// products / starter-kit
Claude Code Starter Kit
The production setup I wish I'd had on day one of running Claude Code as a real engineer. Most people run it with no operating model at all — no charter, no permission boundaries, no guard against the agent leaking a secret. This is the opinionated baseline. Copy it in and start grounded, not from a blank CLAUDE.md.
It's deliberately lean — not 200 files you'll never open. A small set you'll actually adopt. The curation is the point.
// what's inside
template/ # copy into your project root ├── CLAUDE.md # the operating charter ├── .claude/settings.json # permission allow / ask / deny ├── .claude/commands/ # /review, /ship, /promote ├── .agent/ # memory: trust · journal · wisdom ├── .githooks/pre-commit # blocks committing secrets ├── scripts/scan-secrets.sh └── .mcp.json # a curated, minimal MCP set docs/ charter-guide · mcp-guide · walkthrough-script + a ~10-min walkthrough video
- ✓The operating charter — a
CLAUDE.mdbuilt on a trust ladder (the agent earns autonomy Tier 0→3, per domain) and prime directives it can never cross. The centerpiece. - ✓Permission boundaries — a
settings.jsonallow/ask/deny so the agent can't wander into a force-push or read your.env. - ✓A secret-scan pre-commit hook — dependency-free, blocks any commit that contains a credential. The guard this very site ships with.
- ✓A curated MCP set — the two servers worth running by default, plus a guide on what to add and what to skip.
- ✓Two field guides — how to adapt the trust ladder to your team, and how to choose MCP servers. The why, not just the files.
- ✓A worked memory system + a walkthrough video — a filled-in
.agent/example, the/review,/ship, and/promotecommands, and ~10 minutes of the whole thing running on a real repo. Templates you can see work.
Not included, on purpose: prompt packs, 200-file dumps, "10x" promises. This is a setup, not a magic trick.
$49 early-bird price · commercial use · free updates
Secure checkout on Gumroad · instant download
// who it's for
Engineers (solo or on a team) who:
- Run Claude Code on real codebases, not toy demos
- Want boundaries and an operating model, not just more prompts
- Have been one bad agent action away from a force-push or a leaked key
- Would rather copy a grounded setup than trial-and-error their own
// going deeper
This kit is the operating foundation. The course builds a full autonomous pipeline on top of it — kit buyers get a credit toward it.
// not ready to buy?
Get the newsletter — the same engineering, twice a week, free. Subscribers hear when the kit and the course ship.
Subscribe to the newsletter →