Agentic Frameworks and Skills

Production-ready Claude Code skills and frameworks for building, evaluating, and scaling agentic AI systems.

📋 Skills Registry

Complete collection of production-ready skills ordered by importance and usage frequency:

#	Skill	Description	When to Use
1️⃣	Knowledge Framework	Automatic documentation framework using MECE/BFO ontology principles with Mermaid diagrams and Ground Truth attribution	Creating any .md file, documenting systems, building knowledge base
2️⃣	Multi-Agent Orchestrator	Orchestrate parallel execution of multiple CLI agents (Claude Code, Codex, Gemini) for competitive evaluation with objective winner selection	Complex tasks (>7/10), multiple valid approaches, high-stakes solutions
3️⃣	Skill Creator	Guide for creating effective Claude Code skills with YAML frontmatter, Quick Start checklists, and Decision Trees	Building new skills, improving existing skills, skill architecture
4️⃣	Claude Code Sessions	Universal session management - search, resume, and analyze conversations from any directory	Resuming sessions, searching conversation history, tracking work
5️⃣	YouTube to Knowledge Doc	Extract YouTube transcripts and convert to Knowledge Framework documentation with clickable timestamps	Documenting videos, preserving external learning, research archival

🎯 Quick Selection Guide

Choose your skill based on task type:

• 📝 Writing documentation? → Knowledge Framework
• 🤖 Complex task needing best solution? → Multi-Agent Orchestrator
• 🛠️ Building new skills? → Skill Creator
• 🔄 Finding previous work? → Claude Code Sessions
• 🎥 Learning from YouTube? → YouTube to Knowledge Doc

📖 Core Documentation

How to organize documents - Knowledge Framework Guide

Complete guide to MECE/BFO documentation principles - the foundation for all skills in this repository.

🚀 Quick Start

Installing Skills

Global installation (recommended):

# Clone repository
git clone https://github.com/tekliner/improvado-agentic-frameworks-and-skills.git
cd improvado-agentic-frameworks-and-skills

# Install all skills globally
cp -r skills/* ~/.claude/skills/

Project-level installation:

# Copy to specific project
cp -r skills/* /path/to/your-project/.claude/skills/

Installing Dependencies

Some skills require utility scripts and frameworks. Install them to your project:

# Navigate to your project directory
cd /path/to/your-project

# Copy session management utilities (required for claude-code-sessions, knowledge-framework)
mkdir -p data_sources/claude_code
cp -r data_sources/claude_code/* /path/to/your-project/data_sources/claude_code/

# Copy multi-agent framework (required for multi-agent-orchestrator)
mkdir -p algorithms/product_div/Multi_agent_framework
cp -r algorithms/product_div/Multi_agent_framework/* /path/to/your-project/algorithms/product_div/Multi_agent_framework/

# Make scripts executable
chmod +x /path/to/your-project/algorithms/product_div/Multi_agent_framework/*.sh
chmod +x /path/to/your-project/data_sources/claude_code/get_session_id.py

Dependency matrix:

Skill	Dependencies
Knowledge Framework	data_sources/claude_code/get_session_id.py (optional)
Claude Code Sessions	data_sources/claude_code/{get_session_id.py, 21_universal_session_resume.py, 22_list_all_sessions.py} (required)
Multi-Agent Orchestrator	algorithms/product_div/Multi_agent_framework/{run_parallel_agents.sh, run_claude_agent.sh, run_codex_agent.sh, run_gemini_agent.sh} (required)
YouTube to Knowledge Doc	None
Skill Creator	None

Using Skills

Automatic activation - Skills trigger when Claude Code detects relevant context:

• Creating .md files → knowledge-framework
• "Run multi-agent framework" → multi-agent-orchestrator
• "Create new skill" → skill-creator
• "Resume session abc123" → claude-code-sessions
• "Document this YouTube video" → youtube-to-knowledge-doc

Manual invocation:

/skill knowledge-framework
/skill multi-agent-orchestrator
/skill skill-creator
/skill claude-code-sessions
/skill youtube-to-knowledge-doc

🔧 Detailed Skill Information

1️⃣ Knowledge Framework

Purpose: Create structured, maintainable documentation automatically

Key features:

• MECE section organization (Mutually Exclusive, Collectively Exhaustive)
• Continuant (TD) and Occurrent (LR) Mermaid diagrams
• Numbered sections (§1.0, §2.0) and paragraphs (¶1, ¶2)
• Ground Truth attribution with sources and dates
• Quick Start checklist for rapid application
• Author checklist for quality validation

Dependencies: None

Example output:

## 📋 [Title]
**Thesis:** One sentence previewing all MECE sections...

**Overview:** Paragraph introducing each section...

[Mermaid diagrams - structure + process]

## 1.0 First Section
¶1 Ordering principle: [why this order]...

2️⃣ Multi-Agent Orchestrator

Purpose: Competitive evaluation of complex tasks using multiple AI agents

Key features:

• Parallel CLI agent execution (Claude Code, Codex, Gemini)
• Self-evaluation with measurable success criteria (✅/❌/⚠️)
• Automated winner selection based on objective metrics
• Progressive disclosure workflow (location → task file → user edits → launch)
• Artifact placement enforcement (workspace isolation)
• Ready-to-use execution scripts with background monitoring

Dependencies:

