Module 12 — Mastery: habits, mental models, and a checklist

← Module 11 | Index | Next: Module 13 →

You now know every part. Mastery is mostly about judgment — picking the right tool and keeping the work clean. Here’s the distilled version.

The five mental models to keep

Agent, not chatbot. It acts in loops — plans, does, checks, fixes. Give it room to finish.
The desk (context) is finite. Read selectively, clear between tasks, compact when noisy. Protect the desk and quality goes up.
Everything custom is a plain file in .claude. No magic — you can read, edit, share, and version it.
Right trigger for the right job. You trigger slash commands. Claude triggers skills and subagents (by matching descriptions). Events trigger hooks. Pick accordingly.
Descriptions are the steering wheel. Skills and subagents get chosen by their description. Vague description, ignored capability. Sharp description, reliable behavior.

Habits that mark an expert

/clear between unrelated tasks, reflexively. A clean desk is your single biggest quality lever.
Maintain a tight CLAUDE.md. Every correction you make twice belongs in it.
Reach for a hook when a rule must be unbreakable; a CLAUDE.md note when a preference is usually fine.
Delegate big, messy sub-jobs (searches, bulk reads) to subagents to keep the main desk clean.
Be specific in requests, and let the agent run multiple steps before judging.
Give corrective feedback — when it goes wrong, say so; that’s how the setup improves.
Guard write-access connectors and never automate moving money.

”Which feature do I want?” guide

I keep re-explaining the same project facts → put them in CLAUDE.md (Module 3).
I keep re-typing the same instruction → make a slash command (Module 6).
I want Claude to know a procedure and use it on its own → write a skill (Module 5).
I want a rule enforced every single time → write a hook (Module 7).
A sub-job is huge and would clutter things → use a subagent (Module 8).
I need Claude to reach into another app → connect an MCP connector (Module 9).
I want this to run on a schedule → set up a scheduled task (Module 11).
I want a page I can re-open with fresh data → build an artifact (Module 11).
I want a whole capability pack at once → install a plugin (Module 11).

Your learning path from here

Use it daily for real tasks. When you hit friction, ask “which module solves this?” Write your first CLAUDE.md, then your first slash command, then your first skill. Peek inside your .claude folder and read what’s there — it’ll all make sense now. That loop, repeated, is the whole road to mastery.

You started at zero. You now understand the internal .claude files, how skills are stored and chosen, how Claude budgets its context, what hooks and subagents are, and how connectors, plugins, schedules, and artifacts extend the whole system. The rest is practice.

For two deeper, hands-on best-practice modules, continue to Module 13 (token optimization) and Module 14 (context engineering).

← Module 11 | Index | Next: Module 13 →