A minimal, opinionated .claude/ directory you can drop into any repo.
Anthropic publishes excellent official examples. This isn't a competitor — it's the small set of specific patterns I use across the internal repos at GYNERGY: a hook that catches duplicate API routes before they ship, slash commands for the two workflows I run every day (/handoff and /review), a settings template with sensible defaults, and a CLAUDE.md scaffold.
The companion repo, agent-patterns, explains the why behind these. This repo is the how.
.claude/
├── CLAUDE.md.example # project-instructions template
├── settings.example.json # permissions, env, model routing
├── hooks/
│ ├── README.md # how hooks work
│ └── pre-route-grep.example.js # warn before creating duplicate API routes
└── commands/
├── README.md # how slash commands work
├── handoff.example.md # /handoff — session handoff doc
└── review.example.md # /review — cold-context review of current diff
# 1. Copy the .claude folder into your repo
cp -r claude-code-starter/.claude /path/to/your/repo/
# 2. Rename the example files
cd /path/to/your/repo/.claude
mv CLAUDE.md.example CLAUDE.md
mv settings.example.json settings.json
mv commands/handoff.example.md commands/handoff.md
mv commands/review.example.md commands/review.md
mv hooks/pre-route-grep.example.js hooks/pre-route-grep.js
# 3. Edit CLAUDE.md to describe your project
# 4. Edit settings.json — pick the permissions you actually want
# 5. Open Claude Code and run /review or /handoff- Anything project-specific (paths, internal MCP servers, memory contents)
- Anything that would require a paid subscription to use
- Anything decorative. This is the floor — everything else is yours to add.
- agent-patterns — the principles these hooks and commands implement.
- incident-postmortems — the production incidents that made me write the hooks in the first place.
MIT. Fork it, edit it, make it yours.