chrome-devtools

The Problem

You have a large collection of Claude Code resources:

• 249 agents - Specialized AI personas for different domains
• 248 skills - Domain knowledge and code examples
• 216 commands - Workflow instructions and slash commands

Loading all of them into any project blows out your context window. You need a way to:

1. Know what's available without loading everything
2. Pick the right agents/skills/commands for each project
3. Have agents automatically include their related skills and commands
4. Build custom bundles for specific project types

The Solution

| Component | Description |

|-----------|-------------|

| MCP Server | 8 tools for Claude to agentically search, recommend, and install bundles |

| Web GUI | Interactive Bundle Builder with click-to-add and smart suggestions |

| CLI Tool | Command-line interface with rich terminal UI |

| Template Packs | Pre-built bundles for common project types |

Option 1: MCP Server (Recommended for Claude Code/Desktop)

Add to your .mcp.json:

```json

{

"mcpServers": {

"agent-skill-manager": {

"command": "python",

"args": ["C:\\GitHub_Projects\\2026.0109 - Custom Claude Project - Agent Skill Manager\\src\\mcp_server.py"]

}

}

}

```

Restart Claude Code/Desktop, then ask:

> "I'm building a Next.js e-commerce site. What agents and skills should I use?"

Claude will use the MCP tools to search the catalog and provide recommendations.

Option 2: Web GUI (Bundle Builder)

```bash

cd "C:\GitHub_Projects\2026.0109 - Custom Claude Project - Agent Skill Manager"

python launch_gui.py

```

Open http://localhost:8080 in your browser. Click items to add to your bundle, use templates, and export as JSON.

Option 3: CLI Tool

```bash

cd "C:\GitHub_Projects\2026.0109 - Custom Claude Project - Agent Skill Manager\src"

# List available agents

python agent_manager.py list agents

# Search across all items

python agent_manager.py search "frontend"

# Install an agent with related items

python agent_manager.py install frontend-developer --target "C:\path\to\project" --with-related

# Install a pre-built pack

python agent_manager.py install --pack fullstack-web --target "C:\path\to\project"

```

Prerequisites

• Python 3.8+
• pip (Python package manager)

Setup

```bash

# Clone or navigate to the project

cd "C:\GitHub_Projects\2026.0109 - Custom Claude Project - Agent Skill Manager"

# Install dependencies

pip install rich pyyaml

# (Optional) For advanced TUI features

pip install textual

# Verify installation - test the indexes

cd src

python agent_manager.py list agents --limit 5

```

Building Indexes (First Time Only)

If indexes don't exist, build them:

```bash

cd src

python indexer.py # Creates agent/skill/command indexes

python mapper.py # Creates relationship mappings

```

MCP Server Tools

The MCP server provides 8 tools for agentic workflows:

| Tool | Description |

|------|-------------|

| search_catalog | Search agents/skills/commands by name, description, or category |

| get_recommendations | AI-powered suggestions based on project description |

| get_item_details | Get full details including related items |

| create_bundle | Create and save custom bundles |

| install_bundle | Install a bundle to a target project |

| install_items | Install specific items to a project |

| list_packs | List available pre-built packs |

| get_catalog_stats | View catalog statistics |

| analyze_project | Scan a project directory and recommend agents/skills/commands |

Web GUI Features

• Analyze Project: Enter a project path to auto-detect technologies and get recommendations
• Browse: View all agents, skills, and commands with descriptions
• Search: Filter items by name or description
• Click to Add: Click any item to add to your bundle
• Smart Suggestions: When adding an agent, see related skills/commands
• Templates: Quick-start with pre-built templates (Full Stack, Python Data, DevOps)
• Export: Save your bundle as JSON for installation

Project Analysis (NEW)

Automatically analyze any project directory to detect technologies and get tailored recommendations:

Via MCP Server (Agentic):

```bash

# Claude can call this tool automatically when you describe your project

python src/mcp_server.py --test-analyze "C:\path\to\your\project"

```

Via Web GUI:

1. Launch python launch_gui.py
2. Enter project path in "Analyze Project" field
3. Click Analyze
4. View detected technologies (React, TypeScript, Docker, etc.)
5. Click Add All Recommendations to populate your bundle

Detected Technologies (40+):

| Category | Technologies |

|----------|--------------|

| Frontend | React, Next.js, Vue, Angular, Svelte, Tailwind |

| Backend | Express, NestJS, FastAPI, Django, Flask |

| Database | Prisma, Drizzle, TypeORM, Mongoose |

| Testing | Jest, Vitest, Cypress, Playwright, Pytest |

| DevOps | Docker, Kubernetes, Terraform, GitHub Actions |

| Languages | TypeScript, Python, Rust, Go, Java |

| Mobile | React Native, Flutter, Expo |

Example Output:

```json

{

"detected_technologies": [

{"name": "react", "confidence": 1.0},

{"name": "typescript", "confidence": 1.0},

{"name": "jest", "confidence": 1.0}

],

"recommended_agents": [

{"name": "frontend-developer", "relevance_score": 3.0},

{"name": "test-automator", "relevance_score": 2.0}

],

"recommended_skills": [

{"name": "frontend-development", "relevance_score": 2.0}

],

"suggested_pack": {

"name": "my-project-pack",

"agents": ["frontend-developer", "test-automator", ...],

"skills": ["frontend-development", ...],

"commands": ["test-gen", "code-review", ...]

}

}

```

CLI Commands

```bash

# Browsing

agent-manager list agents [--category X] [--limit N]

agent-manager list skills

agent-manager list commands

agent-manager list packs

agent-manager list categories --item-type agents

# Searching

agent-manager search [--limit N]

# Details

agent-manager show

# Installation

agent-manager install --target [--with-related]

agent-manager install --pack --target

# Index Management

agent-manager index # Rebuild indexes

agent-manager map # Rebuild mappings

# Interactive

agent-manager tui # Launch terminal UI browser

agent-manager configure [agent-name] # Configure mappings

```

Pre-Built Packs

| Pack | Contents |

|------|----------|

| fullstack-web | React, Next.js, Node.js, database, testing (9 agents, 8 skills, 4 commands) |

| python-data | Python, data science, ML, analytics (7 agents, 4 skills, 3 commands) |

| devops-infra | Cloud, Kubernetes, CI/CD, deployment (7 agents, 3 skills, 2 commands) |

| mobile-crossplatform | React Native, Flutter, mobile-specific (6 agents, 3 skills, 3 commands) |

```

Agent Skill Manager/

├── indexes/ # JSON catalogs

│ ├── agent-index.json # 249 agents indexed

│ ├── skill-index.json # 248 skills indexed

│ ├── command-index.json # 216 commands indexed

│ └── mappings/ # Relationship mappings

│ ├── agent-skill-map.json

│ ├── agent-command-map.json

│ └── manual-overrides.json

├── packs/ # Pre-built bundles

│ ├── fullstack-web.json

│ ├── python-data.json

│ ├── devops-infra.json

│ └── mobile-crossplatform.json

├── src/ # Python source code

│ ├── agent_manager.py # CLI entry point

│ ├── indexer.py # Index builder

│ ├── mapper.py # Mapping generator

│ ├── installer.py