Your AI deleted the client folder in 0.3 seconds.
It works fast. So do its mistakes. The Tune-Up Kit installs gates that stop them, approval on everything it sends, and a dashboard on everything it spends.
installed in 30 minutes · 14-day money-back guarantee
I run my whole business on Claude Code. So I'll say it plainly.
A written rule is a polite request. The AI can talk itself past it on a confused day. A gate is different: a check that fires before the command runs, outside the AI's reach. It's the only thing that actually stops a deleted folder, overwritten work, or a very strange email to a real customer.
Every skill you install runs as you. Your permissions, your files, your accounts. Most people have no step between "found it on GitHub" and "running it".
The bill grows while you're not looking. Most setups quietly pay top-model prices for grunt work like renaming files. In my workspace that was the single biggest leak, and closing it took one config file.
A rule the AI cannot talk itself past.
This is the guardrail hook doing its job in my workspace. It fires before the command runs, and no amount of clever reasoning on the AI's side gets around it.
Guardrail hook
A hard stop on the worst commands: wiping folders, overwriting saved work, deleting databases, touching folders you marked as protected. It blocks the command before it runs, and the AI cannot argue with it.
Send gate
Emails, posts, and payments need a human to read the full message and type APPROVE before anything goes out. The approval writes a single-use token bound to a fingerprint of that exact message (sha-256), expiring in 5 minutes. Change one word and the fingerprint no longer matches, so it needs approving again.
Skill quarantine
Add-ons you download get scanned for hidden instructions and key-stealing code before they ever run. "Found it online" stops going straight to "running with full access to my files".
Security nudge
If the AI writes risky code, like leaving a password sitting in a file, it gets flagged the moment it happens, not weeks later.
One of my skills burned $552 in four sessions. I'd never have known.
The spend dashboard reads your local session logs and shows where the money actually goes: per day, per model, per project, per skill, and how much the cache saved you. I run it on my own logs every month; the card here shows real numbers from mine.
On a flat subscription it shows what your plan actually does for you, and which skills eat it. Re-run it monthly; the trend is the point.
both numbers measured on my own session logs; the dashboard in this kit is how I caught them
Everyone sells courses about AI. I'm selling the gates I actually run.
01 Working tools, day one
The usual offer is a video course or a community. This is working code: hooks, gates, and a dashboard you install once and keep.
02 Running in my business right now
Every component protects my own client work and my own money as you read this. What you download is my setup, with my paths stripped out.
03 Nothing phones home
Most "AI tools" want your API keys and your data. This kit makes zero network calls, so there is nothing to leak. For a security product, that's the only honest design.
04 A backup vendor, built in
I haven't found another kit that ships this: if your AI vendor has an outage or hikes prices, the cross-agent layer means a second one picks up the same workspace. Your business doesn't wait out somebody else's bad day.
Buy it, unzip it, follow eight steps.
Download lands in your inbox
Checkout below; the zip and the setup guide arrive a minute later.
The guardrail is live
Step one installs the hard stop. From here your AI cannot wipe a folder, even on its worst day.
All gates installed
Eight steps, ordered by payoff, each with a "check it worked" test. Copy a file, paste a snippet: that's the skill level needed. Per-harness notes cover Claude Code, Cowork, and Codex.
Re-run the dashboard
Watch the spend trend and catch the next leak early. If you get stuck, reply to the delivery email and I answer.
Every gate in here exists because my workspace needed it first.
I'm a scientist by training, 15 years in research labs, and for the last three years I've built AI systems for businesses. My whole consultancy runs on Claude Code: my client work, my outreach, my money. And with that much access, hoping the AI behaves stopped feeling like a plan. So I built gates.
This kit is those gates. The actual hooks and patterns protecting my own business, cleaned up so they install in yours. And if a step confuses you, you reply to the delivery email and you get me, not a help desk.
Cheers, Nic
Thirty minutes from now, your AI can't wipe a folder or send an email without you.
Everything in the kit runs in my own production workspace today. Same gates, same patterns, decoupled and documented so they install anywhere.
Try it for two weeks. Not happy for any reason, reply to your receipt and you get every cent back. No forms, no questions.
tune-up-kit/ ├─ SETUP.md the 30-minute guide ├─ settings-snippet.json ├─ hooks/ │ ├─ guardrail_check.py the hard stop │ ├─ security_check.py risky-code nudge │ └─ skill_md_validate.py ├─ lib/ │ ├─ send_gate.py approval tokens │ ├─ validate_skill.py │ └─ check_agents_sync.py ├─ scripts/ │ └─ spend_dashboard.py ├─ args/model-routing.yaml ├─ docs/ 5 pattern guides └─ skill-vet/ quarantine gate + self-test
The actual zip contents. Seven working scripts, one config, five guides, plus the skill-vet suite.
every version-1 update included, free
- Guardrail hook
- Security nudge hook
- Send gate
- Skill quarantine gate
- Automatic skill checker
- Spend dashboard
- Cheap-model routing config
- Instruction-file slim-down guide
- Lazy loading guide
- Token-saving chaining pattern
- Hidden-instruction defense
- Backup-vendor layer + setup guide
This kit is new, so early buyers get me with it. After you install, send me your health-check score and I'll personally reply with what I'd fix first in your workspace, in order of payoff. Normally the only way to get my eyes on your specific setup is the $2,000 audit. This is a launch offer, and it won't be around long.
no accounts · no api keys · nothing leaves your machine
The things people ask before buying.
I just use Claude in a browser tab. Is this for me?
Not yet. The kit is for AI that lives on your computer and can touch your files: Claude Code, Claude Cowork, or Codex. If your Claude lives at claude.ai in the browser, it can't delete your folders, so you don't need gates. The day you move to one of the desktop tools, come back.
Does it work with Cowork and Codex, or just Claude Code?
Claude Code gets everything, including the hooks. Codex gets the scripts, the patterns, and the cross-agent layer (that layer is the whole point: two vendors, one workspace). Cowork gets the scanner, the dashboard, and the patterns; hooks are a Claude Code feature. The setup guide marks each component, so nothing surprises you.
Do I need API keys or new accounts?
No. Nothing in the kit makes a network call. It reads your setup, gates your tools, and renders local reports, all on your machine.
How technical do I need to be?
If you can copy files into a folder and paste a snippet into one settings file, you're covered. Every step in the guide has a "check it worked" test.
Is this a course?
No. It's working tools plus short pattern guides. You install gates, not watch videos.
What does the guarantee cover?
Everything, for any reason, for 14 days. Reply to your receipt, money back, done.
That's exactly what the audit is.
The kit is the self-serve version of what I do for clients. The audit is the done-for-you version: in 2 weeks I identify your number-one manual bottleneck and ship a working AI agent against your real stack. If I don't deliver both, full refund.
Book an introductory call