Full Claude Code Setup prompt

Set up Claude Code from scratch for me. There are four things I want in place before I
write any real code: a memory file so you know who I am, current docs so you stop writing
outdated code, a planning layer so you ask instead of guessing on complex work, and a
connection to my tools (GitHub). Do them in the order below.

Ground rules for the whole setup: take stock before you change anything, only ever APPEND
to my files (never overwrite or delete what is already there), install everything at USER
scope (~/.claude/) so it works in every project, and if anything below already exists and
matches, skip it. If something exists but differs, STOP, show me the difference, and ask
before changing it.

(If a bare `claude ...` command is blocked inside this session, re-run just that command
with CLAUDECODE cleared: `CLAUDECODE= claude ...` on macOS/Linux, or
`$env:CLAUDECODE=''; claude ...` in PowerShell. A couple of steps below are slash commands
that *I* have to type myself - you cannot run those for me, so just hand me the exact line.)

First, take stock - do not change anything yet. Run `claude mcp list` and `claude plugin
list`, and read ~/.claude/CLAUDE.md (on Windows %USERPROFILE%\.claude\CLAUDE.md). Note
what already exists so you can skip it below.

STEP 1 - Give yourself a memory (CLAUDE.md). This is the file you read at the start of
every session, so I never have to re-explain who I am or how I work.

Check ~/.claude/CLAUDE.md first. If it already has a personal "who I am" / rules section,
show me what is there and ask before touching it - only append. If there is no personal
section yet, ask me these three quick questions, then write my answers into
~/.claude/CLAUDE.md (create the file and folder if needed) under a clear heading:
  1. Who am I and what do I build? (name, role, the kind of projects)
  2. What is my default stack? (languages, frameworks, tools you should assume)
  3. Any hard rules for every session? (e.g. ask before big refactors, always write tests,
     keep changes small)
Keep it short and in my words. This is global, so it applies in every project. Then tell me
that for project-specific memory I can run `/init` in any repo (or ask you to write a
./CLAUDE.md there) - do not do that now unless I ask.

STEP 2 - Stop outdated code (Context7). You answer from training data with a cutoff, so the
moment a library ships a new version you are quietly writing against an old one. Context7
feeds you the real, current docs on demand.

If `claude mcp list` already shows "context7", skip to STEP 3. Otherwise add it - free,
hosted, no API key:

  claude mcp add --scope user --transport http context7 https://mcp.context7.com/mcp

Then append this rule to ~/.claude/CLAUDE.md (read it first, do not duplicate, only append):

  ## Context7
  When working with an external library, framework, SDK, or API, use the Context7 MCP
  server to fetch its current documentation before writing or changing code - even for
  well-known libraries. Prefer it over relying on training data.

STEP 3 - Stop the guessing (Superpowers). On anything complex you tend to dive in and
guess; Superpowers is a plugin that makes you slow down, plan, and ask the right questions
first - which is where most wasted debugging time disappears.

This one installs with a slash command that *I* type (you cannot run it for me), so hand it
to me and tell me what to expect. If `claude plugin list` already shows "superpowers", skip
this. Otherwise tell me to type, inside this session:

  /plugin install superpowers@claude-plugins-official

Then I restart Claude Code and run `/help`: I should see new commands
`/superpowers:brainstorm`, `/superpowers:write-plan` and `/superpowers:execute-plan` -
that is how we confirm it worked. If those do not show up after a restart, tell me to run
`/plugin update superpowers` and restart, or to install it from the project's own
marketplace instead with these two commands:

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

STEP 4 - Connect your tools (GitHub MCP). Official MCP servers let you act in the services I
use directly - GitHub's lets you open PRs, read issues, and check CI on my behalf, instead
of me copying things back and forth.

Important: in Claude Code the GitHub server authenticates with a Personal Access Token, not
the browser OAuth flow (OAuth is not supported here yet), so I need to give you a token.

  a) If `claude mcp list` already shows "github", skip this step.
  b) Otherwise tell me to create a fine-grained token at
     https://github.com/settings/personal-access-tokens/new with repository access and
     read/write permissions on the repos you should touch, then paste it back to you. Do
     not echo it back to me or write it into any file.
  c) Check my version with `claude --version`. On Claude Code 2.1.1 or newer, add it at user
     scope (swap in my token where it says MY_TOKEN):

       claude mcp add-json github '{"type":"http","url":"https://api.githubcopilot.com/mcp/","headers":{"Authorization":"Bearer MY_TOKEN"}}' --scope user

     On 2.1.0 or older, use the older form instead:

       claude mcp add --transport http --scope user github https://api.githubcopilot.com/mcp/ -H "Authorization: Bearer MY_TOKEN"

     (On Windows, if `add-json` errors with "Invalid input", use that second
     `--transport http` form.)

Finally - verify everything and tell me where it stands. Run `claude mcp list` and confirm
"context7" and "github" show as connected, confirm whether "superpowers" is installed (or
that you have handed me the slash command to finish it), and show me the sections you added
to my CLAUDE.md. List anything you skipped because it already existed, and anything still
needing me (the GitHub token, the Superpowers slash command). Then remind me to fully
restart Claude Code so the memory file, both MCP servers, and the plugin all load.