boost

name	boost
description	Diagnose and optimize your Claude Code setup. Auto-detects your project, scores your configuration, and guides you through setting up CLAUDE.md, permissions, memory, MCP servers, plugins, hooks, and more.
argument-hint	[level]
allowed-tools	["Read","Glob","Grep","Bash","Write","Edit"]

Boost — Diagnose and Optimize Your Claude Code Setup

Diagnose the user's Claude Code configuration, display their Boost Score, and actively help them set up missing items.

Instructions

Phase 1: Diagnose

Run the exact same diagnostic checks as the boost-check skill. Check all 9 items across 3 levels:

Level 1 — Essentials (3 items)

1. CLAUDE.md — exists in project root? Has > 5 lines? Has model routing section?
2. settings.json — ~/.claude/settings.json has permissions with at least one allow entry?
3. Memory system — project memory directory has MEMORY.md with entries?

Level 2 — Power Features (4 items) 4. CLAUDE.md advanced — has context management / workflow mode sections (2+ keywords)? 5. MCP servers — ~/.claude/mcp.json or .mcp.json has at least one server? 6. Plugins — installed_plugins.json has 2+ plugins (excluding boost)? 7. Git integration — current directory is a git repo?

Level 3 — Expert (2 items) 8. Hooks — settings.json has custom hooks? 9. Keybindings — ~/.claude/keybindings.json exists with custom bindings?

Phase 2: Display Score

Show the Boost Score:

╔══════════════════════════════════════════════════╗
║  Boost Score: X/9                                ║
║  Level 1: a/3 | Level 2: b/4 | Level 3: c/2     ║
╚══════════════════════════════════════════════════╝

Then list all items with ✓/✗ status.

Phase 3: Guide and Act

Start from the lowest level with missing items. For each missing item, offer to set it up.

If the user passed a level argument (e.g., /boost 2), skip to that level.

Setting Up: CLAUDE.md (Level 1)

When CLAUDE.md is missing or sparse:

1. Auto-detect project context:

   • Read package.json → extract name, description, dependencies, scripts
   • Read requirements.txt or pyproject.toml → extract Python dependencies
   • Read go.mod → extract Go module info
   • Read Cargo.toml → extract Rust crate info
   • Read .eslintrc*, .prettierrc*, .editorconfig → extract code style rules
   • Read jest.config*, vitest.config*, pytest.ini, tsconfig.json → extract test config
   • Read .gitignore → infer restricted files
   • Read README.md first 20 lines → extract project description
   • If none of these exist, note "No project config files detected"

2. Generate CLAUDE.md with recommended template:

   Present the complete file with auto-detected values filled in and sensible defaults for the rest:

   # Project Instructions
   
   ## Project Description
   [Auto-detected or "Describe your project here"]
   
   ## Tech Stack
   [Auto-detected from dependency files]
   
   ## Model Routing
   | Level | Model | Use For |
   |-------|-------|---------|
   | Deep reasoning | opus | Architecture planning, complex debugging, cross-system design |
   | Main development | sonnet | Code writing, reviews, research, general tasks |
   | Light tasks | haiku | Simple queries, formatting, quick lookups |
   
   ## Language Preference
   English (change to your preferred language)
   
   ## Code Style
   [Auto-detected from linter/formatter configs, or "Follow existing project conventions"]
   
   ## Testing
   [Auto-detected test framework and conventions, or "Describe your test approach"]
   
   ## Restrictions
   - Do not modify .env or credential files
   [Additional restrictions inferred from .gitignore]
   
   ## Context Management
   - Suggest /compact at phase transitions (research → planning, planning → implementation)
   - Do not compact during active implementation
   - Before compacting, confirm critical context will be preserved
   
   ## Common Commands
   [Auto-detected from package.json scripts, Makefile targets, or "Add your common commands"]
   

3. Present the complete file to the user and ask them to confirm or request changes.

4. If confirmed, write the file with the Write tool.

Setting Up: settings.json (Level 1)

When permissions are not configured:

1. Explain what permissions control: File read/write permissions, Bash command execution, MCP tool access
2. Present recommended permission settings as a complete settings.json snippet. Recommend allowing Read/Glob/Grep/Bash by default for better workflow, with Write/Edit requiring approval.
3. Ask user to confirm, then apply with Edit tool to ~/.claude/settings.json.

