My agent setup, in one paste.
The skills, guards and commands I run in every build. Your agent plans before it codes, proves it's done, remembers your project, and stays away from the dangerous stuff.
Get the kit · $49You've seen your agent do this
None of that means the model is bad. The agent just doesn't have a setup telling it how to work. This kit is that setup.
Real output from my setup
Tap a tab. Red ones are guards: they stop the agent even when everything is on auto-approve. The rest are commands you type.
Small on purpose
Everything you install is sent to the model on every turn. A bloated setup is slower, costs more and works worse, so every piece here earns its place.
7 guards, always on
They run on their own, outside the model. Each one blocks a real mistake; none of them is just a reminder.
No recursive deletes
rm -rf and similar commands are blocked before they run.
No surprise installs
Packages you haven't approved are blocked. Curl-to-shell installs are always blocked.
No leaked keys while it writes
Live keys going into code are blocked, and so is printing your .env values.
No leaked keys at commit time
Every commit and push is scanned, and blocked if a key or a .env file is in it.
Fragile files stay untouched
Lockfiles, .git/ and migrations that already ran can't be hand-edited. And the agent can't switch off its own guards.
No quiet test edits
If a test file was changed behind your back, the tests won't run until you've seen it.
Automatic checks
Your project check runs after every edit and logs PASS or FAIL.
14 skills: the workflow, without reminding it
Skills load when they're needed, so your agent follows the workflow without you typing it out every time.
- Plan before code · asks the questions that matter, then waits for your yes
- Prove it's done · real evidence for every check, not "it should work"
- Project memory · decisions, dead ends and open questions in
MEMORY.md - Tests first · failing tests with exact values before risky logic
- Independent review · read-only, ranked Critical / Important / Minor
- Checkpoint and undo · a save point before risky changes, safe undo after
- Add a package · checks a package before it's installed
- Explain the change · plain English, plus what to try by hand
- Stuck? Stop and explain · no more looping on the same fix
- Fresh conversation handoff · a short handoff so a new conversation can carry on
- Ship check · secrets, settings and payments, before you go live
- Agent-ready prompt · turns a rough idea into a prompt with "done when" checks
- Retro to rules · turns a repeated mistake into one rule or test
- Skill safety check · reviews anyone else's skill before you install it
13 commands
Shortcuts for the moments you'll hit every day.
One paste, no terminal
Unzip the kit
Put the folder anywhere you'll keep it.
Paste one prompt
Into OpenCode, with your project open.
Answer three questions
What you're building, whether it takes payments, and this project or all of them.
Approve the table
It shows exactly which files it will copy, and waits for your yes.
It installs the smallest set that fits your project. Everything is plain text you can read first.
Built from real builds
I use it on camera
It's the setup behind the builds on James NoCode, not a pack put together to sell.
Small, not bloated
7 guards, 14 skills and 13 commands. The ones that matter, nothing that pads the count.
You can read all of it
Plain-text skills and short guard files, with no hidden code. 23 automated tests cover the guards.
Tested where it matters
| Tool | What works |
|---|---|
| OpenCode | Everything: skills, guards, commands and the permissions file |
| Codex | All 14 skills. Codex's own approval settings stand in for the guards (instructions included) |
One price, yours to keep
- 7 always-on guards
- 14 skills and 13 commands
- One-paste installer for OpenCode
- Codex setup notes
- A safe starting permissions file
- Every future update, free
Want to know why each piece works?
The kit is the setup on its own. My course, Mastering Agentic Coding, teaches the thinking behind every guard and skill and how to build your own. It also includes the kit plus step-by-step recipes for payments, sign-in, emails and deploys.
If you want both, the course is the better deal.
See the courseQuestions
I'm not a developer. Is this for me?
Yes. It's built for people who build with AI and don't read code. You install it with one paste, and every skill explains what it's doing in plain English.
Will it slow my agent down?
No. The installer only adds what your project needs. The guards are short files that run outside the model, so they don't use up its context.
Can I turn things off?
Yes. Delete a guard's file and restart OpenCode, or type /uninstall to remove the whole kit. Your agent can't turn guards off by itself; that's on purpose.
Which tools does it work with?
OpenCode gets everything. In Codex, all 14 skills work, and Codex's own approval settings stand in for the guards. I only list tools I've tested.
How do updates work?
They're free for the life of the kit. When there's a new version, you'll get an email and the download in your account is updated.
I already bought the course. Do I need this?
No. The kit is included in Mastering Agentic Coding, so check your course downloads.
What's your refund policy?
See the refund policy.