• Bash, Python 3.8+
• Multiple CLI agents (Claude Code CLI, Codex CLI, Gemini CLI)
• Scripts: algorithms/product_div/Multi_agent_framework/*.sh (see Installation section)

Workflow:

1. User describes complex task
2. Agree on folder location
3. Create task file with success criteria
4. User edits and confirms "Ready"
5. Launch ./run_parallel_agents.sh
6. Compare self-evaluations
7. Declare winner based on criteria met

Success criteria example:

- [✅] Process 1M rows in <5 seconds
- [✅] Handle edge cases (nulls, duplicates)
- [❌] Memory usage <500MB

3️⃣ Claude Code Sessions

Purpose: Universal session management and conversation history

Key features:

• Resume by session ID or text search
• Universal session discovery across all projects
• Automatic path resolution and project switching
• Shell integration (rc command for quick access)
• Session statistics and metadata
• Search by content, date, or participant

Dependencies:

• Python 3.8+
• Scripts: data_sources/claude_code/{get_session_id.py, 21_universal_session_resume.py, 22_list_all_sessions.py} (see Installation section)

Common commands:

# Resume by ID
rc c080fd31-1fea-44e2

# Search by text
rc --text "dashboard implementation"

# Resume latest
rc --last

# List all sessions
rc-list --days 7

4️⃣ YouTube to Knowledge Doc

Purpose: Preserve external learning as structured documentation

Key features:

• Automatic transcript extraction with yt-dlp
• Intelligent folder placement recommendation
• Clickable YouTube timestamp links (MM:SS → ?t=SECONDS)
• Knowledge Framework compliance (MECE structure, diagrams)
• Ground Truth attribution with video source
• Session ID tracking for provenance

Dependencies: yt-dlp, Python 3.8+

Workflow:

1. User provides YouTube URL
2. Extract transcript and metadata
3. Recommend folder location
4. Generate Knowledge Framework doc
5. Add clickable timestamp citations

Timestamp format:

**Quote:** "Direct quote" ([timestamp 23:11](https://youtu.be/VIDEO_ID?t=1391))

3️⃣ Skill Creator

Purpose: Guide for creating effective Claude Code skills with best practices

Key features:

• YAML frontmatter structure and patterns
• Quick Start Checklist creation (5-10 actionable steps)
• 5-Second Decision Tree design
• Practical Workflow examples
• Progressive disclosure architecture
• Reusable resource identification (scripts, references, assets)

Dependencies: None

Workflow:

1. Define concrete usage examples
2. Identify reusable resources
3. Initialize skill structure (SKILL.md, README.md, references/, scripts/)
4. Write YAML frontmatter with triggers
5. Add Quick Start Checklist
6. Create 5-Second Decision Tree
7. Validate against quality checklist

Quality standards:

• Clear "When to Use" section with automatic triggers
• Actionable Quick Start (not documentation, but steps)
• Decision Tree for rapid skill selection
• Real workflow examples (not abstract descriptions)

🎯 Skill Structure

All skills follow consistent architecture:

skill-name/
├── SKILL.md                 # Main documentation
│   ├── YAML frontmatter (name, description, triggers)
│   ├── When to Use (automatic + manual triggers)
│   ├── Quick Start Checklist (5-10 steps)
│   ├── 5-Second Decision Tree
│   ├── Practical Workflow examples
│   └── Detailed documentation
├── README.md                # Technical implementation
├── references/              # Additional documentation (loaded as needed)
│   ├── script_usage.md
│   └── task_templates.md
└── scripts/                 # Executable helpers
    └── create_task_file.sh

📊 Skills Compatibility

Tested with:

• Claude Code Desktop v1.0+
• Claude Sonnet 4.5 model
• macOS (primary), Linux (compatible)

All skills include:

• ✅ YAML frontmatter with explicit triggers
• ✅ Quick Start Checklist (5-10 steps)
• ✅ 5-Second Decision Tree
• ✅ Practical Workflow examples
• ✅ Clear "When to Use" section
• ✅ Automatic trigger descriptions

🛠️ Development

Skill Creation Guidelines

Follow the skill-creator framework:

1. Define concrete usage examples
2. Identify reusable resources (scripts, references, assets)
3. Initialize skill structure
4. Write SKILL.md with YAML frontmatter
5. Add Quick Start and Decision Tree sections
6. Validate against quality checklist

Quality Standards

MANDATORY for all skills:

• ✅ YAML frontmatter with explicit trigger phrases
• ✅ Quick Start Checklist (actionable steps)
• ✅ 5-Second Decision Tree (rapid selection)
• ✅ Practical Workflow (real examples)
• ✅ Clear "When to Use" section
• ✅ No hardcoded paths or credentials
• ✅ Generic examples (no client-specific data)

🤝 Contributing

Skills in this repository are production-tested and follow strict quality standards.

For improvements:

1. Test changes thoroughly in real projects
2. Update Quick Start/Decision Tree if workflow changes
3. Maintain YAML frontmatter accuracy
4. Add examples for new features
5. Ensure no sensitive data (API keys, paths, client names)

📝 License

Internal use. Not for public distribution without authorization.

🔗 Related Resources

• Knowledge Framework Full Guide: How to organize documents
• Multi-Agent Framework: skills/multi-agent-orchestrator/references/
• Task Templates: Available in respective skill references/ folders

---

Repository: https://github.com/tekliner/improvado-agentic-frameworks-and-skills Last updated: 2025-11-13 Skills count: 5 production-ready skills