Claude Launchpad

Getting Started

Install Claude Launchpad and score your coding-agent setup in seconds. Doctor diagnoses Claude Code or Cursor Agent config; eval proves the agent follows your rules; memory persists decisions across sessions and machines.

Claude Code follows CLAUDE.md rules about 80% of the time. Hooks run at 100%. Most projects have zero hooks configured. Claude Launchpad scores and repairs that setup — and does the same for local Cursor Agent projects.

Install

npx claude-launchpad

Or install globally:

npm i -g claude-launchpad
pnpm add -g claude-launchpad
yarn global add claude-launchpad

Claude Launchpad has 4 CLI commands (init, doctor, eval, memory) and 2 skills (/lp-enhance, /lp-migrate-memory). Memory is optional. The npm package name stays claude-launchpad.

Two local harnesses share the same workflow files (TASKS.md, BACKLOG.md) and the same memory store. Scores are never averaged.

CommandClaude CodeLocal Cursor Agent
initdefault (CLAUDE.md, .claude/)--harness cursor (AGENTS.md, .cursor/)
doctor / --fixyesyes (--harness auto detects either)
evalyes--harness cursor
memory installyes--harness cursor (same SQLite + MCP)
/lp-enhance.claude/skills/lp-enhance/.cursor/skills/lp-enhance/

--harness both scaffolds or diagnoses both surfaces in one run. --min-score gates each harness separately. Cursor Cloud Agents cannot use the local memory MCP server.

Why hooks matter: CLAUDE.md rules are ~80% reliable. Hooks are 100% enforced. A SessionStart hook that runs cat TASKS.md means Claude sees your tasks at every session start. A SessionStart hook with the compact matcher re-surfaces tasks after context compression, keeping work on track.

New project

Scaffold

claude-launchpad init

Generates baseline config for the selected harness. Default is Claude Code (CLAUDE.md, .claude/). init --harness cursor writes AGENTS.md and .cursor/. Shared TASKS.md and BACKLOG.md are written once.

Diagnose + fix

claude-launchpad doctor --fix

Catches anything init missed and auto-fixes it.

Let the agent perfect it

/lp-enhance

Inside Claude Code or Cursor Agent, run the skill. It reads your codebase and fills in architecture, conventions, and guardrails on CLAUDE.md or AGENTS.md.

Prove it works

claude-launchpad eval

17 scenarios test whether the selected agent actually follows your rules. Use --harness cursor for Cursor Agent.

Existing project

See your score

cd your-project
claude-launchpad

A typical unconfigured project scores ~42%.

Fix the structure

claude-launchpad doctor --fix

Adds missing hooks/rules/sections, security defaults, and creates /lp-enhance only when no project/global/legacy install already exists. Jumps to ~86%.

Let the agent clean it up

/lp-enhance

Type /lp-enhance in Claude Code or Cursor Agent. The fixer appends sections mechanically — the skill rewrites the instruction file with project-specific content.

Prove it works

claude-launchpad eval

Keep your config healthy

init runs once. Everything else is designed for repeat use as your project evolves.

WhenRun
Added dependencies, changed frameworks, modified hooksdoctor --fix
Major refactor, new modules, CLAUDE.md or AGENTS.md feels stale/lp-enhance
Before a release, after config changes, the agent ignoring ruleseval
You want cross-session memory with local SQLite + MCP toolsmemory

Your codebase changes. Your config should keep up. Run doctor periodically. If the score drops, --fix and /lp-enhance bring it back.

Who does what

Each tool handles one job. No overlap.

Responsibilityinitdoctor --fix/lp-enhanceevalmemory
Detect stack and framework
Generate CLAUDE.md / AGENTS.md, TASKS.md, BACKLOG.md
Generate settings or hooks.json
Generate .claudeignore / .cursorignore
Score config 0-100
Fix missing sections, rules, permissions
Rewrite CLAUDE.md / AGENTS.md with real project content
Suggest hooks, MCP servers, path-scoped rules
Overflow conventions to rules files
Prove the agent follows your rules
Run sandbox test scenarios
Persistent cross-session memory
Memory dashboard, sync, migration

init scaffolds. doctor maintains. /lp-enhance makes it smart. eval proves it works. memory remembers.

Command reference

CommandRuns
claude-launchpad initLocally, free
claude-launchpad / doctorLocally, free
claude-launchpad doctor --fixLocally, free
claude-launchpad doctor --watchLocally, free
/lp-enhance (skill)Inside Claude Code or Cursor Agent
claude-launchpad evalVia Claude CLI or Cursor Agent
claude-launchpad memoryLocally
claude-launchpad memory --dashboardLocally
claude-launchpad memory push/pullVia GitHub API
claude-launchpad memory sync statusLocally
claude-launchpad memory sync cleanLocally

Commands marked Locally run offline with zero API calls. Eval requires the selected harness CLI or SDK (Claude Code or Cursor Agent) and authentication. Memory is optional.

Next Steps

On this page