Claude Code Day-One Setup: CLAUDE.md, Superpowers, Context7 and More

Installing Claude Code takes a minute. Running claude in a project folder and logging in takes another. After that, a lot of people just start firing off prompts, then wonder why the answers sound generic and why the tool keeps asking about a stack it should already know. A short list of installs and config changes, roughly 20 to 30 minutes of work, makes most of that go away. The picks below follow Anthropic's own first-day guidance plus the plugins that keep turning up in real developer setups.

Start with a CLAUDE.md

If you only do one thing, do this. Type /init in the session and Claude scans the repo, then drafts a CLAUDE.md at the project root. Read it, delete whatever it got wrong, and commit it.

Claude loads that file at the start of every session, so it should hold your real build, test and lint commands plus a handful of hard rules. Keep it to about 200 lines. Filler such as "write clean code" burns space for nothing, while a concrete line like "run pnpm test before saying a task is finished" actually sticks.

Skip the file and you'll re-explain the project every session. Write it and the model already knows the entry points and conventions.

Superpowers for a plan-first workflow

From inside Claude Code, run:

/plugin install superpowers@claude-plugins-official

If you want fresher updates, add the author's own marketplace first and install from there:

/plugin marketplace add obra/superpowers-marketplace
/plugin install superpowers@superpowers-marketplace

Superpowers ships skills for brainstorming, writing a plan, test-driven development, systematic debugging and checking before declaring victory. New sessions pick them up on their own. The practical effect is that it breaks the "just start coding" habit behind so many half-built features.

Context7 for current docs

Training data never matches the exact version of every library you use. Context7 fetches version-correct documentation on demand. One common way to install it:

claude mcp add - scope user context7 - npx -y @upstash/context7-mcp - api-key YOUR_KEY

A free key from context7.com/dashboard buys you better rate limits. Once it's in, you can simply tell Claude to "use context7" whenever a package's docs matter. That prevents a good share of invented API calls, which otherwise eat tokens and time.

GitHub access

If your day revolves around issues and pull requests, install either the GitHub MCP server or Anthropic's GitHub plugin. Syntax varies slightly depending on the package version, but the shape stays the same: claude mcp add pointed at the GitHub server, with a token carrying the scopes you need.

Once connected, Claude can list open PRs, read through issue threads and draft a pull request, all without switching windows. The commit-commands plugin from the official marketplace pairs well with it for tidier git workflows.

Loosen permissions carefully

By default, Claude Code asks before every edit and shell command. That's the right call on day one. After a few uneventful sessions, you'll want to relax the routine cases.

Find the settings file, usually under ~/.claude or in the project's .claude folder, and allow the low-risk commands you run all day: the test runner, the linter, git status and git diff. Anything that can delete files or force-push should stay on "ask" or "deny". Also learn Shift+Tab early, since it cycles through the permission modes mid-session.

Code review plugin

Also from the official marketplace:

/plugin install code-review@claude-plugins-official

It runs several review agents in parallel, covering security, correctness, style and historical patterns, then reports findings with confidence scores. Run it before merging. It goes deeper than asking for one review pass in the main conversation.

Frontend design plugin

/plugin install frontend-design@claude-plugins-official

Out of the box, Claude builds interfaces that work but look generic. This plugin pushes a clearer typographic hierarchy, better spacing and visual choices that read as deliberate. For pure backend or infrastructure work, skip it.

Language server plugins

Plugins exist for TypeScript, Python, Go, Rust and more. Install the ones matching the languages you really write. They give Claude symbol navigation and type awareness within the session, so it isn't limited to grepping through files.

A personal CLAUDE.md

Project CLAUDE.md files are shared with the team. Your personal one sits at ~/.claude/CLAUDE.md and applies to every project. Use it for rules that never change for you, like your preferred commit message style, how you want error handling written, and the tools you reach for first. Keep it short as well, because it loads every session.

Token savers, later

Long sessions or many parallel agents might call for skills that compress communication or trim needless tool calls. Caveman-style compression skills come up often in token-optimization threads. Hold off until you've watched your own usage for a few days and know where the spend goes.

One more housekeeping step: run claude update and claude doctor once, to confirm the binary and PATH are clean.

The short version

CLAUDE.md, Superpowers and Context7 on their own already change the quality of the first real feature you build. Everything else on the list just removes friction as the project grows.