For builders using Claude Code

Claude Code that knows when to stop, who to ask, and what you decided yesterday.

Claude Code is brilliant at building. But left to itself, it will push code that breaks your login, commit secrets it didn't know were secret, and say "done" before the tests have run. Claudekit gives Claude a senior developer's instincts — automatic protection, the right specialist for sensitive work, and project memory that doesn't fade between sessions.

Start Pro — $19 / mo Or install the free kit →

Fourteen-day refund, no questions. The server covers LLM costs on every command you run — that's why pricing is a subscription, not a one-time fee.

Live Control Graph · Pro preview · 4 of 21 nodes shown

Project Memory

9 decisions saved

What Claude stops needing to re-learn at the start of every session.

Reflex Guards

5 active

Automatic stops for the moves that cause real damage.

Disaster Cover

6 / 8 covered

How many known failure modes this kit blocks. Gaps shown, not hidden.

Kit Updates

v0.7 — Apr 24

What shipped. No roadmap promises.

Full dashboard unlocks after onboarding — this is what a new project looks like on day one.

What a markdown file cannot do

Five things your instructions can say. Five things only the kit can do.

Claude cannot make big mistakes silently.

Active now

Before any dangerous move — overwriting saved code, committing a secret key, deleting database state — the kit stops Claude automatically. You don't have to remember to tell Claude to be careful. It already is, on every action, every session.

Claude starts every session knowing where you left off.

Shipping in v1

No more “remind me what we were building.” Before you type a word, the kit reads your recent work, what’s in progress, what needs to happen next, and hands Claude a briefing. Claude arrives already caught up — including context from debugging sessions, decisions made, and things that broke last time.

The right specialist shows up automatically.

Active now

The kit ships four specialist subagents — code reviewer, planner, test runner, security reviewer. When your work touches something sensitive, the right one is brought in. Anything touching user accounts or payments goes to the security reviewer first. A new feature starts with the planner. Tests run with the test runner before Claude says done. You don't hire them. They're already there.

Claude verifies its own work before calling it done.

Shipping in v1

Claude says "done" before tests have run more often than you'd like. The kit checks before the final answer lands: did the tests pass, was the plan actually followed, is anything left open? If something's missing, Claude keeps working. The confident-but-wrong answer stops being possible.

Your project's memory stays current — automatically.

Partial now · full in v1

What was decided, what changed, what broke and how it was fixed — the kit keeps track as you work, without you writing it down. Next session, Claude already knows. Start a new conversation, bring in a specialist, come back after a month — the project's memory travels with the project, not with you.

Instructions tell Claude what to do. The kit makes sure it actually happens.

Three real incidents

This is what "Claude made a mistake" actually looks like.

Everyone could read everyone's data — for weeks.

A security setting that protects user data has to be turned on manually in the database. Claude turned it on in the local test environment but didn't know it had to be set again for the live site. The live database ran without it for three weeks before a user noticed something was wrong. By then, four hundred rows of private data had been sitting exposed.

What the kit does: The kit automatically checks every database change that touches user data. If the protection isn't set, Claude flags it and stops — before the change goes live, not after.

A secret key sat on the live site for nineteen days.

A key that gives full access to the database ended up inside a file that ships to every visitor's browser. It sat there for nineteen days before someone noticed unusual traffic. Fixing it meant migrating the entire backend over a weekend.

What the kit does: The kit checks every file Claude is about to save for secret keys. If it finds one somewhere it shouldn't be, it blocks the save and names the exact file — before the file ever leaves your machine.

Force-pushed, lost two weeks of work.

A builder used a shortcut to resolve a code conflict and accidentally overwrote two weeks of work on the shared codebase — login, password reset, the sign-in flow — all gone. They rebuilt from memory and shipped a bug that locked out fourteen real users for six days.

What the kit does: The kit prevents Claude from overwriting saved work on protected branches. You have to deliberately turn that off. It cannot happen by accident.

How it works

Three commands. Sixty seconds.

Free install. No credit card. No email signup. Three slash commands inside any Claude Code session.

