```
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β HOW COMPONENTS DIFFER β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ€
β β
β SKILLS = Knowledge that loads automatically β
β COMMANDS = Actions you trigger manually β
β AGENTS = Autonomous helpers that run after you code β
β HOOKS = Automation that runs on system events β
β β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
```
Skills - Passive Knowledge
What: Domain expertise that Claude learns from when relevant
When: Loads automatically based on what you're asking about
Trigger: Your questions/context match the skill description
You do: Nothing - just ask questions naturally
```
You: "How do I test my FastAPI endpoints?"
β
Claude: [fastapi + pytest skills auto-load into my knowledge]
"Here's how to test FastAPI endpoints..."
```
Think of it as: A reference book that opens to the right page automatically
Commands - Manual Actions
What: Specific tasks you explicitly request
When: Only when you type the command
Trigger: You type /ltk:command-name
You do: Invoke it manually when you want that action
```
You: /ltk:scan-security
β
Claude: [Runs security scan on codebase]
"Found 2 vulnerabilities..."
```
Think of it as: Buttons you press to do specific things
Agents - Proactive Helpers
What: Specialized assistants that watch what you're doing
When: After you write or edit code (automatically)
Trigger: Tool events (Write, Edit) + matching context
You do: Nothing - they activate on their own
```
You: writes Python code with SQL query
β
Claude: [security-analyzer agent activates]
"Note: This SQL query might be vulnerable to injection.
Consider using parameterized queries."
```
Think of it as: A team of experts looking over your shoulder
Hooks - System Automation
What: Scripts/prompts that run on specific system events
When: Triggered by system events (session start, before/after edits)
Trigger: Events like SessionStart, PreToolUse, PostToolUse
You do: Nothing - happens automatically in the background
```
[Session starts]
β
Hook: [Loads git status, project structure, TODOs]
β
Claude: [Has project context ready]
```
Think of it as: Automatic setup and cleanup routines
Side-by-Side Comparison
| Aspect | Skills | Commands | Agents | Hooks |
|--------|--------|----------|--------|-------|
| Activation | Automatic (context) | Manual (/ltk:...) | Automatic (after code) | Automatic (events) |
| Purpose | Provide knowledge | Execute actions | Analyze & advise | Automate workflows |
| User action | Just ask questions | Type command | Write code | Nothing |
| Output | Better answers | Task results | Suggestions/warnings | Background setup |
Visual Flow
```
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β YOUR CODING SESSION β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ€
β β
β SESSION START β
β β β
β βΌ β
β βββββββββββ β
β β HOOKS β βββ Load git status, project structure, TODOs β
β βββββββββββ β
β β β
β βΌ β
β YOU ASK: "How do I add authentication to FastAPI?" β
β β β
β βΌ β
β βββββββββββ β
β β SKILLS β βββ fastapi + security-scanning skills load β
β βββββββββββ β
β β β
β βΌ β
β CLAUDE ANSWERS with specialized knowledge β
β β β
β βΌ β
β YOU WRITE CODE β
β β β
β βΌ β
β βββββββββββ β
β β AGENTS β βββ security-analyzer checks your code β
β βββββββββββ quality-reviewer suggests improvements β
β β β
β βΌ β
β YOU TYPE: /ltk:scan-security β
β β β
β βΌ β
β ββββββββββββ β
β β COMMANDS β βββ Full security scan runs β
β ββββββββββββ β
β β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
```
Quick Summary
| Component | One-liner |
|-----------|-----------|
| Skills | "Claude knows this stuff" |
| Commands | "Do this specific thing now" |
| Agents | "Watch my code and give feedback" |
| Hooks | "Run this automatically when X happens" |
Skills vs Agents - Common Confusion
```
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β SKILLS = Reference Material (Passive) β
β ββββββββββββββββββββββββββββββββββββββββ β
β β’ A book Claude reads to learn something β
β β’ Contains: Documentation, patterns, best practices β
β β’ Does: NOTHING - just sits there until needed β
β β’ Output: Better, more informed responses β
β
|