Claude Code settings.json + statusLine + MCP for Nuxt app

You are a senior staff engineer with extensive technical knowledge in Claude Code customization. You write clear, precise, and implementation-ready content.

Produce a complete Claude Code configuration for a Nuxt app: `settings.json`, a custom `statusLine`, one MCP server registration, and one skill scaffold.

**Project:** Nuxt app
**Team:** 2-5 engineer startup
**MCP integration target:** notion
**Skill to scaffold:** review-pr-diff
**Convention:** public functions require JSDoc with @example

## Output 1: `.claude/settings.json` (project, committed)
Shape:
```json
{
  "model": "claude-sonnet-4-5",
  "env": {
    "NODE_ENV": "development",
    "LOG_LEVEL": "info"
  },
  "permissions": {
    "allow": [
      "Bash(npm test:*)",
      "Bash(npm run build:*)",
      "Bash(git diff:*)",
      "Bash(git status)",
      "Read",
      "Edit",
      "Grep",
      "Glob",
      "WebFetch(domain:docs.Nuxt app.example)"
    ],
    "deny": [
      "Bash(rm -rf *)",
      "Bash(sudo *)",
      "Bash(curl * | sh)"
    ]
  },
  "statusLine": {
    "type": "command",
    "command": "<absolute path to status script>"
  },
  "hooks": { "...": "..." }
}
```
Rules:
- `settings.json` is committed -- no secrets
- `settings.local.json` (gitignored) is for per-dev overrides (tokens, personal tool prefs)
- `~/.claude/settings.json` is for user-global defaults
- Permissions use prefix patterns with `:*`, e.g. `Bash(npm test:*)` allows `npm test`, `npm test -- --ui`, etc.
- Always include a `deny` list for destructive commands

Fill in the full JSON for this Nuxt app, including a realistic `deny` list and an `allow` list matched to public functions require JSDoc with @example.

## Output 2: statusLine script
A short shell (or node) script that prints a single line Claude Code shows at the bottom of the session. Receives JSON on stdin: `{ "cwd": "...", "model": { "id": "..." }, "session_id": "..." }`.

Produce a script that prints, all on one line:
- current git branch (with `*` if dirty)
- current model id
- a compact session cost indicator (e.g. token count from `~/.claude/projects/<hash>/transcript.jsonl` if available)
- a simple project badge (the project name)

Keep execution under 200ms (Claude re-runs it frequently). No network calls.

## Output 3: MCP server registration
Register **notion** by adding a `.mcp.json` at the repo root:
```json
{
  "mcpServers": {
    "notion": {
      "command": "<npx or path>",
      "args": ["<args>"],
      "env": {
        "<VAR>": "${<VAR>}"
      }
    }
  }
}
```
- Use `${VAR}` to reference env without hard-coding secrets
- Document required env vars in a companion `.env.example`
- Add a one-paragraph README snippet explaining what tools this MCP exposes and when to use them

## Output 4: Skill scaffold at `.claude/skills/review-pr-diff/SKILL.md`
Skills are reusable, invocable capabilities. Produce a SKILL.md with:
- YAML frontmatter: `name`, `description` (explicit trigger phrase)
- A role line
- Steps to follow
- An output contract
- A "do not use when" section

Target length for SKILL.md: 50-120 lines.

## Ordering and dependencies
1. Create `.claude/settings.json` first (without hooks pointing to non-existent scripts)
2. Add `.mcp.json` and test the MCP server with `claude mcp list`
3. Add the statusLine script and make it executable
4. Add the skill
5. Commit in that order so each step is independently verifiable

Structure as a playbook with: Overview, Prerequisites, Step-by-step Plays, Metrics to Track, and Troubleshooting Guide.