Fix AI Slop Code

The code AI wrote runs. It does not hold

Each episode is one thing the model leaves out when it writes your code, and the fix

Free, straight to your inbox. No call, no pitch

Episode 23 episodes so far

Row level security switched on is not row level security working

What the model writes
create policy "Enable read access"on public.messages for selectto anon, authenticatedusing ( true );
What it should be
create policy "Users read their own rows"on public.messages for selectto authenticatedusing ( (select auth.uid()) = user_id );
Check it yourself: must print []
Watch the video

Watch how the series works

The same idea in a short video. It plays here, and nothing loads until you press play

Watch it on YouTube

Inside every episode

What you get in every episode

The explanation is free on the notes below. The kit adds the code it wrote, the code it should have written, and a check

The analogy

Something outside code first, so you can repeat it

This is like a bouncer who keeps the guest count in a notebook in the back room.
The code

What the model wrote, and what it should have

const k = `seen:${email.toLowerCase()}`;const k = `seen:${action}:${email.trim().toLowerCase()}:${bucket}`;
The check

One command that tells you if your app has it

# a real limit prints about 5 of 200 and about 45 of 429# a limit that is not limiting prints 50 of 200
The episodes so far

One kit, and every new episode joins it

Read the one you came for, then keep going

  1. Episode 1Your rate limit is probably not limiting anything
  2. Episode 2Row level security switched on is not row level security working
  3. Episode 3A dedupe key carries the intent, not just the identity
  4. NextThe next episode lands here when its video posts
The thesis under all of it

None of these is a bug. They are all omissions

The model does not write bad code. It writes the feature you asked for and leaves out the parts nobody asked for: authorization, ownership checks, rate limits, timeouts, error handling, the lock on the table. Every catalog of vibe-coded failures ranks the same way: missing row-level security, secrets in the bundle, no ownership check on CRUD, no rate limit on auth and write routes. None of those is a bug in what was produced. They are all omissions.

So the practice is never "prompt better." It is: know the list of things it will not volunteer, ask for each one by name, and have something that is not the author check that it arrived.

6 entries

Keys and secrets

Six of the twenty are about credentials because that is where our own code bit hardest and where the numbers in the research are worst: 41 percent of 100 audited vibe-coded apps had an exposed key.

  1. Hit in our code

    Keys go in headers, never in URLs

    A URL is not a private channel

  2. Hit in our code

    A pasted secret is a burned secret

    Chat transcripts persist outside your security boundary

  3. Research, sourced

    The public prefix means public

    NEXT_PUBLIC_ and VITE_ are build instructions

  4. Our code and the research

    Rotate first, then rewrite history. Deleting the file does nothing

    A commit that removes a secret leaves it in every prior commit, every clone, every fork, and GitHub's cached views of deleted forks

  5. Our code and the research

    One key per system, scoped, with a spend cap

    A shared key turns one leak into several rotations and one outage into several

  6. Our code and the research

    IDs can ship. Keys cannot. Know which is which

    A GA4 measurement ID, a Meta Pixel ID, a Supabase publishable key are designed to sit in the browser

5 entries

The limit that limits nothing

The backend equivalent of the omission thesis: the code has a rate limit, a dedupe, a retry, and each one is shaped like the real thing without doing the real thing's job.

  1. Hit in our code

    A read-then-write counter on an eventually consistent store is not a rate limit

    Cloudflare KV, and most edge key-value stores, take up to about 60 seconds to propagate a write

  2. Hit in our code

    A dedupe key carries the intent, not just the identity

    Deduplication keyed on the email alone means the second thing a person does is treated as a repeat of the first

  3. Our code and the research

    Every write that can be retried must be idempotent

    Platforms retry on your behalf: Stripe redelivers a webhook for up to three days, Slack resends if you do not answer in three seconds, Vercel can fire the same cron twice or start a second instance while the first runs

  4. Research, sourced

    Retries are capped and backed off, and every outbound call times out

    Three layers each retrying three times is 27 requests for one failure

  5. Hit in our code

    A form that echoes visitor text inside your signed email is a phishing relay

    If a contact endpoint accepts a recipient and a message, and sends a branded, DKIM-signed email containing that message, anyone can send anything to anyone from your domain

4 entries

Database

Our stack has no Supabase, so these four are from the research and from the one place our own agents touched production. They are here because the most common backend under vibe-coded apps is Supabase, and 83 percent of its exposures are one setting.

  1. Research, sourced

    RLS enabled is not RLS working

    Row-level security is two things: a switch per table, and a policy per command

  2. Research, sourced

    The admin key is not a fix for a policy error

    When a policy blocks a query, the fastest thing that makes the error go away is the service-role key, which bypasses every policy

  3. Our code and the research

    The coding agent never holds production credentials

    An agent with a production database URL will, sooner or later, run a migration, a cleanup, or a "fix" against it

  4. Our code and the research

    Backups the app cannot reach, and one restore you have actually done

    The free plan has no backups

5 entries

Agents and checks

The five that came out of our own pipeline more than the research. They are the ones nobody in the vibe-coding lane is saying, because you have to have run a night of agents to learn them.

  1. Hit in our code

    Never let the thing check itself

    The model that wrote the code, or the tool that produced the measurement, will confirm its own output when asked

  2. Hit in our code

    Scope the tool, not the prompt. Permissions come from keys, not instructions

    "Do not deploy" in a prompt is a request

  3. Our code and the research

    Untrusted text going into a prompt is untrusted code going into an interpreter

    A model with private data, untrusted input, and any way to send something outward can be made to send the private data outward

  4. Hit in our code

    Parallel agents get one worktree each and one owner per file

    Two agents editing the same checkout write over each other, and neither reports an error because each one's write succeeded

  5. Hit in our code

    Verify the output, not the report. Especially before a delete

    Exit codes, summaries, and "9 files matched" are reports

Also true

True, and everyone says it

They are filed here so the list is complete

  • Contrast, labels, alt text, real buttons. 83.9 percent of the top million home pages fail contrast; plain-prompt LLM UIs fail keyboard access 80 percent of the time, accessibility-prompted ones 0 percent. Test with the mouse unplugged.
  • Size the images, do not lazy-load the hero. 62 percent of mobile pages ship an unsized image; only 48 percent pass Core Web Vitals.
  • Disable the button on submit, redirect after post, idempotency key. Double orders from double clicks.
  • Platform limits the model never mentions. Vercel 4.5 MB body and 300 seconds, Workers 10 ms CPU on free, Supabase Edge 150 seconds. Presigned direct uploads and queues for anything long.
  • App Store bounces: no in-app delete account, Google-only sign-in, a web page in a shell, privacy labels that do not match the SDKs, no terms link on subscriptions.
  • Money as a float, sequential IDs in URLs, customer data in logs. Store money as numeric or as integer cents, use UUIDs in URLs, and log IDs instead of whole rows.
  • Slopsquatting. Models invent package names, and some of those names are free for anyone to register. Check when a suggested package was first published (npm view, then its name, then time.created) before you install it, and keep ignore-scripts on.
Longer reads

More on apps AI built

Free, no call

Get the kit, and every episode after it

Your link opens the kit, and every new episode is added to it

Free, straight to your inbox. No call, no pitch

What to do with this

If the checks came back with more than you expected, that is worth a conversation

Book a free 30-minute call