/plugin marketplace add claudecode-kit/marketplace
/plugin install claudekit@marketplace
/reload-plugins

01
Add the marketplace
In any Claude Code session, paste: /plugin marketplace add claudecode-kit/marketplace. One-time per machine.
02
Install the plugin
Paste: /plugin install claudekit@marketplace. Plugin downloads from GitHub and registers itself globally.
03
Reload and verify
Paste: /reload-plugins. On next session start you'll see the greeter: ✧ claudekit v0.5.0 active.
04
Try the commands
/claudekit:about for status. /claudekit:doctor to diagnose your project. /claudekit:self-test to verify integrity. All free, no credentials needed.

Supported stacks

One stack, done properly. Not seven stacks done vaguely.

v1: Next.js + Supabase. The disasters above are real Next.js + Supabase incidents. The kit's protection rules know this stack's specific failure modes by name — database security settings, secret key exposure, authentication overwrites. Not generic advice. Specific stops.

Coming next: Expo + Firebase. On the roadmap. Ship date is when it ships — not when it sounds good to say.

Want your stack? Write to us. We don't ship half-done.

Pricing

Free to try. Pro from $19 / month.

The server covers LLM costs on every slash command you run — that cost is real and ongoing. A subscription is honest about that. A one-time fee isn't.

Free

€0no card

·Core protection rules
·Database safety check
·Recovery playbook for when things go wrong
·— No specialist subagents
·— No /audit, /recover, /explain
·— No ongoing kit updates

Install free

Pro

$19per month — or $149 / year

·7-day Pro trial included — day-8 free fallback
·Full protection + project memory layer
·Four specialist subagents (reviewer, planner, tester, security)
·/audit, /recover, /explain, /cost-check, /teach-me-this
·Live Control Graph dashboard
·Kit updates as we learn what fails in the wild

Start Pro — 7-day trial

Fourteen days, no questions. Cancel any time — your kit keeps working.

A note from the builder

I've watched someone lose two weeks of work to an accidental overwrite, watched a secret key sit on a live site for nineteen days, watched a live database get wiped because someone asked Claude to "tidy things up." These aren't hypothetical risks — they're things I've personally seen happen, mostly to people too embarrassed to admit it. The kit is the answer I'd give them if they called me at midnight. — Mitja

Frequently asked

The actual objections.

Do I need to be a developer to use this?: You need to be comfortable running one command in a terminal — that's the install. After that, everything works inside Claude Code's normal chat window. No configuration files to edit, no code to write.
What does the kit actually add to my project?: A folder called `.claude/` with protection rules, project memory, and specialist agents — all plain text files you can read and edit. A project context file that tells Claude about your codebase. And an optional recovery playbook for when things go wrong at 2am. Nothing goes outside those files without your explicit approval in the install preview.
What if Claude still makes mistakes?: The kit reduces dangerous mistakes, not all mistakes. It can't make Claude smarter — it makes Claude safer, better briefed, and harder to fool into saying "done" when it isn't. Some mistakes are still possible. The ones that cause real damage are the ones the kit is built to stop.
What's in the free kit vs the paid kit?: The free kit ships the core protection rules, the database safety check, and a recovery playbook — the things that prevent the disasters on this page. The paid kit adds four specialist subagents, the slash commands for auditing, recovering, and explaining decisions, and ongoing updates as we learn what fails in the wild.
What if Anthropic ships an official equivalent?: Likely they'll ship something adjacent eventually. We're betting on curated expertise and a standing relationship — the things a built-in default can't replicate. If they ship something that fully replaces us, we'll say so honestly and refund anyone in their refund window.
What if your server goes down — does my Claude Code break?: No. The kit is locally installed in your project. Your protection rules and project memory keep working. Commands that call our server show a clean offline message — nothing crashes, nothing breaks.
I already have a .claude/ folder — will you overwrite it?: No. The install preview shows you every conflict before anything changes. You can reject any specific file without cancelling the rest. Your existing setup is merged, never overwritten.
The dashboard says '6 / 8 covered' — what are the other two?: The full disaster catalog is public. The two gaps are edge cases that need fixture-tested coverage before we'll claim them honestly. We'd rather show the gap than pretend it isn't there.