Episode 11: Foundational Q&A
This episode summarizes the 5 most common questions beginners encounter when exploring the Claude Code extension ecosystem.
Q1: What's the difference between a Tool and a Skill?
The Essence:
- Tool: Atomic capability (e.g., Read/Write/Bash). Think of it as Claude's "Hands." These are built-in features implemented in code.
- Skill: Workflow template (e.g., gsd-quick). Think of it as Claude's "Recipes." These are instruction sets defined in Markdown.
Analogy: A Tool is an action your hands can do (grab, press), while a Skill is a sequence of those actions (how to use your hands to cook a meal).
Q2: What's the difference between MCP and direct API calls?
- MCP: Encapsulates external capabilities as Tools. Claude can call them natively (identifying parameters, handling structured JSON returns).
- Direct API: Usually requires Claude to write
curlcommands and parse raw text responses itself, which is less efficient and less reliable.
Q3: Can Hooks access Claude's internal thoughts?
No. Hooks can only access:
- Tool names and input parameters (PreToolUse).
- Tool execution results (PostToolUse).
- Session status and environment variables. Hooks cannot intercept or read Claude's internal Chain-of-Thought.
Q4: Can one Skill call another Skill?
Yes, but not like a function call.
You can write in a Skill's instructions: "If errors are found, call /gsd-debug." When Claude reads this, it understands that it should initiate a new slash command workflow.
Q5: Is it better to use a Plugin or a manual Skill?
- Go with a Plugin: If you need MCP services (like databases) or automated Hooks.
- Go with a Manual Skill: If you just want to quickly create a simple, personal instruction template.