Master the three features that turn Claude Code from a helpful tool into an autonomous assistant.
Most Claude Code users discover CLAUDE.md and stop there. That's like buying a car and only using the radio.
These three features are where Claude Code gets genuinely powerful:
| Feature | What it does |
|---|---|
| Skills | Reusable slash commands you invoke in one line |
| Hooks | Automatic actions that happen without you asking |
| MCP Servers | Connect Claude to external tools (Notion, Slack, GitHub, etc.) |
You already know skills from Module 2. In Claude Code, they become proper slash commands.
Creating a skill:
Create a file at .claude/skills/my-skill/SKILL.md:
---
name: weekly-report
description: Generate my weekly status report in HivePowered format
---
You are helping a HivePowered Sidekick write their weekly status report.
FORMAT:
## Weekly Report — [Date]
### What I completed this week
[Bullet points of completed tasks]
### What I'm working on next
[Bullet points of upcoming work]
### Blockers or questions
[Any issues that need Prime's attention]
### AI wins this week
[One thing AI helped me do faster or better]
TASK:
Review any notes I give you and generate a complete weekly report in the format above.
Using it:
/weekly-report Here's what I did this week: [paste your notes]
Done. Claude formats everything into the official report structure.
Your skill can accept whatever you type after the command:
---
name: summarize
description: Summarize any document or text into key points
---
Summarize the following content into:
- 3-5 bullet points (key takeaways)
- 1-sentence overall summary
- Any action items mentioned
CONTENT TO SUMMARIZE:
$ARGUMENTS
Use it:
/summarize [paste any article, email, or document]
Claude reads whatever you paste and returns a clean summary.
You can also write skills that Claude uses in the background — you don't call them, Claude does when it recognizes the situation.
Example — a proofread skill Claude applies whenever you ask for content review:
---
name: proofread-checker
description: Check any content for grammar, clarity, and brand voice compliance
user-invocable: false
---
When reviewing content, always check:
1. Grammar and spelling (fix silently)
2. Clarity — flag any sentences that are hard to follow
3. Tone — does it match the project's brand voice from CLAUDE.md?
4. Length — is it within the requested range?
Return the corrected content plus a brief note on what was changed.
Claude applies this automatically whenever you say "proofread this" or "review this content."
Hooks execute shell commands automatically on your machine — without asking you first. Before adding any hook, understand exactly what command it runs.
Stick to simple, obvious commands you can read yourself: running a linter, logging output, sending a known Slack message.
A hook is a command that runs automatically when something happens. No prompting. No asking. It just happens.
Example: Auto-format every file Claude edits
In .claude/settings.json:
{
"hooks": {
"PostToolUse": [
{
"matcher": "Edit|Write",
"hooks": [
{
"type": "command",
"command": "echo 'File updated: ' && cat .claude/last-edit.txt"
}
]
}
]
}
}
Now every time Claude edits a file, your formatting command runs automatically.
What hooks are great for:
- Running tests after Claude writes code
- Logging every change Claude makes
- Sending a Slack message when Claude finishes a task
- Blocking Claude from editing sensitive files
Set up hooks with:
/hooks
Claude walks you through it interactively — no manual JSON editing needed.
An MCP server gives Claude direct, live access to the tool it connects to. This is not a simulation. Claude can read, write, send, and delete — for real.
.mcp.json or ~/.claude.json) may contain API keys — never share these files or commit them to gitMCP (Model Context Protocol) lets Claude talk to external services like Notion, Slack, GitHub, and Google Drive — directly from your conversation.
Without MCP:
You: "Summarize my Notion tasks for today"
Claude: "I don't have access to your Notion..."
With MCP:
You: "Summarize my Notion tasks for today"
Claude: [reads your Notion] "Here are your 4 tasks for today: ..."
Popular MCP servers:
| Server | What it gives Claude access to |
|---|---|
| GitHub | Read issues, create PRs, review code |
| Notion | Read pages, databases, tasks |
| Slack | Read channels, send messages |
| Google Drive | Read and write docs |
| PostgreSQL | Query your database directly |
Adding an MCP server:
claude mcp add --transport http notion https://mcp.notion.com/mcp
Then Claude can read and write Notion — from inside your conversation.
For HivePowered team:
The most useful connections to set up:
1. Notion — so Claude can read your task list and update it
2. Slack — so Claude can post your weekly report automatically
3. GitHub — if you work on code or manage repos
Here's what a fully set-up Claude Code environment looks like for a Sidekick:
Session starts
→ Claude reads global CLAUDE.md (your preferences)
→ Claude reads project CLAUDE.md (client knowledge)
You type: /weekly-report [paste notes]
→ Skill runs → generates formatted report
You say: Post this to #team-updates in Slack
→ MCP (Slack) → posts the report → done
Hook fires: Saves a log of what was posted and when
One conversation. Multiple outputs. Automatic logging. No manual steps.
Don't try to build everything at once. Start here:
/weekly-report or /summarize)Each one takes 15-30 minutes. After a month, you'll have an AI system that runs itself.
Download: The Skills Starter Pack in the Templates folder includes 5 ready-to-use skill files.
HivePowered AI — AI Like a Pro Training