Maker Growth Hub member tool

Agent Clinic

Your maker’s ship’s doctor for misbehaving AI.

  • 21 symptomsWays AI sounds wrong — invented techniques, twee voice, untested safety claims, craft-fair pricing on your $80 work.
  • 35 fix promptsPaste-ready prescriptions across 5 departments. Copy the fix, paste it into ChatGPT or Claude, get the corrected version back.
  • 36 error codesCryptic AI error messages decoded in plain English — what it means, and what to do.
  • 31-point pre-flightThe checklist before you ship a Custom GPT or Claude Project, built from the same checks Christian runs on consulting-client agents.

Walk in with the symptom. Walk out with the fix in your hands.

Walk in with the symptom

When AI sounds wrong, start here

For the days AI hands you back something twee, generic, or full of safety claims you never authorized. These are the 21 cases the Clinic is built around — each one has a root cause, a fix, and pro tips inside.

Five departments

35 prescriptions, by what’s going wrong

Search by what’s going wrong — em-dashes, “Great question!” openers, craft-twee voice, commission overcommits — copy the prescription, paste it into ChatGPT or Claude, and your AI returns the corrected version.

Voice · 18

How it sounds — em-dashes, “Great question!” openers, craft-twee voice, hedging, apologies, and the “we” that should be an “I”.

Format · 5

How it’s shaped — bullet points, emails, social posts, and the follow-up questions you never asked for.

Audience · 4

Who it’s for — plain English, professional, casual, or alt text that describes the image, not the pitch.

Length · 4

How long it runs — shorter without losing the point, longer with depth, or straight to it.

Substance · 4

What’s actually in it — specific examples, visible reasoning, or a clear recommendation.

Try two prescriptions

Take the fix for a spin

Two full prescriptions from the Clinic, exactly as members copy them. Paste either one into ChatGPT or Claude alongside the output AI got wrong. The other 33 live inside.

Stop using em-dashes

Department: Voice · Prescription #19 of 35

Rewrite this without any em-dashes (—). Replace each em-dash with: a period and new sentence, a comma, parentheses, or a colon — whichever fits the rhythm best. Also remove any double-hyphens (--).

Copy it, paste it into ChatGPT or Claude alongside the output AI got wrong, and your AI returns the corrected version.

Fix this restock post — cut the craft-twee

Department: Voice · Prescription #31 of 35

Here's a restock post AI drafted for Instagram / email — rewrite it.

[PASTE THE AI DRAFT]

Cut every:
- "Lovingly," "magical," "artisanal," "with love," "perfect gift"
- "Exciting news!," "Just dropped!," or any exclamation-point hype opener
- Generic hashtag spam — keep 3-5 max, only ones that actually match my work
- "Don't miss out" / "while supplies last" urgency theater

Keep: specific product details, what's new about THIS batch, where to find it.

Voice samples: [PASTE 1-2 SAMPLES]

Copy it, paste it into ChatGPT or Claude alongside the output AI got wrong, and your AI returns the corrected version.

Guided treatment

When the same problem keeps coming back

Five guided workflows for the patterns that come back again and again — each one a five-step diagnosis and resolution, start to finish.

🧠 My agent is making things up

5 steps. Starts with: Identify the type of hallucination

🚫 My agent keeps refusing my requests

5 steps. Starts with: Identify the refusal pattern

✍️ Output doesn't sound like me

5 steps. Starts with: Collect 3–5 voice samples

📋 Output format keeps breaking

5 steps. Starts with: Show an example, don't just describe

🛠️ My AI keeps screwing up my shop content

5 steps. Starts with: Pin down which kind of wrong it's getting

Before you ship

The pre-flight checklist, first 8 of 31

A 31-point checklist before you ship a Custom GPT or Claude Project. Here’s how it starts — the rest lives inside the Clinic.

  1. I've provided 3-5 voice samples in the system prompt
  2. I've listed 5-10 words I use often
  3. I've listed 5-10 banned words/phrases
  4. I've specified my tone (warm/sharp/playful/etc.) — picked one, not 'all of the above'
  5. I've banned em-dashes if I don't use them
  6. I've banned the AI opening fillers ('Great question!', 'Absolutely!', etc.)
  7. I've listed people whose messages always escalate to me
  8. I've listed topics that always require my approval

That’s 8 of 31. The full checklist — Voice & Tone, Boundaries & Permissions, Output Format, Test Cases, Platform Setup, and Ongoing Use — is in the Clinic.

Decoded in plain English

36 error codes, 12 categories

Stop guessing what the model meant: read the diagnosis, take the fix back. The codes span Auth, Billing, Client, Content, Custom GPTs, Files, Generic, Limits, Network, Output, Projects, and Rate Limits. Three samples:

Auth — Invalid API key

Why this happens. Your API key is wrong, expired, or has a typo. This is the most common auth error.

How to fix it. Go to your platform's API settings and regenerate a fresh key. Copy it directly — don't retype it. Make sure there are no spaces before or after the key.

Content — Content policy violation

Why this happens. Your request triggered the platform's safety filters. The model detected content it's been trained to refuse.

How to fix it. Rephrase the request with more context about your legitimate purpose. Add your professional role and use case. If the content is genuinely prohibited, use a different approach.

Limits — Message limit reached

Why this happens. You've hit your plan's message cap for the current period (hourly, daily, or monthly).

How to fix it. Wait for the limit to reset (check when it resets in your account). Upgrade your plan for higher limits. Use a different model that has a separate limit.

🤖 Plus: 24/7 triage in your private Discord.

Drop your case in the #stuck channel any time of day. The Workshop bot sweeps the ward at midnight, 10am, and 4pm America/New York — the med team writes back in your thread, usually within seconds of the sweep. Built on the same maker-tuned frameworks Christian developed from his consulting work. First-response triage so small AI fixes don’t pile up — strategy, custom work, and the hard human calls still come to the live Workshop and the community.

$297 value · included with founding membership

Access granted automatically when you join. Sign in with the same Discord account that gets you into the Hub community — no extra setup, no separate account, no extra cost.

Join the Maker Growth Hub

Already a member? Open the Clinic →

Questions

Before you walk in

What is the Agent Clinic?

Your maker’s ship’s doctor for misbehaving AI. For the days AI hands you back something twee, generic, or full of safety claims you never authorized. Walk in with the symptom. Walk out with the fix in your hands.

Do I need a separate account for the Clinic?

No. Access is granted automatically when you join the Maker Growth Hub. Sign in with the same Discord account that gets you into the Hub community — no extra setup, no separate account, no extra cost.

How do the fix prompts work?

Copy the prescription, paste it into ChatGPT or Claude alongside what AI got wrong, and your AI returns the corrected version. The two samples on this page work exactly that way — the other 33 live inside the Clinic.

What if AI keeps doing something the Clinic doesn’t cover?

Drop your case in the #stuck channel any time of day. The Workshop bot sweeps the ward at midnight, 10am, and 4pm America/New York — the med team writes back in your thread, usually within seconds of the sweep.

Walk in with the symptom

Walk out with the fix in your hands

The full ward — all 35 prescriptions, the complete pre-flight checklist, the workflows, and the patient log — lives inside the Maker Growth Hub, along with the live workshops and the members who’ve already run it on their own shops.

Join the Maker Growth Hub