Onboarding de Claude Code en 30 Minutos

What's in this pack

You just ran npm i -g @anthropic-ai/claude-code, hit /init, and now you're staring at an empty session wondering what to install. Skill marketplaces show 700+ entries. Reddit threads disagree. You don't want to spend Saturday spelunking — you want the opinionated short list that gets you productive before lunch.

This pack is exactly that: ten picks, in install order, that solve the first ten problems every new Claude Code user actually hits. No advanced multi-agent orchestration. No niche language-specific stuff. Just the boring infrastructure that lets every subsequent install pay off.

Who this is for: a developer with a working terminal, a real codebase, and zero patience for browsing. By the time you finish step 10, you'll have project memory, a curated catalog, a connected GitHub, three running agents, automation hooks, a one-keystroke commit-push-PR flow, and the ability to author your own skills.

Install in this order

1. CLAUDE.md — Best Practices Template — The single highest-leverage file you'll ever write for Claude Code. Drop it in your project root and Claude reads it every session: tech stack, conventions, what NOT to touch. Without this, every session starts from zero and you'll repeat the same corrections forever. 5 minutes to fill in. Saves hours every week.
2. Claude Code Templates — 600+ Agents, Commands & MCPs — A curated catalog you can browse instead of grepping marketplaces. This is your reference index for the next 6 months. Skim it once, bookmark the categories that match your stack, come back when you need something specific.
3. Awesome Claude Code — Ecosystem Directory — The canonical community-maintained list of plugins, integrations, and patterns. Pairs with #2: Templates is assets, Awesome is context (what's worth knowing about, blog posts, talks, tools). Read the README once.
4. GitHub MCP Server — Connect Claude Code to GitHub. Now claude can read PRs, file issues, comment on diffs, and check CI status without you context-switching. This is the single highest-frequency MCP for most devs; install it before any niche ones.
5. Claude Code Hooks — Pre/Post Task Actions — Lock in safety + automation early. Run a linter before edits land. Auto-test after Claude touches a file. Block writes to /vendor/. Set up the hooks scaffold now, before you have a real incident — retrofitting is annoying.
6. code-simplifier — Anthropic-Official Cleanup Subagent — Your first subagent. Anthropic ships it. After any session where Claude wrote >100 lines, invoke code-simplifier and it refactors recently-touched files for clarity without changing behavior. The cleanup pass you'd never bother to do manually.
7. verify-app — E2E Test Subagent — Your second subagent. Runs smoke tests after a change so you don't ship something obviously broken. The community equivalent of the verification step Boris Cherny demos publicly. Pairs with #5 (hooks can auto-trigger it).
8. Claude Code Agent: Code Reviewer — Automated PR review. Once you have GitHub MCP (#4) connected, this agent can read a diff and produce a structured review. Catches the obvious stuff so human reviewers focus on architecture.
9. /commit-push-pr — One-Shot Commit + Push + PR — The slash command you'll use 20+ times a day. One command: stages, writes a conventional commit message, pushes, opens a PR. Install it; you'll forget how you lived without it.
10. Claude Skill Factory — Build & Install Claude Code Skills — The graduation step. You've now used 9 skills others wrote — this is how you write your own. Project-specific skills are where Claude Code goes from "helpful" to "my personal senior engineer."

How they fit together

CLAUDE.md (#1)
   │
   └─ project memory, read every session
         │
Catalog: Templates (#2) + Awesome (#3)
   │
   └─ browse instead of guess
         │
GitHub MCP (#4)
   │
   └─ Claude talks to your repo, PRs, issues, CI
         │
Hooks (#5)
   │
   ├─ pre-edit: lint, format
   └─ post-edit: trigger verify-app, run tests
         │
Subagents: code-simplifier (#6) + verify-app (#7)
   │
   ├─ cleanup pass after big edits
   └─ smoke tests after any change
         │
Code Reviewer agent (#8)
   │
   └─ structured PR review, leans on GitHub MCP
         │
/commit-push-pr (#9)
   │
   └─ daily driver: stage → commit → push → PR
         │
Skill Factory (#10) — now build your own.

The CLAUDE.md + Hooks + GitHub MCP trio is the load-bearing wall. Everything else is decoration without those three. If you only have 10 minutes, install #1, #4, #5 and stop there.

Tradeoffs you'll hit

• Templates (#2) vs Awesome (#3) — They overlap. Templates is a runnable asset index (you install from it). Awesome is a reading list (blog posts, talks, ecosystem context). Install Templates first; bookmark Awesome and come back monthly when you want to see what's new.
• code-simplifier vs Code Reviewer — Both "clean up code," but at different stages. code-simplifier runs during development on files Claude just touched (style/clarity). Code Reviewer runs on a finished diff before merge (architecture/correctness). Run both — they catch different bugs.
• Hooks vs subagents — Hooks are deterministic shell commands the runtime triggers (cheap, fast, no LLM). Subagents are LLM-powered specialists you invoke (smart, slow, costs tokens). Use hooks for "always run prettier after edit"; use subagents for "review the architecture of this PR." Don't make hooks do LLM work — you'll cry at the latency.
• GitHub MCP vs the gh CLI — Both can read PRs and issues. MCP gives Claude structured access (typed responses, no shell parsing). gh is faster for you in the terminal. Install both; let Claude prefer MCP, you prefer gh.

Common pitfalls

• CLAUDE.md too long — 30-100 lines is the sweet spot. People paste their entire architecture doc and burn 4K tokens every turn. Keep it terse: stack, commands, conventions, hard NOs.
• Skipping hooks because they feel boring — They are boring. Install them anyway. The day Claude rewrites package-lock.json and you have no pre-edit guard, you'll wish you had.
• Installing 50 subagents at once — Start with the two in this pack (#6, #7). More subagents = more context Claude has to scan to pick the right one = slower + dumber routing. Add subagents only when you hit a real workflow gap.
• Forgetting ~/.claude/CLAUDE.md exists — There's a global CLAUDE.md too (personal preferences across all projects). Don't pollute project-level CLAUDE.md with personal style preferences — put those globally.
• Treating Skill Factory (#10) as optional — It's the multiplier. The first skill you write yourself (even a tiny one) teaches you how the system thinks. Don't skip it because you assume marketplace skills cover everything.