claudecode-starterkit

Install a shared Claude Code baseline globally, then create a thin per-project .claude/ overlay with project-local memory.

---

What this installs

claudecode-starterkit is a bootstrap package for Claude Code.

It installs in two layers:

1. Global baseline — installed once into ~/.claude/
2. Project overlay — installed into each repo you work on

---

Global install

Run once per machine:

npx claudecode-starterkit

This syncs into ~/.claude/:

Directory	What goes there
~/.claude/commands/	Shared slash commands (/init, /plan, /ship, /handoff, /resume, /pr, /review, /status, /research, /health)
~/.claude/skills/	Shared skills (memory-system, context-management, verification-before-completion, systematic-debugging, incremental-implementation, test-driven-development)
~/.claude/hooks/	Lifecycle hooks (post-edit, on-stop)
~/.claude/settings.json	Merged permissions and hooks config
~/.claude/CLAUDE.md	Global rules (only if not already present)

---

Project install

Run in each repo:

npx claudecode-starterkit install

This creates in .claude/:

.claude/
├── settings.json              — project-local config with memory file instructions
├── plans/                     — plan files from /plan command
└── memory/
    ├── project/
    │   ├── project.md         — project identity and goals
    │   ├── state.md           — current position and active work
    │   ├── tech-stack.md      — detected stack and constraints
    │   ├── roadmap.md         — phases and milestones
    │   ├── user.md            — workflow preferences
    │   └── gotchas.md         — footguns and conventions
    ├── handoffs/              — session handoff files
    ├── research/              — research notes
    └── _templates/            — templates for new memory files

Also creates CLAUDE.md at project root (if not already present).

---

Slash Commands

After global install, these commands are available in every Claude Code session:

Command	Description
/init	Create CLAUDE.md and detect tech stack
/plan <task>	Create implementation plan
/start <task>	Begin implementation
/ship	Verify, review, and finalize work
/handoff	Save session state for next session
/resume	Pick up from previous session
/status	Show current project status
/review	Code review
/pr	Create pull request
/research <topic>	Research and save findings
/health	Audit project health

---

Skills

Skills are loaded by slash commands automatically. Available skills:

Skill	Description
memory-system	Memory file structure and session workflow
context-management	Context health across long sessions
verification-before-completion	Verification protocol before claiming done
systematic-debugging	Root cause debugging protocol
incremental-implementation	Vertical slice implementation approach
test-driven-development	TDD workflow

---

Options

# Global install (default — no args)
npx claudecode-starterkit

# Project install
npx claudecode-starterkit install

# Force re-generate memory files
npx claudecode-starterkit install --force

# Skip creating CLAUDE.md
npx claudecode-starterkit install --skip-claude-md

# Explicit global install
npx claudecode-starterkit install-global

---

Architecture

claudecode-starterkit/
├── baseline/              — source of truth for all installed content
│   ├── CLAUDE.md          — global rules
│   ├── settings.json      — baseline permissions and hooks
│   ├── command/           — slash command definitions
│   ├── skill/             — skill definitions
│   ├── hooks/             — lifecycle hook scripts
│   └── memory/
│       ├── _templates/    — memory file templates
│       └── project/       — placeholder project memory files
├── src/
│   ├── constants.mjs      — path constants
│   ├── fs-utils.mjs       — file system utilities
│   ├── templates.mjs      — memory file renderers
│   ├── install-global.mjs — global install logic
│   └── install-project.mjs — project overlay logic
└── bin/
    └── claudecode-starterkit.mjs  — CLI entry point

---

Difference from opencode-starterkit

opencode-starterkit	claudecode-starterkit
~/.config/opencode/	~/.claude/
agents/*.md	Built-in subagent types
command/*.md	~/.claude/commands/*.md
skill/*.md	~/.claude/skills/*.md
opencode.json	settings.json
plugin/*.ts	MCP server (mcp/memory/) + hooks in settings.json
AGENTS.md	CLAUDE.md
.opencode/memory/	.claude/memory/ (markdown templates + memory.db SQLite)

Memory system

The cc-memory MCP server ports the 4-tier temporal memory from opencode-starterkit to Claude Code:

• Tier 1 — Capture: UserPromptSubmit + PostToolUse hooks write to temporal_messages.
• Tier 2 — Distillation: Stop hook runs TF-IDF distillation when ≥10 undistilled messages exist.
• Tier 3 — Curation: Pattern-matching extracts observations (decision/bugfix/pattern/discovery/warning) from distillations.
• Tier 4 — Injection: SessionStart hook injects BM25-scored knowledge as additionalContext.

11 MCP tools become available: observation, memory-search, memory-get, memory-read, memory-update, memory-timeline, memory-graph-add, memory-graph-query, memory-graph-invalidate, memory-compact, memory-admin. Plus /checkpoint <text> for manual capture.

Data lives per-project at .claude/memory/memory.db (SQLite + FTS5).