Setting Up: Memory System (Level 1)

When memory is not enabled:

1. Explain the 4 memory types briefly:

   • user — your role, preferences, expertise
   • feedback — corrections and confirmed approaches
   • project — ongoing work context, decisions, deadlines
   • reference — pointers to external resources

2. Offer to create initial memories by presenting recommended defaults:

   • A user_profile.md with placeholder sections (role, tech stack, preferences)
   • A MEMORY.md index file

3. Present the files for confirmation, then write them.

Setting Up: CLAUDE.md Advanced (Level 2)

When CLAUDE.md exists but lacks workflow/mode sections:

1. Read the existing CLAUDE.md
2. Suggest adding these sections:
   • Context Management — when to compact, phase transition rules
   • Workflow Modes — different behaviors for debugging vs. feature dev vs. review
   • Custom Commands — project-specific shortcuts
3. Present the additions as an Edit (not overwrite) to the existing file
4. Apply after confirmation

Setting Up: MCP Servers (Level 2)

When no MCP servers are configured:

1. Explain MCP concept in one paragraph
2. List popular MCP servers with what they do:
   • Filesystem server — enhanced file operations
   • GitHub server — PR/issue management
   • Web search — internet access
   • Database servers — direct DB access
3. Ask which ones interest the user
4. Guide creating .mcp.json for chosen servers

Setting Up: Plugins (Level 2)

When few plugins are installed:

1. Read the current project type from CLAUDE.md or detected stack
2. Recommend relevant plugins:
   • superpowers — development workflows (if not installed)
   • code-review — PR review automation
   • Other relevant plugins based on project type
3. Provide install commands: /plugin install <name>@claude-plugins-official
4. Do NOT auto-install — only recommend

Setting Up: Git Integration (Level 2)

When not a git repo:

1. Explain benefits of git integration with Claude Code
2. Offer to run git init if appropriate
3. If already a git repo, explain advanced features:
   • Worktree workflows for parallel development
   • Commit conventions Claude Code follows
   • Branch strategies

Setting Up: Hooks (Level 3)

When no custom hooks exist:

1. Explain the hook system: events that trigger shell commands
2. Show practical examples:
   • Pre-commit: run linter before committing
   • Post-edit: auto-format changed files
   • SessionStart: custom welcome message or context injection
3. Guide creating first hook in ~/.claude/settings.json

Setting Up: Keybindings (Level 3)

When using defaults:

1. Show available customizable shortcuts
2. Suggest productivity bindings:
   • Custom key for /compact
   • Quick access to common commands
3. Guide creating ~/.claude/keybindings.json

Phase 4: Update State

After each item is set up:

1. Read current state from ~/.claude/plugins/data/claude-boost/global-state.json
   • If file doesn't exist, create with default schema
2. Add the completed item ID to completedItems array
3. Update boostScore counts
4. Update lastCheckTimestamp
5. If all items in a level are complete, update currentLevel to next level
6. Write updated state back

Item IDs: claude-md, settings-json, memory-system, claude-md-advanced, mcp-servers, plugins, git-integration, hooks, keybindings

Phase 5: Level Transition

When all items in a level are completed:

1. Show congratulations message with updated score
2. Preview what the next level covers
3. Ask if the user wants to continue to the next level or stop here

Important

• NEVER modify existing files without showing the user the changes and getting confirmation
• NEVER auto-install plugins — only recommend with install commands
• When CLAUDE.md already exists, use Edit (not Write) to add missing sections
• Adapt response language to match the user's conversation language
• If the user passes a specific level argument, skip directly to that level's items