Claude Code
Evolution Story
From MCP to Slash Commands, Agents, and Skills — understand the design philosophy behind each tool and how they work together to save your Context.
MCP
Model Context Protocol — An open standard for connecting AI to external systems.
{
"name": "my-api-server",
"version": "1.0.0",
"tools": [
{
"name": "fetch_user",
"description": "Get user by ID"
}
]
}Slash Commands
Custom markdown files in `.claude/commands/` that insert prompts into your conversation.
---
description: Run all tests with coverage
---
Run the test suite and generate coverage report.
Use the test-runner agent for this task.Custom Subagents
Autonomous subprocesses that handle complex, multi-step tasks independently.
{
"description": "Code review specialist",
"role": "You check code quality",
"permissions": ["read", "write"],
"context": "fork"
}Skills
Modular packages that extend Claude's capabilities. Claude auto-detects when to use them.
{
"name": "code-review",
"description": "Reviews code changes",
"context": "fork",
"agent": "code-reviewer"
}Skills + Fork Context
Major upgrade: Skills can now run in forked sub-agent contexts with specified agents.
Commands + Skills Merge
"Merged slash commands and skills, simplifying the mental model with no change in behavior"
The Core Design Principles
📦 Progressive Disclosure
Load details only when needed. MCP keeps API docs external, Commands load workflows on demand, Agents fork execution contexts, and Skills combine them all.
💾 Context Saving
Every tool exists to solve the same problem: limited Context Window vs infinite knowledge needs. Keep Main Context clean for your actual problem.
🔀 Fork & Isolate
Sub-agents work in isolated contexts. Skills can fork execution. Process happens elsewhere, results come back — Main Context stays clean.
📝 It's All Prompts
MCP, Agents, Skills, Commands — different names for the same thing: prompts packaged for different stages of use.
Tool Comparison at a Glance
Quick reference for choosing the right tool for your needs.
| Tool | Best For | Context Impact | When to Use |
|---|---|---|---|
| MCP | External system integration | Low (per use) | Connecting to APIs, databases, file systems |
| Commands | User-triggered workflows | Low | Recurring tasks you initiate manually |
| Agents | Autonomous task delegation | Zero (forked) | Complex, multi-step independent tasks |
| Skills | Reusable capabilities | Zero (forked) | Specialized workflows Claude auto-detects |
* Context Impact: Effect on main conversation when idle (actual usage varies by tool call size)
The Essence
"Put the right prompt, in the right place, at the right time."
Ready to Level Up?
Explore our framework to learn how to apply these principles in your projects.
Explore Director Framework