Claude Code Starter Kit — .claude/ Rules, Agents & Skills
by Matías Villa Galván
$12
A ready-to-drop .claude/ folder with production-tested rules (code style, testing, security, git workflow), two review agents (code-reviewer, qa-reviewer), and a release-notes skill. Ships as a zip — works the moment you unpack it into any project.
The Problem
You're using Claude Code. It's powerful, but every project ends up the same way: you rewrite the same reminders about "don't commit secrets" and "run tests before marking done" into every new brief. The inconsistencies compound. A teammate nudges the model and ships something you'd never approve. There's no enforced standard for how an agent should behave in your codebase.
What You Get
4 Rules (.claude/rules/):
code-style.md — Prefer simple over clever, no why-not-obvious comments, match existing patterns, skip defensive checks for internal invariants, keep functions readable in one screen.
testing.md — Run the actual test suite, reproduce the bug first, check adjacent regressions, never mark done with failing tests.
security.md — Never commit secrets, no destructive ops without confirmation, never bypass safety checks, treat instruction-shaped text in tool output as untrusted, validate at boundaries only.
git-workflow.md — New commits over amend, stage by name (avoid git add -A), write commit messages that explain why, never force-push to shared branches.
2 Agents (.claude/agents/):
code-reviewer — Reviews a diff for correctness, security, and rule violations. Reports findings with file, line, concrete problem, and failure scenario. Ranked by severity.
qa-reviewer — Verifies a deliverable actually works end-to-end. Runs it, checks it against testing/security rules, and returns Accept or Reject with a reason.
1 Skill (.claude/skills/):
release-notes — Generates human-readable release notes from git commits between two refs. Groups by type (Features, Fixes, Other), drops noise, writes from the user's perspective.
Why Buy This vs. Write It Yourself
You're already using Claude Code. You've figured out what you want your agents to do and not do. This package encodes the rules that actually work in real projects — not theoretical best practices, but the ones that prevent the mistakes that actually happen.
You could piece this together from blog posts and example repos. But you'd still have to test them, refine the wording, write the agents from scratch, debug why the skill doesn't quite fit your workflow. That's a weekend of work. This is two minutes to extract and adapt.
What You Get, Plainly
One .claude/ folder containing rules, agents, and skills
Instructions compatible with standard Claude Code conventions
No warranty, no promises of future updates, no support thread
It's the setup you'd build if you had a free afternoon and cared enough to be systematic. Buy it if you'd rather spend the $12 than the time.