian checks accuracy against code)
• •Style Master + Oracle: Consistent styling through remembered preferences
• •Evaluator: Tracks all skill usage and acceptance rates (opt-in)

Example: Guardian reviews code → validates against Oracle patterns → Wizard updates docs → Style Master ensures consistency → Evaluator tracks what worked.

🔄 Community Feedback Loop

ClaudeShack improves through continuous learning from real-world usage. Here's how your feedback shapes development:

How It Works

1. You Use Skills
   ↓
2. Evaluator Tracks (opt-in, anonymous)
   - Which skills are used most
   - Which suggestions are accepted/rejected
   - Session health patterns
   - Integration effectiveness
   ↓
3. Community Submits Feedback
   - GitHub Issues (bug reports, feature requests)
   - Skill Feedback template (what works, what doesn't)
   - Discussions (Q&A, ideas, showcases)
   ↓
4. Data Analysis
   - Aggregate Evaluator metrics (no individual data)
   - Identify patterns in feedback
   - Guardian threshold tuning needs
   - Oracle knowledge gaps
   ↓
5. Skill Improvements
   - Adjust Guardian sensitivity based on acceptance rates
   - Add Oracle patterns from common corrections
   - New Guardian templates for common reviews
   - Summoner MCD refinements
   - Documentation clarifications
   ↓
6. Release Updates
   - Changelog with what changed and why
   - Performance metrics (before/after)
   - Credit to contributors
   ↓
Back to #1 (continuous improvement)

What Gets Improved

From Evaluator Metrics:

• •Guardian thresholds: If suggestion acceptance is low, triggers are too aggressive
• •Oracle patterns: Track which patterns prevent the most errors
• •Skill synergies: Measure which combinations work best
• •Session health: Learn what signals predict degradation

From GitHub Feedback:

• •New templates: Users request reviews for specific domains (database, accessibility, etc.)
• •Documentation gaps: Questions reveal what's unclear
• •Feature requests: What capabilities are missing
• •Bug reports: What's not working as designed

Example Improvements

Real scenario:

Week 1: Evaluator shows Guardian triggers 50 times, but only 20% accepted
  ↓
Analysis: Triggering too aggressively, users find it annoying
  ↓
Week 2: Raise lines_threshold from 50 to 75
  ↓
Week 3: Evaluator shows 40 triggers, 65% accepted (much better!)
  ↓
Record in Oracle: "lines_threshold=75 optimal for this codebase type"

Transparency Reports

We publish quarterly reports showing:

• •Total opt-in users (approximate)
• •Aggregate skill usage statistics
• •Top improvement themes from feedback
• •Changes made and why
• •Performance improvements

Next report: Q2 2025

Your Feedback Matters

Submit Feedback:

• •🐛 Bug Report
• •✨ Feature Request
• •⭐ Skill Feedback
• •💬 Join Discussions

Enable Telemetry (opt-in):

# Help improve skills with anonymous usage data
use evaluator to enable telemetry

# View what's collected locally
use evaluator to show summary

What We DON'T Collect:

• •❌ Your code or file contents
• •❌ Project names or paths
• •❌ Personal information (name, email, IP)
• •❌ Conversation history
• •❌ Anything identifiable

What We DO Collect (if you opt in):

• •✅ Skill usage counts (which skills, how often)
• •✅ Success/failure rates (did the skill work?)
• •✅ Suggestion acceptance rates (Guardian only)
• •✅ Session health metrics (aggregate patterns)
• •✅ Performance timings (how fast skills run)

All data is anonymous, aggregate, and local-first. You control what gets sent.

📚 Resources

• •Marketplace: MARKETPLACE.md - Full installation and usage guide
• •Official Skills Repository: anthropics/skills
• •Claude Code Documentation: docs.claude.com
• •Community Skills: awesome-claude-skills

🤝 Contributing

This is a personal skill center, but ideas and suggestions are welcome! Feel free to:

• •Open issues for skill ideas
• •Suggest improvements to existing skills
• •Share your own skill implementations

📄 License

This repository contains custom skills. See individual skill directories for specific licensing.

Built with ❤️ for the Claude Code community